A calibration and validation process (CAVP) for complex adaptive system simulation

The Calibration and Validation Process (CAVP) for Complex Adaptive System Simulation (CASS) iteratively calibrates constrained simulation agent parameter input sets to produce acceptable simulation measure of performance outputs as compared to target mean values. If all measures of performance are calibrated by a single input agent set, the CAVP validates complex adaptive system simulation. Using response surface methods and data mining techniques to guide each iteration to a better result, the CAVP will find an agent parameter input point (in multiple dimensions) that will validate a CASS if such a point exists.; The CAVP provides an efficient method to calibrate CASS agent input parameters. It is an information-engineering based process that composite maps agent-based simulation outputs to agent parameter inputs (control variables) within a complex adaptive system simulation environment (exogenous variables). This process enables the simulation modeler to calibrate a set of agents against a standard of system output that has been derived either empirically or through expert opinion. Data is generated according to an efficient Nearly{09}Orthogonal Latin Hypercube (NOLH) experimental design to reduce computation expense and efficiently search the constrained search space. The agent parameter inputs are constrained according to reasonable ranges, and the outputs are mapped buck to inputs through data mining techniques such as classification and regression trees, regression, and multiple response surface optimization. Further, the CAVP provides a means to adjudicate the validity of an agent-based simulation by comparing multiple response surfaces of measures of performance. The CAVP extends the capabilities of the Extended Response Surface Method by strengthening Step 5: Calibrate the simulation. The CAVP also presents a novel approach to agent-based simulation validation by determining the fitness of overall simulation output, and then using advanced data mining techniques to determine the influence of heterogeneous agent input parameters.