Abstract: Constraint satisfaction problems (CSP) and the relevant evolutionary algorithms are reviewed first.Then we introduce a new CSP paradigm,elastic CSP model(ECSP),which is the improvement on the existing partial CSP model.We formularize ECSP as 6-tuple .For the purpose of finding the satisfactory solution(s) for decision,an interactive multi-agent architecture is designed,which integrates several technologies including multi-agent,genetic algorithm and distributed parallel computing.Furthermore,we elaborate the GA solver algorithm in the solving agent.How to determine the adaptive function?Which encoding method is chosen?How to define GA operators and initial population?Finally,we give a simple example to demonstrate the validity of our approach.

Key words: elastic constraint satisfaction problem(ECSP), satisfactory solution for decision, agent, genetic algorthm(GA)