Forward Scheduling Optimization Decision-making Method for Power Grid Transmission Channels based on Graph Reinforcement Learning

After replacing the dominant position of deterministic controllable power sources such as coal-fired power with intermittent and fluctuating power sources such as wind power and photovoltaics, the operational characteristics, safety control, and production mode of the power grid have fundamentally changed, resulting in complex and variable power flow in the power system. This poses great challenges to the scheduling and flexible adjustment of the power grid, especially affecting the operation and control of key transmission channels in the power grid. At present, the dispatching system mainly adopts the technical means of sensitivity adjustment and global optimization calculation of physical model. Based on the graph reinforcement learning technology, this paper breaks through the dependence of existing methods on model and prediction data, fully excavates historical data, and combines dispatching experience data to learn and generate a global disposal strategy considering time series characteristics, which effectively reduces the difficulty of solving risk disposal strategies for key transmission channels of the power grid, and improves the online practicality of power grid daily dispatching.