Research On Short-term Traffic Flow Forecasting Based On Path Clustering In Space-time Fusion Model

In recent years,with the acceleration of China's urbanization process and the vigorous development of the automotive industry,the number of motor vehicles has greatly increased,and the contradiction between the capacity of urban road vehicles and the growing demand for vehicles in urban transportation has become increasingly apparent.Therefore,the use of intelligent transportation systems to effectively improve and alleviate urban traffic congestion is currently the focus and important means of vigorous development for countries to deal with the problem of urban traffic congestion.The prediction of urban road traffic flow is the key technology foundation of the intelligent transportation system,so it is of great significance to whether the urban intelligent transportation system can provide accurate and efficient travel services.Among them,the traffic flow data has strong autocorrelation characteristics,showing certain periodic,nonlinear and local non-stationary characteristics in time,and is also spatially affected by the specific urban traffic road network structure.This article aims at these characteristics and considerations In view of some defects of the existing algorithms,new prediction algorithms and models are re-proposed and constructed.First of all,according to the traffic flow data has the spatial characteristics of spatial interaction and influence of traffic,and the traditional traffic flow prediction only considers the temporal autocorrelation of traffic flow data,and ignores the spatial characteristics of traffic flow data,so this thesis It is proposed to construct a spatial path clustering algorithm to extract the traffic flow data of points with strong spatial correlation with the predicted road segment as the spatial sample data,and as the input of the modified particle swarm wavelet neural network(PSO-WNN)Build a spatial prediction model.Secondly,in view of the steady and non-stationary characteristics of traffic flow time sample data,this thesis proposes a new piecewise weighted fitness function to construct a particle swarm wavelet neural network(PWFF-PSO-WNN)to make the algorithm allocate in the optimization process Relatively larger adjustment weights are given to non-stationary data segments,and considering that the particle swarm optimization algorithm is easy to fall into the local minimum during the optimization process,a strategy for inertial weight attenuation suitable for traffic flow data samples in this thesis is proposed,and the fitness curve is compared get conclusion.Then,in order to fuse the temporal and spatial characteristics of traffic flow data with each other and maintain good adaptability,this thesis proposes a weighted fusion method of spatio-temporal prediction,based on the time series composed of several predicted output values of the most recent historical moments of temporal prediction and spatial prediction The cost function composed of sequence mean square error weighting determines the adjustment direction of each weighting coefficient,so as to achieve the purpose of organic integration of space and time.Finally,this thesis encodes each model through the MATLAB experimental platform,and then analyzes the prediction model examples and compares the prediction results.It is concluded that the time-space fusion short-term traffic flow prediction method constructed in this thesis performs better than the prediction of a single time or space sample.To achieve the purpose of improving the optimization results of traffic flow forecast.