Abstract：Traffic flow forecasting is an important part of intelligent transportation. The traffic data to be processed is non-linear, periodic, and random, as a result, when data is predicted, unstable traffic flow data depends on long-term data range. Due to some external factors, the original number often contains some noise, which may further lead to the degradation of prediction performance. Aiming at the above problems, a prediction algorithm EMD-LSTM that can denoise and handle long-term dependence is proposed. The algorithm first uses EMD (Empirical Mode Decomposition) decomposes different scale components in the traffic time series data step by step to generate a series of Intrinsic Mode Function with the same characteristic scale, thereby removing certain noise effects. Further with the help of the long short-term memory neural network LSTM (Long Short Term Memory Network) the algorithm can solve the problem of long-term dependence of data, so it is more outstanding in long-term field prediction. The method is applied to short-term prediction of actual data sets. The experimental results show that the EMD-LSTM algorithm can effectively solve the problem of traffic data, which can significantly improve the accuracy of prediction. The Mean Absolute Error(MAE) of EMD-LSTM is 1.91632 lower than that of LSTM, and Mean Absolute Percentage Error (MAPE) is 4.64545 percentage points lower than that of LSTM. The prediction effect of the hybrid model can significantly improve the prediction accuracy and solve the problem of traffic data effectively.
