Predictive Control Of Grinding Process Based On Improved Particle Swarm Optimization

Grinding is an important process of mineral industry, the quality of grinding mill process productions will directly affect the process indicators and the economic benefits of grinding enterprises. Grinding is a complex system with the characteristics of slow producing process, nonlinear, large time delay, multivariable coupling, unstable production status and big disturbances, which make the system very difficult to control and optimize. The optimization control of the grinding process has attracted great attentions of the scholars in recent years. There are few methods considering the constraints of manipulated variables. It is obvious that, large and frequent variations of the manipulated variables are bad for industrial equipment, which Not only waste the energy but also bring hidden dangers.This paper proposes a nonlinear prediction controller based on the least squares support vector machine method. An improved particle swarm optimization algorithm based on Gaussian searching (G-IPSO) is presented to solve the manipulated variables in nonlinear predictive control. The G-IPSO method initializes the particle swarms by using the Gaussian distribution and takes the last value of manipulated variables as the Gaussian center. The location updating of each particle is improved to make each particle moving towards the center of the particle swarms and enhances the searching strength around the center. The G-IPSO method will decrease the changing rates of the manipulated variables. The proposed algorithm is combined with a LS-SVM based predictive control approach and is capable of simplifying the fitness function of the manipulated variables. A typical grinding process is considered and controlled by using the proposed method, and the optimization algorithm is provided in details.To verify the validity of the proposed method, the experiments of a classic multimodal function and the grinding process are conducted by using the proposed nonlinear predictive controller. The experimental results indicate that they exhibit good application performances.