高原山区高速公路短时交通流预测方法

Short-term Traffic Flow Prediction Method of Expressway in Plateau Mountainous Areas

准确的短时交通流预测是有效避免高原山区高速公路交通事故的关键。高原山区高速公路受高海拔地形影响,短时交通流数据特性较平原区域的更复杂,适用于平原区域的预测模型不一定适用于高原山区。选取SARIMA、GRNN、LSTM模型分别作为数理统计、传统机器学习、深度学习三类预测模型的代表,以四川省阿坝藏族羌族自治州G4217蓉昌高速汶川至马尔康收费站收费数据为样本。结果显示:三种模型均具有较好的预测性能,其中SARIMA和LSTM模型预测效果相当,R2均接近0.97,且较GRNN模型的MAE分别减少了53.12%、57.70%,MAPE分别减少了38.19%、43.72%。研究表明即使数理统计类模型亦可较好预测高原山区高速公路短时交通流,且数据对模型有选择,LSTM模型预测效果最佳,SARIMA模型次之,GRNN模型较差。

Accurate short-term traffic flow prediction is the key to effectively avoid expressway traffic accidents in plateau mountainous areas.However,due to the influence of high altitude terrain,the short-term traffic flow data characteristics of expressway in plateau mountainous areas are more complex than those in plain areas,and the prediction model applicable to plain areas may not be applicable to plateau mountainous areas.The SARIMA,GRNN and LSTM models are selected as the representatives of mathematical statistics,traditional machine learning and deep learning respectively,and the online toll collection data of the G4217 Rongchang Expressway from Wenchuan to Maerkang in Aba Tibetan and Qiang Autonomous Prefecture of Sichuan Province is taken as the sample.The results show that the three models all have good prediction performance,among which SARIMA and LSTM models have the same prediction effect,both R 2 are close to 0.97,and the MAE is reduced by 53.12%and 57.70%respectively,the MAPE is reduced by 38.19%and 43.72%respectively compared with GRNN model.The research shows that even the mathematical statistical models can also have a good prediction effect on short-term traffic flow prediction of expressway in plateau mountainous areas,and the data has a choice of models,the LSTM model has the best prediction effect,followed by the SARIMA model,and the GRNN model comes last.