Using Automated Approximate Satisfaction in Parameter Search for Dynamic Agent Models

Numerical agent models often include a number of parameters. The values of such parameters are usually determined by using some numerical parameter tuning method based on numerical empirical data. However, in many cases no numerical empirical data are available, but properties for dynamic patterns are known that should be fulfilled, as requirements. Classical numerical parameter tuning methods normally cannot work with such dynamic properties, as they can only be true or false. To remedy this, in this paper the notion of approximate satisfaction of dynamic properties is introduced. It adds a numerical measure to the logical notion of satisfaction. By doing this, numerical optimization methods for parameter estimation become applicable to support the design of dynamic agent models for which dynamic properties have been specified as requirements.