Real-time control of urban drainage systems using neuro-evolution

Area of Science:

• Environmental Engineering
• Artificial Intelligence
• Water Resource Management

Background:

• Urban drainage systems face increasing stress due to climate change and urbanization.
• Real-time control (RTC) offers a solution to enhance system performance and mitigate upgrade needs.
• Developing optimal RTC policies is complex due to computational demands and system uncertainties.

Purpose of the Study:

• To present a novel neuro-evolutionary approach for controlling combined sewer overflow (CSO) in urban drainage systems.
• To train a control policy offline, bypassing online optimization challenges.
• To evaluate the effectiveness of the neuro-evolutionary control policy against existing strategies.

Main Methods:

• Utilized neuro-evolution, a method combining neural networks and evolutionary algorithms, for RTC policy development.
• Trained the control policy in advance.
• Simulated performance on the benchmark Astlingen network and analyzed typical CSO events.

Main Results:

• The neuro-evolutionary control policy demonstrated superior CSO volume reduction compared to the equal filling degree strategy.
• The policy exhibited robustness against uncertainties in tank levels.
• Effectiveness was more pronounced in smaller CSO events, particularly during their initial phase.

Conclusions:

• Neuro-evolution provides a viable and effective method for RTC in urban drainage systems.
• The developed policy offers significant improvements in CSO management, especially for frequent, smaller overflow events.
• This research lays the groundwork for future advancements in urban water system control using neuro-evolution.