Real-Time Localization of Zigbee Signals Using Graph Neural Network

With the rapid development of Internet of Things technologies, zigbee-based wireless signal positioning has shown great potential in applications such as smart homes and industrial monitoring. However, traditional positioning methods face challenges such as environmental changes and device diversity, making it difficult to achieve high precision and robustness. This article proposes a zigbee signal positioning system based on graph neural network (GNN), which significantly enhances positioning accuracy and robustness through innovative signal vector embedding, graph structure construction, and message propagation mechanisms. First, we use the received signal strength indicator (RSSI) as the fundamental data for positioning and performing data preprocessing, including time series segmentation, interpolation, normalization, and feature engineering. Then, utilizing a feedforward neural network (FNN) combined with channel attention mechanisms, we map the time series into a high-dimensional space to obtain feature embedding vectors. Subsequently, by calculating the dynamic time warping (DTW) distance and applying the Top-K strategy, we construct graph structures within each time window and form a complete graph topology using the cartesian product method. We introduce a novel attention-based graph neural network, attention-based graph neural network (AGNNConv), for semisupervised learning tasks, which significantly reduces the number of model parameters while improving the accuracy of information aggregation. Moreover, to avoid the over-smoothing phenomenon in graph neural networks, we employ an Inception structure to expand the network’s width and use residual connections to enhance feature diversity. Finally, we feed the results from the graph learning layer into a bidirectional long short-term memory (BiLSTM) to extract temporal features and predict through a regression linear layer. Through a series of experiments, we demonstrate the effectiveness of our system on various datasets, showing a significant advantage over existing methods in terms of positioning accuracy and robustness.