A Multivariate Regression Load Forecasting Algorithm Based on Variable Accuracy Feedback

Load forecasting is a heated discussion point in power system, especially in the field of energy management. There are plenty of researches focusing on the improvement of forecasting algorithm. However, these algorithms are confronted with the computational burden caused by the complex forecasting verification process, as the traditional accuracy verification process is based on the stable relative error between forecasting data and history data. In order to solve this problem, this paper proposes a weather concerning load forecasting method combining with factor analysis feedback, tendency analysis feedback and multivariate nonlinear regression analysis algorithms. The proposed method can reduce the computational burden in forecasting. Tendency analysis about load indexes is demonstrated, and the relationship between weather and load indexes are evaluated by factor analysis. The forecasting equations are calculated by multivariate nonlinear regression analysis algorithm. We apply the variable relative error between the forecasting data and history data as the accuracy verification principle in forecasting verification process. The variable relative error can be calculated by the variable accuracy feedback of tendency analysis and factor analysis. Based on the proposed method, a more effective and convenient accuracy verification process is designed. Case studies are conducted to verify the proposed method. Copyright 2018 Elsevier Ltd. All rights reserved.