An Interval Type-2 Fuzzy Neural Network with Multi-Gradient Learning for Wastewater Treatment Process

Data-driven model plays an important role in monitoring wastewater treatment processes (WWTPs). However, since online data-driven models only focus on modeling the current operational status, it is a challenge to determine an appropriate model to guarantee the prediction performance for WWTPs with multiple operational conditions. Therefore, an interval type-2 fuzzy neural network with multi-gradient learning (MGL-IT2FNN) is proposed for modeling WWTPs in this paper. First, a condition detection mechanism is designed to detect the switching points of operational conditions of WWTPs. This mechanism can guide the learning process of IT2FNN. Second, when operational condition switching occurs, a multi-gradient optimization algorithm is developed to adjust the parameters of IT2FNN, which is able to balance the previous knowledge retention ability and the learning ability of new operational conditions. Third, MGL-IT2FNN is employed to capture the operational features under different conditions. Finally, the presented MGL-IT2FNN is applied to monitor the total nitrogen removal of WWTPs with multiple operational conditions. The experiment signifies that the accuracy and effectiveness of MGL-IT2FNN are greatly outstanding in WWTP modeling.