Anomaly driving speed detection and correction algorithm based on quantiles and KNN

作者：Yanling Guo; Lichen Liu; Meng Gao; Lewen Gao

来源：International Journal of Hybrid Information Technology, 2016, 9(3): 301-310.

DOI：10.14257/ijhit.2016.9.3.28

• Anomaly detection
• Anomaly-detection algorithms
• Correction algorithms
• Driving speed
• Mean relative error
• Model reliability
• Quantiles
• Traffic state identifications

Driving speed is a key parameter for building the traffic state identification model, its precision directly affects the model reliability and the traffic state identification accuracy. Aiming at the standard normal deviation method’s defects in dealing with the extreme noise data, an anomaly driving speed detection algorithm based on quantiles is proposed, use historical data to establish the exception borders which are used to detect whether an unknown data is abnormal;on the basis of the abnormal data detection, a driving speed prediction algorithm based on improved KNN is proposed, use K-means algorithm to clustering the historical data, and predict the next moment’s speed according to the distance between the data to be predicted and the clusters, the predicted speed can be used to correct the abnormal speed. Experimental results show that the detection rate of the proposed anomaly detection algorithm has improved about 4.25% compared with the standard normal deviation method, and the false detection rate has reduced about 25.51%;the mean relative error of the proposed speed prediction algorithm is 13.69%, it can predict the driving speed well, namely, the anomaly driving speed detection and correction algorithm based on quantiles and KNN is feasible and effective.

• 出版日期2016
• 单位东北林业大学