Sensors of different types are used in power systems for different purposes. For example, sensors are attached to a wind turbine to take measurements including real-time power outputs, air pressure, air temperature, etc. These measurements are important for monitoring the operating conditions of the attached device. Effective methods are required for automatically detecting anomalies in the sensor data, especially when many devices in the system need to be monitored simultaneously.
- Anomalies are abnormal and minor patterns emerges in the measurements that distinguish themselves from normal and major patterns. In terms of their durations, these anomalies can be roughly classified into two major categories:
- Anomalous points: the measured values at these points are considerably away from normal values.
- Anomalous intervals: the measured values looks normal if investigated point-wise, while the interval as a whole presents abnormal patterns.
We have developed an online system for anomaly detection for power system devices. The system collects real-time measurements from a diversity of sensors attached to the device. Online data analysis is performed by the monitoring system, based on which online assessments of the device condition are performed and reported to the system operator.
Data analysis is performed based on GOT online monitoring and anomaly detection engine. Device-specific analytic models are adaptively built and updated based on archived measurements. Device condition is assessed based on the deviation of the measurement from the model predictions.
Our system integrates effective machine learning and statistical analytic models. It uses GOT TRUST-TECH/ELITE for optimal analytic model building and ensemble. It is composed of three modules to realize comprehensive and accurate models of normal patterns or behaviors of the device.