Research On Heat Load Prediction Method Based On Wavelet Neural Network And Support Vector Machine

The energy consumption of heating in northern winter is huge.Adopting district heating instead of decentralized heating can effectively improve the efficiency of heating and reduce the waste of resources.However,unreasonable heating plan often cannot meet the needs of the users or causes excess heat dissipation.The accurate heat load prediction can greatly reduce the situation mentioned above,and is very helpful for the balance of supply and demand in district heating.MATLAB software is used to to establish wavelet neural network and support vector machine heat load forecasting model respectively in this paper.The historical weather data and load data of a thermal power plant in Jilin city are taken as the input parameters of the model.The prediction accuracy,prediction speed and generalization ability of different heat load forecasting models are compared.The influence factors including outdoor temperature,wind speed,solar radiation,heat load at the previous one time point and two time point are taken as the input parameters of the model after preliminary screening and correlation analysis.In order to solve the problem that networks convergence is slow or even not-converged due to the random initial parameters in traditional wavelet neural networks(WNN),the genetic algorithm(GA)with fast convergence ability is used to optimize the network structure and initial parameters of heat load prediction model.The results show that when the improved WNN is applied to forecast district heat load,the accuracy and stability of the model are improved remarkably.The simple support vector machine(SVM)model with RBF kernel function has strong local search ability,but its global search ability is poor.In this paper,the grid search algorithm(GS),particle swarm optimization algorithm(PSO)and GA algorithm with strong global search ability are used to optimize the model respectively.The results show that the prediction accuracy and generalization ability of the optimized SVM prediction model are improved,among which the most improved is GA-SVM model.In addition to studying the influence which the change of the prediction model structure on the prediction ability,the change of the input parameter data type isalso taken as the independent variable to study its influence on the prediction ability.The results show that the prediction results of hourly load data are more accurate than those of daily load data under the same model structure.