Research On Improved Clustering Algorithm And Its Application In Traffic Flow Prediction

Traffic flow prediction is essential in modern intelligent traffic system management and intelligent city construction.An improved density-based clustering algorithm(DBSCAN algorithm)is combined with a radial basis(RBF)neural network optimized by genetic algorithm to predict traffic flow.The main work and innovation of this paper are as follows.First of all,in the process of processing traffic flow data set,the longitude and latitude are converted into plane coordinates,which is convenient to calculate geodesic distance accurately and makes up for the shortcoming of calculating geodesic distance directly by using longitude and latitude.Secondly,the improved DBSCAN clustering algorithm is used to process complex traffic flow data.There are two methods of improvement in this paper.First,the scan radius(Eps)parameter in DBSCAN algorithm is selected by calculating the distance between samples and generating the distance matrix.Secondly,DBSCAN algorithm is improved by constructing multi-dimensional search tree(KD tree),so as to achieve the effect of accelerating the clustering speed.Through the analysis and comparison of experimental results,it can be seen that with the increase of data volume,the running speed of the improved DBSCAN algorithm increases more and more obviously.When the amount of data is 1000,the running time of the improved DBSCAN algorithm is reduced by 54 ms and the speed is increased by about 48.65%.When the amount of data is 10000,the running time of the improved DBSCAN algorithm is reduced by 1.419 s and the speed is increased by about 456.27%.When the amount of data is 100000,the running time of the improved DBSCAN algorithm is reduced by 4h26min59.68 s and the speed is increased by about 3708.26%.Experimental results show that the improved DBSCAN algorithm improves the speed of data processing.Thirdly,this paper combines the improved DBSCAN algorithm with the RBF neural network optimized by the genetic algorithm to design a traffic flow prediction model.Compared with other neural network algorithms,radial basis(RBF)neural network not only has simple rules,easy to implement,but also has stronger learning ability and better effect.By analyzing the experimental results of the traffic flow prediction model before and after the algorithm is improved,the effectiveness of the improved prediction model is further demonstrated.The experimental results show that,compared with the RBF neural network prediction model optimized by genetic algorithm without clustering,the root mean square error of the improved prediction model in this paper increases by about 13.93 and 21.57% in terms of accuracy.Compared with ARIMA prediction model,the root mean square error of the prediction model in this paper is increased by about 40.77 and44.59%.In summary,it is proved that the improved traffic flow prediction model in this paper improves the real-time and accuracy of traffic flow prediction,and has certain innovation and good practical value.