Efficient Privacy-Preserving Federated Learning For Electricity Data

Data has become the core driving force for the development of digital economy. Electricity data is also known as weather vane of national economic operating status. There is a huge challenge in terms of mechanism, technology and security on sharing and application of energy data on a larger scale. Based on the characteristics of power data, this paper proposes an efficient model federated-training method with data privacy protected, realizing the security co-construction of analysis models. This approach is combined with federated learning and secret sharing technology, help breaking the barriers between government and power enterprises, as well as among differnet departments of power enterprises. In addition, this paper makes a detailed analysis on the security of the approach, which ensures that the data privacy can be guaranteed in the semi-honest and malicious model of no more than one server being corrupted. Finally, the proposed scheme is verified by experimental simulation, and the experimental results are compared with plaintext training. The results show that the proposed scheme still be highly efficient and practical even if the security computing technology is introduced on.