Early Warning Method of Power Grid Meteorological Disasters Based on Scene Recognition

The current meteorological disaster is one of the initial causes of grid failure. This paper proposes a power grid meteorological disaster forecasting method based on scene classification and recognition for the complex relationship between various meteorological factors and power grids. In this paper, hazard factors are divided into three categories: meteorological factors, equipment factors, and geographical environmental factors. Based on meteorological raw data and power grid historical data, stacked auto-encoder network is used to perform unsupervised feature extraction and supervised fine-tuning to realize the recognition of meteorological scenes, and establish a coupling relationship between factors and power grid faults. The contraction is added to constrain the auto-encoder network, effectively suppress the noise of the input data, and improve the robustness of the auto-encoder network. The calculation analysis shows that the method in this paper can comprehensively and accurately establish the relationship between disasters and power grid faults. It can accurately predict whether the factors will cause grid faults and fault scenarios, and provide more targeted information for the disaster prevention work of the grid.