C18, C44, C45
Electricity markets are undergoing fundamental structural change as they transition from fossil fuel to renewable generation. Imbalance forecasting in electric power systems has received relatively little academic attention despite increasing complexity of the electric power grid system with the intermittent renewable production. This study fills the gap and questions whether the UK power system imbalance forecasting research is valid given the surging renewable generation. To this aim, it applies machine learning binary classifiers based on advanced forecasting methodology across the entire corpus. The present work is the first to perform an extensive imbalance forecasting research for the UK electricity system. Using a 52,560-half-hour dataset that covers 2017 to 2019 with a total of 257 exogenous, endogenous, and lagged variables, the proposed forecasting models exhibit up to 20% higher accuracy compared to the previous forecasting models. Based on various evaluation metrics for comparative assessments with selected benchmarks, it is found that Boosting techniques outperformed Artificial Neural Networks, and strong performance was achievable using Logistic Regression. The findings indicate that the presumption the model complexity is not automatically a desirable characteristic over these simpler techniques.
electricity market; forecasting system imbalance volume; binary classifiers