Application of machine learning at wastewater treatment facilities: a review of the science, challenges and barriers by level of implementation

Abstract

ABSTRACT Wastewater treatment facilities are complex environments with many unit treatment processes in series, in parallel, and connected by feedback loops. As such, addressing prediction, control, and optimisation problems within wastewater treatment facilities is challenging. Machine learning techniques provide powerful tools that can be applied to these challenges. Uncertainties of the treatment process can be quantified and navigated via probabilistic techniques inherent in machine learning. Despite the plethora of literature on the applications of ML techniques to many individual problems within wastewater treatment facilities, a paucity of information remains regarding how those applications can be organised. Hence, the objective of this paper is to provide a systematic review and novel break down of the organisation of ML applications into type and scope. Types of ML applications are classified as prediction, control, and optimisation, and each of these applications is further classified by scope of implementation, ranging from no ML (Level 0) to full facility (Level 4). Based on this analysis, the status of different types and scopes of ML applications is presented, and challenges and key knowledge gaps in ML applications for wastewater treatment facilities are identified. Results show that ML applications to date tend to be focused on prediction rather than control or optimisation, and that full facility applications are limited to prediction applications. However, this study also identified several control and optimisation applications that have demonstrated the ability of ML applications in these areas to balance optimisation of energy and chemical use with effluent quality.