Charging Demand Forecasting of Electric Vehicle Based on Empirical Mode Decomposition-Fuzzy Entropy and Ensemble Learning

A charging demand forecasting method of electric vehicle based on empirical mode decomposition-fuzzy entropy and ensemble learning is proposed. This method decomposes the time series of charging demand for electric vehicle into relatively simple components by empirical mode decomposition. In order to avoid the cumbersome calculation and error accumulation caused by excessive components, firstly, the complexity of each component is calculated by using fuzzy entropy. The components are superimposed and combined to obtain a series of sub-sequences to reduce the number of components. Secondly, long short-term memory (LSTM) neural networks and supported vector regression (SVR) are used as the base learner for prediction of sub-sequences with different frequencies. Finally, the prediction result of base learner, the weather data and time series data of the pre-decomposed charging demand are combined to form the feature set by the Stacking integrated learning strategy. Final forecasting results are obtained through a fully connected neural network. Single-step and multi-step prediction experiments are carried out based on real data of charging demand for electric vehicle in a certain area of a certain city in Southwest China, and the comparison with other algorithms is made, which shows the reliability of the proposed method. © 2020 Automation of Electric Power Systems Press.