Combination Forecasting Technology And The Application In Power Prediction

Currently, there are various PV power station forecasting methods, but single forecasting method often has certain limitation.The combination forecasting method becomes a effective way to eliminate the limitation of single prediction and improve the power forecasting accuracy.At first, this paper proposes a PV power combination forecasting model based on the correlation coefficient. Using the continuous method, SVM method and the similar data method respectively forecast the PV power. By this the correlation coefficients of the actual values and three single forecasting values are respectively calculated, and the average correlation coefficient of the three days before the forecasting day as the single prediction method correlation coefficient of the forecasting day. The single forecasting method which has larger correlation coefficient is given a greater combination weight. By means of simulation and analysis, the prediction accuracy is improved by the combination method.Secondly, this paper puts forward a PV power combination forecasting method based on error distribution. The weight of combination forecasting is determined by the error of the historical single forecasting values and the actual values. Three single forecasting values and actual values are divided into four sections by time. The error distribution of combination forecasting method is close to the standard normal distribution. The weights of three single forecasting methods are respectively determined. PV power combination forecasting model based on the error distribution is set up. Three single forecasting values are segmented according the time. After refactoring, the final prediction result is obtained. The result shows that the combination forecast method has higher accuracy than others obviously.Finally, this paper research a kind of nonlinear combination forecasting method based on clustering. Firstly, the three single forecasting values and actual values have been classified by k_means method, SOM method and average distribution time method. Then the BP neural network is adopted to establish the nonlinear model respectively. According to the different methods, the forecasting day has been classified. According to the combination forecasting model, the final prediction results have been got. By means of simulation and analysis, the PV power prediction accuracy of non-linear combination forecast method based on average distribution time is higher than other method。...