The Research Of Multivariable Decoupling Control Method Based On Neural Network

The controlled system is usually multivariable, coupling and time varying in the field of the industry control. The coupling exists in the circuits of multivariable system. That is to say, the change of some input will bring the change of parts of or all the outputs, which will lower the effect of control system. If the couple is serious, the system will not run. In order to get a satisfying effect, it is necessary to decouple the multivariable system.The coupled multivariable system will be changed into the single-variable system that some output will be only response to some input by decoupling control. But the design method of single-variable can not solve the problem of multivariable, strongly coupled, strong nonlinear, and time varying system. How to achieve the decoupling control projects of such complex process control have become hot issues with major significance in the field of industry process control. The decouple control of multivariable system can be divided into classical decouple control, self-adaptive decouple control, intelligent decouple control (including neural network decouple control and fuzzy decouple control, nonlinear and robust control).Decoupling control based on the neural network is one of the major issues in the field of control theory, particularly applied to the problems which the traditional control method can not solve, so it has more research value. The BP algorithm discussed in the document [9] easily traps into local extreme. For the standard genetic algorithm discussed in the document [10], the diversity of population diminishing with evolution, often get sub-optimal solution close to the global optimal solution.For the above reasons a neural network PID decoupling control Method based on an improved genetic algorithm (GA) was presented. The method adopts the improved genetic algorithm (GA) to train the neural network PID. In this new algorithm, a special fitness function and adapted crossover and mutation probability were designed. And an improved immigration method was introduced so that premature convergence can be avoided and population diversity can be maintained. Combined with the temperature control system of the rectifying tower, the algorithm is verified in simulation. The control effect is satisfactory.