A Multi-batch Fusion Linear Parameter Varying Modeling Method for Batch Process

Due to the multi-batch characteristics of batch processes, a linear variable parameter (LPV) model, based on multiple batch data, is proposed to describe the batch process. The generic LPV modeling method for batch processes is a simple duplication of that for continuous processes, it only uses the information from a single batch and is subject to poor scalability. The proposed LPV method fuses the date of multiple batches and divides all the data into different stages according to the scheduling variables. At each stage, local models are established and then combined according to an exponential weighting function. In this model, an EM algorithm is employed for the fusion of data from multiple batches into parameter estimation, which avoids the influence of the initial value fluctuation of batches. The proposed method is evaluated using a simulated penicillin fermentation process.