Abstract: In recent years, the traffic data has been increasing explosively. Accurate and timely prediction of traffic flow information is very important for intelligent transportation system. Based on LSTM neural network, a prediction method of short-time traffic flow considering the characteristics of passenger car proportion was proposed. The characteristics of passenger car proportion in the traffic flow data were extracted and the power spectrum was drawn by using the fast Fourier algorithm (FFT). The periodicity of the characteristics of passenger car proportion on expressway was verified. According to this, a prediction model of short-term traffic flow based on LSTM with the consideration of the characteristics of passenger car proportion was proposed, and an example of a toll station in Beihuan of Guangzhou was analyzed. The results show that the LSTM prediction model with the characteristics of passenger car proportion can effectively reduce the error of short-term traffic flow prediction and improve the accuracy of prediction.

Key words: traffic and transportation engineering, traffic flow prediction, LSTM neural network, the characteristics of passenger car proportion, traffic flow data of expressway

摘要： 近年来，交通数据呈爆炸式增长，准确、及时的交通流预测信息对于智能交通系统至关重要。基于LSTM神经网络提出了一种考虑客车占比特征的短时交通流预测方法；提取车流数据中的客车占比特征并利用快速傅里叶算法（FFT）绘制其功率频谱图，验证了高速公路客车占比特征的周期性特点；针对该特点，提出了基于LSTM引入客车占比特征的短时交通流预测模型，并以广州北环高速某收费站为例进行分析。结果表明：引入客车占比特征的LSTM预测模型，有效降低了短时交通流预测的误差，提高了预测的正确率。

关键词: 交通运输工程, 交通流预测, LSTM神经网络, 客车占比特征, 高速公路车流数据