Back Propagation Neural Network With Adaptive Differential Evolution Algorithm For Energy Demand Forecasting

Energy is one of indispensable substances in human production and life. Reliable energy demand forecasting can provide decision support for energy enterprise planning development strategy, and establishment a national energy policy. Therefore, it is valuable and meaningful to enhance the energy demand forecasting accuracy. Based on the data used, there are two main types of energy demand forecasting problem. One is that only the historical energy demand data is used to establish the prediction model, namely the time series prediction problem. The other is that the historical data of factors, which affect the demand for energy, such as GDP, population, import and export, are used to establish forecasting model. This thesis mainly discussed the optimization of the Back Propagation Neural Network(BPNN) and the two types of energy demand forecasting problem before-mentioned with the optimized BPNN.Firstly, a self differential evolution algorithm(ADE) is adopted to overcome the shortcoming of BPNN, which easily falls into the local minimum point in forecasting problem, thus to improve the forecasting accuracy of BPNN. In the proposed ADE-BPNN process, ADE is first applied to search for the global initial connection weights and thresholds of BPNN. After that, BPNN is employed to thoroughly search for the optimal weights and thresholds, establishing the forecasting model. Then, the proposed ADE–BPNN is applied to a time series forecasting example of electricity demand with nonlinear characteristics. The effectiveness and high forecasting performance of ADE–BPNN is verified by its comparison results with the autoregressive integrated moving average model(ARIMA), basic BPNN and other hybrid models. Finally, ADE–BPNN is used to multi-factors influenced electricity demand forecasting problems. Both the comparing numerical example and the additional example confirm that the ADE-BPNN on the prediction precision is better than multiple linear regression model, standard BPNN model and other forecasting models. Additional, the total energy consumption of China from 2014-2020 is forecasted to provide possible decision support for future energy supply plan.