Soft-sensing of Wastewater Treatment Process via Deep Belief Network with Event-triggered Learning

Abstract

Due to the complex dynamic behavior of a Wastewater Treatment Process (WWTP), the existing soft-sensing models usually fail to efficiently and accurately predict its effluent water quality. Especially when a lot of practical data is provided and we do not know which data-pair is more valuable, WWTP modeling becomes a time-consuming process. The main reason is that the existing soft-sensing models update their parameters at each data-pair in one iteration, while some update operations are meaningless. To address this thorny problem, this paper proposes a Deep Belief Network with Event-triggered Learning (DBN-EL) to improve the efficiency and accuracy of soft-sensing model in WWTP. First, some events are defined according to different running condition during the process of training DBN-based soft-sensing model. The different running condition is dominated by the fluctuation of error-reduction rate. Second, an event-triggered learning strategy is designed to construct DBN-EL, whose parameters are updated only when a positive event is triggered. Thirdly, we present the convergence analysis of DBN-EL based on the optimization in a Markov process. Finally, the effectiveness of DBN-EL is demonstrated on soft-sensing of total phosphorus concentration in a practical WWTP system. In experiment, DBN-EL is compared with nine different models on soft-sensing of WWTP. The experimental results show that the efficiency of DBN-EL is 27.6%–64.9% higher than that of nine competitive models, which indicates that the proposed model is readily available for industrial deployment.