Research On Short-term Power Load Forecasting Algorithm Based On Machine Learning

In order to realize automatic generation control and economic dispatch control in power grid operation,the first work is to predict the load of power system,and accurate short-term load forecasting is beneficial to the stability of the power system.Therefore,scholars have devoted a lot of research to the short-term power load forecasting,aiming at finding a suitable method to improve the precision of load forecasting.However,with the development of smart grid technology,the scale of power grid monitoring data has increased significantly,the traditional load forecasting model have its limitations.According to this situation,this paper applies the method of machine learning to short-term load forecasting,and the simulation of a certain area is carried out.The specific research content is as follows:1.The relationship between the daily cycle of power load,the relationship between load,and some meteorological factors and the characteristics of the load under different date types are analyzed.On this basis,the characteristics of the influence factors of the load are extracted by the neighborhood rough set theory,so as to avoid the redundant variable making influence on the load forecasting and the training time.2.The least squares support vector machine(LSSVM)model is established.Because of the model can not select the input variables scientifically,the feature extraction results from the neighborhood rough set are introduced.Compared with the conventional LSSVM model and the LSSVM model after neighborhood rough the appropriate influence factors can greatly improve the prediction accuracy.Furthermore,set feature extraction,it is proved that selecting the particle swarm optimization algorithm is used to optimize the kernel width parameters and penalty parameters in the model,and the model is compared before and after optimization.The prediction results show that the precision of the load prediction can be further improved by optimizing the parameters.3.Because the training of large sample data is difficult to implement in the LSSVM model,a BP neural network model is established.A whole year's sample data is used to train the network,but due to the inherent shortcomings of the model,the final prediction effect is not good enough.On this basis,a model of load forecasting model based on the depth confidence network(DBN)network is established.The model is used to extract input variables from multi level restricted Boltzmann machine,and then load is forecasted through top-level BP neural network unit.The simulation results show that the model can effectively improve the accuracy of load forecasting compared with the conventional neural network model.4.The above three models are used to predict the load for a week.The experimental results show that the LSSVM model is suitable for less sample data,while the BP neural network model and the DBN model are suitable for training a large number of samples,of which the DBN model has a higher prediction accuracy and the best generalization performance.