Digital Twins for forecasting and decision optimisation with machine learning: applications in wastewater treatment

Matthew Colwell,Mahdi Abolghasemi

2024-04-23

Abstract:Prediction and optimisation are two widely used techniques that have found many applications in solving real-world problems. While prediction is concerned with estimating the unknown future values of a variable, optimisation is concerned with optimising the decision given all the available data. These methods are used together to solve problems for sequential decision-making where often we need to predict the future values of variables and then use them for determining the optimal decisions. This paradigm is known as forecast and optimise and has numerous applications, e.g., forecast demand for a product and then optimise inventory, forecast energy demand and schedule generations, forecast demand for a service and schedule staff, to name a few. In this extended abstract, we review a digital twin that was developed and applied in wastewater treatment in Urban Utility to improve their operational efficiency. While the current study is tailored to the case study problem, the underlying principles can be used to solve similar problems in other domains.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The main purpose of this paper is to address the cost-efficiency issues encountered during the operation of wastewater treatment plants (using Urban Utilities' wastewater treatment plant as an example). Specifically, the research team developed a new method that combines prediction and optimization, aiming to improve the operational efficiency of wastewater treatment equipment (particularly the Cambi thermal hydrolysis treatment equipment) through a data-driven approach.

The challenges faced by the research team include:
1. **Balancing multiple objectives**: Under the existing intuition-based operation mode, operators need to balance multiple objectives, such as maintaining low levels in upstream storage units to receive new truck deliveries, ensuring that the quality of biosolids meets waste treatment standards, and reducing operating costs.
2. **Complexity and high energy consumption**: The Cambi equipment is complex to operate and has high energy consumption. Especially with the rising natural gas prices, optimizing its operation strategy becomes particularly important.

To address these issues, the research team adopted the following methods:
- **Mixed Integer Programming (MIP) model**: Used to minimize the difference between the level of upstream storage units and the target level, ensuring that the main objectives of the system are achieved.
- **Machine Learning Regression model**: Predicts the behavior under different operational scenarios and selects the most efficient scheme. This model not only considers the quality of biosolids but also optimizes energy use efficiency.

By combining these two methods, the research team built a decision support system that can collect operational data in real-time and provide feedback to operators, thereby helping to reduce operating costs. It is estimated that if Urban Utilities adopts this recommendation, it could save tens of thousands of dollars in natural gas consumption costs annually. Additionally, the paper points out future research directions, such as how to further integrate the prediction and optimization processes to improve decision quality.