Short-term Heating Load Combination Forecasting Based On BP Neural Network

At present,most of the central heating systems are heated by experience,and the production mode of "on-demand heating" has not been realized.Short-term heat load forecasting at the heat source is a prerequisite for heating companies to achieve "on-demand heating".The heat load forecast is derived from historical data and past experience.Although the operational data of the heating site is stored in the industrial information platform,it has not been effectively exploited.Since it is difficult to achieve satisfactory prediction accuracy by using a single prediction method to establish a thermal load prediction model,combined prediction can make full use of the advantages of a single method to achieve higher prediction accuracy.Therefore,this paper uses combined prediction for heat source to conduct short-term heat load research,which is a positive and beneficial exploration for the realization of smart heating.Firstly,the data is processed efficiently from three aspects.Because the thermal load is affected by many factors and the data is complicated,Random Forest(RF)is used to screen out the factors with high degree of influence as the input variables of the prediction model.In order to ensure the reliability of data,K-means clustering method is used to identify abnormal data.The missing values are filled by average,Lagrange,and Support Vector Regression(SVR),respectively,in different cases.Fuzzy C-means clustering(FCM)is used to select similar days based on meteorological and date factors.The experimental results show that the accuracy of the extreme learning machine(ELM)and S VR prediction model is improved after similar days selected.Secondly,based on data processing,a short-term thermal load combination forecasting method based on BP neural network is proposed.Using the BP network self-learning adjustment weights,the SVR that solves the small sample nonlinear problem is nonlinearly combined with the extremely fault-tolerant and simple operation of the Extreme Learning Machine(ELM).Since the prediction accuracy of SVR depends on internal parameters,the particle swarm optimization(PSO)algorithm is used to find the optimal parameters of the SVR model.The experimental results show that the combined prediction model can combine the advantages of a single model.The average absolute percentage error based on the BP neural network combination model is 1.98%,which is 19.5%higher than the PSO-SVR prediction accuracy,19.8%higher than the ELM model prediction accuracy,and 18.5%higher than the linear combination model based on information entropy..Finally,the BP neural network combined forecasting method can meet the needs of heating enterprises,provide guidance for heating production management and scheduling,and achieve more scientific and refined management.