Abstract
The rapid proliferation of EVs is bringing a fast-growing and highly variable load demand source to distribution systems. The demand from workplace and university charging can be particularly concentrated within relatively short periods of time, thus forming a load pattern for which the current distribution feeder design is not prepared. Current research on this problem focuses either on machine learning techniques predicting future EV charging demand based on actual charging data and without evaluating its impact on electrical grid, or on power system analysis assessing the grid impacts of EV charging on benchmark distribution feeders with synthetically generated loads. This work presents ChargeFlow, an open-source reproducible pipeline connecting the two approaches via real EV charging demand. Hourly charging data from Caltech location of the ACN Data dataset for the first half of 2019 (January--June) is processed and aggregated to obtain the actual time series of EV charging demand. Then the hierarchy of forecasting techniques including persistence, seasonal and a simple machine learning approach is employed to produce a short-term EV demand forecast. This forecast is further used in the analysis of the IEEE 33 bus benchmark distribution feeder considering two predefined EV placement scenarios, each evaluated at three penetration multipliers. Finally, the impact of the forecasted EV demand on the bus voltage profiles and line loading is evaluated both with and without the proposed scheduling approach. The scheduling approach only shifts EV demand in the time frame during which the respective vehicle is still connected and allows to deliver enough energy before disconnection. Forecast accuracy, grid impacts under all considered scenarios and efficiency of the proposed scheduling approach in mitigating the identified violations are analyzed based on the results of the experiment. In the course of the pipeline, the actual, predicted, assumed and simulated values are explicitly separated in order to ensure transparency and reproducibility. The novelty of the work does not lie in any new forecasting or simulation technique; instead, it is in the integration of real EV charging demand forecasting with the analysis of its grid impacts and availability aware scheduling in a single reproducible pipeline.