The operation and maintenance data of power grid show that the main causes of the power grid fault have shifted from the level of manufacturing technology of electric equipment and the level of on-site operation and maintenance to natural weather factors such as thunder and lightning

mountain fire

gale and icy disaster. Disaster prevention and mitigation of power grid should also focus on meteorological disaster. Aiming at the characteristics and regularities of association between meteorological and power grid faults

a method of grid weather disaster mitigation based on improved stacked denoising auto-encoder(SDAE)network is proposed. Based on the meteorological historical data and grid operation and maintenance data

the synthetic minority over-sampling technique(SMOTE)is used to reduce the imbalance of the original data set. Auto-encoder network completes the extraction of meteorological information features and the establishment of the relationship meteorological information and grid faults through unsupervised self-learning and supervised fine-tuning

and improves the robustness of the network by incorporating sparse term restrictions and noise-enhanced coding. The case study shows that the proposed SMOTE-SDAE-based meteorological disaster mitigation method can establish the correlation mapping relationship between meteorological information and power grid fault accurately and completely

and can make accurate prediction for whether the given meteorological conditions will cause grid disaster accidents or not.