Research On Short-term Traffic Flow Prediction Method Based On Deep Learnin

As the basis for traffic state discrimination and traffic flow induction,real-time and reliable short-time traffic flow prediction research is of great significance to improve road operation efficiency and service control level.The existing prediction models have the following shortcomings: they don’t fully consider their own influence,and the traffic flow data contains a lot of noise;manual experience adjusts the network parameters,and the feature extraction capability needs to be enhanced;they don’t integrate the influence of external factors and cannot meet the high accuracy requirements of short-time prediction.Therefore,this paper introduces Complete Ensemble Empirical Mode Decomposition(CEEMD),Particle Swarm Optimization(PSO)and Attention Mechanism to address the above shortcomings in turn and effectively improve the prediction accuracy of short-time traffic flow.(1)To eliminate the influence of noisy signals on traffic flow prediction,the CEEMD model was introduced to decompose the data,and the spatio-temporal features within the data were fully explored using Convolutional Neural Networks and Long Short Term Memory Networks(CNN-LSTM)to build a based on CEEMD-CNN-LSTM single-factor traffic flow prediction model.The results show that the RMSE、MAE、MAPE values of the CEEMDCNN-LSTM single-factor prediction model reached 5.699、4.059、0.331 respectively,which significantly improved the prediction accuracy and further validated the effectiveness and accuracy of the CEEMD decomposition algorithm.(2)To accelerate the learning efficiency of the neural network,a based on CEEMDGMPSO-LSTM single-factor traffic flow prediction model is constructed by automatically seeking the structural parameters of the neural network through the PSO optimization algorithm and improving it with dynamically changing inertia weights and genetic algorithm variation operations.The results show that the CEEMD-GMPSO-LSTM single-factor prediction model has the best performance in terms of evaluation metrics,the highest data fit,and has a better fit in the morning and evening peak hours where the variation is large,further validating the effectiveness and reliability of the PSO optimization algorithm.(3)In order to consider the influence of external factors on traffic flow prediction,a multifactor traffic flow prediction model based on CEEMD-CNN-LSTM-Attention(CCLA)is built by combining the attention mechanism to assign different weights to multi-dimensional input features and extract key information for learning.The results show that after incorporating external factors such as weather and holidays,the CCLA multi-factor prediction model achieves the best prediction results on several sets of experiments,further validating the effectiveness and stability of considering multiple external factors and the attention mechanism to improve the prediction accuracy of short-time traffic flow.