关键词: 交通流量预测, 时间序列分析, 卷积神经网络, 长短时记忆神经网络, 特征提取

Abstract: The traffic flow sequence is not stationary, periodic and easily affected by factors such as holidays, thus it is hard to predict. Therefore, a model based on deep learning is designed. Model includes convolution neural network and LSTM（Long Short-Term Memory）. Convolution neural network is used to extract feature components. LSTM uses the extracted features to sequence prediction. The extracted data are used as input of LSTM, and output as a predictive results. This model has better accuracy, real-time and generalization performance, which is testified through the data of freeway traffic in Guizhou province.

Key words: traffic flow forecasting, time series analysis, Convolution Neural Network（CNN）, Long Short-Term Memory（LSTM）, feature extraction