Research On Nonlinear Short-term Traffic Flow Forecasting Based On Time-space Fusion Model

Short-term traffic flow forecasting is at the core of intelligent transportation system.In recent years,the accuracy of short-term traffic flow prediction has gradually been regarded as a key issue in the traffic field.In the face of the increasing requirements of prediction accuracy in complex traffic environment,it is difficult to use a single model to achieve short-term traffic flow forecasting with multiple characteristics,such as time dependence,spatial dependence,non-linear and so on.According to different traffic flow characteristics,this thesis studies the traffic flow time prediction model and time-space prediction model,and then uses variational mode decomposition to transform nonlinear traffic flow into stationary traffic flow components,and a short-term traffic flow combination forecasting model based on time-space fusion is proposed.First of all,in view of the lack and anomaly of traffic data,the Grey correlation analysis algorithm is used to analyze the time periodicity of traffic flow in order to repair abnormal traffic flow.At the same time,according to the time-dependent characteristics of traffic flow,an improved bidirectional long short term memory neural network based on attention mechanism is proposed to improve the accuracy of time prediction model.Finally,experiments show that the ability of extracting time features can be fully improved by using bidirectional long short term memory neural networks to extract time features from top to bottom and from bottom to top,and it is verified that the use of attention mechanism is helpful to improve the fusion ability of time features.At the same time,according to the characteristics of time-space dependent traffic flow,this thesis combines the Grey correlation analysis algorithm with Manhattan distance,proposes a traffic spatial node clustering algorithm,and according to the spatial characteristics of the topological map of the traffic network,studies the use of multi-attention mechanism in the graph structure to mine the time-space dependence of traffic nodes in the graph attention neural network.At the same time,a short-term traffic flow prediction model based on attention mechanism fusion graph attention neural network time-space vector is proposed.Considering the spatial characteristics of traffic network topology can improve the accuracy of one-step and multi-step traffic flow prediction.Then,according to the above research on time-dependent and time-space dependence of traffic flow,a time-space fusion combined forecasting model of nonlinear short-term traffic flow is proposed.First of all,the improved particle swarm optimization algorithm is used to optimize the variational mode decomposition algorithm to decompose the nonlinear traffic flow to ensure that the traffic flow is fully decomposed;then,aiming at the effective decomposition mode in the time-space dimension,the time-space fusion components of different graph attention neural networks and bidirectional long short term memory neural networks with improved attention mechanism are constructed.Finally,the research uses the main time-space characteristics of attention focus to improve the accuracy of traffic flow forecasting;the experimental results show that fully considering the inherent law of traffic flow and flexibly using attention mechanism to build a targeted deep learning model can improve the effect of short-term traffic flow prediction.Finally,this thesis summarizes the research work of the full text,and further points out the further research direction of short-term traffic flow forecasting.