Abstract: In order to improve the accuracy of short-term traffic flow prediction, an adaptive traffic flow real-time prediction model was proposed based on the forgetting factor extreme learning machine (FFOS-ELM) and particle filtering (PF). Firstly, the forgetting factor was introduced and the extreme learning machine with a forgetting factor was deduced. In addition, to avoid the influence of early data on prediction accuracy due to the time-varying of traffic flow, the parameters of the prediction model were updated in real time by an incremental learning method. Secondly, particle filtering was used to eliminate the influence of random noise on prediction accuracy. Through iterative calculation, the optimal estimation and prediction ability of the system state were achieved to improve the prediction accuracy of future traffic flow. Finally, the detector data of a trunk road in Guilin were used for simulation. The predicted results were compared with those of the online models such as the basic extreme learning machine, the forgetting factor extreme learning machine, and the those of offline models such as time series (ARIMA), support vector machine (SVM), long-short term memory neural network (LSTM). The results show that the prediction error index of the proposed adaptive prediction model drops significantly, and the mean square error variation dimension drops to between 0 and 2.5. The traffic flow fitting situation in the whole road segment of the proposed model and its specific prediction accuracy are effectively improved.

Key words: traffic engineering, intelligent traffic, traffic flow prediction, extreme learning machine, forgetting factor, particle filter

摘要： 为提高短时交通流预测精度，提出了一种基于遗忘因子极限学习机（FFOS-ELM）和粒子滤波（PF）的自适应交通流实时预测模型。首先，引入遗忘因子，推导带遗忘因子的极限学习机，通过增量学习方法实时更新预测模型参数，避免由于交通流时变性导致早期数据对预测精度的影响。其次，利用粒子滤波消除随机噪声对预测精度的影响，经迭代计算达到系统状态最优估计与预测能力，实现未来交通量预测精度的提高。最后，利用桂林市某主干路检测器数据进行仿真，将预测结果与基础的极限学习机、带遗忘因子的极限学习机等在线模型以及时间序列(ARIMA)、支持向量机(SVM)、长短期记忆神经网络（LSTM）等离线模型进行比较。结果表明：自适应预测模型预测误差指标明显下降，均方误差变化维度下降到0~2.5之间，模型在路段整体的交通流拟合情况及具体的预测精度上均得到有效提高。

关键词: 交通工程, 智能交通, 交通流预测, 极限学习机, 遗忘因子, 粒子滤波