Towards a cybersecure and privacy enhanced smart grid: A blockchain enabled federated learning framework.

Area of Science:

• Smart Grid Technology
• Cybersecurity
• Machine Learning

Background:

• Smart grids enhance energy efficiency but pose significant privacy and security risks.
• Centralized machine learning and existing federated learning systems struggle with data privacy and identity verification.
• Blockchain solutions often lack robust privacy mechanisms and traceability.

Purpose of the Study:

• To develop a secure and privacy-preserving architecture for smart grids.
• To address vulnerabilities in federated learning and blockchain-based systems.
• To enhance data privacy, security, and forecasting accuracy in energy consumption prediction.

Main Methods:

• A dual-layer blockchain architecture integrated with federated learning (FL) and central differential privacy (DP).
• Secure logging of client interactions using salted cryptographic hashes for traceability and anonymity.
• Gaussian noise addition to aggregated model parameters for central DP, mitigating inference attacks.

Main Results:

• The proposed framework effectively enhances data privacy and security in smart grid environments.
• Random Forest model achieved high accuracy on the PRECON dataset with MAE of 0.153, MAPE of 0.085, and MSE of 0.143.
• The system successfully preserved forecasting accuracy while improving privacy and security.

Conclusions:

• The integrated dual-layer blockchain, FL, and DP architecture offers a robust solution for smart grid privacy and security challenges.
• This approach ensures user anonymity and data integrity without compromising energy demand forecasting performance.
• The framework demonstrates significant improvements in protecting sensitive consumer data within smart grid systems.