Scholarlink

<p>This paper presents a machine learning (ML)-based localization framework for event-driven localization in wireless sensor networks (WSNs) to address the limitations of traditional multi-sequence positioning (MSP) and event-based techniques. The proposed method formulates node localization as a supervised regression problem, where sensor coordinates are predicted from feature vectors comprising event detection times, anchor node proximities, and spatial event information. A lightweight multilayer perceptron (MLP) architecture is employed due to its efficiency and ability to capture nonlinear spatiotemporal patterns without the computational burden of graph- or sequence-heavy models. The model is trained on synthetically generated datasets that emulate enhanced multi-sequence positioning (EMSP) behavior, incorporating noisy event propagation and variable node placements. Large-scale EMSP-style data generation and robustness benchmarking require high-throughput computation; therefore, the proposed evaluation and training pipeline is designed to leverage GPU acceleration and parallel/distributed execution across deployment scenarios. This computing-oriented workflow enables rapid, reproducible localization studies for large WSNs while preserving lightweight inference suitable for real-time edge operation. Extensive experiments are conducted under sparse and dense deployment scenarios, with comparative benchmarking against EMSP, RSSI, ToA, EventMamba, FasterKAN, and GNN-based localization methods. Results demonstrate that the proposed approach achieves superior accuracy, reduced prediction variance, and robust anchor utilization across varying topologies. Notably, under the most adverse tested conditions, including severe timing noise, 10% event packet loss, and imperfect synchronization, the proposed model maintains graceful degradation and continues to outperform EMSP and signal-based baselines. Furthermore, interpretability is enhanced through spatial visualization tools, including 2D true vs. predicted maps and Voronoi-based consistency analysis. Overall, the framework offers a scalable and generalizable solution, positioning event-driven ML-based localization in WSNs as an effective alternative to conventional strategies in resource-constrained environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Event-driven localization in wireless sensor networks: a machine learning approach

  • 期刊论文

  • 发表日期：

    2026年2月9日

查看全文
  • Mansoor Iqbal,
  • Sadia Waheed,
  • Inayat Ali,
  • Syed Zarak Shah,
  • Syed Ahmad Saleem Bokhari,
  • Shahid Sultan

摘要

This paper presents a machine learning (ML)-based localization framework for event-driven localization in wireless sensor networks (WSNs) to address the limitations of traditional multi-sequence positioning (MSP) and event-based techniques. The proposed method formulates node localization as a supervised regression problem, where sensor coordinates are predicted from feature vectors comprising event detection times, anchor node proximities, and spatial event information. A lightweight multilayer perceptron (MLP) architecture is employed due to its efficiency and ability to capture nonlinear spatiotemporal patterns without the computational burden of graph- or sequence-heavy models. The model is trained on synthetically generated datasets that emulate enhanced multi-sequence positioning (EMSP) behavior, incorporating noisy event propagation and variable node placements. Large-scale EMSP-style data generation and robustness benchmarking require high-throughput computation; therefore, the proposed evaluation and training pipeline is designed to leverage GPU acceleration and parallel/distributed execution across deployment scenarios. This computing-oriented workflow enables rapid, reproducible localization studies for large WSNs while preserving lightweight inference suitable for real-time edge operation. Extensive experiments are conducted under sparse and dense deployment scenarios, with comparative benchmarking against EMSP, RSSI, ToA, EventMamba, FasterKAN, and GNN-based localization methods. Results demonstrate that the proposed approach achieves superior accuracy, reduced prediction variance, and robust anchor utilization across varying topologies. Notably, under the most adverse tested conditions, including severe timing noise, 10% event packet loss, and imperfect synchronization, the proposed model maintains graceful degradation and continues to outperform EMSP and signal-based baselines. Furthermore, interpretability is enhanced through spatial visualization tools, including 2D true vs. predicted maps and Voronoi-based consistency analysis. Overall, the framework offers a scalable and generalizable solution, positioning event-driven ML-based localization in WSNs as an effective alternative to conventional strategies in resource-constrained environments.