UKF地铁短时客流预测研究

为获得较为可靠的地铁车站实时客流,提出基于神经网络与无迹变换卡尔曼滤波(UKF)结合的信息融合预测方法。首先利用各站点间进出客流时空相关性,在运行时间约束下组织预测向量,以BP神经网络为函数表达给出目标站点客流的初步预测值。在此基础上,利用无迹变换卡尔曼滤波解决神经网络过学习造成的误差,以提高预测结果精度。最后选取实例验证算法的准确性,结果表明,该改进算法可有效提高预测精度,满足运营需求。

Short-term Passenger Flow Prediction in Subway Using Unscented Kalman Filter

In this paper, a method of data fusion which combines neural networks and unscented Kalman filter is proposed to accurately predict the real-time value of subway station passenger flow. A predictor vector limited by running time is constructed based on space-time relativity of station passenger flow, and a Backpropagation (BP) neural network forecast model is built to obtain the preliminary predictive value of station passenger flow. Moreover, an unscented Kalman filter is used to solve the errors caused by the overfitting of neural networks, to improve the accuracy of the real-time predictive value of station passenger flow. Finally, the accuracy of the algorithm is verified through a case study.