A Big Data Sharing Architecture Based on Federal Learning in State Grid

Along with constructing new power systems and the Energy Internet, much power data is generated and stored at the edge devices, which may contain customer privacy and be challenging to use. Besides, the lack of computing power and the untrustworthy environment in the edge layer prevent further development of the data. How to effectively and securely utilize the accumulated power data and facilitate the construction of new power systems and power business transformation is now a significant challenge for the State Grid Corporation of China (SGCC). This paper proposes a new data sharing architecture based on federated learning and blockchain technology, which exchanges the data model instead of the original data to ensure data security with a privacy protection protocol and a lightweight consensus for edge devices. Meanwhile, an incentive mechanism is established to encourage high-quality data share and training. In conclusion, the architecture has approximate accuracy with acceptable privacy cost compared to traditional methods with high privacy preservation. It can be widely used in big data scenarios in smart distribution or metering areas with medium-sized network scenarios.