Neuro-fuzzy techniques are proposed here to control each light of an intersection, at one-second intervals. Rules, fuzzification and inference are modeled by a neural network. For each signal, the neuro-fuzzy control selects between 'switch on' and 'switch off', and presents the required action to a Petri net. A neuro-fuzzy acceleration of forward dynamic programming (FDP) is obtained by enumerating controls only when there are no rules to apply, or when the rules are conflicting. Simulations on different intersections show decreases in delays with respect to fixed timing from 0 % to 30 % for neuro-fuzzy control, and from 15 % to 35 % for neuro-fuzzy acceleration of FDP.
Neuro-fuzzy techniques for traffic control
Neuro-Fuzzy-Verfahren in der Straßenverkehrsregelung
Control Engineering Practice ; 6 , 6 ; 755-761
1998
7 Seiten, 4 Bilder, 4 Tabellen, 14 Quellen
Article (Journal)
English
Neuro-Fuzzy Techniques for Traffic Control
British Library Conference Proceedings | 1997
|