Optimising water treatment operations with prediction modelling and smart technologies

Abstract

Optimising the raw water intake selection is a critical task for multiple source water treatment plants, as this depends on, among others, water quality, pumping costs, and safety considerations. In case of the Mudgeeraba treatment plant in SouthEast Queensland (Australia), the raw water can be withdrawn by gravity from the small Little Nerang dam, or pumped from the larger Hinze dam. Intake towers with different gates also allow for the plant operators to withdraw the raw water from different depths. Often though, decision on the optimal intake location is taken based on operators' experience, without a real scientific method relying on in-depth considerations on e.g. current water quality at multiple locations and electricity costs. Nevertheless, the large amount of data collected daily at the treatment plant, and at higher frequency in the two reservoirs by remote vertical profilers, provides an opportunity to link water quality with treatment costs and provide numerical evidence for the operators to take a more informed decision. As a consequence, a number of data-driven, chemical and mathematical models linking raw water quality with required chemicals dosages, as well as a model estimating the pumping costs, were developed. As a consequence, it was possible to predict the overall daily variable treatment costs based on raw water quality. Given that the water quality is monitored in real time in the two reservoirs for the whole water column, it was also possible to provide advice on a daily basis on which withdrawal depth and reservoir would lead to the lowest treatment cost. By running the model with historical data, it was found that for almost any given day, withdrawing water from Little Nerang dam would have been a cheaper option due to the associated energy costs (i.e. no pumping required) being much lower than potential extra treatment costs due to poorer raw water quality; however, traditionally operators prefer to keep it as a backup reservoir in cases of, for instance, power outages and inability to use the Hinze dam pumps. Given that Little Nerang dam, despite smaller, would be able to provide enough water for few months, an increased use of this water source was still recommended, especially around wet seasons. Based on this, a medium-term storage volume prediction model was developed, able to forecast the risk of depletion and spill for Little Nerang dam for the upcoming six weeks based on, among others, proposed daily withdrawal rates and weather forecasts. In this way, plant operators would be able to plan the future withdrawals in a way to reduce the costs, but also to avoid excessive spill or depletion risks; thus effectively achieving a better planning and management of water treatment operations. Despite the predicted benefits, it is critical to engage with the relevant stakeholders and potential end-users in order to incentivise the deployment of the end products of research projects; often, lack of proper communication and engagement can lead to the majority of the research outcomes to be...