Predictive Control, Neural Network-based Crystal Growth

Based on the research of the principle of predictive control, the theory of artificial neural network and genetic algorithm, and according the problem that predictive model is not precise enough, the online optimizing algorithm is lack, and that is not suitable for nonlinear system, this paper proposes predictive control based on neural network. For the neural network similarity to nonlinear function, neural network is used to be predictive model, and use the genetic algorithm for the leaning of neural network's connection weights, and it is not only of quick convergence and high precision, but also it can easily escape local optima and initial weights which problem the BP algorithm is facing. Using the genetic algorithm as the online optimizing algorithm, it escape the problem of account the derivative repeatedly at gradient algorithm, and it predigest the problem. At the simulations of the crystal grouping which is time-varying nonlinear system with time-delay, this predictive control strategy obtains good effect.