Cascaded LSTM-Based State Prediction of Equipment in Wastewater Treatment Process

In the wastewater treatment process (WWTP), it is crucial to monitor the state of the equipment for maintaining high-quality automation. When the equipment are subject to multidimensional states, they always, however, suffer from cross-interference among available state indicators due to their unclear causal relationship. To address this issue, in this work, a multidimensional state predictive model (MDSP), based on the cascaded long-short-term memory (Cd-LSTM), is developed to monitor the states of equipment. First, constructing with Cd-LSTM for predicting multidimensional state indicators, the proposed MDSP is able to obtain causal relationships among indicators. Second, several dynamic data storage units embedded into Cd-LSTM are developed to regulate the associative strength between indicators based on the acquired causal relationship, which can alleviate cross-interference. Third, the independent parameters and shared parameters of Cd-LSTM are updated by the algorithm of joint backpropagation gradients to maintain a reliable prediction. Finally, to verify the proposed MDSP, some experiments are carried out on a real-world blower dataset in WWTP to predict the blower states. The experimental results confirmed that the proposed strategy has performed excellently.