The increasing penetration of distributed energy resources in active distribution networks poses significant challenges to voltage regulation and system stability. This article proposes a distributed optimal voltage control algorithm that jointly coordinates active and reactive power regulation to maintain voltage stability while satisfying operational constraints. To enhance convergence speed, a momentum-based acceleration mechanism is incorporated into the traditional primal–dual framework. Furthermore, an event-triggered communication strategy is developed to substantially reduce data exchange among neighboring agents by activating transmissions only when necessary. Rigorous theoretical analysis establishes the convergence of the proposed algorithm, and comprehensive simulation studies verify its effectiveness. The results demonstrate that the proposed algorithm can accurately regulate voltage profiles with faster convergence and significantly lower communication overhead.

The unexplainability and untrustworthiness of deep neural networks hinder their application in various high-risk fields. The existing methods lack solid evaluation metrics, interpretable models, and controllable manual manipulation. This paper presents Manual Manipulation and Decision Visualization (MMDV) which makes Human-in-the-loop improve the interpretability of deep neural networks. The MMDV offers three unique benefits: 1) The Expert-drawn CAM (Draw CAM) is presented to manipulate the key feature map and update the convolutional layer parameters, which makes the model focus on and learn the important parts by making a mask of the input image from the CAM drawn by the expert; 2) A hierarchical learning structure with sequential decision trees is proposed to provide a decision path and give strong interpretability for the fully connected layer of DNNs; 3) A novel metric, Data-Model-Result interpretable evaluation(DMR metric), is proposed to assess the interpretability of data, model and the results. Comprehensive experiments are conducted on the pre-trained models and public datasets. The results of the DMR metric are 0.4943, 0.5280, 0.5445 and 0.5108. These data quantifications represent the interpretability of the model and results. The attention force ratio is about 6.5% higher than the state-of-the-art methods. The Average Drop rate achieves 26.2% and the Average Increase rate achieves 36.6%. We observed that MMDV is better than other explainable methods by attention force ratio under the positioning evaluation. Furthermore, the manual manipulation disturbance experiments show that MMDV correctly locates the most responsive region in the target item and explains the model's internal decision-making basis. The MMDV not only achieves easily understandable interpretability but also makes it possible for people to be in the loop.

CNN (Convolutional Neural Networks) are inspired by the structure of the visual system and are one of the representative algorithms of deep learning. In recent years, Unet has become an acquaintance of Kaggle Challenge for its simplicity, efficiency and ability to extract features from small_scale samples. Since the model is designed for two classifications, there are serious problems of overfitting in using it for multiple classifications. Specifically, the model fits well in the train set, but there are many missed judgments, false judgments, and speckle noise in the test set. And the categories detected in the result graph are not balanced, some categories have better detection results, and some categories are hardly detected. To solve these problems, this paper proposes a new network that improves Unet: (1) Introduce dropout in the feature extraction stage to prevent overfitting; (2) Introduce ACNet to enhanced feature extraction capabilities. The performance of the new network which was trained with our own training dataset after data augmentation was evaluated with FWIoU and accuracy. Experimental results show that the network has higher FIWoU and accuracy rate under the same data.

Semantic segmentation in high resolution aerial images is a challenging task in remote sensing fields. Compared with other scenarios, semantic segmentation of remote sensing images requires larger receptive fields and more global information. The attention mechanism is one of the most effective way to integrate local features. In this paper, a Triple Attention Network (TANet) is proposed to get more global features. In specific, the paper introduce two self-attention module to get position attention and channel attention. And a label attention module, which generated the attention probability map by introducing spatial context information in the label. The experimental results shows that the proposed network has a higher FWIoU and PA scores than other networks.

With the development of deep learning technology, high-resolution remote sensing image change detection based on deep learning has become a hot topic in the field of remote sensing. However, the existing change detection methods based on deep learning only detect the change area of a specific object, and there is no public multi-objects change detection dataset. Focus on these problems, this paper proposed an end-to-end method to obtain the change detection results with change types for high resolution remote sensing images, including sample generation and a deep-learning network, called Res-Unet. Firstly, we obtain the label data by manual annotating. Then, co-registered image pairs are concatenated as an input for the network, and the multi-objects change detection results are directly generated by the network. The experimental results show that the method is effective and Res-Unet has a higher FWloU scores than U-Net.

DlinkNet[1] (LinkNet With Pretrained Encoder and Dilated Convolution) has been proved to be an effective method for road extraction in remote sensing fields as it won the champion in the DeepGlobe's Road Extraction Challenge. However, as the number of hyperparameters increases (such as the number of channels, filter size, etc.), the difficulty and computational overhead of network design will increase. Focused on this problem, this paper put forward effective ideas to improve D-LinkNet: (1) Applying ResNeXt as its backbone instead of ResNet to rebuild D-LinkNet; (2) Replacing initial block with stem block in the beginning of the network. The results of road extraction which was trained with our own dataset was evaluated with IoU scores. The evaluation results shows that the improved network has higher IoU scores than D-LinkNet when maintaining the model complexity and number of parameters.

Extracting building information from remote sensing (RS) images have always played an important role in civil and military. In recent years, many efficient approaches are proposed to detect building in remote sensing images. CNN (Convolutional Neural Networks) has proven to be an effective way of this problem. In this paper, to learn building features better, we propose a convolutional network based on Unet++ and containing dense connections. Our contributions are as follows:(1)Rebuild Unet++ by applying DenseNet as its backbone; (2)Add 1×1 convolution layers that can be introduced as bottleneck layer before DenseBlock to reduce the number of input feature-maps, so as reduce the parameters and improve computational efficiency. The accuracy of building extraction results from the new network which was trained with our training dataset after data augmentation was evaluated with IoU scores. The experimental results show the proposed network has higher IoU scores than Unet++ with fewer parameters.

Extracting road information from remote sensing images is one of the hot topics in image processing. The extraction result is saved as raster data, which is difficult to spatial information query, so it's necessary to convert it into vector data. Existing raster data vectorization methods are difficult to maintain the shape of roads and the connection between them, and don't take the attribute information into account. Focus on these problems, this paper proposed an algorithm for raster data vectorization. The algorithm obtains road by tracking pixels in the image, and the road is segmented by nodes (endpoints and intersections), which ensures that their connections in vector data are not disrupted. Then match them with corresponding attribute information (material and width) by image masking. Finally, vector data with attribute information is obtained. The experimental results show that the accuracy of the connection between roads is 97.5% in the vector data obtained by this method, and it has a correct rate of 95% for attribute matching.