基于CNN-XGBoost的短时交通流预测

Short-term traffic flow forecasting based on CNN-XGBoost

为准确预测短时交通流,缓解交通拥堵提高交通运行效率,提出一种基于CNN-XGBoost的短时交通流预测方法。结合短时交通流数据的时间相关性和空间相关性,将本路段和邻近路段的历史数据一同作为输入进行预测。利用卷积神经网络(convolutional neural networks,CNN)实现特征提取以减少数据冗余性,提出一种参数经果蝇算法优化的XGBoost模型用于交通流量预测。实例验证结果表明,CNN可对时间和空间结合下的交通流数据进行有效特征提取;相比SVR、LSTM等模型,改进的XGBoost模型下的交通流量预测误差明显减小。

To predict short-term traffic flow accurately,alleviate traffic congestion and improve traffic operation efficiency,a short-term traffic flow prediction method based on CNN-XGBoost was proposed.Combining the temporal and spatial correlation of short-term traffic flow data,the historical data of this section and adjacent sections was used as input.Convolutional neural networks(CNN)was used to extract features to reduce the redundancy of data,and an XGBoost model that its parameters were optimized using fruit fly optimization algorithm(FOA)was proposed for traffic flow prediction.The results of example verification show that CNN can effectively extract features from traffic flow data under combination of time and space.Compared with SVR and LSTM models,the prediction error of traffic flow using modified XGBoost model is reduced obviously.