Dynamic resilience for biological wastewater treatment processes: Interpreting data for process management and the potential for knowledge discovery

Abstract

Climate change, population growth and increasing regulation are causing wastewater treatment plants to become increasingly stressed, especially in countries like the UK, where many of these systems date back to the early part of the 20th century. Understanding resilience dynamics for these ageing wastewater assets represents a fundamental step in classifying multi-dimensional water stressors toward preventing severe pollution incidents. This paper explores the potential of a novel dynamic resilience approach to assess and predict the dynamic resilience of biological wastewater treatment based on the separation of stressor events (cause) and process stress (effect) to consider the deviation from reference conditions. The approach presented provides a fundamental link between (1) conventional activated sludge modelling methodologies, (2) actual biological wastewater process instrument data (potential for knowledge discovery) and (3) the characterisation of dynamic resilience in wastewater treatment processes. Results first present the dynamic resilience approach by modelling simulated shock flow conditions on an activated sludge plant, then incorporates ten years of wastewater process instrument data to demonstrate the actual dynamic resilience. The aim is to represent the “dynamic resilience” as self-ordering windows, a visual knowledge base (three dimensional, heat map), which operational staff can easily interpret. The outcomes presented suggest that such an approach is feasible and has the potential for real-time identification of conditions that result in pollution incidents based on actual historical process instrument data (knowledge discovery). Also, the methods presented could be extended to develop an improved understanding of wastewater system resilience under a range of future stressor scenarios.

Introduction

UK water companies are at a crisis point in managing the impact stressors exert on their ageing sewerage infrastructure and wastewater assets. These stressors result from climate change, population growth, changes in consumer behaviour and increasingly stringent discharge permits. In the UK, the water sector financial regulator, Ofwat [1], summarised the crucial challenges to water company performance as (1) environmental impacts, (2) securing long term resilience and (3) keeping water services affordable to customers [2,3]. Since the privatisation of the UK water sector in 1986, shareholders have demanded significant dividends, reducing funding for the proactive maintenance of ageing assets (reduced resilience). Rather than replacement, these assets have been serially updated rather than replaced [4]. This strategy has reduced wastewater process resilience and has been further compounded by rising environmental and population-based stressors [5,6]. More recently, new stressors have emerged, applying tremendous pressure on water resources. For example, the COVID 19 pandemic has changed the demand on water distribution and, in turn, wastewater production patterns, including stressors associated with masks, wipes, and biocides in wastewaters [7]. The UK Meteorological Office [8] estimates a 30% increase in potable water demand in suburban residential areas and a proportional decrease in demand. Comparatively, in large cities, the opposite occurred, with water demand falling by 30%, meaning wastewaters became been more concentrated. These additional stressors exert additional stress on existing wastewater treatment processes, further impacting their long-term resilience, short-term performance and environmental impact.

Many existing resilience methods consider the whole wastewater treatment system boundary rather than discrete unit processes, which may respond differently to external stressors. Research typically follows a lumped parameter approach to measure resilience in wastewater infrastructure, evaluating the whole treatment system at its boundary, known as Global Resilience Analysis (GRA). The GRA methodology has been adapted for wastewater applications as presented by Mugume et al. [9] with state of the art review performed by Juan-García et al. [10]. Other researchers have combined resilience and reliability frameworks such as the Safe and SuRe framework introduced by Butler et al. [11]. However, these approaches also propose analysing the whole Wastewater Treatment Plant (WwTP), using the ‘stressor’ to explain static variations from a benchmarked condition [[12], [13], [14]]. This analogy works if the benchmark condition is fixed but is not suitable for the real dynamic oscillations resulting from diurnal and seasonal changes. For example, biological treatment processes can have numerous complex and dynamic responses to stressor events. Therefore, to understand the dynamics of resilience, it is vital to model biological wastewater process resilience in sufficient detail and avoid the iterative, scenario-based methods used in research focussed on traditional mechanistic models [[15], [16], [17]].

This paper explores the potential of modelling the long term dynamic resilience of biological treatment wastewater assets through decoupling the stressor (cause) from the stresses occurring in a process (effect) introduced by Holloway et al. [18]. As presented in Fig. 1, this approach treats the stressor and process stress as mutually exclusive, with independent characteristics.

The need to understand the dynamics of resilience was confirmed in a survey of water sector stakeholders (including operational staff), revealing that 82% considered a modelling tool for evaluating process stresses as necessary to avoid process-related failures [19]. These outcomes led to the development of an initial ‘state-based’, dynamic resilience evaluation using a heat map of the primary sedimentation process [18]. These methods were limited to the confines of the empirical model and the number of theoretical observations computed by the Monte-Carlo simulations (not based on actual data). To fully explore the dynamics of resilience, it is crucial to consider the complexity of the multivariate transformations in biological wastewater treatment systems, which can only be exploited with the use of mechanistic modelling and actual instrument data. This evaluation could then be analogous to knowledge the discovery from real data, while understanding the dynamics of resilience resulting from external environmental factors such as temperature and dilution (flow) variations that are increasing due to climate change [20].

Knowledge discovery has been proposed for wastewater treatment processes since the early work of Comas et al. [21], which used unsupervised clustering followed by the development of Case-Based Reasoning by Comas et al. [22], which stores events (cases) in a knowledge base. Although these methods are comprehensive, the use of real instrument data and the presentation of a visual knowledge base was not considered, narrowing the application of such systems. Therefore knowledge can be stored easily, but communication after storage to those with an empirical understanding of wastewater treatment plant is a fundamental challenge, which must conform to operational norms such as control charts. More recently, Vasilaki et al. [23] used knowledge discovery and Support Vector Machines to predict N₂O emissions from Sequencing Batch Reactors, which contrasts with the work of Corominas et al. [24]. This research presented a Gartner hype cycle plot, classifying SVM as the ‘peak of inflated expectations’. Conversely, Artificial Neural Networks, Fuzzy logic, regression, Principal Component Analysis, Partial Least Squares, mass balance and control chart were described as the ‘plateau of productivity’, with others, such as Dürrenmatt and Gujer [25], considering data-driven modelling to predict sensor replacement and diagnose faults. Therefore, although previous knowledge discovery methods have provided robust outcomes, no attempt has been made to simplify data (knowledge) through ordering and visualisation to improve operational staff interpretation [26].

With over thirty years of application, the IWA ASM series of models has become a well-accepted ‘white box’ convention for Activated Sludge (AS) modelling [27]. These models have formed the base model for many of the Plant Wide (PWM) and Extended Plant Wide (E-PWM) models, such as those presented by Solon et al. [28] and Mbamba et al. [29]. Based on mechanistic conversions, ASM models can provide a good level of accuracy; however, as described by Regmi et al. [30], experts are required to generate accurate models in practice [31]. This is due to ASM outputs being dependent on a large number of interrelated factors, namely; (1) the quality and quantity of data available for calibration; (2) how the data is cleaned; (3) the accuracy required to validate the model, and (4) the modelling expertise available. Understanding all of these factors is crucial in developing a ‘white box’ model instead of a black box, which forms the basis of data-driven methodologies that use pre-packaged algorithms to predict model outputs [32]. The key concern with using these black-box algorithms (data-driven methods) for unsupervised learning, particularly in wastewaters, is understanding whether the correct model outputs are achieved and how the process risk is managed. A proposed extension to ASM methods is ‘hybrid modelling’, which combines mathematical models and actual data in Artificial Neural Networks [33,34]. These models will likely improve model predictions [35], but first, a means of evaluating and communicating learned process knowledge from real process data must be developed.

Many researchers such as Santos et al. [36] have focussed on the short and long term accuracy of ASM models rather than developing communicable outputs. This has prevented them from being more widely applied for evaluating resilience or knowledge discovery when using actual stochastic data, as described by Regmi et al. [30]. Although, simplifications could be made to reduce the mechanistic rigidity of the ASM models to accept real instrument data and develop a methodology to evaluate the dynamics of resilience and present the knowledge discovered to those operating wastewater treatment processes. Initial interest in this research area was presented by García et al. [37]. However, there is no evidence to suggest that modelling, coupled with actual instrument process data, has been used to communicate process resilience outputs operational staff who, as described by Langergraber et al. [38], make empirical observations. To address this, processed outputs from instrument data must be presented in a versatile yet straightforward visual format that self-orders. Also, to avoid over-processing, modelling and data cleaning must adopt a macro approach rather than processing/modelling each instrument micro-variation, described by Newhart et al. [39] as computationally intensive.

Overall the modelling and simulation of wastewater treatment processes has become a largely iterative program-test exercise for niche process scenarios [40] and has not been adapted to consider the dynamics of resilience. If nothing is done to learn from the vast number of process variants and their resilience then, water companies will not be prepared for the challenges faced by climate change and changes in consumer behaviour through unexpected events, such as pandemics [8].

This study investigates the novel concept of dynamic resilience (stressor and process stress) for biological wastewater treatment processes. Two evaluations are presented to progressively evaluate and visualise process knowledge in ten years of dynamic biological wastewater process data. The first stage simulates hypothetical time-based shock flow conditions (Evaluation 1) and the variation of dynamic resilience while considering the prominence/dominance of events. Evaluation 2 transforms actual process instrument data to detect process-related events (and failures) to present dynamic resilience as three-dimensional self-ordering windows to communicate discovered knowledge (contour-based heat plots). This research extends current resilience methodologies beyond the conventional stressor analogy to incorporate the dynamics of resilience in the context of modelling and real biological wastewater instrument data.