What is it about?
This paper develops a new way for charging stations to forecast their future charging load without sharing raw data. Instead of sending sensitive charging records to a central platform, each charging station trains a local forecasting model using its own historical data and only shares model updates. To capture both short-term patterns and longer-term dependencies in charging behavior, the method uses a hybrid TCN–LSTM deep learning model. The paper goes a step further by removing the centralized aggregation server that most federated learning systems still rely on. A blockchain-based architecture is introduced so that charging stations can collaboratively verify, validate, and aggregate model updates in a fully decentralized way. Different stations are dynamically assigned roles such as workers, validators, and miners, while a model verification mechanism filters out low-quality or malicious updates to improve robustness and security.
Featured Image
Photo by Nik on Unsplash
Why is it important?
As electric vehicle adoption grows rapidly, charging station load forecasting becomes increasingly important for grid operation, infrastructure planning, and charging network optimization. But charging data often contains sensitive information such as travel habits and charging behavior, which makes centralized data collection risky and difficult to implement in practice. This work addresses that challenge by allowing collaborative forecasting while keeping raw data local. This study is also important because it tackles two major weaknesses of conventional federated learning: reliance on a central server and vulnerability to harmful updates. By combining blockchain with decentralized validation and aggregation, the proposed method reduces single-point failure risk, improves trustworthiness, and strengthens resistance to poisoning attacks. The paper’s experiments show that the method achieves strong forecasting accuracy while preserving privacy and improving system security, making it especially relevant for real-world EV charging infrastructures.
Perspectives
What I find most compelling about this work is that it treats forecasting not only as an accuracy problem, but also as a trust and deployment problem. In real charging networks, data privacy, malicious behavior, and system reliability are just as important as predictive performance. By integrating blockchain, federated learning, and model verification into one framework, this paper moves load forecasting closer to real-world implementation rather than staying at the level of an idealized centralized model. I also think the use of a lightweight TCN–LSTM architecture is a practical design choice. It balances modeling power and communication cost, which is essential in decentralized settings with limited resources. Overall, this work points toward a more secure and trustworthy future for AI-enabled EV charging systems, where collaboration is possible without sacrificing privacy or robustness.
Chair, IEEE PES EICC Task Force on AI-Enabled Resilience of CPES|Clarivate HCR|AE: IEEE TSG/TSTE/TII Yang Li
Northeast Electric Power University
Read the Original
This page is a summary of: Blockchain-based fully decentralized stable federated learning charging station load forecasting method for privacy and security, Sustainable Computing Informatics and Systems, June 2026, Elsevier,
DOI: 10.1016/j.suscom.2026.101335.
You can read the full text:
Contributors
The following have contributed to this page
