Stochastic Modelling and Predictive Control of Wastewater Treatment Processes

Abstract

Water Resource Recovery Facilities (WRRF) hold an important responsibility. They use a significant amount of the worlds electrical energy to remove large quantities of nutrients from wastewater before it can be discharged back into environment. Hence WRRFs must ensure that water is treated satisfactory while not using more energy than necessary, as the use of electricity is both costly for operation and related to emissions of Greenhouse Gasses (GHG). In addition, the biological treatment emits nitrous oxide, a strong GHG, and the discharge of nutrients is taxed, adding to the costs. Consequently, the control of WRRFs is an important topic for optimization as it impacts both the operational costs and carbon footprint of a plant. This thesis aims to optimize the aeration in the biological treatment step of a WRRF by suggesting new control strategies. The found strategies builds on a new modelling framework for WRRFs. This framework is based on stochastic differential equations and models ammonium, nitrate and phosphorus concentrations using online data to update the model parameters. This makes it possible to predict concentrations up to 24 hours ahead with a root mean squared error lower than 0.6 mg-N/L. The key is that the model builds on a simple version of well established models of the biology. To account for the simple structure, it is frequently updated, meaning that parameters can adapt to changes in the plant. The model is used for estimating future nutrient concentrations in a Model Predictive Control (MPC) strategy. In this strategy different objectives where investigated. First, total costs estimated as nutrient tax and electricity costs using variable electricity prices from the day-ahead market was investigated. This led to savings of 9-45% compared to currently installed rule-based control strategies. However, when comparing the cost minimizing strategy with a strategy that minimized electricity consumption, the savings where ranging from 1-4 %. This was increased by including additional electricity markets, such as the regulating and special regulating markets, reaching cost savings up to 27.3 %. While these strategies created economic savings, it turned out that in some cases they led to increased GHG-emissions. Hence an objective that reduced GHG-emissions related to electricity consumption and nitrous oxide from the process was suggested. This function could reduce GHG emissions by 35% compared with the current, rule-based control, and 40.9 % compared to optimizing costs. However, the GHG savings resulted in 19.% increased costs and thereby a marginal cost of 0.4 DKK/kg − CO2 − eq. The framework has been tested full-scale on 4 different plants and qualitative investigations imply that it is working. However more run-time and further investigations are needed to evaluate on benefits and savings. In conclusion it is possible to improve the different objectives and switch between them depending on the goals of the utility company, to the benefit of operators, managers, and local environments in future smart societies.