Research On Energy Saving Prediction And Intelligent Regulation Method Of District Heating System

District heating is the main heating method in northern cities and towns in China,and it guarantees the winter heating demand of various building complexes.Accurate heat load prediction is a prerequisite for feed-forward control and on-demand heating in district heating systems(DHS).However,considering that the historical data used to train prediction models are often not optimal or most energy efficient,accurate prediction is difficult to achieve effective energy conservation.On the other hand,the current level of intelligent regulation and control of DHS is low,making it difficult to further improve energy utilization efficiency.This thesis takes the measured data of a heat exchange station in Tianjin as the research object,and uses machine learning and deep reinforcement learning related methods to analyze and study the DHS from the aspects of energy saving prediction of heat load and intelligent control of water supply temperature.In this thesis,we first propose a similar sample evaluation criterion based on weighted Euclidean Distance(WED)to assess the similarity between different samples,and use e Xtreme Gradient Boosting(XGBoost)algorithm to calculate the weighting coefficients of different features.In order to further reduce the energy consumption,the average heat load is calculated from the training samples and similar samples and used as the training label of the prediction model.In addition,the innovative use of low-pass filter(LPF)to filter the heat load prediction results,which obtains a smoother prediction curve and can ensure the smooth operation of the heat network to a certain extent.The experimental results show that the proposed hybrid model can improve the energy-saving rate by about 5.1 percent.On the other hand,considering the limitations of single model in prediction performance,a prediction model pool based on six classical machine learning is constructed,and a deterministic policy gradient algorithm(DDPG)is innovatively used to determine the optimal fusion weights of each model.Through the weighted summation of the prediction results of each model in the model pool,the possible contingencies of the single prediction model in the prediction process can be improved,which further enhances the stability and robustness of the model in the prediction process,thus improving the overall prediction accuracy of the model.The experimental results show that the ensemble optimization strategy can improve the root mean square error by about 2.17 percent.Finally,the intelligent regulation method of water supply temperature based on Proximal Policy Optimization(PPO)is proposed to address the problems of low intelligence level and energy utilization efficiency of the DHS in the regulation process.The simulation environment of the DHS is built using Python and Simulink,and through interactive training with the environment,the agent can feedback a reasonable secondary network water supply temperature in response to outdoor temperature changes,thus ensuring that the indoor temperature is maintained within a reasonable range.The experimental results show that the simulation model can make reasonable responses according to the changes of outdoor temperature and secondary network water supply temperature,and the PPO algorithm can provide stable and reasonable water supply temperature for the changes of outdoor temperature in one interaction,and the convergence speed is faster than the traditional regulation algorithm,which has good potential for application.