Multi-graph Convolutional Network Research Articles

Knowledge-driven multi-graph convolutional network for brain network analysis and potential biomarker discovery

In brain network analysis, individual-level data can provide biological features of individuals, while population-level data can provide demographic information of populations. However, existing methods mostly utilize either individual- or population-level features separately, inevitably neglecting the multi-level characteristics of brain disorders. To address this issue, we propose an end-to-end multi-graph neural network model called KMGCN. This model simultaneously leverages individual- and population-level features for brain network analysis. At the individual level, we construct multi-graph using both knowledge-driven and data-driven approaches. Knowledge-driven refers to constructing a knowledge graph based on prior knowledge, while data-driven involves learning a data graph from the data itself. At the population level, we construct multi-graph using both imaging and phenotypic data. Additionally, we devise a pooling method tailored for brain networks, capable of selecting brain regions that impact brain disorders. We evaluate the performance of our model on two large datasets, ADNI and ABIDE, and experimental results demonstrate that it achieves state-of-the-art performance, with 86.87% classification accuracy for ADNI and 86.40% for ABIDE, accompanied by around 10% improvements in all evaluation metrics compared to the state-of-the-art models. Additionally, the biomarkers identified by our model align well with recent neuroscience research, indicating the effectiveness of our model in brain network analysis and potential biomarker discovery. The code is available at https://github.com/GN-gjh/KMGCN.

Multi-Source Data-Driven Local-Global Dynamic Multi-Graph Convolutional Network for Bike-Sharing Demands Prediction

The prediction of bike-sharing demand plays a pivotal role in the optimization of intelligent transportation systems, particularly amidst the COVID-19 pandemic, which has significantly altered travel behaviors and demand dynamics. In this study, we examine various spatiotemporal influencing factors associated with bike-sharing and propose the Local-Global Dynamic Multi-Graph Convolutional Network (LGDMGCN) model, driven by multi-source data, for multi-step prediction of station-level bike-sharing demand. In the temporal dimension, we dynamically model temporal dependencies by incorporating multiple sources of time semantic features such as confirmed COVID-19 cases, weather conditions, and holidays. Additionally, we integrate a time attention mechanism to better capture variations over time. In the spatial dimension, we consider factors related to the addition or removal of stations and utilize spatial semantic features, such as urban points of interest and station locations, to construct dynamic multi-graphs. The model utilizes a local-global structure to capture spatial dependencies among individual bike-sharing stations and all stations collectively. Experimental results, obtained through comparisons with baseline models on the same dataset and conducting ablation studies, demonstrate the feasibility and effectiveness of the proposed model in predicting bike-sharing demand.

Fault detection and isolation for multi-type sensors in nuclear power plants via a knowledge-guided spatial–temporal model

Sensor faults in nuclear power plants (NPPs) have the potential to propagate negative impacts on system stability, leading to false alarms and accident misdiagnosis. Existing methods seldom concurrently consider complex spatial–temporal correlations among multi-type sensors in the primary circuit. This study presents a novel sensor fault detection and isolation scheme named the knowledge-guided spatial–temporal model (KGSTM), using the knowledge-guided recurrent unit (KGRU) and the concurrent detection strategy. To organically express part and whole interdependencies from inherent sensor layout, several graphs are specifically designed with pertinent domain knowledge. KGRU consists of the multi-graph convolutional network (MGCN) for fusing various spatial information and the gate recurrent unit (GRU) for extracting dynamic temporal features, further obtaining precise reconstructed signals and residuals. The concurrent detection strategy can explicitly quantify abnormal behaviors to detect and isolate faulty sensors by characterizing spatial–temporal signal variation. Numerical results on two real-world datasets from a pressurized water reactor (PWR) with simulated faults illustrate that the KGSTM has superior performance over various state-of-the-art methods in terms of signal reconstruction and fault detection.

Parallel framework of a multi-graph convolutional network and gated recurrent unit for spatial–temporal metro passenger flow prediction

Metro passenger flow prediction is a critical problem in metro transport systems. However, recent studies have either overlooked spatial information on the metro network or primarily focused on modeling spatial dependencies using only the physical topology. To achieve accurate metro passenger flow (inflow and outflow at each station of a network) prediction, this study proposes a joint prediction model that combines the multi-graph convolution network and the gated recurrent unit (GRU). In addition to exploring location topology relationships, the proposed model selects two non-Euclidean spatial dependencies in metro passenger flow prediction to design essential graph elements as part of the stacked spatial block. Three spatial relationships (adjacency, similarity, and correlation) are integrated in parallel with the GRU network. The metro passenger flow prediction framework ASC-GRU (adjacency, similarity, correlation, and gated recurrent unit) is designed to mitigate the distortion of results during the capturing of passenger flow spatial–temporal features. Finally, ASC-GRU is tested using two datasets from the Hangzhou and Shanghai metro networks in China, and the error metrics of different models are compared and analyzed to verify the effectiveness and feasibility of ASC-GRU. The test results demonstrate that the proposed model outperforms other baseline models in passenger flow prediction over long time intervals and large networks. In particular, compared with the best performance of the baselines, the average reduction is around 3%, 12% and 13% in metrics of MAPE, MAE and RMSE, respectively.

Predicting associations between drugs and G protein-coupled receptors using a multi-graph convolutional network

Developing new drugs is an expensive, time-consuming process that frequently involves safety concerns. By discovering novel uses for previously verified drugs, drug repurposing helps to bypass the time-consuming and costly process of drug development. As the largest family of proteins targeted by verified drugs, G protein-coupled receptors (GPCR) are vital to efficiently repurpose drugs by inferring their associations with drugs. Drug repurposing may be sped up by computational models that predict the strength of novel drug-GPCR pairs interaction. To this end, a number of models have been put forth. In existing methods, however, drug structure, drug-drug interactions, GPCR sequence, and subfamily information couldn’t simultaneously be taken into account to detect novel drugs-GPCR relationships. In this study, based on a multi-graph convolutional network, an end-to-end deep model was developed to efficiently and precisely discover latent drug-GPCR relationships by combining data from multi-sources. We demonstrated that our model, based on multi-graph convolutional networks, outperformed rival deep learning techniques as well as non-deep learning models in terms of inferring drug-GPCR relationships. Our results indicated that integrating data from multi-sources can lead to further advancement.

A Data-driven Base Station Sleeping Strategy Based on Traffic Prediction

Due to the rapidly increasing number of base stations (BSs) in the operational cellular networks, their energy consumption is escalating. In this paper, we propose an intelligent data-driven BS sleeping mechanism relying on a wireless traffic prediction model that measures the BSs' capacity in different regions. Firstly, a spatio-temporal cellular traffic prediction model is proposed, where a multi-graph convolutional network (MGCN) is developed to capture the associated spatial features. Furthermore, a multi-channel long short-term memory (LSTM) solution involving hourly, daily and weekly periodic data is used to capture the relevant temporal features. Secondly, the capacities of macro-cell BSs (MBSs) and small-cell BSs (SBSs) having different environment characteristics are modeled, where both clustering and transfer learning algorithms are adopted for quantifying the traffic supported by the MBSs and SBSs. Finally, an optimal BS sleeping strategy is proposed for minimizing the network's power consumption. Experimental results show that the proposed MGCN-LSTM model outperforms the existing models in terms of its cellular traffic prediction accuracy, and the proposed BS sleeping strategy using an approximated non-linear model of the associated capacity function achieves near-maximal energy-saving at a modest complexity.

PLU-MCN: Perturbation learning enhanced U-shaped multi-graph convolutional network for traffic flow prediction

Traffic flow prediction is a challenging task in intelligent transportation systems. To improve the accuracy of traffic flow prediction, graph convolutional neural networks and traffic networks are usually combined to reveal dynamic spatial-temporal correlations. However, the existing graph convolution methods need to repeatedly preprocess data of the same type due to network fluctuations, hardware failures, and other troublesome factors. To address this challenge, a novel Perturbation Learning enhanced U-shaped Multi-graph Convolutional Network (PLU-MCN) is proposed. Specifically, PLU-MCN consists of a Perturbation Learning (PL) module and a U-shaped Multi-graph Convolutional Network (U-MCN) based on encoder-decoder architecture. The PL module is designed based on residual convolution and residual transpose convolution to learn perturbation-resistant features from historical time series in the presence of perturbations. The U-MCN consists of multi-graph convolutional modules based on U-Net to extract dynamic spatial-temporal features from the internal and external information of the perturbation-resistant features and output the final predicted values. Compared with existing state-of-the-art models, the PLU-MCN model demonstrated excellently predictive accuracy and reliability in experiments based on five real-world datasets. The key source code and data are available at https://github.com/Bounger2/PLUMCN.

Prediction of production indicators of fractured-vuggy reservoirs based on improved Graph Attention Network

Various storage and seepage spaces exist in fractured-vuggy carbonate reservoirs composed of multi-scale dissolution pores and fractures. The frequent regulation of the working system causes nonlinear and unstable production data along with complex water breakthrough characteristics, which lead to difficulty in real-time prediction. Traditional methods based on the water drive curve and autoregressive machine learning ignore the spatial correlation among production wells and local geological characteristics. To address these problems, we propose a model based on improved attention, the spatiotemporal multi-graph convolutional network (STMCN), for production prediction. The unit production wells are abstracted as directed graph network nodes to establish adjacency, connectivity and correlation graphs to extract the spatial and semantic features from different perspectives through graph convolutional networks. To depict the law of fluid movement, we realize the fusion of spatial and semantic information through dynamic routing. Aiming at the autocorrelation characteristics of production sequences, the model uses the self-attention mechanism to capture the dependencies in production sequences. The gating mechanism is designed to achieve dynamic production prediction by adaptively aggregating the spatiotemporal characteristics. This paper evaluates the predictive performance of our model by two real-world datasets of the Tahe Oilfield. The results show that the mean absolute error (MAE), root mean squared error (RMSE), and mean absolute percentage error (MAPE) of our proposed model for the prediction of well A2 were 3.19, 4.66 and 0.07, respectively, which were better than the relatively new baseline model.

Model-based real-time prediction of surface roughness in fused deposition modeling with graph convolutional network-based error correction

Additively manufactured parts usually have relatively rough surface finish due to the layer-by-layer process. Extensive research has been conducted on the effect of process parameters on surface roughness as well as on real-time monitoring of surface roughness using sensor data. However, very few studies consider both process parameters and condition monitoring data to perform real-time prediction of surface roughness. To address this research gap, we introduce an augmented deep learning framework that integrates a model-based predictor with a deep learning-based error corrector to predict the surface roughness of parts fabricated via fused deposition modeling. The model-based predictor predicts surface roughness by considering the effects of process parameters on surface roughness, while the deep learning-based error corrector estimates the prediction errors between the true surface roughness and predicted surface roughness using real-time sensor data collected during additive manufacturing. To consider both correlations among features extracted from sensor signals and temporal correlations of each sensor signal, we construct multiple graphs to reveal these correlations and introduce a multi-graph convolutional network to analyze these undirected graphs. Experimental results show that the proposed framework outperforms existing data-driven methods reported in the literature.

A multi-graph convolutional network based wearable human activity recognition method using multi-sensors

A multi-graph convolutional network based wearable human activity recognition method using multi-sensors

Spatio-temporal multi-graph convolutional network based on wavelet analysis for vehicle speed prediction

The research on theoretical techniques and practical applications of vehicle speed prediction holds significant importance in mitigating the increasingly pressing traffic issues and enhancing the operational efficiency of urban transportation systems. This paper aims to improve the accuracy of traffic speed prediction. To achieve this, we have devised a multi-graph convolutional neural network (GCN) framework that utilizes data from three different patterns: recent, daily, and weekly, enabling the extraction of distinct time-related features. Through the integration of multiple GCNs and 3D convolutional neural networks (3DCNN), we have obtained richer spatiotemporal feature representations. Additionally, we have applied wavelet analysis to process the traffic speed data from each node in the road network, unearthing hidden information and eliminating noise interference. Our experiments have demonstrated that the model employing wavelet analysis outperforms the baseline model without wavelet analysis, thus enhancing the prediction accuracy. In conclusion, we propose a wavelet-based spatiotemporal multi-graph convolutional neural network for traffic speed prediction. Our experimental results validate the superiority of this model across different time intervals and real-world datasets, showcasing its advantages in terms of prediction accuracy and robustness. As a result, our approach offers an effective and viable solution for traffic speed prediction tasks.

VGC-GAN: A multi-graph convolution adversarial network for stock price prediction

Not only market signals but also disturbances of related companies influence the stock volatility of a company. Currently, most approaches that utilize inter-stock correlations rely on predetermined graphs, such as industry graphs, shareholder graphs, or event graphs, necessitating supplementary financial data. Consequently, this approach often leads to limited generalization ability of the model and incomplete representation of inter-stock relationships. Thus, this study introduces VGC-GAN, a multi-graph convolutional adversarial framework, for predicting stock prices. Initially, the proposed model generates multiple correlation graphs by analyzing historical stock data, providing a comprehensive depiction of inter-stock correlations from diverse perspectives. Subsequently, a Generative Adversarial Network (GAN) framework, augmented with Mean Square Error (MSE) loss, is constructed to enhance the predictive performance of the model. The framework combines Multi-Graph Convolutional Network (Multi-GCN) and Gated Recurrent Unit (GRU) as the generator. It undergoes supervised and adversarial training with Convolutional Neural Network (CNN), facilitating the in-depth exploration of hidden correlations between stocks and time dependence of stocks. To mitigate the impact of noise, VGC-GAN uses subsequences after Variational Mode Decomposition (VMD) with optimized parameters as input to the generator. The proposed model is evaluated on several real datasets, and the experimental results confirm its effectiveness in stock price prediction.

Trajectory-Based Spatiotemporal Multi-Task Multi-Graph Network for Traffic State Prediction

As a critical application in intelligent transportation systems, traffic state prediction still faces various challenges, such as unsatisfactory capability of utilizing multi-source data and modeling spatiotemporal network relevancies. Therefore, we propose a trajectory-based multi-task multi-graph convolutional network (Tr-MTMGN), a novel spatiotemporal deep learning framework for traffic state prediction on a citywide scale. This method firstly mines the underlying information from vehicle trajectories and designs a multi-graph convolution block to investigate spatial correlations. Sequentially, the multi-head self-attention layer is integrated into the multi-task learning framework to capture the temporal dependencies of the traffic state. The proposed model was evaluated on field data collected in Zhangzhou, China, and demonstrated superior performance when compared with existing state-of-the-art baselines.

Predicting traffic propagation flow in urban road network with multi-graph convolutional network

AbstractTraffic volume propagating from upstream road link to downstream road link is the key parameter for designing intersection signal timing scheme. Recent works successfully used graph convolutional network (GCN) and specific time-series model to forecast traffic flow by capturing the spatial–temporal features. However, accurately predicting traffic propagation flow (tpf) is challenging, since the classical GCN model only considers the influence of adjacent road link. In complex urban road network, specific traffic propagation flow (tpf) is affected by various spatial features, such as adjacent tpf, which influences from tpf with same upstream link and tpf with same downstream link. Thus, we proposed a multi-graph learning-based model named TPP-GCN (traffic propagation prediction-graph convolutional network) in this paper to predict the traffic propagation flow in urban road network. The TPP-GCN model captures not only the temporal features but also multi-spatial features based on multi-layer convolution. We validated the model using real-world traffic flow data derived from taxi GPS data in Shenzhen, China. Finally, we compare and evaluate the proposed model with the existing models across several prediction scales.

PMGCN: Progressive Multi-Graph Convolutional Network for Traffic Forecasting

Traffic forecasting has always been an important part of intelligent transportation systems. At present, spatiotemporal graph neural networks are widely used to capture spatiotemporal dependencies. However, most spatiotemporal graph neural networks use a single predefined matrix or a single self-generated matrix. It is difficult to obtain deeper spatial information by only relying on a single adjacency matrix. In this paper, we present a progressive multi-graph convolutional network (PMGCN), which includes spatiotemporal attention, multi-graph convolution, and multi-scale convolution modules. Specifically, we use a new spatiotemporal attention multi-graph convolution that can extract extensive and comprehensive dynamic spatial dependence between nodes, in which multiple graph convolutions adopt progressive connections and spatiotemporal attention dynamically adjusts each item of the Chebyshev polynomial in graph convolutions. In addition, multi-scale time convolution was added to obtain an extensive and comprehensive dynamic time dependence from multiple receptive field features. We used real datasets to predict traffic speed and traffic flow, and the results were compared with a variety of typical prediction models. PMGCN has the smallest Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) results under different horizons (H = 15 min, 30 min, 60 min), which shows the superiority of the proposed model.

Attention-based spatial–temporal multi-graph convolutional networks for casualty prediction of terrorist attacks

At present, terrorism has become an important factor affecting world peace and development. As the time series data of terrorist attacks usually show a high degree of spatial–temporal correlation, the spatial–temporal prediction of casualties in terrorist attacks is still a significant challenge in the field of counter-terrorism. Most of the existing terrorist attack prediction methods lack the ability to model the spatial–temporal dynamic correlation of the time series data of terrorist attacks, so they cannot yield satisfactory prediction results. In this paper, we propose a novel Attention-based spatial–temporal multi-graph convolutional network (AST-MGCN) for casualty prediction of terrorist attacks. Specifically, we construct the spatial adjacency graph and spatial diffusion graph based on the different social-spatial dynamic relationships of terrorist attacks and determine the multi-scale period of time series data of terrorist attacks by using wavelet transform to model the temporal trend, period and closeness properties of terrorist attacks. The AST-MGCN mainly consists of spatial multi-graph convolution for extracting social-spatial features in multi-views and temporal convolution for capturing the transition rules. In addition, we also use the spatial–temporal attention mechanism to effectively capture the most relevant spatial–temporal dynamic information. Experiments on public datasets demonstrate that the proposed model outperforms the state-of-the-art baselines.

A data-driven model for production prediction of strongly heterogeneous reservoir under uncertainty

Production forecasting plays an important role in the reservoir development process. This task is challenging due to complex spatial and temporal correlations (e.g., inter-well interference and forward-backward linkages temporally). Most existing work tried to solve this problem by mining the impact of historical production conditions on the future production from the perspective of time series analysis. While such works gain satisfactory prediction accuracy, they cannot take the effect of well interference into account. In this study, we revisit the problem of production prediction from strongly heterogeneous reservoirs, aiming to improve the prediction accuracy by leveraging the breakthroughs of deep learning techniques. A new data-driven model is proposed to predict the production of each oil well, where the key novelty is to mine the inter-well interference relationships from the graph perspective. First, based on four constructed inter-well relationship graphs (well pattern structure graph, distance graph, similarity graph, and injection-production relationship graph), a Multi-Graph Convolutional Network (MGCN) is built to capture the spatial interactions of the wells. Then, a Long Short-Term Memory (LSTM) network is used to capture the complex temporal correlations. Finally, to make the model consider more production influencing factors, a Fully Connected Network (FCN) is utilized to fuse spatiotemporal correlations with other production influencing factors to predict production. Besides, considering the uncertainty of the data, an uncertainty evaluation method for predicting production is developed using Latin hypercube sampling and Monte Carlo methods. For evaluation and verification, the proposed model is applied to two case studies of numerical models and actual oilfields. The data of the actual oilfield are preprocessed by the proposed ensemble empirical mode decomposition (EEMD) component threshold denoising method. The optimal hyperparameters in the model are determined by Bayesian optimization algorithm. A series of experiments demonstrated that the proposed method has minimal prediction error and can predict the production with different probabilities, which can reduce the development risk.

A Domain-Adversarial Multi-Graph Convolutional Network for Unsupervised Domain Adaptation Rolling Bearing Fault Diagnosis

The transfer learning method, based on unsupervised domain adaptation (UDA), has been broadly utilized in research on fault diagnosis under variable working conditions with certain results. However, traditional UDA methods pay more attention to extracting information for the class labels and domain labels of data, ignoring the influence of data structure information on the extracted features. Therefore, we propose a domain-adversarial multi-graph convolutional network (DAMGCN) for UDA. A multi-graph convolutional network (MGCN), integrating three graph convolutional layers (multi-receptive field graph convolutional (MRFConv) layer, local extreme value convolutional (LEConv) layer, and graph attention convolutional (GATConv) layer) was used to mine data structure information. The domain discriminators and classifiers were utilized to model domain labels and class labels, respectively, and align the data structure differences through the correlation alignment (CORAL) index. The classification and feature extraction ability of the DAMGCN was significantly enhanced compared with other UDA algorithms by two example validation results, which can effectively achieve rolling bearing cross-domain fault diagnosis.

Fine-Grained Vessel Traffic Flow Prediction With a Spatio-Temporal Multigraph Convolutional Network

The accurate and robust prediction of vessel traffic flow is gaining importance in maritime intelligent transportation system (ITS), such as vessel traffic services, maritime spatial planning, and traffic safety management, etc. To achieve fine-grained vessel traffic flow prediction, we will first generate the maritime traffic network (which is essentially a graph), and then propose a graph-driven neural network. In particular, to represent various correlations among spatio-temporal vessel traffic flow, we tend to extract the feature points (i.e., starting, way and ending points) by utilizing the knowledge of vessel positioning data. These feature points are essentially related to the geometrical structures of massive vessel trajectories collected from massive automatic identification system (AIS) records, contributing to the generation of maritime traffic network. We then propose a spatio-temporal multi-graph convolutional network (STMGCN)-based vessel traffic flow prediction method by exploiting multiple types of inherent correlations in the generated maritime graph. The proposed STMGCN mainly contains one spatial multi-graph convolutional layer and two temporal gated convolutional layers, beneficial for extracting spatial and temporal traffic flow patterns. The main benefit of our graph-driven prediction method is that it takes full advantage of the maritime graph and multi-graph learning. Comprehensive experiments have been implemented on realistic AIS dataset to compare our method with several state-of-the-art prediction methods. The fine-grained prediction results have demonstrated our superior performance in terms of both accuracy and robustness.

False data injection attacks detection with modified temporal multi-graph convolutional network in smart grids

False Data Injection Attack (FDIA) detection can prevent the tampering of important data in the smart grid. This is of great significance to the operation and control of modern power systems. Since the existing FDIA detection methods are limited by the sequential input of data in Euclidean space, they cannot accurately describe the compelling correlation between data components. Therefore, this paper proposes a novel FDIA localization detection method based on graph data modeling and graph deep learning. The proposed approach tries to disaggregate the primary data into graph structured data with graph topological relationships based on graph theory, then designs specialized networks for data with different graph topologies. Moreover, the designed multi-graph mechanism and temporal correlation layer can better fully mine the correlation features between data components, with its attribute characteristics, to construct deep learning on the specific graph topology for FDIA detection. Extensive simulation experiments and visualization show that the proposed scheme is more effective than the conventional detection model, and its overall accuracy in 14-bus, 118-bus and 300-bus systems is 98.3%, 96.4% and 95.8%. It also proves that this scheme has high robustness and generalization ability in different scenarios.