居民出行生成预测的二元logit模型改进及验证

Improvement and Verification of Binary Logit Modeling for Residential Trip Generation Prediction

针对二元logit模型在出行生成预测应用时忽略了变量差异性的缺点,提出了其可观测变量参数β'服从对数正态分布的假设,根据美国PUMS居民出行调查的前300个相关数据,通过参数标定对该假设的合理性进行了验证.采用SAS仿真系统对参数进行标定的结果表明:当部分参数服从对数正态分布时,所有参数的t检验值均在[-1,1]区间之外,从而验证了假设的合理性.分别利用改进前的模型与改进后的模型进行城市居民出行生成预测,结果显示后者的预测精度明显高于前者,进一步验证了应用改进后的二元logit模型进行居民出行生成预测有显著效果.

To overcome the disadvantage of the binary logit model used in existing resident trip models that neglects the difference of known variables, this paper developed a hypothesis that random parameters β＇ obeyed log normal distribution. It was validated by demarcating the parameters, which was based on 300 data of PUSM in USA. These parameters were estimated using SAS software. Its result shows that all the parameters＇ T values are out of [-1,1] interval when assuming that some parameters obey log normal distribution and proves the rationality of the assumption. The resident trip model was forecasted adopting the previous and improved models. The result further shows that the forecasting precision of the latter model is much higher than the former one＇s, which reveals that the improved binary logit model has effective application.