Research On Iterative Optimization Control Methods For Batch Process Against The Model Error

The batch process is characterized by little batch, multiple varieties, serialization, complex synthesis steps, technology intensive and is capable of meeting the demands of modern industry, which has found an increasingly wide application. The demand for production automation and process optimization control with more efficiency and little dissipative has become more and more pressing in the batch production enterprises. So it has great theoretical and practical significance in the optimization control of batch process.The iterative optimum control strategies based on iterative learning control algorithm make full use of the limitation of run-time and the repetitiveness of performing and can get through optimum control problems of systems with strong uncertainty, nonlinearity and strong coupling in a very simple way. The existence of uncertainty and many disturbances in process model become an obstacle in implementing batch-to-batch iterative optimum control strategies based on model. The thesis studied efficient methods for batch process optimum control problems in presence of process model error and even greater error:optimization strategy for batch process considering the effect of plant-model mismatch. In this thesis, the main idea and method of iterative learning control strategy is first introduced and simulation studies illustrate the influence of model error on the performance of iterative learning control. Then the method of iterative learning control containing modification term based on the thinking of gradient approximation (ISOPE) is introduced which closely integrates system optimization and parameter estimation in the two-step approach. In this method, a gradient-modification term is added to the objective function of the optimization problem so that the derivatives of the real process with respect to the controller set-point values are matched exactly with the corresponding derivatives in the model, thus improving quality index. Due to the difficulty in estimating gradients accurately which leads to oscillation, the adjustment of coefficient of controlled variable and terminal condition is proposed and its effectiveness is demonstrated on a alcoholic fermentation process. To solve the plant-model mismatch more effectively, the Gaussian mixture model is used to describe the uncertainty of model prediction. The error GMM can then be exploited to provide the conditional mean and the conditional error variance of model error which are used for outputs compensating and then ISOPE algorithm is combined for optimal control. Its effectiveness is also illustrated by simulation studies.