A just-in-time learning approach for sewage treatment process monitoring with deterministic disturbances

Abstract

Process monitoring for real nonlinear industrial systems becomes a hot research topic in recent years. As an important and typical chemical industry process, sewage treatment process (STP) attracts lots of attention for quality supervision. In recent studies, STP was considered as a non-linear dynamic process, and data-based multivariate statistical methods provide effective tools of process monitoring for STP. However, these methods ignore deterministic disturbances, i.e. weather changes. Due to the existence of deterministic disturbances, the disturbances of STP no longer satisfy the Gaussian distribution and the precondition of multivariate statistical approaches. To solve this problem, a process monitoring strategy for STP called JITL-DD is proposed in this paper. Just-in-time learning (JITL), a nonlinear system model building method, serves as the basis for output prediction, and residuals are processed by a data-driven fault detection approach for linear static processes with deterministic disturbances (DD). Simulation results of STP process monitoring and the comparison with JITL-PCA based strategy show the superior performance of the proposed strategy.