Sparse data reconstruction for industrial IoT networked perception using a regularized fully connected tensor network

Abstract The Industrial Internet of Things (IIoT) has been widely applied in fields such as intelligent manufacturing, factories, and equipment status monitoring. Benefiting from ubiquitous sensing capabilities, wireless networks have become an indispensable cornerstone for various innovative applications. Due to limited sensor deployments and labor costs, it is difficult to obtain complete data from sparse sampling in noisy industrial environments. Therefore, it is crucial to reconstruct as much sparse wireless data as possible. Considering the spatiotemporal correlations in the local network topology, specifically, that wireless nodes in adjacent locations tend to exhibit similar ranging values, a regularized fully connected tensor network is used for sparse reconstruction. First, the multidimensional wireless measurement is mapped into a wireless ranging tensor, which is further decomposed into a low-rank target tensor and a sparse noise tensor via tensor robust principal component analysis. Second, the low-rank target tensor is decomposed to obtain more global low-rank information, and a gradient operator first-order difference matrix is used to apply regularization to enhance local continuity. Third, the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mrow> <mml:msub> <mml:mi>l</mml:mi> <mml:mn>1</mml:mn> </mml:msub> </mml:mrow> </mml:mrow> </mml:math> norm is used to measure the sparsity of the sparse noise tensor to enhance model robustness, which can be solved by the proximal alternating minimization method to reconstruct the low-rank target tensor. Finally, different algorithms related to sparse data reconstruction are evaluated under different noise levels and different outlier rates in extensive simulations. An IIoT-networked perception platform, including wireless bases and mobile nodes, is built. The real-world results confirm that the proposed algorithm has superior reconstruction accuracy compared to relevant algorithms and can provide technical support for state monitoring in industrial sparse measurements.