A Model Predictive Control using a data-driven ASM model for online optimization of WRRFs under different performance objectives

Water Resource Recovery Facilities (WRRFs) need to fulfil new or stricter environmental targets that are continuously updated to cope with global challenges. This requires investment into new processes, or a modification of the current plant operation. In this context, automatic control operations play an important role in ensuring good plant performance. Existing control strategies, however, were developed according to past/current performance criteria, and resources (time for simulating and assessing the impact of new controls, skilled workforce) are thus needed to adapt them to future or new performance indicators. Stentoft et al. (2019) proposed a new approach based on Model Predictive Control (MPC), where a simplified version of ASM1 model is actively used to control the aeration and thus the nitrification/denitrification processes. In this work we tested a general MPC setup using stochastic differential equations allowing WRRF operators to balance between different management objectives:

• Optimal nitrogen removal (Obj. A), where MPC operates solely to comply with effluent requirements for NH4+ and NO3-;
• Minimal electricity consumption (Obj. B), where the MPC focuses on using as little electricity for aeration as possible;
• Reduction of operational costs (Obj. C), calculated as the sum of aeration electricity consumption (with variable electricity prices) and the effluent taxes (defined according to the Danish legislation);
• Reduction of global warming potential (Obj. D), considering both the GWP from electricity production and the N2O emission from N removal (assumed to be directly proportional to NH4+ oxidation).

The performance of the MPC in fulfilling these four objectives were simulated by using data from a small alternating WRRF (cf. Isaacs and Thornberg, 1998) in Denmark (Nørre Snede). Also, we illustrated the flexibility of the MPC in operating in a multiobjective perspective, where WRRF operators can quickly change the control priorities, choosing how to prioritize e.g. cost reduction or reduction of climate change impacts. Our results open new possibilities for a flexible operation
of WRRFs by combining information provided by online sensors with the process knowledge incorporated in the ASM models.