Study on auxiliary operation control of machine learning in multiobjective complex drainage system.

Area of Science:

• Environmental Engineering
• Urban Water Management
• Machine Learning Applications

Background:

• Urban waterlogging and black-odorous river pollution are significant societal concerns.
• Upgrading drainage systems with deep tunnels complicates operational control, traditionally reliant on human expertise or complex simulations.

Purpose of the Study:

• To explore the application of machine learning for real-time operational control of complex drainage systems.
• To determine if real-time operational suggestions can be generated using only real-time data, bypassing traditional simulation methods.

Main Methods:

• Utilized five drainage systems as case studies.
• Selected key variables including initial pipeline water level, key point water level flow, pump station front pool water level, and most unfavorable point water level.
• Applied four machine learning discrimination methods to analyze deep tunnel weir-lowering operations.

Main Results:

• The linear discrimination method achieved an average error rate of less than 10%.
• Demonstrated satisfactory performance in predicting and controlling drainage system operations.
• Validated the feasibility of using machine learning for real-time drainage system management.

Conclusions:

• Machine learning provides a viable alternative to traditional methods for drainage system operation control.
• This approach enhances the efficiency and accuracy of managing complex urban water infrastructure.
• Offers valuable insights for improving the operational strategies of intricate drainage networks.