Group Control And Energy Saving Control Of Refrigeration Station System In Public Buildings

The energy consumption of HVAC(Heating,Ventilation,and Air Conditioning)systems accounts for about 49% of the total energy consumption in building operation,with the chiller plant system accounting for approximately 80% of the total energy consumption of the HVAC system.Therefore,improving the energy efficiency of the chiller plant system in public buildings has become a research hotspot.According to the research conducted by our project team,the high energy consumption of the chiller plant system is mainly due to its long-term operation in a "high flow rate,low temperature difference" state,resulting in a serious mismatch between the cooling capacity and the end load demand,which is caused by poor control strategies.The chiller plant system has many devices and exhibits nonlinear,time-varying,and high inertia characteristics,which pose challenges to the design and implementation of control systems.Currently,most chiller plant systems mainly use PID(Proportional-Integral-Derivative)control methods to regulate individual equipment loops.However,due to the nonlinearity of the system,difficulties in PID parameter tuning result in poor dynamic characteristics of the system and make it difficult to improve the overall energy efficiency of the system.To address the aforementioned issues,this paper proposes a group control optimization predictive control strategy for the chiller plant system.The group control strategy is designed to tackle the coordination and interdependence issues among multiple devices in the system,using association rule mining algorithm to achieve optimized scheduling and control of each device.The predictive control strategy is based on the prediction of cooling load demand,and utilizes rolling optimization to ensure that the system’s cooling capacity meets the load demand while simultaneously improving system energy efficiency.Firstly,to achieve optimal configuration of each device,a large amount of historical data from the chiller plant system is processed and analyzed.A multi-level association rule mining algorithm is applied to mine the association rules between the operation status of each chiller unit and its highest coefficient of performance(COP)level under different operating conditions,including outdoor weather parameters and load demand.This enables the determination of optimal operating schemes for each chiller unit under different conditions,thereby achieving optimized scheduling for the entire system.Next,the Energy Efficiency Ratio(EER)of the chiller plant system is introduced as the evaluation metric for system energy efficiency.Prediction models for "EER-cooling capacity" are built using neural networks,considering different operation statuses of the chiller units.To achieve matching between cooling capacity and load demand,and taking into consideration the influence of multiple factors on load demand,a stacking-based multi-model fusion algorithm is designed to construct the load demand prediction model.Based on these models,a predictive control framework for the chiller plant system is designed,with the optimization of cooling capacity meeting load demand and improving system energy efficiency as the objective function of the predictive control.The dynamic characteristics of the chiller plant system,including inertia,response speed,and time-varying characteristics,are fully considered in determining the appropriate prediction time domain.To address the difficulty of nonlinear optimization,a multi-layer feedforward neural network is used as the optimization feedback controller in this paper.The input and output variables of the neural network controller are determined based on the control objectives of the chiller plant system,and the network structure of the controller is designed accordingly.A combined algorithm of variational method and stochastic gradient descent is used to optimize the weights and thresholds of the controller in real-time,enabling predictive control of complex nonlinear systems with low computational complexity and moderate storage space usage.The experimental results showed that compared to the original system,the implementation of the group control strategy resulted in an average increase of 5.69% in system Energy Efficiency Ratio(EER).The "EER-cooling capacity" prediction model constructed in this study had a relative error of less than 2%,and the relative error of the load prediction model was less than 2.5%.Based on the extensive practical experience of the research team,the accuracy of the models is suitable for engineering applications.Furthermore,the group control optimization and predictive control strategy proposed in this paper,compared to the original system control scheme,achieved an average improvement of21.35% in system energy efficiency,while meeting the cooling load demand.