Abstract:

In order to improve the anomaly detection performance of dam monitoring data, a dam abnormal data detection method based on the improved Prophot-long short term memory-particle swarm optimization Prophet-LSTM-PSO was proposed. Firstly, by improving the Prophet method, the trend component features obtained from the decomposition of abnormal data points were clearly visible. Secondly, the decomposed trend, periodic, and residual components were represented in a three-dimensional space, where the original time series data was substituted with the mean distance of the nearest neighbors in this space. Finally, abnormal data points were identified precisely by combining the LSTM network and PSO algorithm to set and optimize anomaly thresholds. The results show that the method proposed in this paper significantly improves detection performance and exhibits high stability compared with traditional methods. Notably, while maintaining a stable recall rate exceeding 95%, both accuracy and precision surpass 95%, thereby validating the effectiveness and practicality of the proposed method.

Key words: Prophet, long short term memory (LSTM), particle swarm optimization (PSO), anomaly data detection, dam monitoring data