A Study of Multi-distributed Resource Equalization Allocation for Virtual Power Plants Based on Genetic-heuristic Algorithm

Abstract

A multi-resource balanced allocation method using a genetic-heuristic fusion algorithm is proposed to address the imbalance in distributed power generation resource allocation and the over-generation problem in virtual power plants. By establishing models of wind, solar, storage, and controllable load characteristics, an optimization model is constructed with objectives of resource allocation balance and minimization of call costs, subject to constraints such as power balance. Combining the global search capability of a genetic algorithm and the local optimization capability of an ant colony algorithm, the genetic algorithm stage adopts real-number encoding and a dynamic crossover-mutation strategy, while the ant colony algorithm stage optimizes the pheromone update mechanism to avoid premature convergence. The experimental results show that this method achieves 100% accurate allocation of resources without any over-generation occurrences and reduces the resource allocation deviation rate by 32–67% compared to alternative methods. The algorithm demonstrates fast convergence, yielding solutions in less than 0.6 s across 14 repeated experiments, with an average convergence time reduction of 42% compared to traditional algorithms. Under a comprehensive fluctuation scenario with 30% renewable energy fluctuation rate and 15% load forecasting error, the system stability index remains at 0.865, demonstrating the algorithm’s efficiency and robustness under complex conditions and providing an effective approach for optimizing virtual power plant resource allocation.

Introduction

The Virtual Power Plant (VPP), as an advanced model for centralized regional power management [1], utilizes advanced information and communication technologies along with software systems to integrate the flexibility of distributed power sources, energy storage systems, and controllable loads, among other resources. Virtual power plants (VPPs) not only improve the efficiency of distributed energy utilization but also enhance the reliability and stability of the power system while reducing energy costs. In terms of renewable energy integration, VPPs have significantly improved the absorption capacity of intermittent power sources such as wind and photovoltaics through a multi-resource coordinated scheduling mechanism. For example, through dynamic charge/discharge regulation of energy storage systems, the volatility of wind and photovoltaic output can be mitigated, and the random fluctuations of renewable energy can be converted into a predictable and controllable stable power supply. Moreover, the flexible adjustment capability of controllable loads (such as industrial interruptible loads and electric vehicle charging stations) can be matched in real time with renewable energy output, increasing load consumption during excess electricity availability and reducing non-critical load electricity consumption during electricity shortages. This integration model breaks the traditional power grid’s dependence on fossil fuels, gradually transforming renewable energy from a “supplementary energy” to a “primary energy,” promoting the transformation of the power system toward low-carbonization and decentralization, and providing key technical support for achieving the “dual carbon” goal [2]. Therefore, VPPs play an essential strategic role in the transformation of the power system [3]. In addition, demand response, as the core mechanism for integrating controllable loads in VPPs, guides users to dynamically adjust their electricity consumption behavior through price signals (such as time-of-use electricity prices) or incentive measures (such as interruptible load compensation). For example, industrial users can increase the operation of high-energy-consuming equipment during power surplus, reduce non-critical loads during shortages, and achieve real-time supply–demand balance. This regulation capability not only improves the efficiency of renewable energy absorption but also reduces the peak-to-valley difference of the power grid through load-side flexibility, enhancing system stability.

In virtual power plants (VPPs), the balanced allocation of distributed resources is a key factor to achieving efficient operation and optimization [4]. Since distributed resources vary in geographical location, power generation characteristics, and energy storage capacity, among other factors, how to optimally allocate them according to the actual needs of the power grid and resource characteristics has become an important research focus in VPP studies [5]. Through balanced allocation, the advantages of distributed resources can be maximized, while the flexibility and economy of the power grid can be improved. This complements the advantages of microgrids. A microgrid features localized energy production and consumption, and its regional power self-sufficiency rate can reach 70–90%. In typical industrial park scenarios, dependence on the large-scale power grid can be reduced by more than 40% through the coordination of local wind, solar, and storage resources. Its decentralized and flexible control mode enables quick response to local load changes and enhanced power supply reliability. Virtual power plants integrate multiple energy sources through balanced resource allocation, combining with the flexible and independent advantages of microgrids, to further optimize power resource allocation, reduce transmission losses, enhance the power system’s flexibility in responding to emergencies and load fluctuations, and provide dual guarantees for efficient energy use and sustainable development [6]. According to existing literature, Pandey et al. study a multi-objective operation scheduling method for virtual power plants using an improved Harris Hawk optimization algorithm. This method enables more accurate evaluation of various distributed resources (such as distributed power generation, energy storage systems, and controllable loads), thereby achieving multi-objective optimization along with optimal resource allocation, which reduces the overall operating cost of the virtual power plant and improves the system’s economic performance. However, this method may prematurely converge to a local optimal solution during the exploration phase, particularly when dealing with complex optimization problems. This occurs because the Harris Hawk optimization algorithm performs well in the exploitation phase, but may lack efficiency in more complex search spaces during the exploration phase, resulting in suboptimal performance of the algorithm in the global search for multi-objective operation scheduling solutions for virtual power plants [7].

Naughton et al. propose a service-oriented collaborative optimization method for virtual power plants. This method enables virtual power plants to flexibly adjust the output of each distributed resource according to market demands and resource characteristics for optimal resource allocation. However, virtual power plants contain many types of distributed resources with different characteristics. For example, distributed power generation exhibits output volatility, energy storage systems are constrained by limited charge–discharge capabilities, and demand response resources vary in response speed and adjustment range. These differences require the method to fully consider the characteristics and constraints of various resources during balanced resource allocation to maintain allocation balance [8].

Alabi et al. study a data-driven optimization dispatch method for multi-energy system virtual power plants. By collecting and analyzing extensive historical, real-time, and forecast data, they employ advanced algorithms and models to achieve collaborative optimization and dispatch of multiple distributed resources in VPPs. This method is capable of accurately reflecting the actual characteristics of resources and market dynamics, thereby improving resource utilization efficiency while enhancing system stability and economic performance. However, this method typically requires substantial computing resources to support complex algorithms and model operations, potentially resulting in higher computing costs and compromising the real-time performance of scheduling decisions [9].

Ahmadi et al. study a distributed energy resource allocation method based on the multi-objective grasshopper optimization algorithm and establish models for the characteristics of distributed energy resources (such as distributed power generation, energy storage systems, and demand response resources) in virtual power plants, including output characteristics, cost characteristics, and constraints. The optimization goals are set according to the operational needs of the virtual power plant, and the multi-objective grasshopper optimization algorithm is applied to solve the distributed energy resource allocation problem. Through iterative calculations, the optimal resource allocation plan that meets all optimization goals is determined. However, the multi-objective grasshopper optimization algorithm requires evaluating numerous candidate solutions during the search process, potentially resulting in high computational costs, particularly for large-scale, multidimensional virtual power plant resource allocation problems, where this disadvantage becomes more prominent [10].

Taye et al. propose an energy management approach for dispatchable and non-dispatchable energy units in a DC microgrid with energy storage based on fuzzy logic. They develop a detailed model of droop control and a fuzzy logic controller to optimize the allocation of dispatchable and non-dispatchable energy resources and voltage regulation in DC microgrids. This study investigates the fuzzy energy management problem of dispatchable and non-dispatchable energy units in DC microgrid energy storage systems. However, the fuzzy logic control approach relies on a predefined rule library to achieve local adjustment. When applied to the balanced allocation of multiple resources in virtual power plants, the limited rule library results in insufficient global search capability, especially under conditions of significant fluctuations in grid supply and demand or variations in resource characteristics, hindering effective resource allocation [11].

The enhanced Harris Hawk optimization technique introduced in reference [7] exhibits a tendency toward premature convergence when addressing complex optimization problems, hindering its ability to attain globally optimal solutions. The virtual power plant service coordination strategy presented in reference [8] inadequately accounts for the heterogeneity of multiple distributed energy resources, leading to suboptimal accuracy in resource allocation. While the data-driven approach in reference [9] leverages extensive datasets, its substantial computational resource requirements compromises real-time scheduling performance. The multi-objective locust optimization algorithm proposed in reference [10] incurs high computational overhead when applied to large-scale optimization tasks. The fuzzy logic-based control method in reference [11] demonstrates constrained global search capabilities, rendering it ineffective in managing pronounced fluctuations in power grid supply and demand. These individual methodologies exhibit limitations in global exploration, local refinement, adaptability to resource characteristics, and computational efficiency. In light of the limitations identified in prior research, this paper proposes a balanced allocation method for multiple distributed energy resources in virtual power plants (VPPs) based on a genetic-heuristic algorithm. The proposed approach first models the characteristics of distributed energy resources within the VPP framework and subsequently employs the genetic-heuristic algorithm to derive an optimized resource allocation strategy. To address the imbalance between global search and local optimization in traditional single intelligent algorithms, this paper proposes dual improvements through a hierarchical optimization mechanism: First, by employing the chromosome encoding technology of genetic algorithms, multiple types of resource characteristic parameters are mapped into discrete gene fragments, and feasible solutions are searched in parallel through operations such as roulette wheel selection, crossover, and mutation, thereby overcoming the limitations of traditional heuristic algorithms that exhibit heavy dependence on initial solutions; second, through the introduction of the pheromone update mechanism from the ant colony algorithm, where the suboptimal solution output by the genetic algorithm serves as the initial path, enabling dynamic adjustment of the strategy selection probability through pseudo-random proportional rules, and implementing local evaporation and global reinforcement mechanisms to perform fine optimization of candidate solutions, thus resolving the issue of slow convergence in the later stages of traditional genetic algorithms. The experimental results demonstrate that the proposed method achieves 100% accurate resource allocation with zero instances of excess power generation, outperforming the comparison method by reducing the resource allocation deviation rate by 32–67%. The algorithm exhibits fast convergence, obtaining solutions within 0.6 s across 14 repeated experiments, with an average convergence time reduction of 42% compared with traditional algorithms. These results verify the efficiency and robustness of the algorithm under complex working conditions, offering a new approach for resource allocation optimization in virtual power plants. The genetic-heuristic algorithm integrates the global search capability of the genetic algorithm with the local search advantage of the heuristic algorithm to determine the optimal solution for the balanced allocation of multiple distributed resources in virtual power plants by emulating biological evolution processes such as selection and crossover while incorporating specific heuristic rules.

Multi-Distributed Resource Equalization Allocation Methods for Virtual Power Plants

The virtual power plant multi-resource balanced allocation method proposed in this paper achieves comprehensive optimization across the entire process from modeling to policy output through integration of the advantages of both genetic algorithm and ant colony algorithm. The specific process is shown in Fig. 1.

[See PDF for image]

Fig. 1

Flow chart of balanced allocation of multiple distributed resources in virtual power plant

Modeling of Virtual Power Plant Characteristics

The virtual power plant management system is shown in Fig. 2.

[See PDF for image]

Fig. 2

Virtual power plant energy management system

The virtual power plant operates based on information flow and energy flow. The information flow is implemented via the communication system to enable two-way interaction between each unit of the virtual power plant and the energy management center [12]. Initially, the communication system collects operational status information from each distributed generation unit for the current time period and transmits it to the virtual power plant [13]. Subsequently, the virtual power plant generates optimized decisions, and the communication system distributes the plan for the next time period to each power generation unit. Finally, the virtual power plant executes protocol control. The energy flow represents the implementation of the information flow decision-making results, coordinating power allocation among the distributed power generation units in the virtual power plant [14].

Wind Power

The wind power output $p_{\mathit{WT}}$ under balanced allocation of multiple distributed resources in a virtual power plant depends on wind speed:

1

$$p_{\mathit{WT}}=\left\{\begin{array}{cc}0,& A_{\mathit{rt}}<A_r\\ p_{r,WT}\times \left(\frac{A_{\mathit{rt}}-A_{c1}}{A_{\mathit{rt}}-A_{c2}}\right)& A_{c1}\le A_{\mathit{rt}}\le A_r\\ p_{r,WT}& A_r\le A_{\mathit{rt}}\le A_{c2}\\ 0,& A_{\mathit{rt}}>A_{c2}\end{array}\right),$$where $A_{\mathit{rt}}$ indicates wind speed in scenarios $r$ at the moment $t$; $A_{c1}$ and $A_{c2}$ denote the cut-in and cut-out wind speeds, respectively; and $p_{r,WT}$ and $A_r$ represent wind power rating and wind speed rating, respectively.

When the wind speed is between $A_{c1}$ and $A_r$, the output power and the rated power of the wind turbine $p_{r,WT}$ has a proportional relationship. When the wind speed is between $A_{c2}$ and $A_r$, the output power is rated [15]. It is assumed that when the wind speed exceeds $A_{c2}$, the wind turbine will shut down, and its output power will drop to 0 MW.

Photovoltaic Power Generation

For evenly distributed multiple distributed resources in virtual power plants, the output power of photovoltaic generation, $p_{\mathit{rt}}^{\mathit{PV}}$, is calculated using Formulas (2) and (3). It is proportional to the received solar radiation, $F_{\mathit{rt}}$, at any given time and scenario.

2

$$p_{\mathit{rt}}^{\mathit{PV}}=m_{\mathit{PV}}{B}_{\mathit{PV}}{F}_{\mathit{rt}}{\beta }_{\mathit{PV}},$$where $m_{\mathit{PV}}$, $B_{\mathit{PV}},$ and ${\beta }_{\mathit{PV}}$ represent the number of PV panels, area, and power generation efficiency, respectively.

The power generation efficiency of a photovoltaic panel, ${\beta }_e$, can be calculated using the known rated generation efficiency, ${\beta }_{\mathit{PV}}$; the power trimming efficiency, ${\beta }_{\mathit{pc}}$; temperature coefficient, $W_{\mathit{PV}}$; unit temperatures, $W_c$; and the rated temperature, $W_r$.

3

$${\beta }_e={\beta }_{\mathit{PV}}{\beta }_{\mathit{pc}}\left[1-W_{\mathit{PV}}\left(W_c-W_r\right)\right].$$

Energy Storage Systems

The energy storage system plays a crucial role in reducing the fluctuation of distributed generation output power when implementing balanced allocation of multi-distributed resources in a virtual power plant [16]. The energy storage system directly affects the power balance constraints and objective function optimization in the model through its charge–discharge regulation mechanism. At the power balance level, its state-of-charge constraints limit the dynamic range of charge–discharge power, preventing equipment damage or regulation failure caused by excessive charge–discharge operations, and ensuring that the virtual power plant maintains the power demand supply of the grid amid renewable energy fluctuations. In the optimization of the objective function, the adjustable capacity of energy storage is a key parameter of the fitness function, and its cost coefficient and charge–discharge efficiency directly influence the priority of the resource allocation strategy, thereby indirectly improving the economy and stability of the overall allocation strategy. In this study, battery cells serve as the energy storage system, with the state-of-charge constraints specified by Formulas (4, 5):

4

$$R_{\mathit{rt}}=R_{t-1}+p_E{\beta }_l-p_{\partial }/{\beta }_C,$$

5

$$R_{min}\le R_{\mathit{rt}}\le R_{max},$$where $R_{\mathit{rt}}$, $p_E,$ and $p_{\partial }$ represent the state of charge, charging power, and discharging power, respectively; $R_{t-1}$ represents the state of charge of the energy storage system at time t-1; ${\beta }_l$ and ${\beta }_C$ represent the charge–discharge efficiencies, respectively; $R_{min}$ and $R_{max}$ represent the charge minima and maxima, respectively.

Controllable Loads

In the implementation of balanced allocation for multiple distributed resources within virtual power plants (VPPs), the power load can be classified into three priority levels based on criticality: primary and secondary loads necessitate guaranteed power supply and are considered non-dispatchable from the perspective of power system energy management [17], whereas tertiary loads can be strategically curtailed during peak periods to provide flexibility for real-time grid balancing. The portion of tertiary loads that can be reduced or disconnected without substantially affecting end-users is defined as controllable loads, encompassing flexible resources such as power-adjustable heating/cooling systems, electric vehicle charging stations, and other demand-responsive loads. The control of such loads fundamentally represents an operational manifestation of demand response, with their participation being directly influenced by demand response strategies. For instance, by establishing reserve ratios for controllable loads and time-dependent participation windows, VPPs can quantify the adjustable potential of demand response resources. Subsequently, genetic-heuristic algorithms can be employed to optimize dispatch strategies, ensuring minimal resource activation costs while maintaining grid balance.

With the sustained development and growth of the national economy, accompanied by industrial restructuring and increasing residential electricity consumption, the overall structure of power demand and consumption patterns has experienced substantial transformation, resulting in a progressively rising proportion of controllable loads [18]. Regardless of whether the loads are industrial, agricultural, commercial, or residential, each category can be further categorized into controllable and non-controllable loads. This concept should be differentiated from interruptible loads, which are exclusively applicable to large industrial or commercial consumers and are regulated through contractual arrangements to incentivize active user participation [19]. Controllable loads refer to flexible load components derived from load classification that are directly controlled by the energy management center following agreements established with some or all tertiary loads [20].

At a given time $t$, the total load power is $P_Z^t$, where the controllable load power is $P_{\varphi Z}^t$ and the uncontrollable load power is $P_{\phi Z}^t$, satisfying[21]

6

$$P_Z^t=P_{\varphi Z}^t+P_{\phi Z}^t.$$

The controllable load reserve ratio is defined as

7

$${\eta }_t=\frac{P_{\varphi Z}^t}{P_Z^t}.$$

The 24-h controllable load reserve ratio is theoretically represented as a time-varying function that can be derived through historical data fitting. Considering load variation patterns and human behavioral characteristics, this time-varying function is simplified into a piecewise function for two primary reasons: accounting for daily activity patterns, electricity consumption during off-peak periods mainly comprises inelastic base loads with minimal controllable load components; during peak demand periods, controllable loads become more concentrated and exhibit greater elasticity owing to aggregation effects.

Consequently, the daily load profile can be segmented into four distinct time intervals to formulate the piecewise function representing the 24-h controllable load reserve ratio, employing days as the fundamental unit and hours as the measurement scale, with its mathematical expression defined as follows:

8

$${\eta }_t=\left\{\begin{array}{cc}{\eta }_1& t\in \mathrm{\Delta }{t}_l\\ {\eta }_2& t\in \mathrm{\Delta }{t}_m\\ {\eta }_3& t\in \mathrm{\Delta }{t}_n\\ {\eta }_4& t\in \mathrm{\Delta }{t}_f\end{array}\right),$$where $\mathrm{\Delta }{t}_l$, $\mathrm{\Delta }{t}_m$, $\mathrm{\Delta }{t}_n,$ and $\mathrm{\Delta }{t}_f$ are the low hours (0:00–7:00), the morning peak hours (8:00–12:00), the evening peak hours (19:00–21:00), and the slow hours (7:00–8:00/12:00–19:00/21:00–23:59).

If the actual load power is $P_{}^{\prime }_Z^t$, the controllable load participation at that time is defined as

9

$${\mu }_t=\frac{P_Z^t-P_{}^{\prime }_Z^t}{P_{\varphi Z}^t}$$

Considering both load flexibility requirements and power system operational security constraints, the optimal participation range for controllable loads is established at 0.6–0.8. Values below this range would result in reserve capacity underutilization, while exceeding this range may cause backup capacity inadequacy. This specified range demonstrates the practical necessity to maintain equilibrium between load elasticity and grid security by keeping controllable load participation within appropriate bounds (0.6–0.8). Within this optimal interval, controllable loads can effectively support power system dispatch operations while avoiding either excessive resource exploitation or over-reliance on load regulation capabilities [22].

Multi-Distributed Resource Equalization Allocation Model for Virtual Power Plant Based on Genetic-Heuristic Algorithm

After analyzing the operational characteristics of various resources including wind power, photovoltaics, energy storage systems, and controllable loads, a genetic-heuristic algorithm-based balanced allocation model for multiple distributed resources is developed for virtual power plants, where appropriate genetic-heuristic algorithms are employed to determine the optimal allocation strategy and achieve resource optimization.

Multi-Distributed Resource Equalization Allocation Model for Virtual Power Plants

Virtual power plants manage distributed generation resources including wind, photovoltaic, energy storage, and controllable loads. With their increasing adoption in distribution networks, the efficient utilization of these resources has become challenging. Centralized management hinders resource development, while decentralized approaches face difficulties in equitable power allocation to the grid, particularly during supply–demand imbalances. To address this, this chapter incorporates game theory’s “Nash equilibrium” concept to propose an “equilibrium scheduling model.” Nash equilibrium focuses on individual autonomous decision-making in non-cooperative games, whereas balanced scheduling optimizes distributed resource utilization without compromising participant interests, fostering cooperative-conflict balance in grid operations. Within virtual power plants, balanced scheduling maximizes operational effectiveness while meeting grid dispatch requirements, ensuring benefits for all distributed generation resources and achieving equilibrium in both resource utilization and economic returns. The input data for the model are acquired from multiple sources: the total number of distributed resources (Seti) and their adjustable capacity are determined according to equipment nameplate specifications (e.g., wind turbine rated power, photovoltaic panel conversion efficiency) or historical operational records. Real-time output parameters of wind and photovoltaic generation are obtained from meteorological station measurements or satellite remote sensing platform forecasts of solar radiation and wind speed. Energy storage system characteristics, including charge–discharge efficiency and capacity limits, are extracted from manufacturer technical specifications or calibrated through charge–discharge cycle testing. Controllable load classification and participation intervals are derived from historical load curve analysis using smart meter data, with adjustments made in compliance with industry standards such as GB/T 15543-2019. To examine the balanced scheduling strategy for distributed generation resources within the virtual power plant framework, the following assumptions are established in this study.

1. The power demand on the virtual power plant must not exceed the aggregate generation capacity of its multiple distributed energy resources.

2. The distributed energy resources comprising the virtual power plant satisfy the optimal allocation criteria.

The multi-distributed resource balanced allocation problem for virtual power plants is formulated as follows when solved using a genetic-heuristic algorithm:

1. Let $M$ denote the total number of distributed generation resources of wind, photovoltaic, energy storage, and controllable loads inside the virtual power plant, and the number of distributed generation resources inside the virtual power plant is a fixed constant.

2. Let $a_j$ be the $j$-th distributed generation resource, with $M$ being the total number of resources.

   The allocation strategy for each distributed generation resource is the power $p_j$ that generates.

3. Let $p_c$ be the power required by the power system from the virtual power plant, i.e., $\sum _{j=1}^M{p}_j=p_c$.

4. For each distributed energy resource participating in dispatch, the generated power depends on its adaptability—a parameter determined by the resource’s cost coefficient and its dispatchable power capacity (i.e., adjustable power). Consequently, the fitness function evaluating the distributed energy resources can be expressed as follows:

   10

   $$g_j\left(p_j\right)=\frac{1}{o_j}\left(1-\frac{p_j}{p_j^{\prime }}\right),$$where, $g_j\left(p_j\right)$ represents the fitness of policy $p_j$ for the $j$-th individual of distributed generation resources involved in the scheduling; $p_j^{\prime }$ denotes the schedulable capacity (callable power) of each individual; and $o_j>0$ represents the cost coefficient of distributed generation resources. The function implies that higher costs for calling distributed generation resources result in larger adaptation coefficients, while lower costs lead to smaller adaptation values.

The dynamic model ${\overline{p}}_j$ for balanced multi-resource allocation in virtual power plants is expressed as

11

$${\overline{p}}_j=p_j\left[g_j\left(p_j\right)-\tilde{g}\right],$$

12

$$\tilde{g}=\frac{1}{p_c}\sum _{j=1}^M{p}_j{g}_j\left(p_j\right).$$

Among them, $\tilde{g}$ is the average adaptation of distributed generation resources. When Formula (12) reaches steady state, the value ${\overline{p}}_j$ is 0, while $p_j$ exceeds 1. The equilibrium point occurs when $g_j\left(p_j\right)=\tilde{g}$. Moreover, when solving the distributed energy resource allocation strategy, the genetic-heuristic algorithm converges to a state where all individuals (representing distributed energy resource power outputs) achieve identical fitness values (dispatch costs). This equilibrium represents the optimal resource allocation solution as a function of fitness.

Multi-Distributed Resource Equalization Allocation Objective Function for Virtual Power Plants

For each individual distributed generation resource, the fitness function value corresponding to the selected strategy (i.e., the determined generation output) that equalizes benefits for all participants equals the average fitness function value across all strategies from previous individuals. However, directly applying the genetic-heuristic algorithm to the virtual power plant’s distributed generation resource scheduling model would result in excessively slow convergence and difficulty obtaining solutions. Therefore, we modify the genetic-heuristic algorithm by minimizing the absolute difference between each individual’s fitness function and the population’s average fitness function value, which constitutes the core objective of multi-distributed resource balanced allocation in virtual power plants. This approach seeks a resource allocation strategy where the relative merit of each strategy (measured by its fitness function) approximates the population average. Such balanced strategies prevent extreme superiority or inferiority of particular allocation methods, thereby enhancing system stability and operational efficiency. Therefore, the objective function simultaneously ensures optimized resource allocation balance while indirectly promoting the cost of distributed generation resource called $o$ minimized.

13

$$o=min\left|g_j-\tilde{g}\right|,$$where $g_j$ represents the total fitness of distributed generation resource allocation strategies (power generation strategies for distributed resources) and $\tilde{g}$ represents the average fitness across all distributed generation resource allocation strategies.

To mitigate premature convergence risks in genetic-heuristic approaches, dynamic adjustment of crossover and mutation probabilities is implemented during the genetic algorithm phase. As population diversity diminishes, the mutation probability automatically increases to introduce novel genetic material. An elite preservation strategy ensures the direct advancement of top-performing individuals across generations, thereby maintaining solution quality. During the ant colony optimization phase, the pheromone update mechanism is enhanced by increasing the reinforcement coefficient for historically optimal paths while avoiding over-reliance on current optimal solutions, which effectively prevents premature convergence.

Multi-Distributed Resource Equalization Allocation Constraints for Virtual Power Plants

1. Power system power balance constraints.

   All distributed energy resources undergo strategy learning for power generation formulation. The virtual power plant’s power output must consistently satisfy power system requirements, meaning

   14

   $$\sum _{j=1}^M{p}_j=p_c.$$

2. Resource strategy selection constraints

   15

   $$0\le p_j\le p_j^{\prime }.$$

Solving Multi-Distributed Resource Equalization Allocation Strategy for Virtual Power Plants Based on Genetic-Heuristic Algorithm

The genetic-ant colony fusion algorithm represents an increasingly prevalent biologically inspired optimization technique. This hybrid approach combines the strengths of genetic algorithms and ant colony optimization to address the balanced allocation challenge of multiple distributed energy resources in virtual power plants. Conventional methods face limitations when scaling up, necessitating the application of intelligent optimization algorithms. While genetic algorithms demonstrate strong global search capabilities, they tend to suffer from premature convergence. Conversely, ant colony algorithms exhibit slower convergence rates and may encounter stagnation during initial iterations. Therefore, this paper proposes a genetic-heuristic fusion algorithm. Compared with standalone genetic or heuristic algorithms, this hybrid approach leverages the global parallel search capability of genetic algorithms to efficiently explore the solution space while avoiding local optima. Simultaneously, it incorporates the pheromone-guided mechanism of ant colony optimization to achieve local refinement, thereby improving convergence speed while maintaining solution accuracy. Consequently, the multi-resource allocation strategy can rapidly reach equilibrium, enhancing virtual power plant robustness to meet real-time scheduling demands. In the ACO algorithm, pheromone levels indicate path quality or solution superiority. For the distributed energy resource power formulation problem, each potential power allocation strategy corresponds to a “path,” where the pheromone concentration reflects the solution quality. Higher pheromone concentrations denote superior paths (i.e., power allocation strategies), making them more likely to be selected by subsequent ants (i.e., algorithm iterations). Regarding parameter configuration, the genetic algorithm employs dynamic adjustment of crossover probability (initial range: 0.6–0.8) and mutation probability (initial range: 0.01–0.03). When population fitness variance falls below the threshold, the mutation probability automatically increases to 0.05 to maintain diversity. An elite retention strategy preserves the top 5% individuals for direct inclusion in the next generation. For the ant colony algorithm, the pheromone evaporation coefficient is set between 0.1and 0.3, while the global pheromone enhancement coefficient ranges from 1.5 to 2.0. The pseudo-random proportional rule (parameter q₀ = 0.7) balances exploration and exploitation. Initial pheromone concentrations are dynamically assigned according to resource adjustable capacity, ensuring efficient optimal strategy exploration. The genetic-heuristic algorithm enhances virtual power plant resource allocation efficiency through its hierarchical optimization framework, with the core process illustrated in Fig. 3.

[See PDF for image]

Fig. 3

Workflow of genetic-heuristic algorithm

1. Encoding and decoding.

   This study employs real-coded representation, where each distributed energy resource allocation strategy (i.e., generation power formulation strategy) corresponds to an individual chromosome.

2. Adaptation.

   Adaptation quantifies the fitness of individual distributed energy resource power formulation strategies, as mathematically defined in Formula (10) established in the preceding section.

3. Genetic manipulation.

   The initial genetic operation involves selection, where selective advantage constitutes a fundamental characteristic of genetic algorithms. For each individual representing a distributed energy resource power formulation strategy, its inheritance probability to the next generation correlates directly with its fitness value. The selection probability for individual $j$ is expressed as

   16

   $$Q_j=\frac{g_j}{\sum g_j}.$$

   To enhance the search capability of the genetic algorithm, all individuals representing distributed energy resource power formulation strategies are paired consecutively for crossover operations performed with a specified probability, thereby generating two new offspring. This study employs a hybrid crossover method combining single-point and sequential crossover techniques, where $L_1$ and $L_2$ represent two parent individuals as follows:

Assume each individual representing a distributed generation resource power strategy has a coded string of length $n$. A random integer $s_c\in \left[1,n-1\right]$ is generated, with the first $s_c$ genes selected as the crossover segment and stored in gene tables $H_1$ and $H_2$, respectively.

In parent string $L_1$, all genes matching $H_2$ are located left to right and their positions set to 0. Similarly, in string $L_2$, genes matching $H_1$ are identified and zeroed. These genes represent output values of distributed generation units (e.g., wind turbines, solar panels), with their corresponding positions nullified.

For all positions in $L_1$ and $L_2$ where the gene value equals 0, a leftward shift is performed toward the crossover point while preserving the order of non-zero positions. The genes from $H_1$ and $H_2$ are then inserted into the zero-value locations of $L_1$ and $L_2,$ respectively, generating two new distributed generation resource power output strategy individuals $V_1$ and $V_2$. Assuming that $L_1=\left\{3421\right\}$ and $L_2=\left\{1324\right\}$, with a randomly generated crossover point $s_c$ = 2, the resulting offspring individuals after crossover are $V_1=\left\{1324\right\}$ and $V_2=\left\{3421\right\}$.

1. Variation.

   This study randomly selects two chromosome loci based on the predetermined mutation probability and inverts the gene sequence between these points. These steps iterate until reaching the maximum iteration count, at which point the algorithm terminates and outputs multiple near-optimal solutions for distributed energy resource power formulation strategies.

2. The heuristic algorithm (ant colony optimization) performs in-depth refinement of the optimal power formulation strategy for distributed energy resources.

In the ant colony algorithm implementation for distributed energy resource power formulation strategies, the ant population (size $m$) represents searchers exploring different strategy configurations, where ${\lambda }_f$ denotes the set of power allocation strategies discovered by ant $f$ during the search process.

First, the transition probability criterion in the ACO algorithm is examined. This criterion incorporates both the allocation probability from distributed generation resource power output strategy nodes to electricity demand nodes and operates within the virtual power plant’s multi-distributed generation resource suboptimal solution space. Here, variable ${\theta }_f$ represents the suboptimal solution set for current ant $f$’s distributed generation resources at demand node $U_j$. The selection of power output strategies follows pseudo-random rules ${\theta }_j$ for allocation to demand nodes $U_j$, governed by the principles:

17

$${\theta }_j=\left\{\begin{array}{cc}argmax\left\{{\vartheta }^{\sigma }\left({\theta }_j,U_j\right)·{\psi }^{\tau }\left({\theta }_j,U_j\right)\right\}& \mathrm{\Omega }\le {\mathrm{\Omega }}_0\\ 0& \text{other}\end{array}\right),$$where$\sigma$ and $\tau$ denote two parameters that reflect ants’ preference weights for pheromones versus heuristic factors. ${\mathrm{\Omega }}_0\in \left[0,1\right]$ represents a threshold value, where $\mathrm{\Omega }$ is a uniformly distributed random number between 0 and 1. When $\mathrm{\Omega }\le {\mathrm{\Omega }}_0$, within the suboptimal solution space for power output strategies, the strategy ${\theta }_j$ maximizing ${\vartheta }^{\sigma }\left({\theta }_j,U_j\right)·{\psi }^{\tau }\left({\theta }_j,U_j\right)$ is selected for distributed generation resource allocation to node $U_j$. Here, ${\vartheta }^{\sigma }\left({\theta }_j,U_j\right)$ denotes pheromone intensity on path $\left({\theta }_j,U_j\right)$ in ACO algorithm’s selection rules; otherwise, roulette wheel selection determines the pheromone-based allocation to $U_j$. Thus, ${\psi }^{\tau }\left({\theta }_j,U_j\right)$ defines the heuristic information for assigning ${\theta }_j$ to $U_j$.

Pheromone updates consist of local and global components, with the local update rule formulated as follows:

18

$${\psi }^{\tau }\left({\theta }_j,U_j\right)=\left(1-{\pi }_z\right)·{\psi }^{\tau }\left({\theta }_j,U_j\right)+{\pi }_z·\mathrm{\Delta }{\psi }_z\left({\theta }_j,U_j\right),$$where ${\pi }_z$ and $\mathrm{\Delta }{\psi }_z$ denote the pheromone local volatilization coefficient and pheromone initial value, respectively.

The global pheromone update rule is as follows:

• Within the virtual power plant’s multi-distributed energy resource power formulation strategy solution space, global pheromone updating is performed after all ants complete their strategy exploration, with the global pheromone update rule defined as follows:

  19

  $${\psi }^{\tau }\left({\theta }_j,U_j\right)=\left(1-{\pi }_D\right)·{\psi }^{\tau }\left({\theta }_j,U_j\right)+{\pi }_D·\mathrm{\Delta }{\psi }_z\left({\theta }_j,U_j\right),$$where ${\pi }_D$ indicates the global volatilization factor.

The solution steps are as follows:

Initial parameters and pheromone distribution are configured. The genetic algorithm-optimized suboptimal solutions for distributed energy resource power formulation strategies are ranked by fitness value and subsequently utilized as nodal inputs for the ant colony optimization process.

b. Within the near-optimal solution space for multi-distributed energy resource power formulation strategies in virtual power plants, $m$ ants are initially distributed across $m$ demand nodes. For each ant $f$, the following procedure is executed:

For each ant $f$, representing a searcher of power output strategies for distributed generation resources, the following steps are executed: according to the transition probability rules at its current demand node, after completing a distributed generation resource allocation strategy search, it performs local pheromone updates using Formula (18). When ant $f$ has traversed all electricity demand nodes, it records the current allocation policy $p_j$, evaluates the strategy’s fitness, and calculates the dispatch cost $o$. Otherwise, it selects the unallocated demand node with the highest priority as the next resource allocation target and repeats step b.

Upon completion of the strategy search by all ants for distributed energy resources, the current allocation policy with minimal cost $o$ is recorded. This policy is then compared against the global optimal policy, and if superior, replaces the existing global optimum.

The global pheromone is updated using Formula (19). When the iteration count reaches the predetermined threshold, the algorithm outputs the current distributed energy resource generation outputs and allocation results if the difference between each strategy’s fitness and the average fitness is minimized, thereby ensuring balanced power dispatch among multiple distributed resources. Otherwise, the process returns to step (b) and continues until termination criteria are met.

The proposed genetic-heuristic algorithm demonstrates computational compatibility and scalability, with its processing demands manifesting in two aspects: the parallel search operations during the genetic algorithm phase requiring processor capabilities, and the pheromone matrix iterative updates during the ant colony optimization phase demanding memory resources. For virtual power plant scale expansion, the algorithm supports extension through encoding structure enhancement, parallel computation optimization, adaptive parameter tuning, and hierarchical scheduling mechanisms. This architecture maintains solution efficiency for large-scale problems with linearly increasing computational resources, guaranteeing both real-time performance and scheduling accuracy.

Experimental Analysis

Experimental Design

The experimental validation employs an enhanced IEEE 12-bus benchmark system, selected for its typical characteristics, standardized architecture, moderate complexity, and multiscenario adaptability that closely emulates actual grid operation dynamics. This configuration effectively validates the virtual power plant’s multi-resource balanced allocation algorithm in terms of efficacy and robustness, while facilitating practical implementation. As illustrated in Fig. 4, the system features a 12-node bus topology with IEEE C37.15 protocol-enabled real-time data exchange between the virtual power plant’s energy management system, distributed resources, and grid interfaces. The testbed integrates 2 energy storage units and 2 controllable load devices, interconnected with a 61 MW wind farm and 81 MW photovoltaic station to form a hybrid virtual power plant entity. Six generation units (encompassing wind, solar, storage, and controllable load archetypes) are configured, satisfying both minimum functional requirements for coordinated dispatch and simulation of complex operational scenarios through differentiated resource characteristics (e.g., wind power intermittency and storage adjustability). The virtual power plant participates in grid services including peak shaving, frequency regulation, and standby response through dynamic unit output adjustment, while maintaining grid-friendly interconnection via IEEE 1547-compliant power command exchange. Algorithmically, the genetic algorithm employs 0.6–0.8 crossover probability and 0.01–0.03 adaptive mutation probability (dynamically adjusted) to preserve search diversity. The ant colony optimization utilizes 0.1–0.3 pheromone evaporation coefficient and 1.5–2.0 global reinforcement coefficient to balance local exploration with global convergence, accommodating the dynamic optimization demands of virtual power plant multi-resource allocation.

[See PDF for image]

Fig. 4

Improving the IEEE12 system architecture

To validate the feasibility of the internal balanced dispatch model for the virtual power plant, this paper considers a virtual power plant comprising 6 power generation units as a case study, with the specific adjustable capacity presented in Table 1.

Table 1. Adjustable capacity information of distributed generation resources in virtual power plants

Types of distributed power generation resources	Adjustable capacity /MW
Wind farm	61
Photovoltaic power station	81
Energy storage device 1	13
Energy storage device 2	41
Controllable load group 1	31
Controllable load group 2	31

To effectively evaluate the influence of different parameters on distributed generation resource allocation, the simulation is conducted using several representative cases:

1. The call cost factor is set identical for wind farms and PV plants;

2. The adjustable capacity is configured equally for controllable loads 1 and 2, with the participation degree of controllable loads set at 0.7;

3. The cost factor and adjustable capacity are set differently between energy storage device 1 and energy storage device 2;

4. Two scenarios are established to analyze the performance of each distributed generation resource under different power demands, as detailed in Table 2.

Table 2. Scenario design

Types of distributed power generation resources	Power demand value of 258 MW	Power demand value of 55 MW
Wind farm	61	10
Photovoltaic power station	81	25
Energy storage device 1	13	2
Energy storage device 2	41	8
Controllable load group 1	31	5
Controllable load group 2	31	5

These test scenarios are scientifically justified and highly representative. The identical communication cost coefficient setting for both wind farms and photovoltaic power plants reflects the shared communication cost characteristics among certain resources in practice. The configuration of adjustable capacity and participation for controllable loads aligns with the response characteristics observed during peak and off-peak periods in actual power grids. The variation in cost coefficients and adjustable capacities between energy storage devices accurately represents the real-world characteristics of different energy storage equipment. Furthermore, the two electricity demand scenarios encompass typical operating conditions of virtual power plants under both high power demand (e.g., industrial peak loads) and low power demand (e.g., residential nighttime loads), effectively simulating real-world virtual power plant operations and providing a reliable basis for precise analysis of distributed generation resource allocation.

Effectiveness of Distributed Generation Resource Allocation within Virtual Power Plants

Analysis of the Effects of the Allocation of Generation Resources

Comparative experiments are conducted between the proposed method and existing approaches, including the multi-objective operation scheduling method for virtual power plants based on an improved Harris Hawk optimization algorithm (Method in Reference [7]), the collaborative optimization method for virtual power plant services (Method in Reference [8]), the data-driven multi-energy system virtual power plant optimization scheduling method (Method in Reference [9]), and the energy scheduling method based on fuzzy logic (Method in Reference [11]). To ensure experimental scientificity and fairness, identical function evaluation counts are established for all compared algorithms prior to testing. In two scenarios with electricity demand values of 258 MW and 55 MW, respectively, the multiple distributed generation resources within the virtual power plant are evenly distributed, with allocation results presented in Table 3.

Table 3. Distribution of distributed generation resources within virtual power plants

Types of distributed power generation resources	Power demand value of 258 MW					Power demand value of 55 MW
	Proposed method	Multi-objective operation scheduling method for virtual power plants based on improved Harris Hawk optimization method	Collaborative optimization method for virtual power plant services	Optimization and scheduling method for multi-energy system virtual power plants based on data-driven approach	Energy scheduling method based on fuzzy logic	Proposed method	Multi-objective operation scheduling method for virtual power plants based on improved Harris Hawk optimization method	Collaborative optimization method for virtual power plant services	Optimization and scheduling method for multi-energy system virtual power plants based on data-driven approach	Energy scheduling method based on fuzzy logic
Wind farm	61	61	61	61	61	10	9	10	10	10
Photovoltaic power station	81	81	81	81	81	25	22	29	45	20
Energy storage device 1	13	13	13	13	13	2	10	4	7	2
Energy storage device 2	41	41	41	41	41	8	3	16	4	8
Controllable load group 1	31	31	31	31	31	5	13	5	2	5
Controllable load group 2	31	31	31	31	31	5	11	6	0	10
Total power generation /MW	258	258	258	258	258	55	68	70	68	55

As indicated in Table 3, when the power system demand is 258 megawatts, the proposed method, along with the improved Harris Hawk optimization method, virtual power plant service collaborative optimization method, data-driven multi-energy system virtual power plant optimization scheduling method, and fuzzy logic-based energy scheduling method, successfully utilizes the maximum adjustable capacity of each distributed generation resource to meet demand. However, when demand decreases to 55 megawatts, the improved Harris Hawk optimization method produces 13 megawatts of excess capacity, the virtual power plant service collaborative optimization method yields 15 megawatts of excess, and the data-driven multi-energy system virtual power plant optimization scheduling method generates 13 megawatts of excess. In contrast, the proposed method and the fuzzy logic-based energy scheduling method maintain capacity limits. Regarding resource allocation balance, while photovoltaic power generation exhibits significant fluctuations across comparative methods, the fuzzy logic-based energy scheduling method adopts a conservative approach, whereas the proposed method achieves superior balance. In terms of generation resource allocation costs, comparative methods suffer from waste due to overproduction and uneven distribution, and the fuzzy logic-based approach may inadequately account for cost factors. These limitations stem from premature convergence in the exploration phase of the improved Harris Hawk optimization method, insufficient consideration of resource characteristic differences in the virtual power plant service collaborative optimization method, computational resource constraints affecting real-time performance in the data-driven method, and limited global search capability in the fuzzy logic-based method. The proposed method integrates the global search advantages of genetic algorithms with the local search strengths of ant colony optimization, while mitigating premature convergence risks through chromosome encoding, gene manipulation, and pheromone update mechanisms, enabling efficient balanced resource allocation across varying demands without over-generation while maintaining effective cost control.

Furthermore, when the power system demand is 258 megawatts, all methods satisfy the requirement. However, when demand decreases to 55 megawatts, the improved Harris Hawk optimization, virtual power plant service collaborative optimization, and data-driven methods all exhibit over-generation, whereas the proposed method and the fuzzy logic method maintain generation limits. The comparative methods suffer from single-objective or constraint deficiencies, leading to over-generation. The improved Harris Hawk optimization lacks an “over-generation penalty” mechanism, increasing susceptibility to local optima. The collaborative optimization method employs oversimplified resource characteristic modeling, causing strategy-actual adjustable capacity mismatches. The data-driven method faces real-time performance limitations and inadequate cost constraints. The fuzzy logic method depends on a fixed rule library and exhibits insufficient global search capability. The proposed method implements a dual mechanism combining “genetic algorithm global encoding with resource constraints” and “ant colony algorithm local embedding with cost penalties,” which dynamically maps resource physical boundaries via chromosomes, integrates over-generation penalties into pheromone path evaluation, and quantifies balance through fitness functions, thereby achieving precise demand-resource characteristic matching. This approach eliminates over-generation caused by traditional methods’ objective/constraint shortcomings while enhancing allocation accuracy and cost control performance.

Analysis of the Effects of Genetic-Heuristic Algorithms

For the genetic-heuristic algorithm, the number of iterations required to converge to the optimal solution for the balanced allocation strategy of distributed generation resources serves as an effective indicator of the algorithm’s convergence speed. A smaller number of iterations indicates faster convergence, meaning the algorithm can solve the problem in less time. When the genetic-heuristic algorithm proposed in this work converges to the optimal solution for the multi-distributed generation resource balanced allocation strategy, the objective function iteration curve is shown in Fig. 5.

[See PDF for image]

Fig. 5

Genetic-heuristic algorithm iteration curve

Figure 5 demonstrates that the genetic-heuristic algorithm proposed in this study requires 50 iterations to solve the balanced allocation strategy for multiple distributed generation resources in virtual power plants. This result confirms the algorithm’s excellent convergence performance. By simulating biological evolution processes through selection and crossover operations, the algorithm performs effective searches in the solution space of virtual power plant allocation strategies. Furthermore, the heuristic algorithm component provides search direction guidance, enabling the identification of optimal solutions for distributed generation resource allocation in virtual power plants.

The solution time statistics for the distributed generation resource balanced allocation strategy, obtained from 14 repeated experiments using the genetic-heuristic algorithm, are presented in Fig. 6.

[See PDF for image]

Fig. 6

Solution time for balanced allocation strategy of distributed power generation resources

As illustrated in Fig. 6, the genetic-heuristic algorithm demonstrates superior performance with both reduced solution time and diminished time fluctuation range. This enhanced efficiency originates from its hybrid optimization approach: initially employing genetic algorithms for global exploration followed by ant colony algorithms for localized refinement within the genetic-heuristic framework. The algorithm consistently maintains these performance advantages across various operational scenarios, including both high and low power demand conditions, when implementing its multi-distributed generation resource balancing strategy. Notably, the solution process displays minimal temporal variations and improved stability, with all computational intervals consistently remaining below 0.6 s.

Distributed Generation Resource Utilization Analysis

Taking wind power resources in a virtual power plant as an example, the analysis of wind curtailment status reveals the actual utilization of these resources, identifying specific periods or regions where wind power potential remains underutilized. This analysis enables power system dispatchers to adjust generation plans and optimize resource allocation, thereby maximizing wind power utilization. In this paper, the proposed method is presented in Table 2 for two scenarios: balanced allocation of multi-distributed generation resources in a virtual power plant, and wind power variations as shown in Table 4.

Table 4. Usage status of distributed generation resources

Analysis content	Power demand value of 258 MW	Power demand value of 55 MW
Remaining wind power /MW	0	40
Maximum remaining wind power /MW	0	51
Maximum residual wind power rate	0	0.836

Table 4 demonstrates that when implementing the balanced allocation method in virtual power plant dispatching strategy, under high power demand conditions (258 MW), the residual wind power, maximum residual wind power, and maximum residual rate of wind power all register 0. Conversely, under low power demand scenarios (55 MW), these values measure 40 MW, 51 MW, and 0.836, respectively. This indicates the proposed method’s capability to fully utilize wind power generation resources during peak demand periods while maintaining balanced generation capacity during low demand, thereby reducing distributed generation resource dispatch costs. Although the maximum wind power residual rate reaches 0.836 under low demand, this does not signify inefficient utilization or wastage of wind resources. Rather, it reflects a conservative allocation strategy designed to preserve power supply stability and predictability while maintaining reserve capacity for potential future demand increases or sudden load variations in power generation capacity.

To validate the performance of the proposed method under varying renewable energy variability and demand uncertainty levels, sensitivity analysis experiments are performed. These experiments establish different renewable energy variability conditions, demand uncertainty scenarios, and benchmark cases, while employing resource allocation deviation rate, algorithm convergence time, and system stability index as evaluation metrics. The resulting sensitivity analysis data are presented in Table 5.

Table 5. Sensitivity analysis results

Scenario variables	Resource allocation deviation rate (%)	Convergence time (s)	System stability index
Renewable energy volatility
Wind-PV 10% fluctuation	3.2 ± 0.5	0.48	0.921
Wind-PV 20% fluctuation	6.3 ± 1.2	0.55	0.876
Wind-PV 30% fluctuation	8.1 ± 1.5	0.58	0.852
Demand uncertainty
Load forecasting error ± 5%	2.7 ± 0.4	0.45	0.935
Load forecasting error ± 10%	5.1 ± 0.9	0.51	0.902
Load forecasting error ± 15%	7.4 ± 1.3	0.56	0.881
Comprehensive fluctuation scenarios
Wind 20% + PV 20% + Load ± 10%	6.8 ± 1.1	0.54	0.889
Wind 30% + PV 10% + Load ± 15%	8.3 ± 1.4	0.59

As presented in Table 5, the proposed method demonstrates strong robustness when handling renewable energy variability and demand uncertainty. While the resource allocation deviation rate increases with growing renewable energy volatility and demand uncertainty, accompanied by a marginal rise in convergence time and a moderate decrease in system stability index, the algorithm consistently maintains convergence times below 0.6 s and preserves the stability index at relatively high values. Under comprehensive fluctuation scenarios, the method ensures reasonably balanced resource allocation while sustaining satisfactory performance across various interference levels, thereby guaranteeing effective multi-distributed resource allocation in virtual power plants.

Contributions

1. Algorithmic framework innovation: This study proposes a novel genetic-heuristic algorithm that synergistically combines the global search capability of genetic algorithms with the local optimization strength of ant colony algorithms. By implementing chromosome encoding and pheromone update mechanisms, it effectively addresses the global–local optimization imbalance in conventional algorithms while enhancing search efficiency and solution accuracy for virtual power plant resource allocation.

2. Multidimensional modeling and balanced allocation methodology: A comprehensive virtual power plant characteristic model incorporating wind power, photovoltaic generation, energy storage, and controllable loads is developed. The designed fitness function quantitatively evaluates resource allocation balance, enabling coordinated optimization between virtual power plant resource utilization efficiency and grid demand requirements.

3. Multiscenario verification and performance evaluation framework: Using the IEEE 12-bus system as a foundation, typical test scenarios are constructed with sensitivity analyses conducted considering renewable energy fluctuations and load uncertainties. An integrated evaluation system assessing convergence speed, allocation deviation rate, and system stability is established to validate algorithm robustness and real-time performance under complex operational conditions.

4. Over-generation mitigation and balanced allocation strategy: This research introduces an innovative balanced allocation strategy for multiple virtual power plant resources. Comparative experiments confirm the proposed method’s effectiveness in preventing over-generation across various power demand scenarios while improving distributed resource allocation balance and power grid operational stability.

Conclusion

To mitigate the negative impacts of distributed power generation on grid operations while maximizing the utilization of renewable energy, virtual power plants have drawn considerable attention. By establishing a unified framework to integrate diverse distributed power generation resources, virtual power plants enable coordinated scheduling of generation resources, ensuring power system stability while enhancing energy utilization efficiency, offering both environmental benefits and economic advantages. During virtual power plant operations, accounting for the technical characteristics of various distributed generation resources and their influence on the power system, an appropriate algorithm is employed to develop an optimal resource scheduling model. This paper investigates a balanced allocation approach for multiple distributed resources in virtual power plants utilizing a genetic-heuristic algorithm. The genetic algorithm, which emulates natural selection and genetic mechanisms, conducts comprehensive searches within the solution space for balanced allocation strategies of distributed resources in virtual power plants, effectively preventing convergence to local optima. This capability facilitates the identification of solutions approaching global optimality when addressing the intricate multi-resource allocation challenges in virtual power plants. The integration of genetic and heuristic algorithms further refines solution quality by applying heuristic-based local optimization to the globally optimal solutions identified by the genetic algorithm. The experimental results demonstrate that this method possesses distinct advantages in handling complex non-linear problems efficiently while enhancing power system stability and reliability, rendering it particularly valuable for resource management and optimal scheduling in virtual power plants. However, practical implementation of the proposed method in real-world power grid environments may encounter challenges including real-time performance deviations due to communication delays and interface compatibility issues among equipment from different manufacturers. Future enhancements will focus on improving the method’s robustness and real-time performance through algorithmic refinements and system integration. At the algorithmic level, a dynamic adaptive parameter adjustment mechanism will be implemented, complemented by reinforcement learning techniques to boost the algorithm’s adaptability and self-tuning capability under complex operating conditions. On the system level, standardized communication interfaces will be established to minimize latency, while an IoT and big data-based real-time monitoring and prediction system will be developed to ensure accurate data provision for resource allocation, thereby enabling truly dynamic and efficient resource distribution.

Author Contributions

Haifeng Li contributed to conceptualization, resource, and writing. Tao Jin and Lin Shi were involved in methodology and writing. Xian Xu contributed to supervision and resource.

Funding

This work was supported by the SGCC Jiangsu Electric Power Company project, “Research on the key technologies of online risk assessment, early warning and intelligent control of dual-high AC/DC hybrid power system,” (Project Number: J2023078).

Data Availability and Access

The raw data can be obtained on request from the corresponding author.

Declarations

Conflict of Interest

The authors declare no competing interests.

Ethical Approval

This article does not contain any studies with human participants or animals performed by any of the authors.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Shafiekhani, M; Ahmadi, A; Homaee, O; Shafie-khah, M; Catalao, JPS. Optimal bidding strategy of a renewable-based virtual power plant including wind and solar units and dispatchable loads. Energy; 2022; 239, 122379.1-122379.15.

2. Zhang, M; Cai, S; Xie, Y et al. Supply resilience constrained scheduling of megs for distribution system restoration: a stochastic model and FW-PH algorithm. IEEE Trans. Smart Grid; 2025; 16, 1 pp. 194-208.

3. Ahmet, D; Demet, CD. A review on machine learning models in forecasting of virtual power plant uncertainties. Arch. Comput. Methods Eng. State Art Rev.; 2023; 30, 3 pp. 2081-2103.

4. Subhasis, P; Sarthak, M; Pravat, KR; Binod, KS. A conceptual review on transformation of micro-grid to virtual power plant: issues, modeling, solutions, and future prospects. Int. J. Energy Res.; 2022; 46, 6 pp. 7021-7054.

5. Joakim, B; Karl, H; Johansson, FD. Dynamic virtual power plant design for fast frequency reserves: coordinating hydro and wind. IEEE Trans. Control Network Syst.; 2023; 10, 3 pp. 1266-1278.4661297

6. Zhang, T; Zhao, W; He, Q et al. Optimization of microgrid dispatching by integrating photovoltaic power generation forecast. Sustainability; 2025; 17, 2 648.

7. Pandey, A. K., Jadoun, V. K., Ns, J.: Real-time and day-ahead risk averse multi-objective operational scheduling of virtual power plant using modified Harris Hawk’s optimization. Electric Power Syst. Res. 220, 109285.1–109285.15 (2023)

8. Naughton, J; Wang, H; Cantoni, M; Mancarella, P. Co-optimizing virtual power plant services under uncertainty: a robust scheduling and receding horizon dispatch approach. IEEE Trans. Power Syst. Public. Power Eng. Soc.; 2021; 36, 5 pp. 3960-3972.

9. Alabi, TM; Yang, Z; Lu, L. Data-driven optimal scheduling of multi-energy system virtual power plant incorporating carbon capture system (CCS), electric vehicle flexibility, and clean energy marketer (CEM) strategy. Appl. Energy; 2022; 314, 118997.1-118997.23.

10. Ahmadi, B; Ceylan, O; Ozdemir, A. Distributed energy resource allocation using multi-objective grasshopper optimization algorithm. Electr. Power Syst. Res.; 2021; 201, 107564.1-107564.10.

11. Taye, BA; Choudhury, NB. Fuzzy logic-based energy management of dispatchable and non-dispatchable energy units in DC microgrid with energy storage system. SN Comput. Sci.; 2023; 4, 5 576.

12. Phani, R. L., Seshu, K. R., Koteswara, R. D.: A comprehensive review of soft computing algorithms for optimal generation scheduling. Int. J. Energy Res. 45(2 Pt.1), 1170–1189 (2021)

13. Kansal, V; Dhillon, JS. Ameliorated artificial hummingbird algorithm for coordinated wind-solar-thermal generation scheduling problem in multiobjective framework. Appl. Energy; 2022; 326, pp. 1129-1146.

14. Su, SW; Ji, YX; Zou, Y; Zhi, L. The optimized scheduling of virtual power plant is considered. Computer Simul.; 2022; 39, 2 pp. 56-61.

15. Fenghui, Z; Michael, MW; Xuecai, B; Weirong, L. Centralized resource allocation and distributed power control for NOMA-integrated NR V2X. IEEE Internet Things J.; 2021; 8, 22 pp. 16522-16534.

16. Flores-Quiroz, A; Strunz, K. A distributed computing framework for multi-stage stochastic planning of renewable power systems with energy storage as flexibility option. Appl. Energy; 2021; 291, pp. 116736.1-116736.15.

17. Rokhforoz, P; Gjorgiev, B; Sansavini, G; Fink, O. Multi-agent maintenance scheduling based on the coordination between central operator and decentralized producers in an electricity market. Reliab. Eng. Syst. Saf.; 2021; 210, pp. 107495.1-107495.13.

18. Appino, RR; Hagenmeyer, V; Faulwasser, T. Towards optimality preserving aggregation for scheduling distributed energy resources. IEEE Trans. Control. Netw. Syst.; 2021; 8, 3 pp. 1477-1488.4333869

19. Sambeet, M; Cletus, JC; Chiara, B; Jordi, M. Smart contract formation enabling energy-as-a-service in a virtual power plant. Int. J. Energy Res.; 2022; 46, 3 pp. 3272-3294.

20. Sreenivasulu, G; Sahoo, NC; Balakrishna, P. Low economic risk operation of transactive energy markets with renewable sources and virtual power plants using self-adaptive particle swarm optimization. Electr. Eng.; 2022; 104, 4 pp. 2729-2755.

21. Wu, L; Zhang, B; Jiang, Y. Output-power-controllable parity-time-symmetric wireless power transfer system with soft switching for kitchen appliances. IEEE Trans. Consumer Electron.; 2024; 70, 4 pp. 6630-6640.

22. Arman, A; David, P. Performance-based virtual power plant offering strategy incorporating hybrid uncertainty modeling and risk viewpoint. Electr. Power Syst. Res.; 2022; 203, 107632.1-107632.9.