Research On Power System Short-Term Load Forecasting Model And Optimum Selsction

Power system short-term load forecasting (STLF) is an important task of power utilities, a foundation of power system programming and decision, and a precondition of electricity market. It is also important for power department to improve economic benefit. With the development of electricity market, the method of short-term load forecasting is becoming more and more important.At first, this paper discusses the constituents and characteristics of the electric load, analyzes and compares the merits and shortcomings of some forecasting methods. Then aiming at the load of Chengdu region, ARIMA (Auto-Regressive Integrated Moving- Average) model and ANN (Artificial Neural Network) of an improved BP algorithm model are presented. According to the forecast results, the precision and adaptability of ANN is improved to ARIMA. But the weekend prediction load precision is also poor. In different date type the load characteristics are different. Based on the second model, this paper proposes the third model, which build ANN training and forecasting models respectively with working day and general holiday. This model improves the precision of Monday and general holiday's load prediction to a certain extent, but the peak and valley prediction load is notimproved obviously. The load serials can be considered as a linear combination of sub-serials characterized by different frequencies. Therefore, the fourth method based on Artificial Neural Network and wavelet transformation for short-term load forecast is presented in this paper. The load serials are first decomposed into.different sub-serials by using the proper wavelet function and the resolution level. Each of them varies with specific periodicities and regularization. Therefore different ANN models are designed to capture each sub-serial's characteristics. After all sub-serials are forecasted, the whole predicted load series would be composed or reconstructed. The outcome of the study clearly indicates that the proposed composite model has higher accuracy and stability and reduces the error of peak and valley load remarkably.We need consider the high precision as well as quickly calculating rate and accuracy to the short-term load forecasting. So a multi-objective decision method is proposed to select the optimum model on the four given forecasting models. The results show that the wavelet transform and ANN based approach is the optimum model. We should choose the right wavelet and decomposed level according to the load, because in different regions the load characteristics are different. Then the multi-objective decision method is also built to select the optimum wavelet and decomposed level automatically.This paper applies the method to the different regions to compare and analyze. The outcome indicates that the wavelet transform has strong adaptability and application potential in load forecast and has remarkable superiority than other models. In different areas and different time the load characteristics are different. This paper puts forward the thought of optimum selections to prediction models. Then the suitable forecasting model can be selected automatically according to the practical application of power system by the method.