Intelligent system to optimally trade at the interference of multiple crises

Abstract

In this paper, we aim to create an intelligent system to predict the wholesale market prices and provide the input variables for producers to optimally approach the electricity Day-Ahead Market (DAM) and Balancing Market (BM). We create a data set that includes open-source fundamental features collected from ENTSO-E, OPCOM, etc. and predict the DAM price using a univariate Long Short-Time Memory (LSTM) model and the sign (the deficit or surplus) of the system using ten Machine Learning (ML) classifiers. These new features are inserted as input to predict the imbalance prices with a stacked multivariate LSTM model. The predicted prices on DAM, system sign, and prices on BM are required to estimate the income from trading on DAM/BM and create optimal bidding strategies. This approach that combines classification to obtain a new significant feature (namely the imbalance sign) that is included into the input data set to further predict the imbalance prices using recurrent neural networks is an original method that can bring competitive advantages by creating optimal bids. Compared with the baseline, the proposed forecasting method for BM price improves Mean Absolute Error (MAE) by 14.63%, RMSE by 20.45%, MAPE by 9.5% and R² by 1.76%. Probability of the sign is used as probability to trade on BM for increasing or decreasing the output, whereas MAE from the predicted BM prices is used to enhance the optimal bidding pairs of prices and quantities on DAM. Therefore, in this paper, we propose a complex method that integrates variables and results from both DAM and BM markets and combines prediction and optimization models to attain competitiveness in the electricity markets.