Optimizing Electric Vehicle Charging Costs Using Machine Learning

The article presents a novel information system for optimizing urban EV charging costs.Algorithms for machine learning identify trends and make decisions in real time based on patterns in diverse data sources.The criteria for the decision model include electricity pricing, demand, driving habits and historical data.The analysis employs public data on electricity pricing, energy consumption, temperature, and transportation habits.This article compares three distinct strategies for charging Electric Vehicles (EVs): the Always Charge (AC) model, the optimal charging strategy using Dynamic Programming (DP), and the gasonly strategy.The optimization algorithm employs Q Learning with reinforcement technique allows the system to learn and adapt to dynamic conditions by making decisions based on past experiences and rewards (cost reductions).Deep Neural Networks (DNNs) can identify complex patterns in various data sources (electricity pricing, demand, driving habits) to predict optimal charging times.The results indicate that the AC model achieved significant cost reductions during the summer, ranging between 28% and 74% across vehicles.The optimal charging strategy based on dynamic programming achieved extraordinary summer and winter median gains of 95% and 88%, respectively.Particularly, the Deep Neural Network (DNN) approach showed promise, approximating the global optimum attained by DP.Standard machine learning techniques were used to evaluate the system, and the results were promising.EV charging station operators can use it to dynamically adjust charging prices based on real-time electricity costs and demand.