Improved Error Detection for Inductive Loop Sensors

摘要

The report describes the development of an algorithm to detect anomalies in the time series from inductance loop sensors. The anomalies may arise from traffic incidents or loop detector system malfunction. The algorithm uses a statistic produced with inductance loop data to make an optimal prediction of the volume and occupancy values that will occur at the next time step. To guarantee the optimality of this prediction, a Kalman predictor, for use with inductance loop data, is developed. To detect variations from the normal state the optimal prediction is compared with the observed value. Anomaly detection is accomplished by applying thresholds to the difference between the predictions and the observed values. The report demonstrates the use of the anomaly detection algorithm with inductance loop data gathered on Interstate 5 in Seattle, Washington. The report also discusses the scaling and values of thresholds necessary for anomaly detection. This type of dynamic prediction and threshold can be valuable to traffic management systems that rely heavily on inductance loop data.