Exploration of Model Parameter Optimization Strategies for Improved SWMM Model Calibrations

Wade Trim has leveraged optimization algorithms to enhance collection system model calibration efforts. An optimization framework consisting of the SWMM engine, optimization algorithms, and observed data currently aims to evaluate (1) the time efficiency gains of applying optimization algorithms as a pre-processing step to calibration and (2) the reduction in prediction error of previously calibrated models through parameter refinement. An overview of previous efforts and lessons learned will outline the start of this current initiative.

The following optimization algorithms were tested in both combined and separate sanitary sewer network models: Genetic Algorithm (GA), Pattern Search (PS), Reinforcement Learning (RL), Tree-Structured Parzen Estimator (TPE), and Inversion (a Wade Trim purpose-built algorithm). Respective optimization runtimes, convergence plots, and resulting model prediction errors were evaluated and compared for two (2) collection system models in the City of Pittsburgh.

The optimization iteratively adjusts various model parameters, such as subcatchment hydrological properties, hydraulic minor losses, and RTK, while continuously evaluating against multiple objectives, including model-to-meter prediction error (e.g., Euclidean norm) and curve fit metrics (e.g., correlation).

As model complexity and parameter space expands, the time required to maximize the objective functions also increases. To achieve reliable and quick convergence, a multi-step optimization approach, progressing from coarse to fine is implemented.

This presentation explores how optimization algorithms and the SWMM engine were used within an optimization framework to improve existing model calibrations in combined and sanitary sewer systems. Performance and evaluation time are also compared to traditional calibration methods. The next step in this initiative will include discussion on the ideal sequence in the calibration process to apply these optimization tools and how that will inform future development.