Core Vector Regression with Particle Swarm Optimization Algorithm in Short Term Load Forecasting

Abstract

Short term load forecasting is very essential to the operation of electricity companies. However, the methods of complexity of training time and space can not be acceptable when using a large dataset for forecasting a period of power loads. This paper proposes a new method for short term load forecasting using particle swarm optimization (PSO) and Core Vector Regression (CVR), PSO is applied for determining the parameters of CVR. The features of load data is analyzed for finding factors which may have great influence on forecasting, at the same time, it will create several training sets in diffident size for observing if a larger training data set could include more accurate results. Using PSO-CVR model is very efficiency to continuously predict one week loads. Experiments show that the PSO-CVR model has comparable performance with SVR (Support Vector Regression), where produces much faster and fewer support vectors on very large data sets. It also has good stability.