Research On Optimization Of Urban Road Short-Time Traffic Flow Prediction Model

In order to alleviate and manage traffic congestion better,intelligent transportation systems(ITS)have been used more and more widely.Short-term traffic volume predicting is an indispensable part of the continued development and improvement of ITS.Accurate traffic volume prediction can help road managers control and coordinate traffic volume better,also can provide drivers with a smoother road,avoid traffic jams effectively and save time.Based on the research background and realistic meaning,combined with the research results of domestic and foreign scholars in short-term traffic volume prediction.The traffic prediction model is proposed based on wavelet denoising and phase space reconstruction(WD-PSR-GA-BP),based on the current research problems that need to be improved.The model uses wavelet denoised technology to decompose,denoised,and reconstruct the original traffic volume data,puts the reconstructed data into the phase space reconstruction algorithm,and calculates the embedding dimension and time delay(?).Using these two parameters to reconstruct the denoised traffic volume,and map the one-dimensional traffic volume to high-dimensional.Finally,the BP neural network is used as the prediction model.However,because the weights and thresholds of BP neural network use gradient descent method,it is easy to fall into the minimum value,so the genetic algorithm is used to optimize the weights and thresholds.In the early theoretical discussion,the various parameters of each algorithm were discussed and simulated,and finally the choice of parameters was explained and determined.In the experimental part,in order to ensure the accuracy,robustness and timeliness performance of the proposed algorithm,four error evaluations were added,namely: MAPE,RMSE,MAE and Time.To ensure the validity of the proposed model,the selected data source is from California Pe MS.Regarding whether the prediction model chooses single-step prediction or multi-step prediction,this paper specifically conducts independent experiment comparison.The results of the above four error evaluations show that,in this experiment,the result of single-step prediction is better than multi-step prediction.In order to compare the superiority of the proposed combined model,a single prediction model radial basis function neural network(RBF)and random forest(RF)were used as comparison model in the experimental;GA-BP,PSR-GA-BP and EEMD-GA-BP were also selected as comparison model in the experiment.The experiments of each model were independently run 20 times,and the average results of each error evaluation was recorded.It can be seen from the final experimental results,The combined model is more superior in accuracy and robustness than a single prediction model.The WD-PSR-GA-BP model has the most excellent performance among the indicators MAPE,RMSE and MAE compared to other comparison models.From a timeliness point of view,the single prediction model takes the shortest time,and among the combined prediction models,EEMD-GA-BP needs the longest time.