A Behavior-Aware Approach for Deep Reinforcement Learning in Non-stationary Environments without Known Change Points