Multi-model Identification Of Batch Process And Performance Evaluation Method Based On PCA

In this paper, the main content is applying the PCA to multi-model division of the batch process and the performance evaluation of controller. Products of batch process is often complex and sophisticated, and the entire production process often requires many operation modes to complete, so how to correctly distinguish the transitional mode and stable mode from various modes is very important. After the mode is divided, there are also multiple controller design methods in each mode, so we have to choose the best controller. Then, the rest of the question is performance evaluation for each mode.Firstly, the paper introduces the multivariate statistical methods in industrial processes, which mainly introduces the PCA method. However, PCA can only deal with variables submitting to multivariate normal distribution, when variables does not submit to multivariate normal distribution PCA can’t be used. Then, this chapter proposes a new PCA method which combines with ICA to deal with this situation. The simulation result verifies the correctness of the new method.Secondly, the paper describes the identification and division of multi-mode for batch processes in detail. Division method of the traditional multi-modal batch process can not achieve the soft partition or too complicated so that it is not conducive to online application requirements. Combing with the above problems, this paper put forward a new dividing method of multi-modal batch process based on PCA. First, use the PCA to set K sub-model along the time axis(K is the sampling time), and then transplant the main point of view which is used in subspace similarity to the comparisons of the principal component model, the intermittent process is divided into different stable phase (ie, mode) and transition phase. The simulation results show the effectiveness and practicality of the method.Finally, the content is performance evaluation method of controller. The paper introduces widely used performance evaluation method based on the minimum variance criteria. Minimum variance criterion can provide a practical reference point for the performance of the control system, and it only requires routine operating data of the process and the time delay. However, when there is a strong correlation between the output variables, the evaluation results based on the minimum variance criteria are not ideal. Because the PC A has the characteristics of dimensionality reduction and elimination of the correlation, the paper put forward a new minimum variance criteria based on the PCA. Further on the new criteria, evaluate the performance of multiple controllers in the same system, and select the excellent controller to control. Matlab simulation is used to verify the validity of this evaluation benchmark.