Research On Decision-making Method For Intelligent Operation Of Smart Heating System

Driven by clean heating,peak carbon dioxide emissions and carbon neutrality,as well as heating technology transformation,district heating systems（DHSs）in China present the following development trends:diversified heat sources,cross-region interconnected and long-distance heating networks,and increasing requirements of comfort heating of users.The dynamic and complexity of heating operation conditions are significantly enhanced.These changes require DHSs to have robust global coordination and dynamic control abilities.As a result,the traditional decision-making mode of heating production and operation based on domain knowledge and experience has been unable to ensure clean,low-carbon,safe and efficient heating processes.Developing smart heating and realizing intelligent operation of DHSs are of great significance to improve energy utilization efficiency,reduce operation cost and ensure operation safety for DHSs.This dissertation studies modeling and simulation,machine learning and deep learning methods for smart heating,constructs a decision-making support engine for intelligent operation of DHSs,and promotes the transformation from the traditional operation decision-making mode driven by experiences and processes to the intelligent operation decision-making mode driven by mechanism models and data models.Thus,machine intelligence can be used to assist and improve human intelligence to solve key problems in heating processes including online fault detection and diagnosis of sensors,hydraulic and thermal simulation of DHSs,online detection and diagnosis of leakage faults in district heating networks（DHNs）,short-term heating load forecasting,and coordinated optimal scheduling of heat and electricity.Firstly,cheking the working status of sensors in substations and ensure the credibility of their observation data.For each target sensor,prediction models of its observation data are built based on Long Short-Term Memory（LSTM）network.A dynamic thresholding algorithm is proposed.It can adaptively determine the anomaly threshold according to the error vector of model predictions and sensor observations,so as to identify whether the target sensor is in normal condition.In case study,five types of sensors in the primary network of a substation are used for fault detection and diagnosis including flow,supply pressure,return pressure,supply temperature,and return temperature.For each sensor,three prediction models with different structures are developed.The performance of each prediction model and its impact on the accuracy of sensor fault detection and diagnosis are analyzed and evaluated.Moreover,taking the fixed threshold algorithm as the benchmark,the superiority of dynamic thresholding algorithm is verified.Next,the modeling and simulation methods of hydraulic and thermal conditions for DHSs are studied.For the hydraulic condition,the response time of hydraulic parameters is usually much shorter than the period of hydraulic regulation.Based on graph theory,a DHS can be simplified into a directed graph,which contains two basic components,i.e.,pipes and nodes.Using the node flow balance equations and loop pressure balance equations,a steady-state hydraulic simulation model is developed,and the numerical calculation method of the model is given.A pipe impedance identification algorithm based on genetic algorithm is utlilized to calibrate the pipe impedances of the hydraulic simulation model.In terms of the thermal condition,because of the delay and attenuation of heat distribution,a thermal transient simulation model including heat sources,pipe networks,substations and equivalent heating buildings is established according to the law of energy conservation.Based on the finite difference method,the numerical calculation method of the model is proposed.Using observation data of the primary sides of substations in a DHS,the accuracy and effectiveness of the proposed modeling and simulation methods are verified.Then,with the function that the heating Internet of Things can monitor flow and pressure prarameters of heat sources and substations in real time,a data-model fusion driven method is proposed.The data-model fusion driven method uses the steady-state hydraulic simulation model to generate a leakage fault data set of the DHN,and then train a machine learning model,which can utilize the tiny changes of flow and pressure parameters to identify the leakage pipe in the DHN online.In case study,the proposed data-model fusion driven method is applied to detect and diagnose leakage faults in a loop DHN.The accuracy and macro-F₁ score of the diagnosis results are85.85%and 0.99786,respectively.The results prove that the data-model fusion driven method can timely and accurately detect and diagnose leakage faults in DHNs online.To predict short-term heating load multi-step ahead,this dissertation adopts three time series forecasting strategies,namely recursive strategy,direct strategy and multi-input multi-output strategy,to develop four prediction models based on LSTM network.The input variables of each prediction model are historical heating load and outdoor temperature time series.In case study,the four models are used to forecast the hourly heating load of a DHS in the next day.The advantages and disadvantages of each forecasting strategies as well as the performance of each prediction model are comprehensively analyzed and compared.Finally,in the context of networked long-distance heating mode with multi-source and multi-energy complementarity,a short-term coordinated optimal scheduling method is proposed for the heat and electricity integrated energy system（HE-IES）composed of DHS and power system to improve its operation flexibility and wind power consumption capacity,and minimize its operation cost.Considering the accessibility and thermal inertia of DHSs,the proposed method can be divided into two decision-making steps,namely,accessibility analysis of DHS and coordinated optimization of heat and electricity.In the first step,the steady-state hydraulic simulation model is used to optimize the hydraulic condition of DHS through the coordinated regulation of pumps and valves.In the second step,according to the operation characteristics of plant units,the load profiles of users,and the thermal transient simulation model of DHS,the coordinated optimization model of heat and electricity is developed.Then,the model can be solved by an optimization solver.In case study,six typical days with different relative heating load ratios are selected,and the short-term coordinated optimal scheduling of a HE-IES is carried out.The results show that,compared with the conventional heat-led operation mode,the proposed method can reduce the wind abandonment rate by 1.18%～6.17%and the operation cost by 0.47%～2.30%.To sum up,this dissertation establishes multi-scene and multi-level mechanism models and data models of DHSs by comprehensively using modeling and simulation,machine learning,and deep learning methods.These two kinds of models complement each other,and constitute an intelligent decision-making system with mutual mapping,timely interaction and efficient cooperation between physical space and information space,which can provide theoretical guidance and technical methods for the intelligent operation of smart heating systems.