Machine learning parallel system for integrated process-model calibration and accuracy enhancement in sewer-river...

Area of Science:

• Environmental Engineering
• Water Resource Management
• Computational Science

Background:

• Urban water system management is increasingly digitalized, leading to complex, multi-functional process-based models.
• Increased model complexity introduces significant uncertainty and computational demands.
• Conventional calibration methods struggle with long computation times and scarce monitoring data.

Purpose of the Study:

• To introduce a novel machine learning system (MLPS) for expedited parameter optimization of integrated process-based models.
• To enhance the performance, efficiency, accuracy, and stability of complex water models.
• To address the limitations of conventional calibration methods in handling computational intensity and data scarcity.

Main Methods:

• Developed a machine learning parallel system (MLPS) integrating model surrogation with Ant Colony Optimization (ACO) and Long Short-Term Memory (LSTM).
• Employed ACO and LSTM for efficient parameter search and optimization within the MLPS framework.
• Validated MLPS using an integrated sewer network and urban river model.

Main Results:

• Achieved a significant reduction in parameter calibration time by 89.94%.
• Substantially improved prediction accuracy for river pollutant concentrations, decreasing average absolute percent bias from 124.3% to 8.8%.
• Demonstrated close alignment between MLPS-optimized model outputs and monitoring data.

Conclusions:

• MLPS enables efficient optimization of complex, integrated process-based models, even with limited data.
• Facilitates the practical application of highly precise environmental management models.
• Offers valuable insights for optimizing complex models across various scientific and engineering fields.