Modeling And Predictive Control For Refrigerating Station Systems Of Public Buildings

At present,the energy consumption of buildings in China accounts for about 21.7% of the total energy consumption of society,of which the energy consumption of air conditioning and refrigeration station systems accounts for about 40% or more.The freezing station system is a complex thermal process with many devices and complex characteristics such as inter-loop coupling,nonlinearity,large hysteresis and time-varying,which makes it difficult to model the mechanism,while the traditional data-driven modeling method is difficult to balance the performance of both offline modeling and online correction,making it difficult to optimize the overall system control.On the other hand,because the refrigeration station system is a complex nonlinear system with multi-loop coupling,the traditional PID control parameters are difficult to adjust,causing the system to oscillate easily under changing conditions,and the PID control cannot take into account the performance of both meeting the cooling capacity demand and energy saving.The research team found that,because the system is prone to oscillation,the actual engineering of the freezing station control system is mostly changed to manual operation after 1~2 cooling seasons of operation,with sloppy management and control,resulting in a great waste of energy.Therefore,it is very necessary to study the modeling and intelligent optimization control strategy of the refrigeration station system in order to achieve the control goal of saving as much energy as possible while meeting the building cooling demand.In order to achieve the best energy efficiency of the whole refrigeration station system under the premise of satisfying the built load requirements,this thesis proposes a predictive control strategy for the energy efficiency ratio of refrigeration station system based on load prediction.The basis of predictive control is predictive model,rolling optimization and feedback correction.Given that the freezing station is a nonlinear system,it is necessary to study the construction of the model and online correction method of the freezing station system with strong anti-interference ability and easy engineering implementation,as well as the rolling optimization algorithm of the nonlinear system with small computation and less storage space occupation.In order to improve the model online correction and model portability performance,this thesis proposes a dynamic modeling method for load and energy efficiency ratio of refrigeration stations based on an improved Takagi-Sugeno(T-S)fuzzy model.Firstly,the structure and parameters of the two models of freezing station load and energy efficiency ratio were determined;then the fuzzy C-mean clustering method was optimized based on particle swarm algorithm to solve the problem that the traditional fuzzy C-mean clustering algorithm requires high selection of initial values during iterative solution and is easy to fall into local optimum,so as to complete the structure identification.On this basis,the extended Kalman filter algorithm is used for online correction of fuzzy model posterior parameters to solve the problem of poor parameter identification caused by noisy field measurement data and to improve modeling efficiency at the same time.The performance of the constructed load and energy efficiency ratio models was verified in this thesis using the operating data of a refrigeration station in a public building in Guangzhou,the subject of the study,and the relative errors of the two models were 2.75% and 2.25%,respectively,meeting the industrial control requirements.On this basis,this thesis uses data from another public building to verify the model portability of the developed model,and the results show that tshe error of the ported model is within ±6%,which meets the industrial control requirements.Due to the nonlinearity of the freezing station system,the optimization solution is difficult,and the Hamilton-Jacobi-Bellman(HJB)(dynamic programming)algorithm is usually used in related fields for the optimization solution of nonlinear systems,but the HJB algorithm is computationally intensive,occupies a lot of storage area,and is not easy to be implemented in engineering.To this end,the research team proposes a nonlinear predictive optimization algorithm based on neural networks,using a three-layer feedforward neural network as the optimal feedback controller,and taking the control system optimization objective function,i.e.,meeting the building load and the optimal system energy efficiency ratio,as the neural network optimization performance index.Based on the Euler-Lagrange algorithm and the stochastic gradient descent method,the team performs rolling optimization of the neural network controller weights to solve the problem of Euler-Lagrange open-loop control which is sensitive to random disturbances and uncertainties,and to avoid the "dimensional disaster" problem of dynamic programming algorithms.This thesis applies the above algorithm to the control of a freezing station system in a public building in Guangzhou,the object of study.The experimental results show that the neural network predictive control strategy proposed in this thesis,compared with PID control,improves the average energy efficiency of the system by about32.5% under the condition of meeting the building cooling demand,and is able to overcome the influence of various changes and uncertainties in the control process,with better dynamic and steady-state performance.The algorithm takes up moderate storage space,is computationally small and easy to implement in engineering.On this basis,given that the system optimization objective is to meet the demand of building load and system energy efficiency ratio,in order to solve the problem of adaptive adjustment of performance index weights in the optimization objective function,this thesis designs a weight adaptive fuzzy controller and adds it to the optimization objective function of the predictive control system.The fuzzy controller queries the fuzzy control table in real time to obtain the objective function according to the time-varying building load and system energy efficiency ratio Optimal weights.The experimental results show that,based on the use of neural network predictive optimization control method,compared with the adaptive fuzzy controller without adding weights,the refrigerating station system prepares the cooling capacity more closely to the cooling load demand,and improves the energy efficiency ratio of the system by about 6.93% under the premise of ensuring the demand cooling capacity.