AutoGMM-RWKV: A Detecting Scheme Based on Attention Mechanisms Against Selective Forwarding Attacks in Wireless Sensor Networks

Integrating wireless sensor networks (WSNs) technologies into Internet of Things amplifies their monitoring capabilities by enabling real-time data collection and analysis across various environments. Nevertheless, the selective forwarding attacks of malicious nodes and diminished forwarding rates among normal nodes due to degraded channel conditions in harsh environments collectively undermine WSNs’ reliability. Previous methods chose cumulative forwarding rates (CFRs) as input features that struggle with increased error rates under such conditions. This study presents a novel scheme for detecting selective forwarding attacks in harsh environments using an integration of autoencoder, the Gaussian mixture model (GMM), and K-means. Instead of CFRs, the scheme focuses on node single-round forwarding rates (SFRs) time series for analysis. The autoencoder extracts features by reconstructing the SFRs series of all nodes. GMM classifies the features and node reconstruction errors, constructing nodes’ evaluation values assuming normal and malicious nodes belong to different Gaussian distributions. A linear attention mechanism called receptance weighted key value (RWKV) is introduced into the autoencoder and GMM, significantly improving our detection accuracy. In the initial detection round, K-means is employed to identify potential malicious nodes. Adaptive thresholding is then used to detect malicious nodes by computing the mean and standard deviation of nodes’ evaluation values. Simulation results demonstrate the effectiveness of the proposed scheme, with an average false detection rate (FDR) of 0.09% and an average missing detection rate (MDR) of 0.56% under mobile harsh environmental conditions. This research offers a robust solution for enhancing the resilience of WSNs against selective forwarding attacks.