Hierarchical Temporal Memory Based Anomaly Detection for Hydrological Monitoring of Unmanned Surface Vehicle

Abstract

Intelligent hydrological automatic monitoring of the target area under all-weather and unattended conditions is of great significance for preventing water disasters and protecting water resources. An application of intelligent hydrological monitoring is developed based on unmanned surface vehicle (USV). Abnormal detection is an important part of hydrological monitoring. Hierarchical temporal memory (HTM) network imitates the structure of mammalian cortex, and it is used for abnormal detection of hydrological sensor data collected from USV. The prediction HTM network consists of initialization, computation of cell states, and updating of synapses on dendritic segments. The hydrological data of rivers, such as water depth, velocity and flow, can be easily obtained by USV. And a hydrological dataset is established with the mark of normal or abnormal sample. The HTM based abnormal detection method is detailed and it is compared with other machine learning based methods. The experimental results indicate that the HTM based method has the best anomaly detection performance.