Research For Short-Term Traffic Flow Prediction Of Wavelet Neural Network Based On Cuckoo Algorithm

Traffic flow forecasting is the hot spot of the intelligent transportation system. In recent years, researchers in various fields have proposed many traffic flow prediction methods. Among all the methods, the intelligence learning is the most active method. But due to the complexity and highly nonlinearity of transportation system, there is no algorithm can provide accurate results for the traffic flow prediction until now, the research on traffic flow prediction has never been stopped.At the beginning of research, on the basis of identification of traffic flow data from the traffic data research laboratory of the Minnesota Duluth university, pretreatments includes vacancies filling, correction and noise reduction were completed. Then from the perspective of the chaotic characteristic of traffic flow data, phase space reconstruction was used to dig more hidden information. In order to speed up program convergence, normalization processing was applied after pretreatment. Based on the analysis of the advantages and disadvantages about various traffic flow forecasting models, the short-term traffic flow prediction model based on Wavelet Neural Network (Wavelet Neural Network, WNN) was established. Based on above, simulation experiment was completed. The results show that short-term traffic flow prediction based on WNN is in good performance, but the accuracy and stability remain to be further improved.Then, aimed at the error randomness of a single wavelet neural network, a short-term traffic flow forecasting model based on Bagging-WNN was proposed, inspired by the thought of integrated study. The Bagging-WNN model used the difference between models, to improve the generalization ability of the system. The simulation experiments show that the short-term traffic flow prediction based on Bagging-WNN model has higher accuracy compared with the WNN model.Thirdly. Wavelet neural network based on gradient descent algorithm is sensitive to the initial value, so a new intelligent learning algorithm -- the Cuckoo Search (Cuckoo Search, CS) algorithm was introduced to optimize the parameters of wavelet neural network. Then, the CS-Bagging-WNN model was established. Compared with results from GA-Bagging-WNN and PSO-Bagging-WNN model, the accuracy of short-term traffic flow prediction based on CS-Bagging-WNN model is improved.Finally, in order to improve the prediction accuracy, the adaptive cuckoo algorithm (ACS) in which adaptive adjustment was used for the bird’s nest update rate was proposed. The ACS-Bagging-WNN model combined ACS and Bagging-WNN was established. Simulation shows that, compared with other models above, the ACS-Bagging-WNN model has a higher accuracy and better overall performance.