Integrated Energy System Load Forecast Based on Information Entropy and Stacked Auto-Encoders

The integrated energy system (IES) has the characteristics of complex coupling and huge demand for precise load forecasting (LF). This paper proposes an IES LF method based on the information entropy theory and stacked auto-encoders (SAEs) to forecast the power, cooling, and heating load. Firstly, all necessary primary input parameters are presented and described in detail. Then, the information entropy theory is introduced to select the most important parameters among the primary ones, which helps to reduce the complexity of the proposed model. Besides, the SAEs are utilized to learn the high order characteristics of input data and forecast the load demand of each load type with the wake-sleep algorithm to improve the training efficiency. The case study shows that compared with classic back-propagation (BP) artificial neuron networks (ANNs), the prosed method is of higher prediction accuracy, which provides references for comprehensive LF of IES.