Data Driven Approach to Achieve Coordinated Charging Among Electric Vehicles

Abstract

AbstractIt is estimated that petroleum-based vehicles release about 1.5 billion metric tons of greenhouse gases into the earth’s atmosphere, which contributes to rising temperatures, breathing problems among other things. Electric vehicles are an eco-friendly alternative to petroleum fuel-based vehicles. Electric vehicles, due to recent improvements in battery technology, have comparable capabilities as conventional automobiles in terms of distance they can travel and speed they can achieve. But due to two major pain points, the electric vehicles are not going mainstream as soon as expected. The first major pain point is the time required to fully charge an electric vehicle is much higher when compared to the refueling process of a petroleum-based vehicle. This issue is somewhat tackled by the invention of fast charging technology, which charges an electric vehicle up to 50% in a comparatively short period of time, thus enabling the user to juice up and go when required. But even that time to reach 50% is usually in the range of 30–50 min, which is significantly higher compared to refueling of traditional vehicles. The second major pain point is the excessive load on the power grid due to the charging of electric vehicles. The traditional power grid load expectancy calculation and maintenance are done using weather forecast data. But due to the mobile nature of electric vehicles, it is highly difficult to estimate the load on the power grid and start the preservation procedures to avoid any damage. Hence, a coordinated approach for charging of electric vehicles is extremely necessary. This paper explores data driven approaches to determine load on the power grid beforehand, to enable the chance of maintenance work to avoid the damage of the power grid. We used random forest (RF) and K-nearest neighbor (KNN) machine learning techniques to determine the load on the power grid at any given point of time using historical data. The results obtained are about 90% accurate, thus proving data driven approaches can be taken to solve this problem.KeywordsDecision tree (DT)Electric vehicles charging stations (EVCS)K-nearest neighbors (KNN)Machine learning (ML)Power rating (PR)Random forest (RF)