Study On Daily Load Forecasting Method Based On Wavelet Transform

Daily load forecasting is the basis of the power system stability and safety operation. And it is important for power quality improving, energy saving and cost reducing. The daily load forecasting accuracy is influenced by two factors: The one is the orignal load data processing; the other is the forecasting mode building. The purpose of data processing is to check leaks, fill the vacancy, discard the false and retain the true. The basis of the forecasting model selecting is the features of the orignal load data, which is the foundation of choosing the suitable method. So there are two tasks should be considered in load forecasting, which are orignal data processing and forecasting model establishing.Disorder data processing and data continuation are two tasks of orignal load data processing. In this paper, it is combined the features of the orignal load data serial with the disorder data classification. And the searching and processing methods are provided respectively. Connected with the boundary effect of the wavelet transform, this paper the difference compensating is used as data continuation method by analyzing the favorites means.The load forecasting model which based on the orignal load varying regularity is established. The more obvious orignal load varying regularity is, the more accuracy forecasting result is. Wavelet transform is an analyzing method, which has the good localization nature both in the time-domain and frequency-domain at the same time. And the time series can be break down into different subsequences which have different frequencies, relative concentration energy and more obvious regularity than the original load serial. the prediction model which based on the wavelet transformis established. The model can improve the accuracy of forecasting result more effectively. In this paper, wavelet transform is analyzed in detail and Mallat algorithm is selected by comparing the decomposing and reconstructing sequences. And a wavelet function as well as the decomposition method of scale are selected. Orignal load serial is a time serial, in this paper, time serial model is selected as the forecasting model, and gives a number of commonly used time serial models and the model identification, parameter estimation and model validation methods.At last, the daily load forecasting software is developed. And the structure, functional modules and software features are introduced. And the error of a practical calculated example is analyzed. The result shows that the algorithm is effective and practical.