Privacy-aware electric vehicle load forecasting via blockchain-based federated transfer learning

Electric vehicle (EV) load forecasting has become critical with the rapid growth of EVs and fast-charging technologies. However, privacy concerns pose challenges to traditional forecasting methods. Our work proposes a novel blockchain-based federated transfer learning framework for privacy-aware probabilistic EV load forecasting. Our approach uses secure aggregation via blockchain to prevent gradient leakage, with smart contracts aggregating masked vectors to produce a global model without exposing individual data. Our work introduces transfer learning to adapt the global model to varying charging scenarios by transferring layer-wise weights. The framework integrates machine learning, pattern recognition, and big data analytics methodologies. Experimental results with datasets of varying client numbers demonstrate that our method outperforms baselines in prediction accuracy and communication efficiency, particularly for large-scale scenarios. By combining advanced techniques in data mining, machine learning, and blockchain, this framework provides a scalable solution for privacy-preserving EV load forecasting while ensuring robust performance in dynamic environments. Our innovations include secure parameter aggregation using homomorphic encryption, blockchain-based secure aggregation, and transfer learning for model adaptability, resulting in significant improvements in privacy protection, communication overhead, and prediction accuracy.