A lightweight drift sensing method for UASN based on sliding window-LSTM cooperative strategy

As the rapid growth in marine resource development and underwater national defense security demands, the large-scale deployment of underwater acoustic sensor networks (UASN) has made them critical information infrastructure, with dynamic detection of node vulnerabilities becoming a core challenge for ensuring network reliability. In complex underwater environments, the time-varying nature of acoustic channels, equipment performance degradation, and the continuous evolution of new network attacks lead to significant spatiotemporal heterogeneity in node behavior data. Traditional static detection models struggle to distinguish between environmental noise interference and malicious attack behaviors, leading to severe misjudgments and missed detections. To address the challenges of intrusion detection in dynamic environments within underwater acoustic sensor networks (UASN), this paper proposes a hybrid mechanism based on sliding window dynamic threshold adjustment and LSTM drift trend prediction. By statistically analyzing recent data distribution characteristics, it dynamically adjusts the drift detection threshold to balance sensitivity and stability. At the same time, a lightweight LSTM model is used to learn historical drift patterns, predict future drift trends, and trigger model updates in advance. The prediction results are combined with the sliding window threshold to achieve “prediction-response” collaborative decision-making, significantly reducing detection latency. This enhances the real-time performance and robustness of the Adaptive Intrusion Detection System (AIDPS).