Exploration And Simulation of Neural Network PID In Temperature Control System Abstract ：This paper presents a new kind of intelligence PID control method on BP neural network and some of basic concepts about BP neural network . Neural network intelligence PID controller has many advanced properties compared with traditional PID controller. The BP neural network PID control method is applied to temperature control system in industry field. The simulation results show that the control method has high control accuracy ,strong adaptation and excellent control results. Key words ：Neural network , PID controller , Temperature control system1 ForewordIn industrial process control, PID control is a basic control method, its robustness, simple structure, easy to implement, but the conventional PID control also has its own disadvantage, because the parameters of conventional PID controller is based on being mathematical model of controlled object identified, when the mathematical model of the object are changing, non-linear time, PID parameters is not easy in accordance with its actual situation and make adjustments, the impact of the quality control so that the control of the quality control system decline. Especially in the pure time-delay characteristics with the industrial process, the conventional PID control more difficult to meet the requirements of the control accuracy. Because of neural networks with self-organization, self-learning, adaptive capacity, In this paper, based on BP neural network PID controller, so that artificial neural network PID control with the traditional combination of each other and jointly improve quality control and to the method in the temperature control system using the simulation language Matlab application.2 BP neural network model and algorithm constitute2.1 BP neural network model constituteBP neural network learning process constituted mainly by two stages: The first phase (forward propagation), the input signal through the input layer, hidden layer after layer-by-layer treatment, in the output layer is calculated for each neuron the actual output value. The second stage (the process of error back-propagation), if not in the output layer the desired output value, the actual layer-by-layer recursive output and desired output of the margin, and the right to adjust the basis of this error factor.2.2 The neural network PID controller structure and algorithmIn the traditional PID control, classical incremental PID control forms:u(k)=u(k-1)+e(k)-e(k-1)+e(k)+e(k)-2e(k-1)+e(k-2) K: proportional coefficient =: Integral coefficient : Differential coefficientSet up BP neural network PID controller structure:PlantPIDNNr(k) e(k) u(k) y(k) +Arithmetic _ y(k)Adaptive in order to achieveof the purpose, the output layer for the three neurons, corresponding to. Input layer, hidden layer neurons, the number of charged objects in accordance with the complexity of fixed. Hidden layer activation function used for the positive and negative symmetrical sigmoid function :Output layer activation function of the use of non-negative sigmoid function:We assume that ， is the output of output layer, which correspond to，. We take the performance index function as follows: When the actual output and the deviation between the desired output, then the error back-propagation. Reverse the spread of the substance is by adjusting the weights so that the smallest deviation, it can use the steepest descent method, error function by a negative gradient direction to all levels of neuron weights to adjust or amend. Then have:= -: Learning rate : Momentum of Available by the chain rule:= =-e(k+1) One: = 1, 2, 3 So BP neural network can be the output layer weights of the calculation formula: Of which:Because of the PID control algorithm in normal circumstances are unknown, can be used to replace function symbols, and through adjustments to correct errors. Empathy can be hidden layer weight coefficient calculation formula: Of which: In the above various types, the S corner (1), (2), (3) express, respectively, input layer, hidden layer, output layer, : The number of output layer neurons : The number of hidden layer neurons : The number of input layer neurons Based on the above can be BP neural network control algorithms: (1) determine the neural network architecture, initialized weights on each floor. Control the volume of output, error check the initial value 0. (2) of the sampling system has been 、. Calculated by the error . And under the incremental PID algorithm to the error component input layer as input. (3) According to al