Bayesian Network-Based Modeling and Operational Adjustment of Plantwide Flotation Industrial Process

Abstract

The operational variables of the flotation industrial process (FIP) are controlled by the operators and are usually not adjusted in time. This makes it difficult to control the technical indexes such as concentrate grade within acceptable ranges. To resolve this problem, a Bayesian network (BN)-based modeling and operational adjustment method is investigated. Considering the complexity of modeling, a modular BN modeling framework for plantwide FIP is proposed. First, the plantwide FIP is decomposed into several related submodules, and corresponding local BNs are established through structure learning and parameter learning. Then, on the basis of the process knowledge and associated variables, each local BN is fused into the global BN. In the application, a novel operational adjustment framework for plantwide FIP is proposed. First, a global operational adjustment is inferred by setting the index of concentrate grade. Then, according to the obtained operational adjustment of each submodule from local to global, the concentrate grade is further predicted. Once the predicted result meets a certain condition, the current operational adjustment will be implemented right away. Data experiments evaluate the performance of the proposed method in the decision-making of operational adjustment.