Yang Tao, Jiang Xuefeng, Li Wei, Liu Peiyu, Wang Jinming, Hao Weijie, Yang Qiang
China Tobacco Zhejiang Industrial Co., Ltd, Hangzhou, China.
College of Electrical Engineering, Zhejiang University, Hangzhou, China.
PLoS One. 2025 Jun 11;20(6):e0324543. doi: 10.1371/journal.pone.0324543. eCollection 2025.
Industrial sensor networks exhibit heterogeneous, federated, large-scale, and intelligent characteristics due to the increasing number of Internet of Things (IoT) devices and different types of sensors. Efficient and accurate anomaly detection of sensor data is essential for guaranteeing the system's operational reliability and security. However, existing research on sensor data anomaly detection for industrial sensor networks still has several inherent limitations. First, most detection models usually consider centralized detection. Thus, all sensor data have to be uploaded to the control center for analysis, leading to a heavy traffic load. However, industrial sensor networks have high requirements for reliable and real-time communication. The heavy traffic load may cause communication delays or packets lost by corruption. Second, there are complex spatial and temporal features in industrial sensor data. The full extraction of such features plays a key role in improving detection performance. Nevertheless, the majority of existing methodologies face challenges in simultaneously and comprehensively analyzing both features. To solve the limitations above, this paper develops a cloud-edge collaborative data anomaly detection approach for industrial sensor networks that mainly consists of a sensor data detection model deployed at individual edges and a sensor data analysis model deployed in the cloud. The former is implemented using Gaussian and Bayesian algorithms, which effectively filter the substantial volume of sensor data generated during the normal operation of the industrial sensor network, thereby reducing traffic load. It only uploads all the sensor data to the sensor data analysis model for further analysis when the network is in an anomalous state. The latter based on GCRL is developed by inserting Long Short-Term Memory network (LSTM) into Graph Convolutional Network (GCN), which can effectively extract the spatial and temporal features of the sensor data for anomaly detection. The proposed approach is extensively assessed through experiments using two public industrial sensor network datasets compared with the baseline anomaly detection models. The numerical results demonstrate that the proposed approach outperforms the existing state-of-the-art models.
由于物联网（IoT）设备数量的增加以及传感器类型的多样化，工业传感器网络呈现出异构、联邦、大规模和智能的特点。对传感器数据进行高效准确的异常检测对于保证系统的运行可靠性和安全性至关重要。然而，现有的关于工业传感器网络传感器数据异常检测的研究仍然存在一些固有局限性。首先，大多数检测模型通常考虑集中式检测。因此，所有传感器数据都必须上传到控制中心进行分析，这导致了沉重的流量负载。然而，工业传感器网络对可靠和实时通信有很高的要求。沉重的流量负载可能会导致通信延迟或数据包因损坏而丢失。其次，工业传感器数据中存在复杂的时空特征。充分提取这些特征对于提高检测性能起着关键作用。然而，大多数现有方法在同时全面分析这两种特征方面面临挑战。为了解决上述局限性，本文提出了一种用于工业传感器网络的云边协同数据异常检测方法，该方法主要由部署在各个边缘的传感器数据检测模型和部署在云端的传感器数据分析模型组成。前者使用高斯和贝叶斯算法实现，有效过滤工业传感器网络正常运行期间产生的大量传感器数据，从而减轻流量负载。只有在网络处于异常状态时，才会将所有传感器数据上传到传感器数据分析模型进行进一步分析。后者基于GCRL，通过将长短期记忆网络（LSTM）插入图卷积网络（GCN）中开发而成，能够有效提取传感器数据的时空特征以进行异常检测。与基线异常检测模型相比，通过使用两个公共工业传感器网络数据集进行实验，对所提出的方法进行了广泛评估。数值结果表明，所提出的方法优于现有的最先进模型。
