Short Term Load Forecasting Of Multiple Time Scale Power System Based On Machine Learning

The basic requirements of power system operation are reliability,good power quality,economy and flexibility of power network operation and dispatching.In order to meet the requirements of power system operation economy and power network scheduling flexibility,the short-term load forecasting of power system is particularly important.Scholars at home and abroad have done a lot of research on power system load forecasting.According to the literature research,there is no generally accepted research method in the field of power system load short-term forecasting,which can be applied to all practical backgrounds.In this paper,machine learning based short-term load forecasting of multiple time scale power system is proposed.The specific research contents are as follows:(1)The load history data of power system is preprocessed.The complex Morlet wavelet is used to preprocess the multiple time scale analysis and normalization.Time series data have different periodic characteristics under different scale frequency.In this paper,the appropriate period is selected according to the data size and other factors.In this paper,the original power system load time series data is divided into several sections,and a section of data is selected as the model input.Wavelet processing can extract the most accurate periodic characteristics of the data,so as to lay a good foundation for the model training of power system load forecasting.(2)The support vector machine model based on grid optimization is used to predict the power load.By using the grid optimization algorithm to optimize the kernel width parameters and penalty parameters of the model,the grid optimization algorithm is an exhaustive search method.The optimal learning algorithm is obtained by optimizing the parameters of the estimated function through cross validation.Experimental results show that the model is suitable for data training and prediction of small sample power system.(3)RBF neural network is used to forecast the power load.Before RBF networktraining,input vector p and target vector t need to be given.The purpose of training is to obtain the weight W1 and threshold B1 between the first and second layers,and the weight W2 and threshold B2 between the second and third layers.When creating RBF network,the number of hidden layers is automatically selected.Experimental results show that compared with the general neural network model,the model effectively improves the prediction results and training speed.It is suitable for small sample power system data,and the speed requirement is not particularly high.(4)In order to overcome the disadvantage that SVM is more suitable for small samples,this paper uses BP neural network for power system load forecasting.Therefore,the BP neural network model is constructed as a comparison.When 446 *24 data set is used in the experiment,the BP model runs slowly,the forecasting accuracy is good,and the load data volume of the power system is large,which has no requirement for the forecasting speed Load forecasting site.(5)Width learning framework has a significant advantage in prediction speed.In this paper,width learning algorithm is constructed and compared with SVM experiment.The results show that the experimental speed is the fastest and the accuracy is high,which is suitable for the field of short-term prediction of real-time power system with small amount of data.Through the validation experiment on the European Intelligent Technology Network(EUNITE)competition load forecasting sample data set,it is proved that the four machine learning based load forecasting methods adopted in this paper have the characteristics of accuracy and rapidity in the application field of power system load forecasting.The research content of this paper can provide a good theoretical basis and practical application significance for the actual power grid scheduling and operation.