Application of Fuzzy Logic and Neural Networks to Automatically Detect Freeway Traffic Incidents

Abstract

To date, efforts to manage freeway congestion have been seriously impeded by the inability to promptly and reliably detect the presence of traffic incidents. Traditional incident‐detection algorithms distinguish between congested and uncongested operation by comparing measured traffic‐stream parameters with predefined threshold values. Given the range of possible operating conditions in the traffic stream, selecting a single threshold value, and the suitability of that selected threshold, is full of uncertainty. This inherent uncertainty makes fuzzy logic a promising approach to incident detection. A Fuzzy Logic Incident Patrol System (FLIPS) is proposed to solve many of the problems inherent in traditional incident‐detection algorithms. The FLIPS combines fuzzy logic with the learning capabilities of neural networks to form a connectionist model. The system can be constructed automatically from training examples to find the optimal input/output membership functions. Threshold values, implicitly obtained by fuzzy‐logic rules and membership functions, are treated as dependent variables, which change according to prevailing traffic‐stream parameters measured by detectors. The FLIPS avoids the rule‐matching time of the inference engine in the traditional fuzzy‐logic system. The potential effectiveness of the FLIPS is evaluated using an empirical database collected in Toronto, Canada. Future refinement to the FLIPS are also discussed in this paper.