基于SARIMA-GPR模型的短时交通流预测应用研究

Research on the Application of Short-Term Traffic Flow Prediction Based on SARIMA-GPR Model

交通流量数据具有周期性、不平稳性、复杂性等特点,若使用单一模型对其进行预测,则预测效果不是很好,因此提出一种组合的SARIMA-GPR模型。SARIMA (Seasonal Autoregressive Integrated Moving Average)模型与GPR (Gaussian Process Regression)模型分别很好拟合交通流量的线性部分与非线性部分,且GPR模型考虑到数据的噪声,能更好地抓取到数据信息。对原数据进行特征提取与分析,训练SARIMA模型与GPR模型,得到两个预测模型,根据模型的MAE得到两个模型的权重值,得到最终的预测值。将该组合模型与SARIMA、GPR、SVM、SARIMA-SVM组合模型进行预测效果对比,实验结果表明,SARIMA-GPR模型预测效果要优于单一模型,预测结果平均绝对百分比误差(MAPE)减少到4.51%,预测结果更接近真实数据。

Traffic flow data have the characteristics of periodicity, instability and complexity. If a single model is used to predict it, the prediction effect is not very good. Therefore, a combined SARIMA-GPR model is proposed. The SARIMA (Seasonal Autoregressive Integrated Moving Average) model and the GPR (Gaussian Process Regression) model fit the linear part and the nonlinear part of the traffic flow well respectively, and the GPR model takes into account the noise of the data and can better capture the data information. Perform feature extraction and analysis on the original data, train the SARIMA model and the GPR model, and obtain two prediction models. According to the MAE of the model, the weight values of the two models are obtained, and the final prediction value is obtained. The combined model is compared with SARIMA, GPR, SVM, and SARIMA-SVM combined model. The experimental results show that the prediction effect of SARIMA-GPR model is better than that of a single model, and the mean absolute percentage error (MAPE) of the prediction results is reduced to 4.51%, the prediction effect is closer to the real data.