Enhancing the Electrical Load Forecasting Efficacy for V2G Scheduling by Using LSTM Model and a Specific Loss Function

As the number of electric vehicles (EVs) increases, vehicle-to-grid (V2G) technology is widely adopted to help the power system manage the additional charging load. To minimize load variance using V2G technology, it is necessary to rely on dayahead electrical load forecasts. Although current electrical load forecasting (ELF) models exhibit excellent statistical accuracy, their forecasts do not necessarily translate into superior V2G scheduling performance. Therefore, this paper proposes a novel loss function to achieve value-oriented ELF, aiming to enhance the operational value of the forecasting model. Firstly, the relationship between forecast errors and V2G scheduling performance is revealed by an experiment. Secondly, a loss function is proposed to help forecasting model learn the pattern of relative magnitudes of the actual load, therefore realising value-oriented ELF. Finally, extensive experiments validate that the forecasts generated by the value-oriented ELF are with less statistical accuracy than those generated by the quality-oriented ELF, but they consistently achieve significantly superior V2G scheduling performance under various numbers of EVs.