For batch industrial intelligent control of the process, this paper proposes a data-driven batch process control scheme (KICR-ILC) based on kernel independent component regression, the control strategy are two technical problems need to be solved, kernel independent component regression (KICR) modeling and batch iterative learning control (ILC) algorithm designing.Based on kernel independent component regression independent component analysis (KICA), a nonlinear multivariate regression method, kernel independent component regression (KICR), is proposed for nonlinear statistical regression prediction. In order to eliminate the nonlinear correlation, KICA was firstly performed on the nonlinear input data for preprocessing to get the kernel-independent component(KIC), then multiple least-squares regression (MLR) was performed using the extracted KIC in place of original data for a calibration targets. The numerical simulation results of a nonlinear-equations model shows superior performance than the vogue fast independent component regression (Fast-ICR) for nonlinear modeling.Then,KICR model can be converted to a formula related with the output in the form of an offline-train corresponding Lagrangian function and the observation input of the control variables kernel matrix, design the corresponding iterative learning control (ILC) algorithm via discreting the control variables, the optimization using quadratic performance function as control index, the quadratic programming method to get optimal solution of control variables, the error between batches eliminated by multiple iterations. This control scheme can avoid the accurate model to calculate, it has small samples and strong robustness, effectively make up for the traditional intelligent control scheme in data modeling control deficiency of non-gaussian distribution. Batch reactor case of numerical simulation shows superior performance than classical support vector machine-iterative learning control (SVM-ILC) scheme. |