Energy Demand Forecasting for Electric Vehicles Using Blockchain-Based Federated Learning

Abstract

The widespread adoption of electric cars (EVs) can be attributed to their many advantages over conventional gas-powered automobiles. However, there may be difficulties in incorporating EVs into the grid due to increased energy demand and peak load. We propose a blockchain-based federated learning scheme using different linear regression algorithms for energy demand prediction for EVs. The information gathered from EVs is stored on the blockchain network. Only those with the proper credentials can decrypt the data from its encrypted storage. Data from EVs is utilized to train a machine learning model with the use of a federated learning algorithm. Each EV is used to train a model, and then the models’ parameters are distributed throughout the blockchain. Our approach is innovative in analyzing of BCFL communications overhead and latency issues, while delving deeper into its dynamics to measure and reduce communication delays to maximize system efficiency. The implementation results verify the effectiveness of our system in anticipating EVs’ energy requirements. For the training of the BCFL model, a huge real-world dataset was used from over 60,000 transactions at EV charging stations in Boulder city, Colorado. The results show that the framework is reliable, since all the models have R2 values above 0.91, which indicates a high degree of accuracy in predicting energy use.

Keywords

• Federated learning
• linear regression
• blockchain
• electric vehicle
• energy consumption