AI-based smart load balancing for sustainable electric vehicle charging hubs using LSTM forecasting and reinforcement learning

Abstract

The rapid growth of electric vehicles (EVs) on a global scale has put a significant burden on existing electricity distribution structures. Traditional charging methods at a public charger, utilizing fixed rates, cannot adjust to the continually changing supply and demand for renewable electricity or the effect of peak loads on the electric power grid system. To solve this very important limitation, this research presented an AI-Based Smart Load Balancing Electric Vehicle Charging Hub (AI-SLBECH). The proposed framework includes a simulation-validated, multi-layer intelligent ecosystem that uses all of the following components: a 200 W solar photovoltaic (PV) array, a lithium iron phosphate (LFP) battery energy storage system (BESS), grid interconnectivity, IoT monitoring system, all being operated under an adaptive energy management architecture. The proposed framework consists of an integration of three complementary AI components: (1) an AI-based Finite State Machine (FSM) for tri-modal energy dispatch, (2) a two-layer Long Short Term Memory (LSTM) neural network for predicting 15–30 min ahead, demand (NRMSE = 7.2%, MAPE = 6.8%), and (3) a Proximal Policy Optimization (PPO) reinforcement learning (RL) scheduling agent, trained over 10,000 episodes with a composite reward function, achieved convergence from episode 8,200 on and produced a Jain’s Fairness Index of 0.94. The complete control pipeline was implemented on an ESP32 microcontroller (operating under constraints) with TensorFlow Lite Micro, on-device LSTM was using only 78KB of flash memory. In comparison with the leading AI-powered EV charging systems that have been examined in a variety of published materials, this AI-SLBECH has unique design features. It offers an effective method for integrating all three service modalities (dispatch mode, predictive scheduling using LSTM models and optimal policies through PPO-RL algorithms), as well as real-time monitoring via Internet of Things (IoT) devices and was designed to operate on edge-deployable, hardware-constrained systems. When tested over 30 days using actual NASA POWER solar irradiance data for Chennai, India, the AI-SLBECH produced several remarkable results compared to a traditional grid-based solution. It reduced the peak demand of the grid by 38.1% (42.3 KW-26.2 KW), lowered the average session cost for EV charging from 88.43 INR to 61.20 INR (a 30.7% reduction), decreased the total amount of carbon dioxide CO₂ released during the month by 61.3% (921 kg to 356 kg), achieved a 48.4% use of renewable energy and exhibited a Jain’s Fairness Index score exceeding 0.94. A preliminary techno-economic feasibility analysis indicates that this project will require a capital investment of about Rs. 2.8 Lakh (approximately $3,360US) and will generate substantial returns on investment approximately every 28–34 months, based on current utility tariff rate structures.