Grounding grid corrosion can alter current dissipation characteristics and reduce the safety and reliability of grounding systems. Since traditional excavation-based inspection methods are costly and disruptive, non-excavation identification of local corrosion is of practical significance. To achieve this goal, a corrosion localization model based on surface potential features and a BP neural network is established. First, CDEGS is used to establish simulation models of a straight-line grounding grid under normal and different corrosion conditions, and the surface potential distribution characteristics under impulse current are analyzed. Then, the peak values of surface potentials at 20 observation points are extracted as input features to identify 15 single-point corrosion conditions, and three training algorithms, namely LM, BR, and SC, are compared. The results show that local corrosion causes local distortion of the surface potential, and the BP neural network can effectively identify different corrosion locations. Among the three algorithms, the BR algorithm is more suitable for the corrosion localization problem under the small-sample condition considered in this paper.

Bird nesting is an important ecological factor threatening the safe operation of transmission lines. In areas with rich biodiversity, bird damage has become a major hidden danger for power grids. To reveal the disaster-causing mechanism and evaluate its effect on power-system reliability, this study uses 234 bird-nest fault records and meteorological data of the Shanghai Power Grid from 2018 to 2022. Three factors, namely ecological niche overlap, nest-building behavior, and environmental amplification, are integrated into a three-level risk model. The spatiotemporal distribution characteristics and ecological mechanisms of faults are analyzed, and the comprehensive impact on power-supply reliability and grid criticality is quantified. The results show that bird-nest faults exhibit a bimodal seasonal distribution from February to April and from July to August. High-incidence areas are concentrated in Fengxian, Jinshan, Chongming, and Pudong New Area, whereas no such faults are recorded in the city center. All faults are caused by nest-building behavior and are closely related to bird breeding phenology as well as meteorological factors such as spring humidity, rainfall, summer typhoons, and strong winds. The monthly risk ranking produced by the Niche-Behavior-Environment (NBE) model is highly consistent with the actual fault ranking (Spearman $\rho=0.867, p\lt 0.01$). Independent spatial comparison with bird-watching records further validates the model’s ecological plausibility. By introducing a weighted correction based on grid criticality, the risk levels of key links such as radial lines and important-user feeders are significantly improved. This study extends ecological mechanism analysis to power-system reliability management and provides a quantitative and risk-stratified decision-making basis for bird-damage prevention and control in urban power grids.

Recognizing the coreference relationship between different event mentions in the text (i.e., event coreference resolution) is an important task in natural language processing. It helps to understand the association between various events in the text, and plays an important role in information extraction, question answering systems, and reading comprehension. Existing research has made progress in improving the performance of event coreference resolution, but there are also some shortcomings. For example, most of the existing methods analyze the event data in the document in a serial processing mode, without considering the complex relationship between events, and it is difficult to mine the deep semantics of events. To solve these problems, this paper proposes a cross-document event co-reference resolution method (HGCN-ECR) based on hypergraph convolutional neural networks. Firstly, the BiLSTM-CRF model was used to label the semantic role of the events extracted from a number of documents. According to the labeling results, the trigger words and non-trigger words of the event were determined, and the multi-document event hypergraph was constructed around the event trigger words. Then hypergraph convolutional neural networks are used to learn higher-order semantic information in multi-document event hypergraphs, and multi-head attention mechanisms are introduced to understand the hidden features of different event relationship types by treating each event relationship as a set of separate attention mechanisms. Finally, the feed-forward neural network and the average link clustering method are used to calculate the coreference score of events and complete the coreference event clustering, and the cross-document event coreference resolution is realized. The experimental results show that the cross-document event co-reference resolution method is superior to the baseline model.

With the implementation of national greening policies, bird related faults in transmission lines have become increasingly serious, and have become the third largest transmission line fault problem after lightning strikes and external damage. In previous studies on electric birds, strong stimuli such as electricity, sound, and light were mostly used for bird prevention, and there was a lack of research on the relationship between bird activity patterns and line fault data. This study is based on public collection of bird watching data and compares the situation of power facility failures in Shanghai. The results show that the high incidence areas of bird related failures coincide with bird activity areas, and bird related failures in transmission lines are closely related to bird habits and bird diversity, both of which are directly affected by the surrounding environment. By analyzing these survey data, the distribution and activity patterns of these birds were studied, providing a basis for bird line symbiosis and achieving a combination of ecological protection and power grid construction.

At present, event causality extraction still relies heavily on intra-sentence causal clues to infer intersentence causal relations. However, many existing approaches fail to effectively capture the essential information of event structures and their causal interactions. To address these limitations, this paper introduces a dual-grid graph pairwise network for event causality extraction. Specifically, a MoE-MLP hybrid expert network is first employed to encode the text and obtain refined event semantic representations, which enhances prediction accuracy. Next, dual-grid labeling is utilized to identify causal event pairs along with their argument roles, while simultaneously learning parameter information associated with event types. Finally, a graph-based pairwise attention network is constructed to detect intra-sentence causal relations and further infer implicit inter-sentence causality. Experimental results indicate that the proposed approach substantially enhances the performance of event causality extraction.

Text-to-SQL generation is an important area of natural language processing. It can help non-specialists interact with databases using natural language, simplify the database query process, improve efficiency, enhance the user experience, etc. Existing work on Text-to-SQL mainly utilizes large-scale pre-trained models to improve the performance of model generation. Despite progress, Text-to-SQL still has some shortcomings in some respects, such as the discrepancy of words in natural languages, inaccurate scheme links, and inadequate domain generalization capabilities. In this paper, we present an SQL generating framework,which enhancing the interaction graph of data schema and syntactic structure with pre-trained language model for Text-to-SQL(SGIS), aimed at improving the domain generalization of models and the ability of models to deal with cross-cutting questions. Specifically, we first introduce a model linking method based on a pre-trained model, extracting input NL question and database scheme relationship structures to solve the scheme linking question between the NL questions in the model and database models. On this basis, the sentence in the input NL question is extracted from the reliant information and integrated into the well-structured chart data to solve the incomplete question of the relationship characteristics embedded in it. At the same time, in order to prevent the over-adaptation of embedded sides during the training process during the optimization process, we use a type-coding method to help the model effectively differentiate between the type of relationship that the sentence depends on when embedding sides, thereby reducing unnecessary entanglement. Numerous experiments have proven that SGIS's performance on both data sets Spider and Spider-SYN under standard settings is due to all comparative base lines.

The three-dimensional model of medical sequence image can display human tissues and organs in stereo, and effectively overcome the insufficiency of two-dimensional examination results for disease diagnosis. In this paper, the three-dimensional reconstruction of sequential medical images is studied from the perspective of image super resolution and semantic segmentation. In view of the problem of ignoring global information association in existing image super-resolution reconstruction methods and the problem of long distance dependence due to lack of information in existing image semantic segmentation methods. In this paper, the method of fusion of global and local features is introduced into the super-resolution reconstruction task, and the method of cross-level fusion compensation is introduced into the semantic segmentation task of medical images, so as to realize the three-dimensional reconstruction of sequential medical images. Finally, the reliability of the proposed method is verified by experiments. In the experiment of 3D reconstruction of medical sequence images, we can observe that some details are preserved completely and the clarity is improved significantly after reconstruction.

Visual question-answering is an important application of the fusion of vision and language in multimodal learning. Its basic task is to understand the input 2D image or 3D point cloud and answer text questions based on it. At present, the development of visual question and answering technology is facing some problems. In some fields, it is difficult to obtain accurate answers by inferring only the information contained in images and problems. The data imbalance in practical application scenarios also limits the ability of models to handle rare or complex problems. In this article, we propose an image question-answering model based on problem-guided hybrid learning and knowledge embedding. This model queries relevant knowledge in the knowledge graph and integrates the queried knowledge with the problem text to form new textual information. During the training process, tuples of similar problem types (v, q, a) are mixed to generate new data samples. Then, feature extraction and cross modal fusion are performed on the new samples, and the samples are fed into the answer prediction network to obtain the answers. The optimization model narrowed the gap between predicted answers and mixed answers, ultimately enabling the model to generate more accurate answers. A series of comparative experiments conducted on the OKVQA dataset and the SLAKE dataset has verified that the model can effectively improve the accuracy of image question-answering.

The point cloud data structure is characterized by disorder and spatial irregularity, which makes it impossible to apply 2D convolutional neural networks directly to extract features like regular data such as images and text, so the point cloud classification task faces a significant challenge. This study aims to classify the inferior mesenteric artery (IMA) in the form of point clouds, the key of which is to analyze the correlation between its branch origin and morphological changes. Therefore, we propose PointSGLN, a point cloud classification network based on normalization after sampling and grouping operations. Unlike previous point cloud classification networks, normalization is performed after point cloud sampling and grouping operations, which reduces the complexity of the point cloud representation while preserving the geometric properties of the original point cloud to ensure that the branching structure of the inferior mesenteric artery (IMA) is adequately preserved and improves the classification accuracy by optimizing the training strategy to ensure the inference speed while improving the classification accuracy. The proposed PointSGLN is tested on the ModelNet40, ScanObjectNN and IMA. From the experimental results, it is shown that the advanced performance of this model on the point cloud classification task and its good robustness to point sparsity are verified.

In big data era, multi-source heterogeneous data become the biggest obstacle to data sharing due to its high dimension and inconsistent structure. Using text classification to solve the ontology construction and mapping problem of multi-source heterogeneous data can not only reduce manual operation, but also improve the accuracy and efficiency. This paper proposes an ontology construction and mapping scheme based on hybrid neural network and autoencoder. Firstly, the proposed text classification method uses the multi-core convolutional neural network to capture local features and uses the improved Bidirectional Long Short-Term Memory network to compensate for the shortcomings of the convolutional neural network that cannot obtain context-related information. Secondly, a similarity matching method is used for ontology mapping, which integrate autoencoder to improve anti-interference ability. We have carried out several sets of experiments to test the validity of the proposed ontology construction and mapping scheme.

In community question answering, many questions have no topic labeling or the topic labeling is very diverse, which has become the biggest obstacle to building the bridge between users and posts. Topic clustering methods could alleviate this issue. However, existing research employed words as topic representation units and could not express topic semantic relevance. In this paper, we propose a novel Topic Clustering framework based on the Graph Neural Network (called TCGNN) to alleviate topic diversity in Community Question Answering. Firstly, we separately consider the relationship representation of existing topics and unlabeled topics. For manually labeled topics, we count the frequency of topics in community questions and construct a topic co-occurrence matrix to represent the topic relation. For unmarked topics, we extract the core phrases from community questions and employ them to indicate the topics of questions. Then, we transform the topic co-occurrence matrix into a topic relation graph, optimizing the topic relevance and improving presentation efficiency. Next, we employ a graph neural network for embedding the topic connection graph and get the vector representation of each topic. Finally, an improved K-mean method is proposed for topic clustering based on the distance of topic vectors. Additionally, we briefly discuss the extended effect of topic clustering methods in other domains (bibliographic information and reviews). In the literature we have, it is a primary work that conders topic clustering in multiple situations and offers innovative cogitation to apply graph neural networks in topic clustering. Our experiment compared prevalent clustering methods and some combination methods of text representation and graph embedding. The outcome of experiments on four extensive and varied datasets (Stack Overflow, DBLP, Yelp, and Zhihu) illustrate that TCGNN leads the prevalent baseline in Entropy and Purity.

Climate change is a major global issue of general concern to the international community. As a developing country focusing on environmental protection, China has adopted a series of relevant policies and measures to address global warming, namely, achieving carbon peak by 2030 and carbon neutrality by 2060. It can be said that carbon peaking is China's national policy to actively respond to climate change, and is also an independent commitment based on scientific demonstration; Carbon neutrality is an action goal based on China's national conditions and is also a long-term development strategy with a long-term vision. In order to realize the important long-term strategy of “carbon neutrality” at an early date, low-carbon buildings will gradually come into people's lives. This paper first analyzes the crisis and challenges we are facing under the background of the current era, and expounds how to solve the energy crisis. In this paper, we propose that green and low-carbon buildings will make a significant contribution to solving the problem of carbon pollution. Then, we studied the low carbon scheme of science and technology parks in southern cities through literature review, and case study. Next, taking the project of Block H01-01, Unit PDC1-0401, Lingang New Area of Shanghai Free Trade Zone as an example, we 、analyzed the technical characteristics of the project's climate responsive design and the integrated design of renewable energy buildings, and elaborated its project functions. While satisfying people's demands for scientific research and comfortable living, it also responded to the call of the national “double carbon” goal and completed the task of “strengthening building energy conservation and continuously improving building energy efficiency” in the key work of energy conservation, emission reduction and carbon reduction in Shanghai. Finally, it is concluded that green and low-carbon buildings can make outstanding contributions to the realization of “double carbon” goals in the future, and at the same time provide a more powerful guarantee for human living environment and quality of life.

Knowledge reasoning is a key technology for constructing knowledge graphs and realizing knowledge graph completion. A single category of reasoning has its own advantages and disadvantages, so hybrid reasoning, which integrates the advantages of multiple single reasoning methods, has become a hot research direction. This paper will briefly introduce the background and definition of hybrid reasoning, introduce the classification of knowledge reasoning oriented to knowledge graph in detail, and then focus on the research progress, future research directions and possible challenges of hybrid reasoning in recent years.

In recent years, the related research of entity alignment has mainly focused on entity alignment via knowledge embeddings and graph neural networks; however, these proposed models usually suffer from structural heterogeneity and the large-scale problem of knowledge graph. A novel entity alignment model based on graph isomorphic network and compressed sensing is proposed. First, for the problem of structural heterogeneity, graph isomorphic network encoder is applied in knowledge graph to capture structural similarity of entity relation. Second, for the problem of large scale, key node and community are integrated for priority entity alignment to improve execution speed. However, the exiting node importance ranking algorithm cannot accurately identify key node in knowledge graph. So the compressed sensing is adopted in node importance ranking to improve the accuracy of identifying key node. The authors have carried out several experiments to test the effect and efficiency of the proposed entity alignment model.

At present, the relation completion mainly study the influence of single-path or first-order information, but ignoring the more complex relation information widely existing between entities. Meanwhile, recommendation systems based on collaborative filtering algorithms are susceptible to data sparsity issues, leading to a cold start of the system. Therefore, a novel embedding learning framework for relation completion and recommendation based on graph neural network and multi-task learning is proposed. The graph neural network relation completion model GNNRC predicts the relation between entities by embedding learning, which fuses the high-order semantic features of the two target entities’ subgraph based on graph neural network. A comparison with TransD model on embeding learning for relation completion is reported. On this basic, the multi-task learning relation recommendation model MLRR add the bridging unit into deep end-to-end recommendation model, which realizes the alternating learning of knowledge graph embedding and recommendation algorithm. Experimental results show that the performance of the proposed recommendation model is significantly better than other baselines. Moreover, strong recommendation performance can be maintained in cold start scenarios where data are sparse.

With the development of science and technology, science and technology policies are increasing year by year. Science and technology policies are literature existing in the form of texts, which are characterized by rigorous structure, clear hierarchy, and standard language. Mining template information from policies can optimize data templates and improve the efficiency of recommending data to users. This paper proposes a joint entity relation extraction model based on capsule networks and part-of-speech weighting. In order to learn more feature information from word vector, capsule network based on bidirectional gated cyclic unit is used to replace the traditional convolutional neural network. In view of the phenomenon of imperfect semantic expression of word vector, part-of-speech features are added to enrich text information. Meanwhile, in order to solve the weight distribution problem of word features and part-of-speech features, an artificial fish swarm algorithm is proposed to optimize the two feature weights by iterative optimization, and the effectiveness of the proposed model is proved by experiments.

Relation extraction is the core mission and a great significant part of natural language processing. This paper briefly expound the development of relation extraction and introduces the commonly used document-level relation extraction datasets and evaluation indexes of model effects. According to the different representation of entity by model, documentlevel relation extraction can be divided into sequential method and graph method. Besides, we conduct contrastive analysis concerning different relation extraction models, and compared the effects of various relation extraction models. Finally, we summarize the key research contents of document-level relation extraction in the future and forecast the development trend.

This paper summarizes the main methods of knowledge representation learning. Representation learning represents the entity information of the knowledge graph as a low dimensional vector. Its vector dimension is low, which helps to improve the computational efficiency and make full use of the semantic information between entities. In order to embed two KGs into a unified semantic space, joint embedding is used to achieve this goal. With the development of research, there are many improved embedding methods, such as iteration, multi view embedding, knowledge graph entity alignment based on graph neural network and so on.

Big data is massive and heterogeneous, along with the rapid increase in data quantity, and the diversification of user access, traditional database, and access control methods can no longer meet the requirements of big data storage and flexible access control. To solve this problem, an entity relationship completion and authority management method is proposed. By combining the weighted graph convolutional neural network and the attention mechanism, a knowledge base completion model is given. On this basis, the authority management model is formally defined and the process of multilevel trust access control is designed. The effectiveness of the proposed method is verified by experiments, and the authority management of knowledge base is more fine-grained and more secure.

With the increasing complexity of scientific research, it has gradually turned to a collaborative approach, which can promote knowledge sharing, resource sharing and improve the efficiency of scientific research achievements. Therefore, It is of great significance to study the internal organizational structure and evolution mechanism of scientific research collaboration, which plays a crucial role in the management of scientific research work and the formulation of scientific and technological policies. This paper focuses on three aspects: core node evaluation, community detection and visual layout algorithm of scientific research collaboration network, which is constructed based on the network embedding of the scientific research achievements' attributes. Considering network topology and node heterogeneity, a core node evaluation method is proposed, and a community detection algorithm and a visual layout algorithm is improved to display the community structure of scientific research collaboration network from many aspects. The experimental results show that the proposed method can more clearly show the internal structure of scientific research collaboration community. (c) 2021 Elsevier B.V. All rights reserved.