DynamicanalysisofmultiagentQ-learningwithε-greedyexploration