A new soft computing method for integration of expert's knowledge in reinforcement learn-ing problems