Predictive Control For Central Heating Substation Systems

Centralized heating systems,as the primary heating modality in northern China,necessitate continuous and stable operation throughout the entire heating season.According to statistics,in 2020,China’s centralized heating systems consumed 214 million tons of coal equivalent,accounting for 20%of the nation’s total building energy consumption,with high energy intensity.Consequently,the elevated energy consumption of centralized heating in China is a significant factor influencing the country’s pursuit of carbon neutrality.At present,centralized heating system operation and regulation primarily rely on manual control,with a limited number of systems employing PID control methods.However,manual control is overly dependent on the experience of the operators,and PID control,as a lagging control mechanism,cannot promptly adjust to large-lag systems.Moreover,PID control can only achieve single-loop control,precluding holistic system energy efficiency improvements.Therefore,there is an urgent need to introduce efficient operational control technologies for centralized heating systems to enhance heating quality and conserve energy.Addressing the aforementioned issues,this study focuses on a residential community heating system in Beijing and proposes a heat station system model predictive control strategy aimed at conserving energy consumption of secondary circulation water pumps while satisfying users’heating demands.To implement model predictive control,it is necessary to construct a heat station system prediction model.Considering the cyclical and stochastic nature of the secondary return water temperature in heat stations,this paper designs a two-layer artificial neural network(ANN)prediction model.The feature selection layer employs the XGBoost algorithm to filter input features,deriving a feature set affecting the secondary return water temperature;the load prediction layer utilizes ANN for secondary return water temperature prediction and employs the Grey Wolf Optimizer(GWO)algorithm to optimize the ANN model’s initial weights and thresholds,enhancing its robustness and generalization capabilities.Due to the nonlinear,strongly coupled,and highly lagging characteristics of heat station systems,the original control system’s variable temperature difference and fixed flow rate PID control is ineffective.Based on the aforementioned secondary return water temperature prediction model,this paper introduces a variable temperature difference and flow rate model predictive control strategy.An optimization objective function is designed according to the control target,while the prediction and control domains are determined through a temperature delay identification algorithm.The rolling optimization controller is realized using the Particle Swarm Optimization(PSO)algorithm to obtain the optimal control quantity sequence.On the other hand,fixed-weight optimization objective functions cannot adjust control targets according to operating conditions.Consequently,this paper proposes an adaptive adjustment method for the optimization objective function’s weight,devising a fuzzy logic-based weight coefficient adaptive module through extensive experimentation to further enhance the control strategy’s potential to improve the system’s dynamic performance and conserve energy.The experimental findings reveal that the constructed secondary return water temperature prediction model yields an RMSE of 0.24 and an R~2 of 0.91,outperforming the XGBoost-ANN model with an RMSE enhancement of 20%and an R~2 improvement of 3.4%.Compared to the original system’s PID control method,the proposed model’s predictive control strategy significantly ameliorates both the dynamic and steady-state characteristics,reducing the secondary circulation pump’s energy consumption by approximately 9.3%.When juxtaposed with fixed weight coefficients,the adaptive optimization of objective function weight coefficients further diminishes the energy consumption of the secondary circulation pump by around 5.9%.