Fuel consumption of vehicles at signalized intersections is directly impacted by congestion. Accurately characterizing fuel consumption at signalized intersections gives additional insights into how signal timing strategies and intelligent transportation systems impact traffic. Cost effective sensing technologies in vehicles have introduced new ways to collect vehicular trajectories ubiquitously, which can help characterize fuel consumption at intersections at scale. This study investigates fuel consumption prediction with a Long Short-Term Memory (LSTM) neural network using onboard diagnostic (OBD) and crowdsourced probe vehicle trajectory data. The samples represent both free flow and stopped traffic as well as different driving conditions like acceleration, deceleration, and idling. The results of the LSTM model indicate that total fuel consumption at an intersection is related to traffic operation parameters, i.e., delay. Accurately predicting fuel consumption will pave the way for improvement of fuel economy.

交通拥堵直接影响信号交叉口车辆的油耗。准确表征信号交叉口的油耗可以进一步了解信号配时策略和智能交通系统如何影响交通。车辆中具有成本效益的传感技术引入了新方法来无处不在地收集车辆轨迹，这有助于大规模表征十字路口的燃料消耗。本研究使用车载诊断 (OBD) 和众包探测车辆轨迹数据，通过长短期记忆 (LSTM) 神经网络调查燃料消耗预测。样本代表自由流动和停止的交通以及不同的驾驶条件，如加速、减速和怠速。LSTM 模型的结果表明，交叉口的总油耗与交通运行参数即延误有关。准确预测燃油消耗量将为提高燃油经济性铺平道路。