Due to high voltage gain and easy-to-install structure, multilevel DC/DC boost converter, have become very popular in distributed generation systems incorporating renewable energy sources. Constant Power Loads (CPLs), on the other hand, lead to converter instability due to their highly nonlinear nature. In this regard, the use of advanced control techniques to stabilize the output voltage of the DC/DC multilevel boost converters by increasing their robustness to undesirable effects of CPLs is crucial. This research aims to overcome this problem by using a novel combination of Nonlinear Terminal Sliding Mode Control (NTSMC) technique and model-free control based on Deep Reinforcement Learning (DRL) for a DC/DC multilevel boost converter in the presence of a non-ideal CPL. Moreover, a Digital Twin (DT) of the controller is created to improve the accuracy of the model implemented on a Digital Signal Processor (DSP). In the proposed control approach, the NTSMC parameters are the dynamic controller coefficients that are adaptively created by the Deep Deterministic Policy Gradient (DDPG) agent through the online learning of Actor-Critic Neural Networks (NNs). To validate the effectiveness of the proposed methodology, software-in-loop (SIL) and hardware-in-loop (HIL) testing procedures have been brought up. The results obtained showed that the proposed methodology can provide satisfying outcomes since DDPG algorithm is tuning the feedback control coefficients.