Evolutionary Strategies as a Verification and Validation Tool

We describe the use of evolutionary algorithms to improve verification and validation (V&V) of a model simulation. Evolutionary algorithms (EAs) have two characteristics useful for V&V. First, they search a broad range of values in the model’s parameter space. This allows testing of unusual combinations of parameter values that might not be found by more conventional bounds checking and sensitivity analysis. Second, they exploit with Darwinian ruthlessness any slight evolutionary advantage shown by these combinations, whether or not these combinations of parameter values were intended or anticipated by the designer. Two approaches to evolutionary V&V are described: genetic algorithms and scatter search. One test involved use of a hand crafted genetic algorithm working with a logistics simulation written in a general purpose language. The other involved use of a commercial optimizer that employed scatter search /tabu search in conjunction with a commercial simulation tool. In both cases, the evolutionary V&V approach detected intentional coding and design errors. Shervais & Wakeland Rev. D 3 Shervais & Wakeland Page 4