Abstract: Aiming at the problem of people's electric shock and fall in power application scenarios, this paper proposes an electric shock fall detection method based on AlphaPose and adaptive feature weight LightGBM algorithm. Firstly, the bone key points are detected by AlphaPose. Then, eight space-time features of human posture are calculated by the data of skeleton keypoints and human body detection frame data. Next, traditional LightGBM algorithm is improved, and the feature vector is weighted according to the importance of feature, and the adaptive feature weight LightGBM classifier is obtained by training. Finally, eight space-time features of human posture are input into the adaptive feature weight LightGBM algorithm to detect the falling behaviors. Through comparisons with other detection models on Le2i public fall dataset and selfmade dataset, the experimental results show that this method has a significant improvement over the traditional detection model in terms of complexity and judgment accuracy.