Optimized Deployment of Intelligent Multi-Modal Sensor Network Based on Graph Transformer

To address the issues of node redundancy, uneven distribution, and high energy consumption caused by random deployment in intelligent sensor networks, this paper proposes an optimization deployment algorithm for intelligent multimodal sensor networks based on graph transformer. Firstly, a multidimensional graph construction algorithm based on spatiotemporal fusion is proposed to construct the raw sensing data of the multimodal sensor network into a multidimensional graph. On this basis, a graph transformer with overlapping local and global attention is designed to perform feature representation on the time series of each dimension of the graph. Additionally, a graph convolutional network and a multilayer perceptron are introduced to achieve contribution inference and fusion. The entire inference process is iteratively optimized under the constraints of feature reconstruction error and feature semantics. Experimental results show that the proposed algorithm effectively identifies important nodes and outperforms other graph neural network models across various metrics.