Efficient fault monitoring in wastewater treatment processes with time stacked broad learning network

Abstract

Process monitoring models play an increasingly indispensable role in promptly differentiating faults within the wastewater treatment process to maintain a safe state. The accuracy and time overhead are crucial indicators in judging whether models results are excellent or not. In terms of accuracy, deep learning methods excel, because of their generalized approximation and strong learning capabilities that enable them to infinitely approach an ideal level of monitoring accuracy. Meanwhile, they also have large computational loads that generates significant time overhead, which poses challenges for real-time monitoring of the wastewater treatment process. Hence, a novel process monitoring method called Time-Stacked Broad Learning System (Time-SBLS) is developed to alleviate these concerns in an effective and accurate manner. Time-SBLS leverages the efficiency of the BLS while incorporating the residual mechanism and time lag idea to capture necessary nonlinear features and time correlation associated with the fault detection accuracy better. Furthermore, another advantage of Time-SBLS is its inherited efficiency from BLS regarding time overhead, which is minimized by utilizing the pseudo-inverse technique to calculate weight parameters. Experimental findings on the Benchmark platform reveal that Time-SBLS offers considerable advantages in terms of accuracy and network time overhead compared to other state-of-the-art approaches.