Constraint programming applications in designing electronic agents: An experimental study

This paper explores the designing of effective artificial agents. It reports some real experiments with artificial agents working on a realistic problem: exploring suppliers´ pricing strategies and consumers´ consumption strategies in a deregulated stochastic electric power marketplace. Several domain specific constraint-handling techniques in genetic algorithms, which are the core for agents learning, have been implemented and tested in a unified experimental setting, with promising results. The approach can be generalized to other network industries, such as gas, water, financial service and telecommunications.