Development of Self-Organizing Traffic Control System Using Neural Network Modelby Takashi Nakatsuji, Hokkaido Univ, Sapporo, Japan,
Syunich Seki, Hokkaido Univ, Sapporo, Japan,
Terutoshi Kaku, Hokkaido Univ, Sapporo, Japan,
Abstract: This paper is concerned with developing a self-organizing traffic control system using a neural network model, a multilayer network model. The neural network inputs the control variables, such as split lengths and offsets, and outputs the measures of effectiveness, such as Performance Indexes. The operation is separated into two processes: a training process that builds up an input-output relationship between the control variables and the measures of effectiveness and an optimization process that optimizes the control variables. First, a single split model that optimizes only splits of an isolated intersection was introduced. Using this model, the relationship between the structure of neural networks and the training abilities were studied. Second, to deal with the increase of synaptic weights for roads that consist of several intersections, a multiple split model, in which only neurons that were related to an intersection were connected to each other, was proposed. This model improved not only the computation time but also the estimation precision for untrained patterns. Finally, to deal with the optimization of offsets and the variation of traffic situations, a complicated neural network, a multi-channel model, which has three input sources, one for splits, one for offsets and the other for inflow volumes, was proposed. This model was applied to a road network and split lengths and offsets were optimized so as to minimize the total sum of the Performance Indexes on inflow links.
Subject Headings: Traffic models | Neural networks | Optimization models | Control systems | Intersections | Computer models | Training
Services: Buy this book/Buy this article
Return to search