Graph Convolutional Network Research Articles

Novel graph convolutional network for geological profile prediction using non-equidistant borehole data

ABSTRACT Accurately predicting geological profiles from sparse borehole data is challenging due to limited data availability and complex spatial correlations within strata. This study presents an advanced graph convolutional network designed to model flexible graph structures and predict geological profiles using unequally spaced borehole data. In this framework, each graph node corresponds to a stratum unit which represents the transformed coordinate features, with edges indicating spatial relationships among these units. Initially, the method was applied to a case study in Australia, demonstrating a 30% improvement in boundary accuracy compared to traditional methods. Subsequently, the approach was employed in a foundation project in Hangzhou, achieving mean measurement accuracies above 0.8 for validating boreholes. The study further explored the impact of borehole quantity and unequal spacing on prediction accuracy. Results indicate that significant accuracy gains occur only when new boreholes add critical spatial features; otherwise, the improvement remains minimal. Additionally, flexible borehole positioning results in a 44% disparity between the most and least accurate predictions.

Graph-Root: Prediction of Root-Associated Proteins in Maize, Sorghum, And Soybean Based on Graph Convolutional Network and Network Embedding Method

Background: The root system plays an irreplaceable role in plant growth. Its improvement can increase crop productivity. However, such a system is still mysterious for us. The underlying mechanism has not been fully uncovered. The investigation on proteins related to the root system is an important means to complete this task. In the previous time, lack of root-related proteins makes it impossible to adopt machine learning methods for designing efficient models for the discovery of novel root-related proteins. Recently, a public database on root-related proteins was set up and machine learning methods can be applied in this field. Objective: The purpose of this study was to design an efficient computational method to predict root-associated proteins in three plants: maize, sorghum, and soybean. Method: In this study, we proposed a machine learning based model, named Graph-Root, for the identification of root-related proteins in maize, sorghum, and soybean. The features derived from protein sequences, functional domains, and one network were extracted, where the first type of features were processed by graph convolutional neural network and multi-head attention, the second type of features reflected the essential functions of proteins, and the third type of features abstracted the linkage between proteins. These features were fed into the fully connected layer to make predictions. Results: The 5-fold cross-validation and independent tests suggested its acceptable performance. It also outperformed the only previous model, SVM-Root. Furthermore, the importance of each feature type and component in the proposed model was investigated. Conclusion: Graph-Root had a good performance and can be a useful tool to identify novel rootrelated proteins. BLOSUM62 features were found to be important in determining root-related proteins.

Extended multi-stream temporal-attention module for skeleton-based human action recognition (HAR)

Graph convolutional networks (GCNs) are an effective skeleton-based human action recognition (HAR) technique. GCNs enable the specification of CNNs to a non-Euclidean frame that is more flexible. The previous GCN-based models still have a lot of issues: (I) The graph structure is the same for all model layers and input data. GCN model's hierarchical structure and human action recognition input diversity make this a problematic approach; (II) Bone length and orientation are understudied due to their significance and variance in HAR. For this purpose, we introduce an Extended Multi-stream Temporal-attention Adaptive GCN (EMS-TAGCN). By training the network topology of the proposed model either consistently or independently according to the input data, this data-based technique makes graphs more flexible and faster to adapt to a new dataset. A spatial, temporal, and channel attention module helps the adaptive graph convolutional layer focus on joints, frames, and features. Hence, a multi-stream framework representing bones, joints, and their motion enhances recognition accuracy. Our proposed model outperforms the NTU RGBD for CS and CV by 0.6% and 1.4%, respectively, while Kinetics-skeleton Top-1 and Top-5 are 1.4% improved, UCF-101 has improved 2.34% accuracy and HMDB-51 dataset has significantly improved 1.8% accuracy. According to the results, our model has performed better than the other models. Our model consistently outperformed other models, and the results were statistically significant that demonstrating the superiority of our model for the task of HAR and its ability to provide the most reliable and accurate results.

AGCNAF: predicting disease-gene associations using GCN and multi-head attention to fuse the similarity features

Abstract Association prediction between diseases and genes is a critical step in revealing the molecular mechanisms of diseases and developing drug treatment strategies. With the explosive growth of data in the biomedical field, how to effectively utilize these data for accurate prediction has become a hotspot and challenge in current research. To overcome the limitations of current prediction methods in dealing with complex biological network structures and feature extraction, this study proposes AGCNAF, a method that combines an unsupervised Graph Convolutional Network (GCN) and a multi-head attention mechanism. The metagraph-guided random walk strategy enables AGCNAF to capture local and high-order topological structures in the graph, while GCN is responsible for realizing deep feature extraction of these structures. By incorporating similarity features through the multi-attention mechanism, AGCNAF achieves effective integration of global and local features, which significantly improves the prediction performance. By incorporating similarity features through the multi-attention mechanism, AGCNAF achieves effective integration of global and local features, which significantly improves the prediction performance. By utilizing the machine learning binary classification model for prediction, the experimental results through five-fold cross-validation show that AGCNAF demonstrates significant advantages in prediction performance compared to existing methods, with its AUC and AUPR reaching 0.9686 and 0.9709, respectively, and the AUC up to 0.9812 under specific conditions. 
To verify the practical application value of AGCNAF, this study also conduct case studies on Alzheimer's disease, lung cancer, and breast cancer. The results further confirm the excellent performance of AGCNAF in identifying potential disease-gene associations, which opens up new possibilities for future disease-gene research.

Predicting variable-length ACE inhibitory peptides based on graph convolutional network

Traditional molecular descriptors have contributed to the prediction of angiotensin I-converting enzyme (ACE) inhibitory peptides, but they often fall short in capturing the complex structure of the molecule. To address these limitations, this study introduces molecular graphs as an advanced method for peptide characterization. Peptides containing 2–10 amino acids were represented using molecular graphs, and a graph convolutional network (GCN) model was constructed to predict variable-length peptides. This model was compared with machine learning (ML) models based on molecular descriptors, including Random Forest (RF), Support Vector Machine (SVM), and k-Nearest Neighbor (kNN), under the same benchmark. Notably, the GCN model outperformed the other models with an accuracy of 0.78, effectively identifying ACE inhibitory potential. Furthermore, the GCN model also demonstrated superior performance, exceeding existing methods with an accuracy rate of over 98 % on an independent test set. To validate our predictions, we synthesized peptides VAPE and AQQKEP with high predicted probabilities, and their IC50 values of 2.25 ± 0.11 and 3.75 ± 0.17 μM, respectively, indicating potent ACE inhibitory activity. The developed GCN model presents a powerful tool for the rapid screening and identification of ACE inhibitory peptides, offering promising opportunities for developing antihypertensive components in functional foods.

Identification of mild cognitive impairment using multimodal 3D imaging data and graph convolutional networks

Abstract Objective. Mild cognitive impairment (MCI) is a precursor stage of dementia characterized by mild cognitive decline in one or more cognitive domains, without meeting the criteria for dementia. MCI is considered a prodromal form of Alzheimer's disease (AD). Early identification of MCI is crucial for both intervention and prevention of AD. To accurately identify MCI, a novel multimodal 3D imaging data integration graph convolutional network model is designed in this paper. Approach. The proposed model utilizes 3D-VGGNet to extract three-dimensional features from multimodal imaging data (such as sMRI and FDG-PET), which are then fused into feature vectors as the node features of a population graph. Non-imaging features of participants are combined with the multimodal imaging data to construct a population sparse graph. Additionally, in order to optimize the connectivity of the graph, we employed the Pairwise Attribute Estimation (PAE) method to compute the edge weights based on non-imaging data, thereby enhancing the effectiveness of the graph structure. Subsequently, a population-based graph convolutional network (GCN) integrates the structural and functional features of different modal images into the features of each participant for MCI classification. Main results. Experiments on the ADNI demonstrated accuracies of 98.57%, 96.03%, and 96.83% for the NC-EMCI, NC-LMCI, and EMCI-LMCI classification tasks, respectively. The AUC, specificity, sensitivity, and F1-score are also superior to state-of-the-art models, demonstrating the effectiveness of the proposed model. Furthermore, the proposed model is applied to the ABIDE dataset for autism diagnosis, achieving an accuracy of 91.43% and outperforming the state-of-the-art models, indicating excellent generalization capabilities of the proposed model. Significance. This study demonstrate the proposed model’s ability to integrate multimodal imaging data and its excellent ability to recognize MCI. This will help achieve early warning for AD and intelligent diagnosis of other brain neurodegenerative diseases.

Enhancing Portfolio Optimization: A Two-Stage Approach with Deep Learning and Portfolio Optimization

The portfolio selection problem has been a central focus in financial research. A complete portfolio selection process includes two stages: stock pre-selection and portfolio optimization. However, most existing studies focus on portfolio optimization, often overlooking stock pre-selection. To address this problem, this paper presents a novel two-stage approach that integrates deep learning with portfolio optimization. In the first stage, we develop a stock trend prediction model for stock pre-selection called the AGC-CNN model, which leverages a convolutional neural network (CNN), self-attention mechanism, Graph Convolutional Network (GCN), and k-reciprocal nearest neighbors (k-reciprocal NN). Specifically, we utilize a CNN to capture individual stock information and a GCN to capture relationships among stocks. Moreover, we incorporate the self-attention mechanism into the GCN to extract deeper data features and employ k-reciprocal NN to enhance the accuracy and robustness of the graph structure in the GCN. In the second stage, we employ the Global Minimum Variance (GMV) model for portfolio optimization, culminating in the AGC-CNN+GMV two-stage approach. We empirically validate the proposed two-stage approach using real-world data through numerical studies, achieving a roughly 35% increase in Cumulative Returns compared to portfolio optimization models without stock pre-selection, demonstrating its robust performance in the Average Return, Sharp Ratio, Turnover-adjusted Sharp Ratio, and Sortino Ratio.

Siamese Graph Convolutional Split-Attention Network with NLP based Social Sentimental Data for enhanced stock price predictions

Predicting stock market behavior using sentiment analysis has become increasingly popular, as customer responses on platforms like Twitter can influence market trends. However, most existing sentiment-based models struggle with two major issues: inaccuracy and high complexity. These problems lead to frequent prediction errors and make the models difficult to implement in real-time trading systems. To address these challenges, this paper proposes a new method called Siagra-ConSA-HSOA (Siamese Graph Convolutional Split-Attention Network with NLP-based Social Sentiment Data). Two data sources feed the model: specifically, NIFTY-50 Stock Market and real-time Twitter sentiment. Through Natural Language Processing (NLP), the raw data is pre-processed and key features are extracted before they are fused into a unified dataset using a cross-domain transformer, namely CDSFT, and then Circle-Inspired Optimization Algorithm (CIOA) selects the most important features from this dataset. This decreases the complexity of the model without losing essential information. Finally, a Graph Convolutional Split-Attention Network (SGCSAN) for promisingly predicting whether the stock prices are going to hit the ground and fly high again or is going to nosedive with Humboldt Squid Optimization Algorithm (HSOA) is introduced to further improve accuracy with lesser error generation. The proposed model Siagra-ConSA-HSOA achieved 99.9% accuracy and 99.8% recall in the testing stage, meaning that such a model performs better than the current approaches both in prediction accuracy and efficiency. Thus, this is a glimmer that the model shall be able to overcome some of the main problems with the current techniques used in predicting the behavior of the stock market.GitHub Repository: https://github.com/jramans2/Siamese-GCN-SplitAttention-Stock-Prediction.git

A Study of Recommendation Methods Based on Graph Hybrid Neural Networks and Deep Crossing

In the face of complex user behavior patterns and massive data, improving the performance of recommender system models is an urgent challenge. Traditional methods often struggle to effectively handle feature interactions and complex user-item relationships. Combining the advantages of graph neural networks and the Deep Crossing network, this paper proposes a recommendation method based on hybrid neural networks with Deep Crossing (Deep Crossing with Graph Convolution and GRU, DCGCN-GRU). First, by constructing the graph structure of users and items, higher-order feature representations are extracted, and node features are updated using a multilayer graph convolution operation. Then, the higher-order features learned by the graph convolution network are spliced and weighted with the original features to form new feature inputs. Next, a Gated Recurrent Unit (GRU) is introduced to capture the inter-feature temporal dynamic relationships and sequence information. Finally, the Deep Crossing model is utilized to learn the interactions between the fused features at multiple levels and enhance the interactions between the features. Comparative experiments on three public datasets, MovieLens-ml-25m, Book-Crossings, and Amazon Reviews’23, show that the model achieves significant improvements in accuracy, mean square error (MSE), and mean absolute error (MAE).

Fully exploring object relation interaction and hidden state attention for video captioning

Video Captioning (VC) is a challenging task of automatically generating natural language sentences for describing video contents. As a video often contains multiple objects, it is comprehensively crucial to identify multiple objects and model relationships between them. Previous models usually adopt Graph Convolutional Networks (GCN) to infer relational information via object nodes, but there exist uncertainty and over-smoothing issues of relational reasoning. To tackle these issues, we propose a Knowledge Graph based Video Captioning Network (KG-VCN) by fully exploring object relation interaction, hidden state and attention enhancement. In encoding stages, we present a Graph and Convolution Hybrid Encoder (GCHE), which uses an object detector to find visual objects with bounding boxes for Knowledge Graph (KG) and Convolutional Neural Network (CNN). To model intrinsic relations between detected objects, we propose a knowledge graph based Object Relation Graph Interaction (ORGI) module. In ORGI, we design triplets (head, relation, tail) to efficiently mine object relations, and create a global node to enable adequate information flow among all graph nodes for avoiding possibly missed relations. To produce accurate and rich captions, we propose a hidden State and Attention Enhanced Decoder (SAED) by integrating hidden states and dynamically updated attention features. Our SAED accepts both relational and visual features, adopts Long Short-Term Memory (LSTM) to produce hidden states, and dynamically update attention features. Unlike existing methods, we concatenate state and attention features to predict next word sequentially. To demonstrate the effectiveness of our model, we conduct experiments on three well-known datasets (MSVD, MSR-VTT, VaTeX), and our model achieves impressive results significantly outperforming existing state-of-the-art models.

Accelerating spin Hall conductivity predictions via machine learning

AbstractAccurately predicting the spin Hall conductivity (SHC) is crucial for designing novel spintronic devices that leverage the spin Hall effect. First‐principles calculations of SHCs are computationally intensive and unsuitable for quick high‐throughput screening. Here, we have developed a residual crystal graph convolutional neural network (Res‐CGCNN) deep learning model to classify and predict SHCs solely based on the structural and compositional information. This is enabled by having access to 9249 instances of SHCs data and incorporating extra residual networks into the standard CGCNN framework. We found that Res‐CGCNN surpasses CGCNN, achieving a mean absolute error of 115.4 (ℏ/e) (S/cm) for regression and an area under the receiver operating characteristic curve of 0.86 for classification. Additionally, we utilized Res‐CGCNN to conduct high‐throughput screenings of materials in the Materials Project database that were absent in the training set. This led to the prediction of several previously unreported materials displaying large SHCs exceeding 1000 (ℏ/e) (S/cm), which were validated through first‐principles calculations. This study represents the inaugural endeavor to construct a machine learning model capable of effectively capturing the intricate nonlinear relationship between SHCs and crystal structure and composition, serving as a useful tool for the efficient screening and design of materials exhibiting high SHCs.

Graph Neural Network-Based Speech Emotion Recognition: A Fusion of Skip Graph Convolutional Networks and Graph Attention Networks

In speech emotion recognition (SER), our research addresses the critical challenges of capturing and evaluating node information and their complex interrelationships within speech data. We introduce Skip Graph Convolutional and Graph Attention Network (SkipGCNGAT), an innovative model that combines the strengths of skip graph convolutional networks (SkipGCNs) and graph attention networks (GATs) to address these challenges. SkipGCN incorporates skip connections, enhancing the flow of information across the network and mitigating issues such as vanishing gradients, while also facilitating deeper representation learning. Meanwhile, the GAT in the model assigns dynamic attention weights to neighboring nodes, allowing SkipGCNGAT to focus on both the most relevant local and global interactions within the speech data. This enables the model to capture subtle and complex dependencies between speech segments, thus facilitating a more accurate interpretation of emotional content. It overcomes the limitations of previous single-layer graph models, which were unable to effectively represent these intricate relationships across time and in different speech contexts. Additionally, by introducing a pre-pooling SkipGCN combination technique, we further enhance the ability of the model to integrate multi-layer information before pooling, improving its capacity to capture both spatial and temporal features in speech. Furthermore, we rigorously evaluated SkipGCNGAT on the IEMOCAP and MSP-IMPROV datasets, two benchmark datasets in SER. The results demonstrated that SkipGCNGAT consistently achieved state-of-the-art performance. These findings highlight the effectiveness of the proposed model in accurately recognizing emotions in speech, offering valuable insights and a solid foundation for future research on capturing complex relationships within speech signals for emotion recognition.

Federated Learning-Based Traffic Flow Prediction Model in Intelligent Transportation Systems

The existing Intelligent Transportation System (ITS) achieves high success. As an essential component of ITS, Traffic Flow Prediction (TFP) has attracted tremendous attention. It is a critical but challenging task to improve the robust convergence and high accuracy of TFP in real-world scenarios. This study presents an optimized Federated Learning (FL)-based ChebNet model, FedproxChebNet, to realize the highly effective and accurate TFP. By selecting the best penalty constant in the proximal term to optimize the objective function, this model can achieve fast and stable convergence. Using the ChebNet model to aggregate neighbor nodes’ characteristics, more hidden information underlying the spatio-temporal traffic data can be taken into consideration for training the global and local models. All the superiority of the FedproxChebNet model makes it outperform other FL models with the Graph Convolutional Network (GCN), Graph Attention Network (GAT) and Spatio-Temporal Graph Convolutional Network (STGCN). We designed a series of experiments on various FL-based GNN model comparisons, parameter sensitivity tests, and on verifying the performance of FedproxChebNet with different heterogeneous systems with [Formula: see text], [Formula: see text] and [Formula: see text]. Based on four real-world data sets from the cognitive network, the experimental results demonstrate that the presented FedproxChebNet provides the best convergence and the highest accuracy in TFP, and achieves the best performance in a highly heterogeneous system ([Formula: see text]). Specifically, the accuracy of FedproxChebNet is at least improved by [Formula: see text] on PeMS07 than other FL-based GNN models. This proposed FedproxChebNet model may be preferable for different scenarios in ITS such as route planning, traffic congestion control and reversible lanes.

Tree sequences as a general-purpose tool for population genetic inference.

As population genetics data increases in size new methods have been developed to store genetic information in efficient ways, such as tree sequences. These data structures are computationally and storage efficient, but are not interchangeable with existing data structures used for many population genetic inference methodologies such as the use of convolutional neural networks (CNNs) applied to population genetic alignments. To better utilize these new data structures we propose and implement a graph convolutional network (GCN) to directly learn from tree sequence topology and node data, allowing for the use of neural network applications without an intermediate step of converting tree sequences to population genetic alignment format. We then compare our approach to standard CNN approaches on a set of previously defined benchmarking tasks including recombination rate estimation, positive selection detection, introgression detection, and demographic model parameter inference. We show that tree sequences can be directly learned from using a GCN approach and can be used to perform well on these common population genetics inference tasks with accuracies roughly matching or even exceeding that of a CNN-based method. As tree sequences become more widely used in population genetics research we foresee developments and optimizations of this work to provide a foundation for population genetics inference moving forward.

DGCNN approach links metagenome-derived taxon and functional information providing insight into global soil organic carbon

Metagenomics can provide insight into the microbial taxa present in a sample and, through gene identification, the functional potential of the community. However, taxonomic and functional information are typically considered separately in downstream analyses. We develop interpretable machine learning (ML) approaches for modelling metagenomic data, combining the biological representation of species with their associated genetically encoded functions within models. We apply our methods to investigate soil organic carbon (SOC) stocks. First, we combine a diverse global set of soil microbiome samples with environmental data, improving the predictive performance of classic ML and providing new insights into the role of soil microbiomes in global carbon cycling. Our network analysis of predictive taxa identified by classical ML models provides context for their ecological significance, extending the focus beyond just the most predictive taxa to ‘hidden’ features within the model that might be considered less predictive using standard methods for explainability. We next develop unique graph representations for individual microbiomes, linking microbial taxa to their associated functions directly, enabling predictions of SOC via deep graph convolutional neural networks (DGCNNs). Interpretation of the DGCNNs distinguished between the importance of functions of key individual species, providing genome sequence differences, e.g., gene loss/acquisition, that associate with SOC. These approaches identify several members of the Verrucomicrobiaceae family and a range of genetically encoded functions, e.g., related to carbohydrate metabolism, as important for SOC stocks and effective global SOC predictors. These relatively understudied but widespread organisms could play an important role in SOC dynamics globally.

Personalized movie recommendation in IoT-enhanced systems using graph convolutional network and multi-layer perceptron

Exploring the optimization of communication strategies for animation films in the context of cross-cultural communication, this research integrates the Internet of Things (IoT) and convolutional networks. The research constructs a collaborative filtering (CF) movie recommendation model based on a graph convolutional neural network (GCN) and investigates its application in cross-cultural communication. The fusion of IoT and convolutional networks in movie communication is also analyzed, and the effectiveness of the proposed GCN-CF model is validated through comparative experiments. The results indicate that, compared to other models, the GCN-CF model achieves the lowest Root Mean Square Error (RMSE) on the MovieLens 100 K and MovieLens 1 M datasets, with values of 0.8762 and 0.8275, respectively. Compared to traditional models, the GCN-CF model exhibits significantly superior performance in terms of RMSE, with reductions ranging from 0.6 to 5.2%, highlighting its heightened detection accuracy and overall performance. Moreover, the performance of the GCN-CF model is enhanced after introducing attention mechanisms and auxiliary information on both datasets, showing an improvement of 0.4% compared to the scenario without these additions. This data demonstrates the effectiveness of attention mechanisms and auxiliary information. Finally, the research presents an animation film communication strategy based on IoT and convolutional networks, offering novel insights for film production and communication, along with positive implications for cultural exchange and the advancement of the global media industry.

SOAMC: A Semi-Supervised Open-Set Recognition Algorithm for Automatic Modulation Classification

Traditional automatic modulation classification methods operate under the closed-set assumption, which proves to be impractical in real-world scenarios due to the diverse nature of wireless technologies and the dynamic characteristics of wireless propagation environments. Open-set environments introduce substantial technical challenges, particularly in terms of detection effectiveness and computational complexity. To address the limitations of modulation classification and recognition in open-set scenarios, this paper proposes a semi-supervised open-set recognition approach, termed SOAMC (Semi-Supervised Open-Set Automatic Modulation Classification). The primary objective of SOAMC is to accurately classify unknown modulation types, even when only a limited subset of samples is manually labeled. The proposed method consists of three key stages: (1) A signal recognition pre-training model is constructed using data augmentation and adaptive techniques to enhance robustness. (2) Feature extraction and embedding are performed via a specialized extraction network. (3) Label propagation is executed using a graph convolutional neural network (GCN) to efficiently annotate the unlabeled signal samples. Experimental results demonstrate that SOAMC significantly improves classification accuracy, particularly in challenging scenarios with limited amounts of labeled data and high signal similarity. These findings are critical for the practical identification of complex and diverse modulation signals in real-world wireless communication systems.

Pre-connected and trainable adjacency matrix-based GCN and neighbor feature approximation for industrial fault diagnosis

Industrial fault diagnosis methods based on graph convolution network (GCN) becomes a hot topic for its great feature extraction ability to multivariate time-series data. However, GCNs ignore inter-sample temporality when constructing the adjacency matrix (AM), leading to low prediction accuracy. A novel fault diagnosis method based on pre-connected and trainable AM-based GCN and neighbor feature approximation (PTGCN-FA) is proposed at the node-level task. Firstly, PTGCN-FA introduces the temporal nearest neighbors into spatial nearest neighbors to pre-connect and construct the AM. Then, the AM is trained only where the samples are connected, which makes the best weights obtained and reduces the time complexity of the model. Finally, after the GCN layers, the trained AM is introduced into the approximation of features, which are neighbors in the original sample space. Two process industry cases are carried out, and the simulation results including diagnosis accuracy, confusion matrix, study to the ratio of labeled data and an ablation experiment verify PTGCN-FA has more efficient and accurate diagnostic performance than related methods. Additionally, the analysis of the temporal neighborhood weight parameter shows that the performance of fault diagnosis can be improved by considering both temporal and spatial information between samples.

Sparse graph structure fusion convolutional network for machinery remaining useful life prediction

Effective prediction of machinery remaining useful life (RUL) is prominent to achieve intelligent preventive maintenance in manufacturing systems. In this paper, a sparse graph structure fusion convolutional network (SGSFCN) is proposed for more accurate end-to-end RUL prediction of machine. A novel node-level graph structure called time series shapelet distance graph (TSSDG) is designed to convert the time series to node feature. The SGSFCN model is proposed to learn degradation information from the graph structure. In SGSFCN, a sparse graph structure (SGS) layer and a fusion graph structure (FGS) layer preceding the graph convolutional network (GCN) are designed to learn the SGS from node representation and fuse the original graph structure, enabling the graph structure and node update iteratively in subsequent layers. Concurrently, a bidirectional long short-term memory network (BiLSTM) layer is integrated to capture the global temporal dependencies. The method is validated by two test rig data, and results demonstrate that the proposed method offers significantly higher prediction accuracy of RUL compared to several state-of-art methods.

MSA-GCN: Multistage Spatio-Temporal Aggregation Graph Convolutional Networks for Traffic Flow Prediction

Timely and accurate traffic flow prediction is crucial for stabilizing road conditions, reducing environmental pollution, and mitigating economic losses. While current graph convolution methods have achieved certain results, they do not fully leverage the true advantages of graph convolution. There is still room for improvement in simultaneously addressing multi-graph convolution, optimizing graphs, and simulating road conditions. Based on this, this paper proposes MSA-GCN: Multistage Spatio-Temporal Aggregation Graph Convolutional Networks for Traffic Flow Prediction. This method overcomes the aforementioned issues by dividing the process into different stages and achieves promising prediction results. In the first stage, we construct a latent similarity adjacency matrix and address the randomness interference features in similarity features through two optimizations using the proposed ConvGRU Attention Layer (CGAL module) and the Causal Similarity Capture Module (CSC module), which includes Granger causality tests. In the second stage, we mine the potential correlation between roads using the Correlation Completion Module (CC module) to create a global correlation adjacency matrix as a complement for potential correlations. In the third stage, we utilize the proposed Auto-LRU autoencoder to pre-train various weather features, encoding them into the model’s prediction process to enhance its ability to simulate the real world and improve interpretability. Finally, in the fourth stage, we fuse these features and use a Bidirectional Gated Recurrent Unit (BiGRU) to model time dependencies, outputting the prediction results through a linear layer. Our model demonstrates a performance improvement of 29.33%, 27.03%, and 23.07% on three real-world datasets (PEMSD8, LOSLOOP, and SZAREA) compared to advanced baseline methods, and various ablation experiments validate the effectiveness of each stage and module.