Short-Term Traffic Flow Prediction of Expressway Based on CEEMDAN-PE-BiGRU Combined Model Optimized By Attention Mechanism

Short-term traffic flow is an essential guarantee for realizing traffic flow prediction and control. Because of the randomness and complexity of traffic flow, this paper proposes a CEEMDAN-PE-BiGRU combined model based on the optimization of attention mechanisms. Firstly, this paper uses the CEEMDAN algorithm to decompose the unsteady original traffic flow data into multiple relatively stable modal components. Then, the PE algorithm is used to calculate the permutation entropy value of the IMF components, and the components with similar entropy values are superimposed to form a new sequence; BiGRU is adopted to extract the hidden features of the time series data in each component. After the attention mechanism highlights the key features, each sequence is predicted separately; finally, the prediction results of each component are superimposed to obtain the final predicted value. Moreover, use the measured traffic flow data near the British M25 high-speed Heathrow Airport, for example, verification. The results show that the prediction effect of the CEEMDAN-PE-BiGRU combination model is significantly better than that of the BiGRU model, the BP model, and the ABiGRU model. With the prediction error greatly reduced, the fitting accuracy was improved by 0.055, 0.048, and 0.000056 compared with BiGRU, BP, and ABiGRU, respectively, and the average prediction accuracy was improved by 10.96% and 0.48% compared with BiGRU and ABiGRU, respectively. The CEEMDAN-PE-BiGRU combined model can accurately extract the latent features of the traffic flow time series data, improve the prediction accuracy, and provide scientific and technical support for the intelligent transportation system.