Investigation and Demonstration of an Evolutionary Computation-Based Model Correction Procedure for a Numerical Simulation Model

Traditional model calibration attempts to correct a model so that the model output will match a set of system observations by tweaking a set of model parameters. Potential model structural error limits, however, the effectiveness and accuracy of calibration, undermining the predictive capabilities of the calibrated model. An evolutionary computation-based model error correction procedure that couples an evolutionary algorithm and a genetic program was previously developed and tested for two analytical models. Due to nonuniqueness in the solution space, numerous forms of correction terms that similarly fit the observation data were found. This procedure is further investigated to explore and identify alternative correction terms that not only provide a good fit but also results in good prediction performance. This approach is then demonstrated using a numerical groundwater contaminant transport simulation model.