Short-term Load Forecasting Based On Data Mining

Load forecasting is an important basis for formulating the power generation plan,and is also an important work to ensure the normal operation of the power system.Early in the middle and late of the last century,the study of load forecasting began.Statistics was first applied to load forecasting,and the most classical method is regression analysis.With the development of computer technology,artificial neural network was born,which is a commonly used load forecasting method.In recent years,with the explosive growth of data,data mining technology has been applied to load forecasting.Data mining is a process of discovering hidden rules from a large number of data.Its commonly used means include statistics,information retrieval and neural networks.In this manuscript,the short-term load of a coastal city in Hebei is taken as the prediction object,and the prediction of the 24 hour load value in the future is carried out.Firstly,the characteristic variables of short-term load forecasting are selected and sample data are preprocessed.Then,on the basis of load forecasting based on traditional neural network and principal component analysis neural network,an adaptive small wave RNN method based on principal component analysis is proposed,and the short-term load forecasting is completed.The experiment was carried out and the performance was compared.The main work of this article is as follows:Considering the diversity of the load factors,the 24 hour load,the maximum minimum temperature,the weather condition and the date type are used as the characteristic variables of the prediction.In order to avoid the impact of error data on the prediction results,the sample data are preprocessed: to inform the clustering and filter the error data and then use the linear fitting to repair the data.In order to eliminate the dimensional difference between different types of data,all data is normalized.In order to compare the prediction effect of traditional neural network,BP neural network and wavelet neural network are used to predict the load,and the selected feature variables are used as input data.The experimental results show that the wavelet neural network is superior to the BP neural network.In view of the large number of feature variables and possible correlations,principal component analysis is used for sample data.According to the principle of cumulative contribution rate greater than 95%,new feature variables are selected.The prediction effect of BP neural network and wavelet neural network have both improved after the use of principal component analysis.In order to improve the performance of the wavelet neural network,the cyclic structure is introduced into the wavelet RNN.Then the principal component analysis and the wavelet RNN are combined to construct an adaptive wavelet RNN based on the principal component analysis.The adaptability of the method is reflected in the automatic component analysis and selection of feature variables.The experiment shows that the prediction performance of this method is the best.In this manuscript,a short-term load forecasting software is designed.The software has a good user interface,and the load prediction is carried out by calling the script files of MATLAB.The prediction results can be finally displayed in the user interface.