DECENTRALIZED ENERGY DEMAND FORECASTING FOR ELECTRIC VEHICLES VIA BLOCKCHAIN AND FEDERATED LEARNING

Abstract

The many benefits that electric vehicles (EVs) provide over traditional gas-powered vehicles are the main reason for their increasing use. However, because of the higher energy consumption and peak load, integrating EVs into the grid can be challenging. We suggest a blockchain-based federated learning system for predicting EV energy consumption that makes use of several linear regression techniques. The blockchain network stores the data collected from EVs. The data is stored in encrypted storage, and only those with the right credentials may decode it. A federated learning approach is used to train a machine learning model using data from EVs. A model is trained on each EV, and the parameters of the model are then dispersed throughout the blockchain. In order to quantify and minimise communication delays and optimise system performance, we use a novel technique to the analysis of BCFL communications overhead and latency concerns. The outcomes of the deployment confirm how well our approach predicts the energy needs of EVs. A massive real-world dataset including more than 60,000 transactions at EV charging stations in Boulder, Colorado, was utilised to train the BCFL model. Since all of the models had R2 values over 0.91, which indicates a high degree of accuracy in predicting energy usage, the findings demonstrate the framework's dependability.