Deep Recurrent Q-learning Method for Area Traffic Coordination Control