Learning internal representations by error propagation
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
Essentials of fuzzy modeling and control
Essentials of fuzzy modeling and control
| Hi-index | 0.98 |
Incident detection involves both the collection and analysis of traffic data. In this paper, we take a look at the various traffic flow sensing technologies, and discuss the effects that the environment has on each. We provide recommendations on the selection of sensors, and propose a mix of wide-area and single-lane sensors to ensure reliable performance. We touch upon the issue of sensor accuracy and identify the increased use of neural networks and fuzzy logic for incident detection. Specifically, this paper addresses a novel approach to use measurements from a single station to detect anomalies in traffic flow. Anomalies are ascertained from deviations from the expected norms of traffic patterns calibrated at each individual station. We use an extension to the McMaster incident detection algorithm as a baseline to detect traffic anomalies. The extensions allow the automatic field calibration of the sensor. The paper discusses the development of a new novel time indexed anomaly detection algorithm. We establish norms as a time dependent function for each station by integrating past ''normal'' traffic patterns for a given time period. Time indexing will include time of day, day of week, and season. Initial calibration will take place over the prior few weeks. Online background calibration continues after initial calibration to continually tune and build the global seasonal time index. We end with a discussion of fuzzy-neural implementations.