Research On Optimization And Control Methods For Process Based On Data-Driven

Data driven method is an effective way to solve the problem of optimization and control for the complex system when rich data is available for data statistics, analysis, evaluation and utilization, which provides feasible optimization strategy. Therefore, it is of great importance in terms of theoretical and practical purpose.Firstly, the basic principle and algorithm of model-free adaptive control is introduced in the dissertation. Next, the algorithm is applied to the process level on the basis of the idea used in the control loop, and we put forward the process optimization and control method which based on model-free adaptive control, in which a series of dynamic linear time-varying models is established to estimate the relationship between the production index and set points of the control system whose parameters (pseudo gradient vectors) are obtained by the online estimation based on the batch index information. During the estimation of pseudo-gradient vectors, iterative learning algorithm is proposed to update those vectors. The simulation results validate the effectiveness of the proposed method. At the same time, the key parameters of process optimization and control are analyzed, and fuzzy rules are proposed to tune those parameters. Compared with original algorithm, the algorithm is more efficient in terms of improving the performance indexes. The rich optimization data generated by adaptive model-free control makes it available for mining the useful rules by the way of the rough set. When the operating condition changes, initially we check whether there is feasible rule that matches the present condition. Once matched, the available rule is taken to optimize the process. Otherwise, adaptive model-free control is utilized, and then the optimization solution is preserved to add to the case database with the rule updating correspondingly. Overall, an algorithm combined with adaptive model-free control and case reasoning is proposed to optimize the process on the basis of rich data. The effectiveness of proposed methods is validated with the simulation of feeding optimization control for alcohol batch fermentation.