Cooling Load Forecasting And Control Strategy For Water Storage Systems

With the continuous development of society,the use of air conditioning systems has become more and more widespread.In first-tier cities,the electricity consumption of air conditioning has accounted for more than 25% of all peak electricity consumption.In addition,the consumption of air-conditioning and industrial power both concentrates in the daytime,resulting in increased peak-valley load difference in power systems.The cold storage air-conditioning systems can shift the electric load from daytime to night,which by storing the cold in the tank at night and releasing the cold during the daytime.Government departments also have issued policies to encourage applications cold storage technology in order to achieve the ‘peak load shifting' effect of electricity.However,it is found that current storage air-conditioning systems in Shenzhen generally suffer from poor control which failed to achieve optimal control between the cold storage tank and the chiller under different loads.The water storage air-conditioning systems have the distinguishing features of large delay,nonlinearity and strong coupling,while the traditional PID control is difficult to deal with the complex coupling relationship between variables.Predictive control is essentially the optimal control of the operating state of the system.It can predict the target value at the next time in advance.Through the solution of the system performance optimization indicators,the goal of energy-saving and optimal control of the cold storage system can be achieved,and the predictive control algorithm is applied to the model.On the other hand,accurate prediction of building load is a basic prerequisite for water storage air-conditioning systems.The systems determine the amount of cold stored at night based on the load forecast.Chiller and water tank are optimally distributed according to hourly load,and the system can achieve the goal of energy saving and cold capacity matching.At present,there are two difficulties in the study of predictive control: First,there are many devices in the systems and it is difficult to establish the mechanism model;while the existing HVAC simulation platforms(such as TRNSYS,EnergyPlus,etc.)lack dynamic module of water tank and plate heat exchanger.So it is necessary to develop dynamic module to cooperate with pump,piping and other modules.Second,the predictive control algorithm is complex,and it is difficult to implement in actual project.In addition,the existing predictive control optimization algorithms usually use dynamic programming(HJB)and Euler-Lagrange(EL)when solving nonlinear systems,but both have disadvantages.The HJB equation will easily cause "dimensional disaster" problems.And the EL equation can only solve the optimal control vector sequence in a certain initial state,and there is no feedback function.This paper takes the water storage air-conditioning systems of Shenzhen Guoyin Building as the research object,and carries out load forecasting and control strategy research.This paper first uses the sensors installed in the actual project to collect the actual data of the air-conditioning season,and then uses BP neural network to establish the load forecasting model of the cold storage systems.In order to improve the generalization ability of the network,a modified Bayesian normalization method was used to train the network;and using part of the sample data to test the established model,using the data of one week to test the model and the error only get 1.21%,it indicates that the accuracy of the model can meet the actual requirements,it can be used in practical projects.In this thesis,the neural network predictive control strategy is studied.This strategy combines HJB and EL equations.Artificial neural network is used as the optimal controller to solve the nonlinear MIMO time-varying systems to optimize the feedback control quantity.And using rolling optimization to continuously find the optimal solution.Excellent,this algorithm can optimize multiple performance indicators,also will overcome the influence of coupling and interference signals in time.The advantages of this algorithm are small calculation and effectively overcomes ‘dimensional disaster' problem,and it is easy to implement in actual projects.On the basis of constructing the load forecasting model,this paper uses rapid development function of TRNSYS to build models of water tanks and plate heat exchanger.According to the function of MATLAB which can be invoked by the type155 module of TRNSYS,a joint simulation platform of TRNSYS and MATLAB is built.The neural network predictive control strategy is used to control the water storage air conditioning system and the optimal feedback solution of the cool storage system is solved.The neural network predictive control strategy has been tested on a co-simulation platform.The model is established to control the temperature of the hot end of the plate heat exchanger to achieve the purpose.The experimental results show that using the predictive control strategy adopted in this thesis,the system will control the hot end outlet temperature of the plate heat exchanger at 5.5? after 190 seconds,while the adjustment time of the PID control will be as long as 20 minutes.Through comparative experiments,it can be concluded that the neural network predictive control strategy adopted in this thesis can control chilled water at 5.5? under the condition of load change in the control of chilled water for cold storage air conditioner.And through the control of the cooling capacity of the water tank,it can achieve the function of ‘peak shift valley',and it also can reduce the operating electricity fee and save the energy consumption of the chiller.