• WANG Beibei ¹

  WANG Beibei

  School of Electrical Engineering, Southeast University, Nanjing 210096, China
  在知网中查找
  在百度中查找
  在本站中查找
  
• ZHU Jing ¹

  ZHU Jing

  School of Electrical Engineering, Southeast University, Nanjing 210096, China
  在知网中查找
  在百度中查找
  在本站中查找
  
• WANG Jiale ²

  WANG Jiale

  School of Cyber Science and Engineering, Southeast University, Nanjing 211189, China
  在知网中查找
  在百度中查找
  在本站中查找
  
• MA Jinjie ³

  MA Jinjie

  Marketing Department of State Grid Jiangsu Electric Power Co., Ltd., Nanjing 210000, China
  在知网中查找
  在百度中查找
  在本站中查找
  

1.School of Electrical Engineering, Southeast University, Nanjing 210096, China;2.School of Cyber Science and Engineering, Southeast University, Nanjing 211189, China;3.Marketing Department of State Grid Jiangsu Electric Power Co., Ltd., Nanjing 210000, China

On the background of electricity market reform, the electricity consumption acquisition system of power supply companies is not open to the active distribution network operators. Meanwhile, end users prefer to keep user information locally to protect their privacy in the future, and active distribution network operators need to carry out power business such as load forecasting without meter reading right. Therefore, by selecting the weather and time factors as the correlation factors of load, a federated-learning based load forecasting framework for the protection of meter reading data of industrial users is proposed. On this basis, the industrial user data set is constructed; the load forecasting model is established based on the long- and short-term time-series network (LSTNet); and the sub-industry load forecasting framework based on the federated learning is established by using FedML framework. The case analysis shows that the proposed method can enable users in the same industry to conduct federated training without sharing load data, and support active distribution network operators to carry out relevant business on the premise of protecting users’ electricity privacy. It has better prediction performance, fewer models and shorter time consumption.

long- and short-term time-series network (LSTNet);load forecasting;federated learning;FedML framework;privacy protection