This research investigates a new Deep Q-Network (DQN) based approach to manage Dynamic Voltage Frequency Scaling (DVFS) on a multiprocessor architecture, such that it would guarantee the balance between energy consumption minimization and application feasibility. This paper also addresses software periodic real time applications with time constraints. The proposed DQN formulation operates in two steps: on offline and on online configuration. It calculates the optimal number of activated homogenous cores and their frequency and reconfigures the platform with these parameters. We perform an experimental investigation on different parameters in a simulation environment executing periodic tasks that are generated randomly with different system charge. The results suggest that the proposed method reduces energy upto CC-EDF and Static-EDF and guarantees schedulability test when compared to state of the art feedback that addresses the same applications.