| Central heating is a basic undertaking related to people’s livelihood.In recent years,with the development of economy and technology and the requirements of green and low-carbon policies,it has been developed rapidly in northern China.However,the relative slow development on operation management and control technology results in many problems,such as the mismatch between the heat supplied from heat sources and the heat required by users,the poor thermal comfort,and the lower energy utilization rate of the system.Smart heating is the only way to achieve on-demand heating at the heat source,improve the thermal comfort of user and the energy efficiency of heating system.Heat load forecasting is the basic requirement for realizing smart heating.The heat exchange station is in the central link of the central heating system,and the accuracy of heat load forecasting directly affects the thermal comfort of users and the accurate implement of heat supply from heat sources.The heat load forecasting model of the heat exchange station and its influencing factors were researched by this paper.The main contents include:(1)The principles and characteristics of several prediction models of neural networks were compared and analyzed,and a convolutional neural network was proposed to build the heat load forecasting model for heat exchange stations.Using multiple hidden layers to perform nonlinear transformation on the original input variables,with its weight sharing and local connection characteristics,The problem that shallow neural networks are easy to fall into local minima was effectively solved.Moreover,The hyperparameters in the convolutional prediction model are optimized by Bayes,and the convolutional neural network prediction model with the best combination of hyperparameters is determined,the prediction accuracy and generalization ability of the model were furtherly improved.(2)Considering that among the many factors affecting the heat load,there is a lack of self-adjusting factors,the change law of the heat load cannot be fully reflected.Taking the heating system of the residential area in Handan City,Hebei Province as an example,The heating system was constructed by using MATLAB/Simulink,and its operation was simulated under conditions of adjusting the room temperature.The results provide a multi-dimensional influence factor for the research on the heat load forecasting of the heat exchange station.(3)The factors affecting the heat load were divided into three dimensions:meteorological factors(outdoor temperature),heating system factors(the circulating water flow of the primary network,the circulating water flow of the secondary network,the water supply temperature of the secondary network,the return of the secondary network Water temperature,heat load value of the previous three days,heat load value of the previous two days,heat load value of the previous day)and random factors(building use intensity).The correlation analysis was carried out on the nine influencing factors,and the results show that the influencing factors have obvious correlation with heat load.In order to reducing the information overlap between the influencing factors,the principal component analysis method was used to obtain the new variables,which can effectively reduce the coupling of the original data and the information loss as much as possible,and these provide a basic guarantee for the accuracy of the heat load forecasting model.(4)The proposed heat load forecasting model,Bayesian optimizing convolutional neural network,was used in the heat exchange station,the processed data set were adopted as the input variables of the model,and the prediction results was compared and analyzed with the prediction results from the unoptimized convolutional neural network prediction model.the prediction model proposed is more reasonable than others in prediction accuracy and generalization ability;moreover,the superiority of the proposed model were furtherly demonstrated through comparing with the inverse neural network,radial basis neural network and long short-term memory neural network prediction models.Finally,this model was used to predict the heat load of different time lengths,which shows the universality of the model,and provides the basis for the energy saving of the heating system,the improvement of thermal comfort of user and the realization of smart heating. |
