The Short-Term Load Forecast Using Evidence Theory And Support Vector Machine

With the development of the electric power market, the load forecasting has become power system management members'correct policy-making the basis and the guarantee. Accurate load forecast can help arrange the internal electric generating sets start or stop reasonably, reduce the generating cost effectively, raise the economic efficiency and the social benefits. Therefore the study on how to do the load forecast accurately has the vital practical significance.Many algorithms for load forecast are available, yet there exist the problem that only considered the weather effect factors but neglect the daily load level. In view of this, this article conducts the deep research to the load forecast, enhances the forecast precision through support vector machine in the foundation of considering the weather effects and daily load level.Firstly, this paper presents select the average load size, load curve shape and the temperature difference as the amount of the selected features to the selected similar days, reoccupy fusion of evidence theory, avoid the traditional method only considering various factors affecting or load curve, selection of similar day with the load forecast, have very big difference, the results show that the prediction of the day, and similar load have better similarity.Secondly, By using the evidence theory pre-processing the original data, a large number of data which have a huge difference of the original data and forecast daily load characteristics is eliminated, enabling the changes of the historical data series have a greater simplification, and contains weather information. The establishment of support vector machine prediction model to predict, a large number of experiments carried out to select the optimal parameters for this sample to predict.Finally, study the neural network algorithm and the adaptive chaotic. Using the above three methods forecast load, using d-s evidence fusion, then compare with standard support vector machine, the neural network and adaptive chaotic, the comparison results show that the algorithm proposed method can obviously improve the prediction results.