Using Hierarchical Temporal Memory to Detect City Infrastructure Anomalies

Abstract

Cities today utilize a wide range of sensors to monitor the status of various utilities such as water distribution and the electricity grid. The gathered data can be analyzed in real-time to detect anomalies, allowing the city to rapidly identify and resolve the issue. Currently existing anomaly detection methods such as operate on the assumption of a static source environment [1, 2]. However, sensitivity of the sensors may degrade with time and background noise levels may vary as well. This leads to the generation of costly false positives and negatives, which we aim to minimize through the use of Hierarchical Temporal Memory (HTM). HTM is particularly suited for anomaly detection due to its ability to recognize patterns in a chronological stream of values [1]. In this paper, we analyze the efficacy of HTM anomaly detection for a series of 5 angular sensors measuring the alignment of a well. Additionally, we compare the results of feeding raw versus delta values from our sensors into an HTM algorithm, finding that the delta values produced a more consistently accurate result. Finally, we discuss recommendations for the applications of HTM in real city infrastructure situations.