(371d) Process Monitoring Based on Mutual Information–Multiblock Rolling Pin Vine Copula

Modern chemical processes often have many different operational units, making the relationship between the process variables complex and increasing the difficulty of process monitoring. To overcome this problem, the multiblock method has received increasing attention. In this paper, a new monitoring method based on mutual information–multiblock rolling pin vine copula (MI-MRPVC) is proposed. This method divides the correlated variables into the same subblock according to mutual information, avoiding the acquisition of prior knowledge. In each subblock, the root node of vine copula is set as a standard variable, and rolling pin transformation is adopted to make other variables linear according to this variable. The vine copula-based dependence description models in different subblocks are used to capture the local behavior. The multiblock strategy not only reduces the impact of global variable local faults during the monitoring process but also reduces the number of parameters that need to be estimated and the number of copula trees. Multiblock Rolling Pin transform can avoid the coverage of irrelevant variables of global transformation and help vine copula model describe the relationship between variables more accurately. The MI-MRPVC model provides a new method for describing the dependence of high-dimensional complex variables and improves the accuracy of the monitoring process.