A Reconfigurable Intelligent Surface is an eminent approach for improving the net data rate and maximizing the energy efficacy of wireless Base Stations (BS). Due to the vast number of surface elements, the task of optimizing the BS’s transmission and Reconfigurable Intelligent Surface (RIS) element’s configuration is incredibly challenging. In principle, to enhance energy conservation and diminish the BS power consumption, it is essential to optimize the transmission power of the BS and phase configuration of the RIS. This paper proposes a Long Short-Term Memory (LSTM) based scheme which performs decision-making using dynamic information of the wireless networks following channel intricacy and RIS’s energy harvesting while increasing the energy efficacy. Once trained in a real-time environment, the proposed LSTM model foretells optimal RIS configuration for each transmission. The transmissions considered are designated for users located in various regions in the corresponding wireless network. The LSTM model and Adam optimizer are used to build the RIS-aided downlink system model and explore its energy efficiency and robustness. The results achieved after performing various simulations determine that the LSTM framework raises energy efficacy to 35.42% while increasing the RIS elements from 9 to 25. In addition, the model can achieve more than 100 bps $/$ Hz net data rate.