“With the increase of distributed source load penetration rate, sensor monitoring data is growing rapidly, and power grid operation and maintenance services have put forward a rapid response demand for power quality data analysis. To achieve fast response and high-precision identification services for power quality disturbances, this paper proposes an image analysis based edge cloud collaborative identification framework for power quality disturbances. With the latest advances in image analysis, the concept of biphasic Lissajous trajectories is proposed to convert power quality disturbance signals into trajectory images with special shapes. Deploy lightweight convolutional neural networks with the same structure on the edge and cloud, respectively, to perform fast response and training tasks. By sharing model weights between edge and cloud, this framework can achieve fast and high-precision identification of power quality disturbances. To continuously improve model performance, design a deep convolutional neural network deployed to the cloud for data labeling to assist in model updates. The experimental results show that the framework can provide more accurate identification of power quality disturbances and meet the real-time response requirements in engineering practice.”