A Novel LSTM Deep Learning Approach for Short-Term Traffic Flow Prediction Driven by Cross-Sectional Data

To accurately predict short-term traffic flow, considering the time characteristics of traffic flow and combining the characteristics of DFT, KNN and LSTM models, a DFT-KNN-LSTM hybrid model is proposed. The model first uses the DFT method to decompose the traffic flow data into trend term and residual term data to remove the influence of residual term data on traffic flow prediction. Second, the KNN algorithm based on Euclidean distance is used to screen the traffic flow data with high similarity between K days and target forecast days in the data. Furthermore, the filtered data are used as the training set, and the target day’s data are used as the test set, which is substituted into the LSTM model for prediction; finally, the Mean Absolute Error (MAE), Mean Square Error (MSE) and Root Mean Square Error (RMSE) were used as evaluation indexes to analyze and evaluate the prediction results. Taking the traffic flow data collected from Xizhaosi Street in the Dongcheng District of Beijing as an example, the prediction performance of the combined model is analyzed. The results show that compared with other commonly used prediction models, the MSE of the DFT-KNN-LSTM combined model is improved by 3.24–19.05%, RMSE is improved by 1.54–9.98%, and MAE is improved by 3.05–8.97%. It can be seen that the combined model has better prediction performance than the traditional single model and other combined models, and can be better applied to short-term traffic flow prediction.