基于梯度推进决策树的日维度交通指数预测模型

GBDT Method Based on Prediction Model of Daily Dimension Traffic Index

城市交通运行监测和预测是掌握交通运行变化特点,制定缓解交通拥堵策略的重要工作,其结果能为公众提供有效的路况信息,亦为政策措施的制定和效果评估提供重要支撑.有别于传统的短时交通预测,本文提出的预测模型不是针对相邻时段的运行状态预测,而是更长跨度上,针对日级别高峰时段交通运行状态的预测.构建了包含时间周期、特殊天气、节假日、限行、大型活动等因素的多维度影响因素集;以长期历史交通指数构建数据训练集,提出了基于梯度推进决策树的日维度路网状况预测模型.应用最优模型进行验证,结果表明,模型预测精度可达90%以上,与其他4种回归模型的对比分析也显示,本文所提出的模型在各项评分中均表现最优,说明其更适合于大样本、多因素的回归分析.本文所提出的日维度预测模型对提升城市路网运行质量、缓解交通拥堵具有重要的应用价值.

Monitoring and forecasting of urban traffic operation is an important task to grasp the characteristics of traffic operation changes and formulate strategies to alleviate traffic congestion. The forecasting results can provide effective road information for the public, and also provide support for the formulation of policy measures and the evaluation of the effect. Different from the traditional short-term traffic forecasting, the forecasting model proposed in this paper is not for the operation state prediction of adjacent periods, but for the long-span, for the daily traffic operation state prediction. This paper constructs a multidimensional factors set including time period, weather, holidays, vehicle restriction, large events and so on;builds a data training set based on long- term historical traffic index, and proposes a daily road network condition prediction model based on gradient boosting decision tree;validates the model by using the optimal model. The results show that the prediction accuracy of the model can reach more than 90%. The comparative analysis of the regression model also shows that the model presented in this paper performs best in all scoring items, indicating that it is more suitable for regression analysis with large samples and multiple factors. The daily prediction model proposed in this paper has important application value in improving the quality of urban road network operation and alleviating traffic congestion.