Operating Cost and Peak Load Driven Day-Ahead Multivariate Load Forecasting For Integrated Coal Mine Energy Systems

Day-ahead multivariate load forecasting is the foundation of establishing scheduling strategies for integrated coal mine energy systems. However, conventional evaluation metrics cannot reliably indicate whether the forecasting model can reduce the system’s operating costs, and peak load plays a significant role in the day-ahead load forecasting errors, which is crucial for the system’s stable and profitable operation. This paper proposes an operating cost and peak load driven day-ahead multivariate load forecasting method for integrated coal mine energy systems. First, we use an improved temporal convolutional network and a long short-term memory network to extract and merge deep load features so as to get the preliminary prediction of dayahead multivariate loads driven by operating costs. Following that, we predict the multivariate loads’ peak using the peak attention mechanism. Finally, based on the results of the peak load prediction, the day-ahead multivariate load preliminary prediction is corrected to obtain the day-ahead multivariate load prediction driven by peak load and operating costs. The proposed method is applied to an integrated coal mine energy system and contrasted with previous methods. The experimental results show that the proposed method can reduce the system’s operating costs while increasing the multivariate loads’ prediction accuracy.