An Optimum E-Vehicle Energy Management System using Deep Reinforcement Learning

A growing body of evidence indicates that incorporating onboard computer vision hardware and software into modern automotive systems aids in the pursuit of eco-driving goals. Automotive engineers face a lengthy and tedious task
when developing Energy Management Strategies (EMSs) for various hybrid electric vehicle configurations. By capitalizing on similarities between various hybrid electric vehicle EMSs, experienced engineers can shorten the development cycle. This automated EMS development framework aims to speed up the production of hybrid electric vehicles. The study presented here combines computer vision with deep reinforcement learning, which leads to an improvement in the fuel economy of
hybrid electric cars. The proposed method can autonomously learn the best policy for control based on observed data. We
employ the cutting-edge convolutional neural networks-based object detection technique to glean useful visual data from
onboard cameras. A continuous deep reinforcement learning model takes the detected visual data as a state input and
generates policies for conserving power. To be more precise, the sharing of information among four very different hybrid
electric vehicle types is investigated. In this paper, we propose a transfer learning-based tactic to automate the improvement of hybrid electric vehicle EMSs through the exchange of cross-type knowledge between EMSs that employ various flavors of deep reinforcement learning. According to the findings, the proposed method achieves the highest possible fuel efficiency of the global optimization programming, and the depth reinforcement learning-based system with image perception uses less
fuel than the one without visual information. Moreover, the system without visual information uses less fuel than the one with
visual information. Battery modeling, accurate battery state of charge and state of health estimation, and the development
of other advanced EMS in EVs can solve most of the problems, allowing for more precise driving range estimates and more
efficient charging and discharging strategies. The proposed strategy was shown to be effective and reliable in reducing
losses and increasing safety during training and validation. The proposed energy management strategy performed better
than the methods that were based on deep learning in terms of the amount of time needed for computation and the amount
of energy lost in the combination battery bank. This provides support for the utilization of this method in the development of future systems for managing energy