Research On Combined Forecasting Method In Short-Term Load Forecasting

Short Term Load Forecast (STLF) plays an important role in electric power management. The degree of accuracy forecasting is not only safe and important solution for power supply but also brings the benefit for electric power. Many factors such as the time, weather, random disturbance and the classification of customers are affected to the short term load forecast. Nowadays, for short term load forecast people use a lot of methods such as Artificial Neural Network (ANN), Time sersies, Fuzzy logic, Arima, wavelet analysis. However, these methods still have many disadvantages and not ideal for prediction and the results of load forecast are not high. Therefore to adapt the requirements of load forecast, this thesis applied the combined method for STLF. It is the combination of two single methods. Moreover, there are a lot of different combined methods such as Fuzzy-neural network, neural network-wavelet transform, Fuzzy-RBF...Each of combined method has different characteristic and depend on different factors. For example, the characteristics of combined forecast method will have more effective and flexible than single forecast method through out the information analysis, reducing the disadvantages of single forecast method for example in the short-term power load forecasting model based on fuzzy RBF neural network, it has overcome the BP algorithm’s disadvantage of slow convergence rate. Therefore, the combined model improves the ability of forecasting with the small prediction error.One of the best methods which are recently used is the combination between two methods BoxJenkin method and Wavelet analysis method. It is also call WARIMA, the Box Jenkin method is based on the ARIMA model and the intergration of wavelet with generalized autoregressive conditional heteroskedasticity (GARCH) model is proposed. Firstly, the time series will be decomposited by wavelet analyis, the proposed approach divides the electricity series into low frequency part and high frequency part; secondly, on this basis the GARCH models for each sub-series are established respectively to carry out the forecasting; thirdly, the forecasting results of sub-series are reconstructed by wavelet theory to implement the load forecasting. The data for experiment in this thesis belongs to historical data of Hunan province on05January2002to14January2002and forecast the load in the next day (24hours) on15January2002. The proposed forecasting approach is verified. Verification result shows that the proposed forecasting approach is feasible and effective.The theory and concepts which relate to the Arima model and Wavelet analysis, are also presented in this thesis. In addition, some of the applications of GARCH Tool Box and WAVELET Tool Box are reviewed in chapter3and chapter4. Finally, the method which is used in this thesis is forecasted the daily and weekly load in the short-term prediction and the results are showed in the simulations.