Dimensions of learning in a real-time knowledge-based control system

✍ Scribed by N.V. Findler

• Publisher: Elsevier Science
• Year: 1992
• Weight: 493 KB
• Volume: 17
• Category: Article
• ISSN: 0066-4138
• DOI: 10.1016/s0066-4138(09)91027-8

✦ Synopsis

We first describe and justify the domain in which cooperating and learning real-time distributed expert systems perform control operations of urban street traffic signals. We then present the general design of the ultimate system as well as the simplifying assumptions used in a running prototype. The latter, proving the technical feasibility of the approach, has attained a 42% improvement in the traffic flow under non-saturated conditions. Finally, we draw some conclusions concerning the methodology of distributed planning and problem solving systems. Keywo rds. Street traffic signal control; learning, collaborating and real-time expert systems; predictive control; optimization of rule bases. easy and inexpensive to make, when the "permanent" traffic environment changes.

Research in Artificial Intelligence has identified a variety of learning modes in which computing systems can adapt themselves to improve their level of performance on the basis of experience. The range of machine learning covers different levels of abstraction, from rote learning (e.g., data gathering and utilizing pre-arranged classification schemes) to procedural learning (e.g., automatic program writing systems).