This paper presents a neural network direct self-tuning PID controller. Its main feature is that the control structure no longer contains an independent PID controller, but integrates the neural network and the PID control law. The learning algorithm of the neural network controller and the stability analysis of the control system are given. The simulation results show that the control system has strong adaptability and robustness. PID control is one of the earliest developed control strategies. Its algorithm is simple, robust and reliable, and it is widely used in industrial process control.
However, with the continuous development of science and technology, the controlled object becomes more and more complex. The actual industrial production process is often non-linear and time-varying.
There are many uncertainties.
The parameters and environment of the object often change with time, and various uncertain disturbances will also affect the control effect. At present, people are demanding higher and higher control quality, and the defects of traditional P1D control are gradually exposed. It is mainly manifested in the control of the unsuitable uncertain system, the control of the unsuitable non-linear system, the control of the unsuitable time-varying system and the control of the unsuitable multi-variable system.
In recent years, the rapid development of artificial neural networks has attracted people’s attention.
It can approximate arbitrary continuous nonlinear functions with arbitrary precision, and has the ability of adaptive and self learning for complex uncertain problems. It can handle processes that are difficult to model and rule, and has strong information integration ability. According to its characteristics, many PID control algorithms based on neural network are proposed, but the contradiction between the rapidity and overshoot of “linear combination” can not be overcome.
In this paper, a direct self-tuning PID controller based on neural network is proposed. The characteristic of this controller is to put the proportion, integral, differential operation of error signal and the adaptive tuning of PID parameters into a feedforward neural network. Experiments show that it has good adaptive and self learning functions, and is of great significance for improving the control effectiveness and robustness of the control system with strong nonlinearity and fast time-varying performance. The neural network direct self-tuning PID control system is shown in Figure 1. This paper takes the network structure of Figure 2 as an example to illustrate its working principle. The input of two nodes in the input layer is the given value and the output value of the controlled object. The hidden three nodes are used to realize the proportion, integral and differential operation of the error signal. The output layer completes the synthesis of the non-linear PID control law. The P, I and D coefficients are reflected by the weights of the network.
The weights of the network are adaptively corrected by the system errors according to the selected rules. The adaptive learning of the weights from the input layer to the hidden layer is used to filter the output of the system with random interference, and the adaptive adjustment of the weights from the hidden layer to the output layer is used to find the appropriate PID parameters. The weights from the first unit of the input layer to the first unit of the hidden layer are set as follows: 1 corresponding in the input layer unit, = 2 corresponding; the weights from the first unit of the hidden layer to the unit of the output layer; the total input of the first neuron of the hidden layer, and the total output of the first neuron of the hidden layer. The neural network direct self-tuning PID control system presented in this paper is used for simulation. The initial values of the weights coefficients of each layer of the network are: equal 1, equal – 1, random numbers on intervals [- 0.5, 0.5] and learning rate. The reference input signal takes square wave with a period of 100 s. The tracking curve of the system is shown in Figure 3. As can be seen from Figure 3, thermostatic element the output of the simulation object can well track the input of the system. From the design method and simulation results, it can be seen that the neural network direct self-tuning PID control system proposed in this paper has the advantages of simple structure, easy realization and clear physical meaning of the conventional PID controller. It also has the ability of self-learning and self adaptation of neural network.
It avoids manual adjustment, improves the stability of the system, and has strong adaptability and robustness. With the further research, this control method will have a broad application prospect in the field of industrial control.