Study of artificial neural network based short term load forecasting

With more and more renewable energy integrated into the power grid and demand response in smart grid environment, electric load forecasting becomes more important. Accurate load forecasting facilitates better renewable energy integration and electricity market operation. Over the years, different load forecasting methods have been developed and applied. Multiple linear regression and artificial neural network based methods are well accepted by industries. This paper focuses on ANN-based method and provides detailed steps of load forecasting including data processing and neural network design.

I. Introduction

Electric load forecasting has been an important topic for electric utilities. With the growing penetration of variable power generation and demand response projects in the context of smart grids, there is a pressing need for better load forecasting. This will help system operators to better accommodate fluctuating wind and solar power generations which effectively changes the “net load” behavior. Over the years, regression-based and artificial neural network-based forecasting methods have been widely adopted in utility industry. Multiple linear regression method has been used for electric load forecasting, considering the interactions between variables that affect electric load behaviors [1]–[3]. The method is based on the general linear regression model. By introducing quantitative, qualitative, and transformed variables, complex relationship between forecast output (load) and input variables can be captured by polynomial regression model. The model has satisfactory accuracy, which is comparable to ANN based method.