一种结合光伏组件表面灰尘图像的污染功率损失范围量化模型OA

To effectively quantify the power loss range of PV modules resulting from soiling,an improved multimodal model named PLQ-Net that combines PV module image and environmental infor-mation with dual branch structure is proposed.Initially,dynamic convolution is used to enhance the ability of backbone to extract soiled PV module image features.Then,data normalization and non-linear function transformation are carried out independently for environmental data at different scales.This targeted approach enables the model to extract the characteristics of environmental data more effectively.Lastly,combining recursive architecture and iterative learning strategy,a multimodal iterative fusion network based on dynamic multilayer perceptron is proposed to achieve effective feature fusion between soiled PV module images and environmental data.To verify the effectiveness of PLQ-Net,ablation experiments and comparative experiments are conducted on the PV-Net dataset.The experimental results show that the accuracy of PLQ-Net reaches 87.18%,which is higher than those of the image-only and multimodal models.