Neural Network-based Load Forecasting Model for Efficient Charging of Electric Vehicles

In the modern era, electric vehicles are considered as the main solution for green transport and controlling the emerging environmental challenges. Intelligent scheduling scheme for optimal charging of electric vehicles at minimum cost is inevitable due to the increased burden on the electric power grid due to excessive consumption of electric power. The charging model conventionally requires the information of future base load and arrival rate of electric vehicles at charging station for the scheduling scheme. In this paper, we investigate machine learning-based forecasting models to predict base load for day ahead, and we model the arrival rate of electric vehicles as Uniform distribution and Poisson distribution. The charging cost is designed as linear function of total load in respective interval. The load forecasting model uses the temperature information and electric load history of the region. We performed experimental analysis on a benchmark dataset and compared the performance of linear regression, bagged tree regression and artificial neural network for predicting the electric load. Artificial neural network achieved the lowest prediction error for load forecasting, especially during the the peak hours of the day, which depicts the efficacy of neural networks in load forecasting for efficient charging of electric vehicles.