2D-PID Adaptive Iterative Learning Control Method For Batch Processes

Development of efficient control methods for process equipments is of great importance. However, the proportional integral derivative(PID) controller with non-adaptive parameters is widely applied to process equipment unit. The batch operation with wide range and strong flexibility is more significant to meet many varieties of small volume product. Additionally, comparing with the continuous process, batch processes often show dynamic, time-varying and nonlinear characteristics which make them difficult to control.The improved research focuses on the PID controller of the fermentation process and the stirred tank reactor, is from two aspects of within-batch and batch-to-batch. After an overview of the status of batch processing control methods, the auto-tuning neuron PID(ANPID) controller is introduced to the application of batch processes combining with the particle swarm optimization(PSO) algorithm. Moreover, considering the repetitive nature, an adaptive control method using the two-dimensional PID(2D-PID) iterative learning control(ILC) is proposed for batch processes.The major contributions of this dissertation are given as follows:(1) In order to satisfy the practical applications, ANPID with simple structure is applied to nonlinear batch processes with uncertainties. The parameters of PID are first optimized using the PSO algorithm in a quick manner. The obtained result on a simulated fermentation process and a stirred tank reactor show that the proposed PSO-ANPID is suitable for nonlinear batch processes with time-varying parameters.(2) Considering repetitive nature of batch processes, an adaptive 2D-PID control approach is proposed. First, in order to study the control of steady expectation processes between batches, the PID control is integrated with ILC to design a PSO-based 2D-PID control structure. Then, an ANPID controller is further adopted to adaptively tune the parameters within the batch operation. Furthermore, the PID-type ILC is used to capture the useful information in historical processes. Consequently, the verified simulation results on a fermentation process and a stirred tank reactor show that the control performance can be gradually improved from batch to batch.