Time Series Prediction Model Based on Deep Learning for Traffic Flow

In order to address the issues of high random volatility, low reliability, and poor prediction accuracy in short-term urban traffic flow, a prediction model is developed by integrating variational modal decomposition (VMD) with long-short-term memory neural network (LSTM). Initially, the historical traffic flow data is decomposed using VMD, following which each component is input into LSTM for prediction. Subsequently, the prediction outcomes are combined linearly to derive the anticipated traffic flow value. The study focuses on the traffic flow data of a specific road segment in Shanghai spanning from September 4 to September 25, 2018. The findings indicate that the VMD-LSTM model outperforms conventional prediction models like ANN, gaos, and single LSTM in terms of error rates. This model proves to be more effective for short-term traffic flow prediction and can serve as a valuable reference for road traffic management purposes.