Case Study Of Hybrid Prediction Model Based On BP Neural Network

The prediction problem is an important issue in statistics, accurate forecasts is of great significance for the design of policies, programs, and to work out plan ahead of time to prevent the occurrence of adverse circumstances. BP neural network is a kind of artificial intelligence method, which is applied in the field of prediction inclusively since it possesses non-liner mapping, self-learning, self-adaptive, generalization and fault tolerance abilities. In this paper, three hybrid forecasting models are constructed based on BP neural network because of its good characteristics and applied to predict the wind speed in Hainan province, China, the electricity market price in New South Wales, Australia, and the electricity market power load in New South Wales, Australia, which made a good prediction.The first model is a hybrid forecasting model based on outlier detection and fuzzy time series, which is constructed by the part of serial data preprocessing component and part of prediction component, in which BP neural network is used to estimate the fuzzy relationship in binary variable high order fuzzy time series of predict component part. The model is applied to the daily scale prediction of four winds sites wind speed series in Hainan province, China, datasets collected from 2008 to 2012. In case that the basic ARMA model has reached a higher prediction accuracy, the hybrid model still increase wind speed forecasting accuracy significantly, while it is found in case studies that excluding outliers from the wind speed serial data is necessary to improve model accuracy.The main bodies of the second and third hybrid forecasting models are BP neural network model optimized by PSO and BP neural network model optimized by CS respectively. Two models are applied in the study of NSW electricity market price and electricity load forecasting, the instance of the study were half-hour electricity and electricity load data in June 2011. The results show that the hybrid model can improve the prediction accuracy of the price and electricity load to a certain extent, it is also found in specific instances that SVM reached higher accuracy than hybrid model prediction.Through a comprehensive analysis of three case studies, it is found that the hybrid model can improve forecasting accuracy in varying degrees, it is indicated that the hybrid forecasting model based on BP neural network has practical significance in improving the prediction accuracy, it also can be seen that, the prediction model based on BP neural network cannot reach optimal effect in all cases, which requires researchers to analyze according to the specific circumstances, the analysis of specific issues, and then proposed the solution fit to the specific issues, in order to reach better prediction results.