Deep learning in wastewater treatment: a critical review.

Modeling wastewater processes supports tasks such as process prediction, soft sensing, data analysis and computer assisted design of wastewater systems. Wastewater treatment processes are large, complex processes, with multiple controlling mechanisms, a high degree of disturbance variability and non-linear (generally stable) behavior with multiple internal recycle loops. Semi-mechanistic biochemical models currently dominate research and application, with data-driven deep learning models emerging as an alternative and supplementary approach. But these modeling approaches have grown in separate communities of research and practice, and so there is limited appreciation of the strengths, weaknesses, contrasts and similarities between the methods. This review addresses that gap by providing a detailed guide to deep learning methods and their application to wastewater process modeling. The review is aimed at wastewater modeling experts who are familiar with established mechanistic modeling approach, and are curious about the opportunities and challenges afforded by deep learning methods. We conclude with a discussion and needs analysis on the value of different ways of modeling wastewater processes and open research problems.

A generalised chemical precipitation modelling approach in wastewater treatment applied to calcite.

Process simulation models used across the wastewater industry have inherent limitations due to over-simplistic descriptions of important physico­chemical reactions, especially for mineral solids precipitation. As part of the efforts towards a larger Generalized Physicochemical Modelling Framework, the present study aims to identify a broadly applicable precipitation modelling approach. The study uses two experimental platforms applied to calcite precipitating from synthetic aqueous solutions to identify and validate the model approach. Firstly, dynamic pH titration tests are performed to define the baseline model approach. Constant Composition Method (CCM) experiments are then used to examine influence of environmental factors on the baseline approach. Results show that the baseline model should include precipitation kinetics (not be quasi-equilibrium), should include a 1st order effect of the mineral particulate state (Xcryst) and, for calcite, have a 2nd order dependency (exponent n = 2.05 ± 0.29) on thermodynamic supersaturation (σ). Parameter analysis indicated that the model was more tolerant to a fast kinetic coefficient (kcryst) and so, in general, it is recommended that a large kcryst value be nominally selected where insufficient process data is available. Zero seed (self nucleating) conditions were effectively represented by including arbitrarily small amounts of mineral phase in the initial conditions. Both of these aspects are important for wastewater modelling, where knowledge of kinetic coefficients is usually not available, and it is typically uncertain which precipitates are actually present. The CCM experiments confirmed the baseline model, particularly the dependency on supersaturation. Temperature was also identified as an influential factor that should be corrected for via an Arrhenius-style correction of kcryst. The influence of magnesium (a common and representative added impurity) on kcryst was found to be significant but was considered an optional correction because of a lesser influence as compared to that of temperature. Other variables such as ionic strength and pH were adequately captured by the quasi-equilibrium description of the aqueous-phase and no further kinetic corrections were required. The baseline model is readily expandable to include other precipitation reactions. For simple representations, large values for kcryst with n = 2 (or n = 2 or 3 for other minerals, as appropriate) should be selected without corrections to kcryst. Where accuracy is required (e.g., in mechanistic studies), machine estimation of kcryst should be performed with robust process data and kcryst should at least be corrected for temperature.

Effects of ionic strength and ion pairing on (plant-wide) modelling of anaerobic digestion.

Plant-wide models of wastewater treatment (such as the Benchmark Simulation Model No. 2 or BSM2) are gaining popularity for use in holistic virtual studies of treatment plant control and operations. The objective of this study is to show the influence of ionic strength (as activity corrections) and ion pairing on modelling of anaerobic digestion processes in such plant-wide models of wastewater treatment. Using the BSM2 as a case study with a number of model variants and cationic load scenarios, this paper presents the effects of an improved physico-chemical description on model predictions and overall plant performance indicators, namely effluent quality index (EQI) and operational cost index (OCI). The acid-base equilibria implemented in the Anaerobic Digestion Model No. 1 (ADM1) are modified to account for non-ideal aqueous-phase chemistry. The model corrects for ionic strength via the Davies approach to consider chemical activities instead of molar concentrations. A speciation sub-routine based on a multi-dimensional Newton-Raphson (NR) iteration method is developed to address algebraic interdependencies. The model also includes ion pairs that play an important role in wastewater treatment. The paper describes: 1) how the anaerobic digester performance is affected by physico-chemical corrections; 2) the effect on pH and the anaerobic digestion products (CO2, CH4 and H2); and, 3) how these variations are propagated from the sludge treatment to the water line. Results at high ionic strength demonstrate that corrections to account for non-ideal conditions lead to significant differences in predicted process performance (up to 18% for effluent quality and 7% for operational cost) but that for pH prediction, activity corrections are more important than ion pairing effects. Both are likely to be required when precipitation is to be modelled.