Content area
As a global transition towards renewables is underway, proper management and scheduling of long duration energy storage (LDES) technologies is essential to maintain grid reliability, and address uncertainty within Variable Renewable Energy (VRE) generation. This thesis employs a dynamic Echo State Network (ESN) that generates price forecasts used in an optimization problem to optimize the schedule of varying sizes of LDES devices. The objective function of the model is to maximize energy arbitrage by choosing when to charge and discharge the storage devices. The optimization model makes use of a rolling horizon, optimizing over a period with extended foresight. The ESN is trained with price, VRE and load data from NREL’s 118-Bus system to generate realistic price forecasts that are used as foresight in the optimization model. This work creates a framework that is better than deterministic models, by incorporating foresight of realistic price forecasts to inform the model. A multitude of simulations were conducted analyzing various lengths of foresight and sizes of LDES devices under two different market structures, a wholesale electricity market and an ancillary services (A/S) market. The operation dynamics of devices with shorter discharge durations were captured better with less foresight, while devices with longer discharge durations required more foresight. Smaller devices participating in the A/S market were influenced minimally by foresight horizon. Whereas larger devices saw a significant increase in value when simulated with a longer foresight. A vital takeaway is the impact on the value of storage devices of varying sizes when forecast error is present.
