Neural Networks Decoupling Control And Research Based On Immune Algorithm

In present practical industry process, most are nonlinear, multivariant and strong coupling plants, which math model often is difficult to get, so the utilization of the traditional decoupling control methods in its control is limited. The intelligent decoupling control, which is represented by the neural networks decoupling control, can efficiently resolve the control problem of the complex industry process because it don't need the math model of the plants. But because the present neural networks decoupling control algorithms are still imperfect not only in the aspect of algorithm but also theory, so the neural networks decoupling algorithms are studied here. Besides, when the additional optimization algorithms train the neural networks, which exist some shortcoming such as low precision, long convergence time and bigger possibility of getting into local extremum. For overcoming these shortcomings, the immune algorithms are used to train neural networks. But because the immune algorithms develop late and are still immature not only in the aspect of theory but also utilization, so the immune algorithms are studied here.Main research works in this paper are as follows:1. Aiming at the nonlinear, multivariant and strong coupling plant, a PID decoupling control strategy based on the diagonal recurrent neural network is designed on the basis of the PID multivariant decoupling control strategy. Thereinto, the diagonal recurrent neural network is used as the identifier of plant. a PID controller is constructed based on the diagonal recurrent neural network and several PID controller are adopted in parallel as the neural network decoupling controller for achieving the function of decoupling and control at the same time. Finally, the convergence characteristic of the identifier and the stability characteristic of all the control system are analyzed based on the Lyapunov stable theory.2. On the basis of analyzing the characteristic of the basic immune algorithm and utilizing the ergodic property of the chaotic sequence, an adaptive immune evolutionary algorithm combined with chaotic sequence is proposed and used at the training of the neural networks. Finally, utilizing the method of probability analysis, the algorithm is proved to be convergent.