Adaptive Iterative Control In Batch Process Based On Integrated Data And Model Driven

Competition of chemical products becomes increasingly sharp in consumer market with the development of economy and society. Batch process was concerned highly because of production property of fine and customization, so reinforcing the level of automation becomes an important research topic issue. For the repeat characteristics of batch process, the key of our research force on the performance improvement of iterative learning control. The new adaptive iterative learning control strategy integrally driven by process data and system model is proposed.Firstly, the unfalsified iterative learning control driven by process data is proposed to solve the control problem of measurable process variable, such as temperature in batch process. The proposed algorithm is sample and fast execution, which is satisfied the control requirement of real-time and fast. The math description of unfalsified iterative learning control and adaptive law of parameters in condition of finite parameters set and infinite parameters is given. Simulation results which obtained from typical batch process show that the proposed control algorithm is effective for temperature control problem with good performance in both convergence speed and stability. Secondly, bi-layer iterative learning control structure was proposed for macroscopic quality control in batch polymerization process. Adaptive terminal iterative learning control based on model-driven was developed for the outer layer of quality control. Adaptive iterative learning control based on data-driven for the inner layer of process variables control was proposed to solve the problem of batch-varying reference temperature. Because of complex mechanism and un-measurable online of product quality, the recursive SISO least square support machine was used to build the model of product quality, which is updated batch-wise. Then an adaptive iterative learning control law is derived based on the optimization, and the convergence is further discussed. The simulations show that the integrated iterative learning control strategy between batch-to-batch and batch-in can effectively improve control performance.Finally, the adaptive iterative learning control for microcosmic quality in batch process, i.e. molecular weight distribution (MWD), is preliminarily studied. Recursive MIMO least square support machine regression algorithm was given for building the update model between temperature/feed ratio and measurable low order moments of MWD. The adaptive iterative learning control algorithm and its convergence are considered. The simulation results in polystyrene process show that the precise tracking to given MWD trajectory can be achieved.