An intelligent optimization method for communication monitoring alarm threshold based on BeiDou satellite positioning data

This paper proposes an intelligent optimization method for BeiDou communication monitoring alarm thresholds based on Representation Learning and Adaptive Thresholding for Anomaly Detection (RLAT-AD), aiming to address the issue that under complex environments, the positioning-communication joint time-series data of BeiDou terminals produce anomalies, leading to misjudgment of communication quality and failure of alarm thresholds. First, a sliding window is employed to construct supervised learning sample pairs, and a stacked autoencoder (SAE) is used to perform nonlinear dimensionality reduction and noise suppression on the original time-series data to extract low-dimensional robust features. Second, the neural circuit policy (NCP) model is introduced to capture non-stationary temporal dependencies through its sparse connectivity and hierarchical topology, achieving high-precision time-series prediction. Third, a dynamic threshold optimization (DTO) strategy based on a multi-objective evolutionary algorithm is proposed, which aims to maximize precision and recall to solve for the optimal threshold coefficient, thereby achieving an adaptive balance between precision and recall. Experimental results demonstrate that the proposed RLAT-AD method achieves superior performance across five evaluation metrics, with accuracy of 0.892, precision of 0.856, recall of 0.839, F1-score of 0.847, and AUC of 0.908, outperforming all baseline models. Additional ablation, robustness, and generalization experiments further validate the effectiveness and practicality of the proposed method.