Neural Network Modeling And Control Of Lithium Bromide Chiller

The current status of China’s energy is becoming more and more serious,and building energy consumption is high.Environmental problems caused by the consumption of fossil fuels have also become increasingly serious.Absorption refrigeration system to meet the needs of society,but there are still many problems to be solved.This paper starting from the single-effect lithium bromide absorption refrigeration system,the modeling and control of the chiller are mainly studied.In this paper,it summarizes the characteristics and development of the absorption chiller in recent years and studies the operating mechanism of single-effect LiBr absorption chiller.Because of the uncertainty of the internal parameters and the strong non-linearity of the absorption chiller,the artificial neural network identification method is chosen to establish the unit model.By studying the cooling principle of the absorption chiller,several external conditions that affect the output temperature of the chilled water are analyzed,and the input and output variables are selected to establish the model.Using hot water type single-effect lithium bromide chiller platform,it collects a large number of valid experimental data.Neural network is a kind of mathematical model that simulates human brain thinking mode.It has the characteristics of nonlinear,self-learning and powerful mapping ability,which is suitable for the identification of nonlinear models.Among them,BP network is one of the most commonly used recognition neural network.In this paper,the advantages and disadvantages of BP neural network are summarized and an optimization strategy is proposed.Using the LM method as the learning algorithm of error back propagation,it reduces the probability of the network falling into the local minimum,and increases the stability and the convergence speed.Meanwhile,the rapid pruning method is used to determine the number of hidden layer nodes in BP network,which can improve the modeling efficiency.The optimized network is used to identify the experimental data of the absorption chiller and establish a model with "5-6-2" network structure.Because of the characteristics of nonlinearity,large inertia and susceptible toexternal disturbances,an improved neural network controller is proposed to control the absorption chiller in this paper.Firstly,it improves the single neuron PID controller,making its parameters online adjustment.Subsequently,this single neuron PID controller is used to replace the traditional PID controller in RBF supervisory control strategy.The controller is used in the temperature control of the LiBr chiller.The chiller can achieves the steady state effect more quickly,and can recover the steady state more quickly when exposed to outside interference,and has better effect.