graph-based sensor modeling

Despite the growing importance of Graph Neural Networks (GNNs) in IoT and 6G networks, no systematic survey exists to date. Method: This paper presents the first comprehensive review of GNNs for wireless systems, analyzing their technical evolution, application paradigms, and cross-layer coordination mechanisms. It synthesizes foundational architectures—including GCN, GAT, and GraphSAGE—within key wireless use cases: data fusion, intrusion detection, spectrum sensing, network optimization, and tactical communications, while integrating wireless channel modeling, distributed graph learning, and edge-coordinated inference. Contribution/Results: We propose a novel taxonomy for GNN applications in wireless networks, distill their structural modeling advantages, and systematically identify core challenges alongside 12 promising research directions. The work establishes the first authoritative, structured, and implementation-oriented theoretical framework and technical guideline for deploying GNNs in 6G intelligent air interfaces and autonomous networks.

To address the trade-off between sensor deployment cost and monitoring quality in structural health monitoring (SHM), this paper proposes the Time-Vertex Mutual Learning (TVML) framework—a novel integration of graph signal processing, temporal dynamic modeling, and machine learning. TVML overcomes the limitation of conventional methods that neglect the time-varying nature of structural behavior by jointly optimizing spatiotemporal correlations. This enables simultaneous redundancy suppression and fidelity preservation of critical information, yielding interpretable and computationally efficient sensor placement strategies. Evaluated on two real-world bridge datasets, TVML achieves equivalent or superior monitoring performance using 30–50% fewer sensors compared to baseline approaches. Moreover, it significantly improves damage detection accuracy and dynamic graph signal reconstruction fidelity, demonstrating its effectiveness for cost-sensitive, high-fidelity SHM applications.

To address the low accuracy and high computational cost of radio signal propagation modeling in real-world scenarios, this paper proposes a data-driven coverage map generation method based on Graph Neural Networks (GNNs). The approach automatically constructs a heterogeneous graph from environmental imagery, jointly encoding spatial topology and ray-based propagation relationships, and learns propagation patterns end-to-end using sparse, real-world signal measurements as supervision. This work is the first to apply GNNs to empirical radio propagation modeling, eliminating reliance on traditional physics-based solvers. Experiments demonstrate that the method surpasses classical numerical solvers (e.g., FDTD) and heuristic models in coverage map reconstruction accuracy, achieves 10–100× faster inference, and exhibits strong generalization—enabling high-fidelity signal coverage prediction from only a few measurements. These advances significantly enhance the efficiency and practicality of wireless network deployment.

Real-time state monitoring in complex systems is hindered by high sensor costs, deployment constraints, and unmeasurable critical parameters; existing virtual sensing approaches fail to model temporal couplings among heterogeneous sensors with mismatched sampling rates, disparate dynamic scales, and varying operating conditions. Method: We propose the Heterogeneous Temporal Graph Neural Network (HTGNN), the first framework to explicitly model the joint dynamic relationship between multimodal signals and operational conditions, overcoming limitations of conventional time-alignment paradigms. HTGNN integrates four core techniques: graph-structured modeling, temporal-adaptive encoding, multi-scale dynamic aggregation, and operating-condition-aware embedding. Contribution/Results: Evaluated on two newly constructed benchmark datasets—bearing load monitoring and bridge live-load estimation—HTGNN significantly outperforms state-of-the-art methods, achieving over 32% reduction in prediction error under highly variable operating conditions.

Existing topological methods struggle to jointly model directed (e.g., water flow) and undirected (e.g., pipe diameter) edge signals, while failing to distinguish the intrinsic directionality of edges themselves. To address this, we propose a direction-aware edge signal modeling paradigm that formally defines and simultaneously satisfies both direction equivariance and direction invariance—establishing the first principled framework for unified edge-level representation learning. Based on this, we introduce EIGN, the first general-purpose edge-level topological graph neural network, equipped with an algebraic-topology-inspired, direction-aware edge-level graph shift operator that rigorously preserves theoretical guarantees. Extensive experiments across traffic, hydrological, and power systems demonstrate consistent superiority over state-of-the-art methods: EIGN achieves up to 23.5% reduction in RMSE for flow simulation tasks.

Existing studies lack quantitative theoretical analysis of the differences in optimization and generalization performance between graph neural networks (GNNs) and multilayer perceptrons (MLPs). Method: Leveraging feature learning theory, this paper provides the first rigorous analysis of two-layer graph convolutional networks (GCNs) trained via gradient descent, modeling ReLU^q activation functions and spectral properties of the expected degree matrix D, under a structure-guided signal-noise separation framework. Contribution/Results: We theoretically prove that graph convolution explicitly exploits graph structure to significantly enhance signal learning while suppressing noise memorization, thereby expanding the benign overfitting regime by approximately √D^{q−2} compared to CNNs. This constitutes the first quantitative characterization of the fundamental generalization advantage of GNNs over MLPs. Extensive simulations corroborate the superior generalization and robustness of GCNs predicted by our theory.

This study addresses the challenge of uniformly processing heterogeneously sampled signals in sensor networks by proposing a graph neural network-based sampling-invariant embedding method. Through complex-valued radio frequency signal encoding and embedding learning, the proposed approach projects heterogeneously sampled signals under arbitrary topologies into a fixed-dimensional vector space. This effectively eliminates the influence of sampling discrepancies and enables topology-aware signal processing. Experimental results on synthetic signal datasets demonstrate that the method significantly enhances discriminative capability and successfully achieves waveform separation free from sampling bias, thereby overcoming the constraints inherent in traditional signal processing paradigms.

This work addresses the challenge of generalizing data-driven damage localization methods to unseen locations under sparse piezoelectric sensor networks, where limited training coverage hinders performance. The authors propose a graph learning framework that jointly performs inverse localization and forward wave response prediction. By modeling sensor layouts with graph neural networks and incorporating a physics-consistency regularization term, the method ensures that predicted damage locations adhere to guided wave propagation principles. Integrating spectral features with energy deviation prediction, the approach achieves high-precision damage localization on carbon fiber reinforced polymer plates. It significantly outperforms both non-graph and existing graph-based baselines, demonstrating notably enhanced robustness in extrapolation scenarios beyond the training domain.

This study addresses the critical yet underexplored impact of sensor placement on the performance of graph neural networks (GNNs) for leak detection in water distribution networks. To fill the gap in systematic optimization approaches, the authors propose a novel sensor placement strategy that leverages PageRank centrality to prioritize nodes based on their topological importance—a first in this domain. The work systematically evaluates how different placement schemes influence GNN performance across pressure reconstruction, forecasting, and leak detection tasks. Integrating EPANET-based hydraulic simulations with GNN models, experiments on the Net1 benchmark network demonstrate that the proposed method significantly enhances pressure reconstruction accuracy, prediction stability, and leak detection precision. These findings offer a new topology-aware perspective for optimizing sensor deployment in intelligent water infrastructure.

Traditional numerical simulations for interferometers such as LIGO are computationally expensive, struggling to balance accuracy and efficiency. To address this, this paper pioneers the application of graph neural networks (GNNs) to precision optical instrument modeling, proposing a physics-guided graph-structured approach: optical paths are explicitly encoded as graphs, with optical parameters serving as node and edge features, and training is supervised by fundamental physical constraints. Key contributions include: (1) introducing the first high-fidelity optical simulation benchmark dataset covering three canonical interferometer topologies; (2) enabling end-to-end prediction of optical responses with accuracy matching state-of-the-art numerical tools (e.g., FINESSE) and achieving an 815× speedup in inference; and (3) demonstrating strong cross-topology generalization and open-sourcing the dataset to advance the emerging “AI for Instrumentation” research direction.

This study systematically investigates the effectiveness boundaries and practical value of Graph Neural Networks (GNNs) across twelve application domains. Building upon a unified design space, it derives both spectral and spatial formulations of GNNs from first principles, analyzes their expressive power through the lens of the Weisfeiler–Leman test, and evaluates domain-specific graph construction strategies and architectural choices. The work establishes the first cross-domain analytical framework that disentangles genuine performance gains from baseline biases, uncovering common challenges such as heterophily, scaling effects, and deployment gaps. It clarifies the applicability limits of GNNs, highlights the discrepancy between leaderboard-topping models and deployable ones, and offers constraint-aware practical guidelines to address issues including oversmoothing, over-squashing, and distributional shifts.