Anomaly Detection of Sensor Data Based on Similarity

Anomaly detection in time series refers to the process of identifying abnormal events or behaviors from normal time series data. Sensors, due to inherent sensitivity issues and the influence of various environmental factors, may transmit anomalous data, leading to problems in data quality. High-density sensor data is often characterized by time series, which has increasingly led to the adoption of time series anomaly detection techniques for sensor anomaly detection. However, most sensor anomaly detection methods focus primarily on individual sensors, neglecting the relationships between multiple sensors, which makes it difficult to detect collective anomalies in sensor data. This paper proposes a sensor anomaly detection method to address the issues of point anomalies, context anomalies, and collective anomalies in sensors. The method differentiates sensors based on various sensor types and classifies sensor anomalies accordingly. Point and context anomalies are handled using the Isolation Forest model, which can successfully distinguish outliers and isolated points in the sensor data. The resulting similarity matrix is then used to identify anomalous sensor time series. The experimental results demonstrate that this approach effectively detects various types of sensor anomalies.