Multi-Target Cloud Trajectory Prediction Method Based on Improved YOLOv5s and DeepSORT
Jinwei Liao, Shuang Chen, Wen Liu, Hang Wang, Xiang Yan, Yang Yong · 2024
Ineffective tracking caused by cloud shading is an important factor affecting the service life and power generation efficiency of photovoltaic tracking systems. To solve this problem, a cloud trajectory prediction model based on improved YOLOv5s and DeepSORT is proposed. As for the improvement of YOLOv5s, firstly, 9-Mosaic algorithm was selected in the data preprocessing stage to enhance the robustness of the model. Secondly, CA module is added to Backbone to increase the positioning and recognition ability of the model. Finally, DeepSORT module was added based on the improved YOLOv5s to realize multi-target tracking. The location coordinates of the historical anchor frame fed back from the tracking process were used to fit and predict the moving trajectory of the tracking cloud through the Kalman filtering algorithm. The experimental results show that the network model can effectively predict the trajectory of the cloud cluster, and the prediction error is small, which can provide a favorable reference for optimizing the control strategy of the photovoltaic tracker.