Research On Chaotic Characterization And Prediction Of Traffic Data

Traffic flow prediction is one of the key components of intelligent transportation system.Accurate and efficient traffic flow prediction can provide strong support for macroscopic road network regulation and control by traffic control departments,and provide reasonable suggestions for route planning of urban travelers,so as to fine-grained road network resources deployment,which is one of the effective means to alleviate traffic congestion on urban roads and distribute the utilization rate of urban road network in a balanced manner.The gradual increase of locomotive ownership and the expanding scale of urban road networks have promoted the development of traffic flow big data research,and the rapid emergence of data sensing technology and machine learning technology also provides technical support for the feature processing capability of traffic flow big data.In view of the complex nonlinear dynamics of traffic flow affected by various non-stationary real factors,how to use feature analysis methods to mine the internal features of data from chaotic traffic flow with complex nonlinear characteristics and combine deep abstraction ability of deep learning to sense the future nonlinear evolution of chaotic traffic flow,so as to provide reliable technical support for urban intelligent transportation system,has become one of the research hotspots in the field of intelligent transportation.The paper takes measured chaotic traffic flow as the data object,realizes the mapping of singular attractors collapsed in the low-dimensional space to the highdimensional phase space domain,and combines hybrid deep learning modeling theory and intelligent algorithm optimization strategy to propose a research framework of nonlinear evolutionary trend analysis of measured traffic flow,chaotic phase space reconstruction and traffic flow time series prediction based on chaotic features inside the data.Firstly,the research on the prediction model of chaotic systems is reviewed in detail,the shortcomings of the current field of chaotic system prediction are summarized,and the research direction and focus of the paper are elaborated,the specific research angle is planned and the structure of the paper is constructed.Secondly,a chaotic data pre-processing stage is conducted to propose a chaos-free singular value smoothing scheme to filter the data noise without damaging the nonlinear characteristics of the data,and to verify the chaotic nature of the data and the losslessness of the noise removal scheme based on the chaos determination theory,so as to provide a data basis for the subsequent chaotic characterization of the data.Then,the highdimensional reconstruction operation of chaotic traffic flow data is carried out to map the lowdimensional chaotic traffic flow time series to the high-dimensional space by the phase space reconstruction algorithm,and to address the problem of low accuracy and efficiency of the phase point spacing metric in the calculation of the correlation dimension of the reconstruction algorithm,a fused parametric search domain is proposed based on the topological equivalence principle to achieve a game balance between efficiency and accuracy.Finally,a hybrid deep learning model suitable for the organization of high-dimensional phase space tensor data is used for the prediction of chaotic traffic flow.The contrast scattering algorithm in the two-layer restricted Boltzmann machine adopts a layer-by-layer complete pre-training training scheme,which can effectively avoid the problem of sparse feature information in the long-distance transmission process.The convolution component of the model can also efficiently process the high-dimensional phase space tensor data and facilitate the extraction of phase point space partial row features in the phase space.At the same time,there is a time-dependent relationship between the phase points in the chaotic phase space,so a simple gated cyclic unit with longtime memory capability is introduced to learn the temporal features.In order to improve the robustness of the hybrid model,a genetic algorithm optimization component with dynamic probability reproduction mechanism is proposed to address the problem that the training process is prone to local optimality due to the complex data structure of the high-dimensional phase space and the deep layers of the hybrid model.The above research idea provides a new modeling idea for constructing a hybrid prediction model of measured traffic flow based on chaotic feature analysis.This study provides a systematic feature analysis idea and prediction model construction method for chaotic phase space reconstruction and nonlinear spatio-temporal feature extraction of measured traffic flow,and builds a highly robust traffic flow prediction model based on internal chaotic feature extraction of traffic flow data,which promotes the development of analysis of internal chaotic evolutionary dynamics of data in the field of intelligent transportation and contributes to the alleviation of urban road congestion.