Short-term Power Load Forecasting Based On Improved CNN-SVM

With the further development of modern industry and technology,the power industry has been closely linked to people’s daily lives.Among them,power load forecasting is a key technology in the power system,used for safety assessment and timely dispatch of electricity in the power system.Accurate load forecasting can not only serve as a guarantee for the safe,reliable and stable operation of the power system,but also reduce the waste of limited resources,which is beneficial for environmental protection and improves the economic benefits of electricity.Therefore,improving the accuracy of power load forecasting has become an important goal of load forecasting technology.In order to reduce the error of short-term power load forecasting and improve its prediction accuracy,this paper studies a load forecasting model based on a combination of convolutional neural networks(CNN)and support vector machines(SVM).The main method is to extract features through convolutional neural networks,then input the extracted results into the support vector machine model for prediction,and finally output the prediction results to achieve load forecasting.The CNN-SVM model mainly combines the advantages of both CNN and SVM models.CNN can achieve better results in feature extraction,while SVM can better handle problems such as small samples,nonlinearity,and high dimensionality.Considering that power loads are susceptible to various factors such as time and environment,it is necessary to use these factors as feature input variables for load forecasting.In this paper,firstly,random forest(RF)algorithm is used to analyze the importance of multiple characteristic variables that affect the power load.After the characteristic variables are preprocessed,the characteristic variables that have less impact are removed,and the more important characteristics are selected as the input characteristics of the CNN SVM model together with the load data.Then the experiment is compared with the CNN SVM model and SVM model without random forest algorithm.The results show that the proposed RF-CNN-SVM model has better prediction effect.In order to further improve the accuracy of the prediction model,the optimization algorithm is used to optimize the parameters of the CNN-SVM model.At the same time,in order to avoid the limitations of the particle swarm optimization(PSO)algorithm,the simulated annealing(SA)algorithm is introduced to improve and optimize the particle swarm algorithm.Based on the random forest algorithm to optimize the input variables,the penalty factor and core parameters of the SVM in the CNN-SVM model are optimized,and the optimal parameters are given,Better load forecasting.Then,comparative experiments were conducted with the RF-CNN-SVM model and the RF-CNN-PSO-SVM model,and the results showed that the RF-CNN-SAPSO-SVM model had higher prediction accuracy.