Identification and interpretation of patterns and relationships in the vast amounts of data collected by water utilities has the potential to drive better operating decisions and drive reductions in cost, however, engineers and decision makers are sceptical of using artificial intelligence (AI) or machine learning (ML) as standalone approaches.
Algorithms are often presented and sold as ‘black boxes’ that produce non-transparent, unexplainable outcomes, and require constant oversight and supervision. Large amounts of clean, historical and granular data are needed to be accurate. More importantly, experienced operators and practitioners in the industry see the lack of connection to the industry itself as a weakness. An ML model on its own does not care about the industry that it is applied to and makes no connection to the accepted physics, chemistry and biology of the process. How should the water industry realise the benefits that AI can deliver, while managing the risks that come with the approach?
Digitalisation is driving change in the water and wastewater treatment sectors, just as it is in other industrial and corporate environments. Water and wastewater treatment plants collect more data than ever before. Control and treatment equipment is increasingly augmented with sensors that collect data to support control of the process.
Deterministic and AI learning approaches should be used together as the industry continues along its journey of digitalisation
As engineers and plant managers, we know and trust deterministic models. These models have proven reliable over the years to represent and simulate processes such as chemical reactions in drinking water or industrial wastewater treatment. If the inputs into a process are known, the model can produce a set of outputs that we use to make decisions. It’s accepted that best practice models can represent a process accurately enough to be used to understand the implications of changes to the process. However, these approaches are manual and traditionally require substantial technical expertise to drive. Deterministic models are also by their very nature not conducive to forecasting.
The water industry needs to change to meet increasingly stringent operational, regulatory and environmental requirements, as well as the needs of surrounding communities. Understanding the implications of what doing something different would look like is, however, a key challenge still facing water utilities. It is currently difficult to measure or understand the applicability of alternative operating scenarios. Operators and engineers must either live test and see what happens, take part of the plant offline to run tests or run offline simulations or studies. This is slow and impractical. Leaders in the sector are looking to alternative options such as digital twin technology as a way of creating a virtual representation of their operations, where changes can be simulated - and implications understood - quickly and with no impact to business-as-usual operations. Automated deterministic models, connected in near real-time, are ideal for this purpose, but they are reactive and cannot deliver useful projections on what you should do to avoid an issue.
Hybrid approaches that combine deterministic and AI approaches can overcome the limitations of both methods. Deterministic modelling can be used to represent total process performance, grounded in accepted science of the industry with AI used to drive accurate forecasts and recommendations for short-term actions to improve performance. Keeping personnel involved to oversee and implement recommendations remains important while trust is built in the system. At Envirosuite, we see the value in both deterministic and AI learning approaches and believe that they should be used together as the industry continues along its journey of digitalisation. As approaches evolve and improve, we look forward to the further opportunities that AI could deliver to the water industry, as long as we keep the humans in the loop.