Short-term traffic flow prediction based on CEEMDAN-CNN-LSTM

Abstract

Traffic flow is characterized by nonlinearity, volatility and randomness. To further improve the accuracy of short-term traffic flow prediction, a combined short-term traffic flow prediction model (CEEMDAN-CNN-LSTM) based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), convolutional neural network (CNN), and long short-term memory network (LSTM) was used. The model utilizes CEEMDAN to decompose the original traffic flow data into k smooth intrinsic mode functions (IMFs), inputs each modal function into the CNN-LSTM model for prediction respectively, and aggregates and accumulates the predicted values to obtain the short-term traffic flow prediction results. In the model, CNN is used to better capture the spatial characteristics of the traffic flow. The experimental results show that the combined prediction model has a high prediction accuracy compared to the ARIMA, LSTM, CNN-LSTM, CEEMDAN-LSTM, and EMD-CNN-LSTM models with reductions of 50.7%, 44.6%, 39.7%, 20.7%, and 9.7% in terms of MAE, respectively.