Application And Research On Informer Of District Heating System Load Forecasting

District heating system(DHS)is a basic social service.Heating load forecasting is one of the critical steps of DHS,which plays an important role in ensuring the effective production,distribution and rational utilization of energy.In order to discuss the performance of Informer,a new time series prediction model,in the district heating system,based on the historical operation data and relevant weather data of a district heating system in Tianjin,this thesis conducts research from three aspects:model application,positonal encoding,data optimization and energy-saving prediction.The specific contents of the thesis include:Firstly,Informer is applied to the field of heating load forecasting.The historical heat load and weather data of DHS are regarded as the input characteristics,and other four commonly-used prediction models are established for comparison.The experimental results show that Informer has the smooth and stable prediction curve,and obtained the best prediction result.Secondly,according to the importance of position coding in time series prediction task,several common position encoding methods are explained and analyzed.These position codes are introduced into Informer respectively.The experimental results show that the XL position encoding can improve the performance of Informer.Thirdly,a data optimization and energy-saving prediction strategy based on similar hour(SH)method is proposed.Light GBM(Light Gradient Boost Machine)and Euclid norm(EN)are used to select the SH dataset.In experiment 1,the data feature dimension was expanded to improve the learning ability of the models.In experiment 2,the minimum load selected by SH is used as the prediction target to achieve energy-saving prediction.The experimental results show that(a)the hybrid model based on SH can further improve the prediction performance;(b)SH＿Informer achieves the highest energy-saving rate,reaching 11.09%,10.05% and 10.36% in the prediction length of 24,48 and 168 h,which demonstrate the feasibility of the new prediction strategy.Finally,the research contents and achievements of this thesis are summarized,and the future development of district heating load forecasting and the application of the model are prospected.