Abstract: Transit ridership forecast is an important component of transit planning, and is the basis to determine the scale of transit development, locations of stops and transit routes. The paper analyzes the mechanism of neural network and influential factors of transit generation. A three-layer BP neural network model is developed to forecast transit riders, in which the input nerve cells include land use type (8 nerve clells), population and section location coefficient, the output nerve cells include transit generation and attraction, and the number of nerve cells of concealed layer is calculated with the model in which the maximum fractional error reaches the minimum value. The model is calibrated and testified with one-day field data from Harbin, which shows that the forecast precision of the model is relatively high.

Key words: transit generation., back-propagation neural network., traffic generation., traffic attraction

摘要： 公交客流预测是城市公共交通规划的重要内容，是确定城市公共交通发展规模、布置场站及布设线路的根本依据，也是制定城市公共交通发展政策的重要依据.本文通过分析神经网络的作用机理和公交客流生成量的影响因素，建立了公交客流生成量预测的三层BP神经网络模型，以土地利用(8个神经元)、人口数以及区位系数作为输入神经元，输出神经元为公交客流的产生量和吸引量，隐层神经元数根据最大相对误差最小为目标试算求得.以哈尔滨市一日的调查数据对模型进行了标定与检验，结果证明，模型具有较高的预测精度.

关键词: 公交生成量, BP神经网络, 产生量, 吸引量