Electric Load Forecasting Based on Locally Weighted Support Vector Regression and Hybrid Wavelet Neural Network

Very short-term load forecasting forecasts the loads 1 h into the future in 5-min steps in a moving window fashion based on real-time data collected. Operative forecasting is significant in area generation control and resource communication. It is still difficult in view of the noisy data collection process and complex load features. In this paper wavelet neural networks with data pre-filtering, Hybrid wavelet neural network (HWNN) and support vector regression (SVR) methods are used to solve the load forecasting problem. The key idea is to use a spike filtering technique to detect spikes in load data and correct them. Wavelet decay is then used to decompose the filtered loads into multiple components at different frequencies, discrete neural networks are applied to detention the features of individual components, and outcomes of neural networks are then united to form the final forecasts. Testing results over MATLAB R2014a demonstrate the effects of data pre-filtering, the correctness of wavelet neural networks, the effectiveness of hybrid wavelet filters for taking different features of load components, and the accuracy of resulting prediction intermission approximations, based on a data set from ISO New England. The Hybrid Wavelet Neural Network based solution of STLF is proposed that provides a better outline for building a more accurate solution To perform moving forecasts, 12 dedicated wavelet neural networks are used based on test results. Numerical testing establishes the effects accuracy of wavelet neural networks based on a data set from ISO New England.