Control of wastewater treatment plants using economic-oriented MPC and attention-based RNN disturbance prediction models

Abstract

In this work, we introduce a nonlinear economic-oriented model predictive control framework that can optimize the economic operation of wastewater treatment plants (WWTPs), while accounting for inlet flow disturbances. The proposed method utilizes an attention-based recurrent neural network (RNN) model to predict influent flow rate variations, and a WWTP reduced-order model specifically tailored for MPC integration. At each sampling instant, the proposed scheme recursively solves an optimal control problem, where the objective is to minimize the plant energy consumption. The inlet flow rate RNN predictions are integrated within the scheme and critical controller parameters, such as the prediction horizon, are optimized by considering the best RNN multi-step ahead prediction horizon. The proposed framework is applied to a modified benchmark simulation model no 1 (BSM1) representation that corresponds to an actual WWTP and its performance is compared against different control schemes, outperforming the alternative methods in terms of optimizing WWTP performance.