High-speed traffic flow high-precision prediction method based on multi-source disturbance characteristics

Abstract

The invention provides a high-speed traffic flow high-precision prediction method based on multi-source disturbance characteristics, and relates to the field of data prediction, and the specific steps are as follows: firstly, a multivariable entropy driving interaction field module maps the multi-source disturbance characteristics into a unified energy field, calculates joint information entropy density and constructs a joint interaction field; processing the original feature sequence; secondly, the collaborative disturbance reconstruction module adopts a learnable mapping matrix and a multi-scale mechanism to extract dynamic differences of features under different time scales, and generates enhanced disturbance response features through a decoupling network after global disturbance collaborative response is fused; then, a spatial manifold mapping and partitioning module realizes spatial expression and partitioning modeling of a traffic flow tension evolution trend; and then, the prediction module constructs an asymmetric prediction structure in combination with the disturbance amplitude factor and the weighted disturbance characteristics, adopts a mean square error, introduces a disturbance constraint term to train the model, and outputs a final traffic flow prediction result through the trained high-speed traffic flow prediction model.

Description

Technical Field

[0001] The present invention belongs to the field of data prediction, and in particular relates to a high-precision prediction method for high-speed vehicle flow based on multi-source disturbance characteristics. Background Art

[0002] With the continuous expansion of the expressway network and the continuous increase in traffic density, traffic congestion and frequent emergencies have become important factors restricting road operation efficiency and traffic safety. As a core link in the intelligent transportation system, highway traffic flow prediction has a direct impact on traffic induction, signal optimization and road resource scheduling. However, traditional traffic flow prediction methods mostly rely on historical average flow models, regression fitting and time series analysis, and are often inferred based on a single vehicle detector and historical traffic curves. They have problems such as limited data sources, insufficient response to complex disturbances, and strong prediction lag. Especially in scenarios with drastic fluctuations in traffic conditions, it is difficult to meet the needs of high-precision and real-time predictions.

[0003] In recent years, with the development of vehicle networks, road sensors, and meteorological monitoring systems, traffic forecasting methods based on deep learning have emerged. Some studies have introduced neural networks to model the temporal dependence and spatial correlation of traffic flow, which can significantly improve forecasting performance. Some scholars have also combined attention mechanisms with graph neural networks to realize the mining of road topology and spatiotemporal correlations. This type of method has made certain progress in alleviating the problem of traditional models relying on a single data source and enhancing feature expression capabilities.

[0004] However, current highway traffic flow prediction still faces many challenges: first, the multi-source heterogeneous disturbance features are inconsistent in spatial and temporal scales, and traditional feature fusion methods are difficult to accurately reflect real traffic dynamics; second, existing models fail to adequately model the asymmetric response mechanisms of traffic flow during surges and decreases, making it difficult to capture the sharp fluctuations in high-speed scenarios; in addition, prediction models generally lack the ability to adaptively control the evolution trend of traffic flow under a disturbance background, making it difficult to achieve high-frequency and high-precision predictions. Therefore, it is urgent to propose a new highway traffic flow prediction method to improve the ability to perceive traffic operation status and provide core support for the construction of smart highways and intelligent transportation systems. Summary of the Invention

[0005] The present invention provides a high-precision prediction method for high-speed vehicle flow based on multi-source disturbance characteristics. A prediction model is proposed for complex multi-source feature data of high-speed vehicle flow. The model consists of a high-order disturbance encoder module, a dynamic gated feature generator module, a cross-variable interaction modeling module and a prediction module.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose specifically includes the following steps: S1. Collect traffic flow data on highways and construct an original data set, which includes vehicle speed, weather, static road conditions, and dynamic event multi-source disturbance features; S2. Mapping the multi-source disturbance features into a unified system energy based on the original data set, constructing an energy field, using the energy field to simulate the joint information distribution and calculate the joint information entropy density, constructing a joint interaction field based on the energy field and the joint information entropy density, and processing the original features to obtain a first data set; S3. Capturing characteristic disturbance responses in the first data set through an asymmetric disturbance channel, introducing a multi-scale mechanism to perform sliding window residual calculation on the characteristic disturbance responses, performing a weighted sum operation, and obtaining a second data set using a decoupling network; S4. Calculating a structural tension tensor using a tension attenuation coefficient according to the second data set, constructing a deformation response model based on the structural tension tensor, introducing a dynamic scaling factor to calculate spatial grouping weights, and performing a weighted aggregation operation to obtain a third data set, wherein the third data set is divided into a training set and a prediction set; S5. Mapping the training set into a prediction vector using a compression mechanism based on characteristic disturbance intensity, constructing an asymmetric prediction structure based on the prediction vector, and training the model using mean square error and introducing a disturbance constraint term. S6. The prediction set is input into a trained high-speed traffic flow prediction model, and finally a traffic flow prediction value is output.

[0007] Preferably, the method for constructing the original data set of highway traffic flow collects multi-source information under the highway operation environment, and the original data set includes vehicle operation speed, weather characteristics, static road condition characteristics and dynamic event characteristics, wherein weather characteristics include temperature, wind speed, rainfall, snowfall, humidity and haze; static road condition characteristics include road speed limit, number of lanes, road length and number of entrances and exits; dynamic event characteristics include traffic accidents, road construction and holiday information, and vehicle operation data is collected in real time by speed monitoring equipment, traffic flow monitoring cameras and ground sensing coil sensors installed on key sections of highways; weather characteristic data is collected by meteorological monitoring stations and Internet of Things environmental sensors along the route, and temperature and humidity are obtained by automatic weather station sensors with a resolution of 0.1°C and 0.1%; wind speed is collected by ultrasonic anemometer with an accuracy of 0.1 m / s, and rainfall and snowfall are recorded by rain gauge and snow depth sensor with an accuracy of 0.1 mm; haze concentration is collected by PM2.5 / PM10 laser particle monitor; static road condition characteristics are obtained from the road infrastructure database; dynamic event characteristics are obtained from the traffic accident automatic detection system, construction management platform and holiday calendar database, and finally the original data set of highway traffic flow is obtained for subsequent module processing.

[0008] Preferably, in step S2, based on the original data set, the multi-source disturbance features are mapped into a unified system energy, an energy field is constructed, the energy field is used to simulate the joint information distribution to calculate the joint information entropy density, a joint interaction field is constructed based on the energy field and the joint information entropy density, and the original features are processed to obtain the first data set; Mapping the multi-source disturbance characteristics into a unified system energy to construct the energy field , which is composed of power energy, environmental energy, static road condition energy and dynamic event energy. The feature mean at each moment is obtained, and the square of the Euclidean distance between the feature vector at the current moment and the feature mean is accumulated to obtain the overall deviation value. The deviation value is weighted and scaled using the preset kernel width. The scaled deviation value is negative and then exponentially converted and normalized to obtain the density value of the probability distribution. , after taking the negative value of the density value, and then performing logarithmic transformation, the joint information entropy density is obtained , calculate the sample average joint information entropy density and the average energy field , the joint information entropy density and energy field at each moment are subtracted from the average joint information entropy density and average energy field, and the weight coefficient is introduced The differences are proportionally fused to obtain the joint interaction field , for the four types of variables in the original data, namely vehicle speed, weather, static road conditions and dynamic events, correction coefficients are assigned respectively, and after weighting the joint interaction field according to the corresponding coefficients, the weighted result is subtracted from the original feature vector to finally obtain the processed feature sequence As the first data set.

[0009] Furthermore, in view of the characteristics of diverse disturbance factors, uneven feature distribution, and drastic changes in traffic conditions in high-speed traffic flow, the present invention proposes a multivariate entropy-driven interaction field module. First, by mapping the multi-source disturbance features of high-speed vehicle speed, weather, static road conditions, and dynamic events into a unified system energy, a system energy field is constructed, and the dynamic energy, environmental energy, static road condition energy, and dynamic event energy are calculated separately and fused to express the disturbance intensity. This can effectively solve the inconsistency problem between multi-dimensional heterogeneous disturbance features and improve the physical interpretability and model adaptability of feature fusion; then, the joint information entropy density is further introduced to characterize the dense distribution degree of multi-source disturbance features, and a joint interaction field of the energy field and the joint information entropy density is constructed to dynamically perceive the disturbance significance, realize explicit correction and reconstruction of the original features, and suppress inefficient disturbances while retaining key information, thereby improving the robustness and generalization ability of the model.

[0010] Preferably, in the S3 step, the feature disturbance response in the first data set is captured by an asymmetric disturbance channel, a multi-scale mechanism is introduced to perform sliding window residual calculation on the feature disturbance response, a weighted sum operation is performed, a decoupling network is used to obtain a second data set; The external interference feature sequence in the first data set is input into the asymmetric disturbance channel, and the feature sequence is linearly transformed The transformed result is added to a bias vector , and an intermediate activation value is obtained through hyperbolic tangent function processing, and a learnable mapping matrix The activation value is linearly transformed again to output the feature disturbance response The multi-scale mechanism is introduced, the dynamic difference of each disturbance channel at different time scales is extracted by performing sliding window residual calculation on each disturbance path The cross-scale weighted fusion mechanism is introduced, and the learnable weight The dynamic differences at different time scales are weighted and summed to obtain a global disturbance collaborative response The decoupling network is used to retain effective disturbance features, and a decoupling transformation matrix is used for linear transformation, and a bias vector is superimposed, and a feature matrix is obtained through activation function processing as the second data set.

[0011] Further, in view of the problem that the feature disturbance response in high-speed traffic flow is difficult to fully represent multi-dimensional dynamic changes, the application proposes a collaborative disturbance reconstruction module. First, based on the nonlinear evolution process of multi-dimensional disturbance features in the time and space scales, an asymmetric disturbance channel is designed, and a channel-by-channel transformation is performed to obtain a disturbance response value, thereby improving the model's ability to capture disturbance change trends in different disturbance directions. Then, a multi-scale mechanism is introduced to perform sliding window residual calculation on each disturbance path to obtain the dynamic difference of each channel, thereby enhancing the model's sensitivity to disturbance amplitude and direction changes, effectively capturing disturbance instability driven by weather, road conditions and events under high-speed traffic conditions, and establishing a linkage relationship between time scales and disturbance expression. Finally, a cross-scale weighted fusion mechanism is introduced to perform weighted fusion on the dynamic differences at different scales to form a unified global disturbance collaborative response, thereby significantly improving the integration and expression ability of multi-scale disturbance information. Through a decoupling network, the dominant component in the disturbance signal is further extracted, redundant disturbance is effectively suppressed, feature refinement is realized, and a second data set is obtained, thereby providing accurate input for downstream prediction tasks.

[0012] Preferably, in step S4, a structural tension tensor is calculated based on the second data set using a tension attenuation coefficient, a deformation response model is constructed based on the structural tension tensor, a dynamic scaling factor is introduced to calculate spatial grouping weights, and a weighted aggregation operation is performed to obtain a third data set, wherein the third data set is divided into a training set and a prediction set; Each feature vector As a spatial node in the tension field, the L1 norm between the eigenvectors of any two time points and the square of the L2 norm of the sum of the two vectors are calculated, and the number of feature dimensions is used to introduce the tension attenuation coefficient. The influence degree of the feature is controlled exponentially to obtain the structural tension tensor Based on the structural tension tensor, the attenuation degree of the tension effect is regulated in the form of the inverse of the distance, and a positive number is introduced To prevent numerical anomalies caused by extremely small distances, the deformation response model is constructed to obtain the deformation response vector , mapping the deformation response vector to The spatial grouping is performed, and the square of the L2 norm of the deformation response vector and the learnable center vector is calculated. The dynamic scaling factor is introduced to construct the The weight of the spatial grouping , weighted aggregation is performed based on grouping weights, and the aggregation representation combination of the output spatial groups is output As the third data set, the third data set is divided into a training set and a prediction set in a ratio of 7:3.

[0013] Furthermore, to address the problem that disturbance features in high-speed traffic flow fluctuate dramatically over time and are easily diffused in space, the present invention proposes a spatial manifold mapping partitioning module. First, based on the relative position relationship between features and the coupling of tension potential energy, a structural tension tensor is constructed, which can dynamically perceive the attraction and repulsion between feature points and accurately characterize the propagation intensity and direction of the disturbance in the local space; then, a deformation response model is constructed under the drive of the structural tension tensor to simulate the vector offset behavior under local tension disturbance, and dynamic clustering and center update of features are realized through dynamic scaling factors and weighted fusion mechanisms, thereby enhancing the model's ability to resolve deformation responses to different types of disturbances and improving its adaptability to complex road conditions and multi-source disturbances.

[0014] Preferably, in step S5, the training set is mapped into a prediction vector using a compression mechanism of characteristic disturbance intensity, an asymmetric prediction structure is constructed based on the prediction vector, and the model is trained using mean square error and introducing a disturbance constraint term; Based on the compression mechanism of the feature perturbation strength, each group of aggregation representations The perturbation amplitude factor is formed with the learnable center vector, processed by the logarithmic function, and multiplied element by element with the original aggregate representation to obtain the weighted perturbation feature. Finally, the normalized weighted average of all spatial groups is performed to obtain the predicted vector , according to the trend difference The positive and negative of the dynamic selection weight branch is linearly weighted prediction output to obtain the asymmetric prediction structure, based on the mean square error, using the disturbance suppression strength factor The disturbance constraint term is constructed, and the difference between the predicted value and the true label is measured and optimized to obtain a trained high-speed traffic flow prediction model.

[0015] Furthermore, the prediction module first constructs a prediction vector based on the aggregated spatial grouping representation, improves the anti-disturbance ability after feature compression, reduces prediction noise interference, and enhances prediction robustness; then, it constructs a historical direction symmetry judgment structure based on the asymmetric trend judgment strategy, effectively enhancing the model's sensitivity and response intensity to sudden traffic change trends, and further improving the ability to accurately characterize traffic trends; finally, the prediction error is measured based on the mean square error, and the disturbance suppression regularization term constraint is combined to improve the model stability and generalization ability, thereby significantly enhancing the prediction accuracy and dynamic adjustment performance of the high-speed traffic prediction model in complex scenarios.

[0016] Preferably, in step S6, the prediction set is input into a trained high-speed traffic flow prediction model, and finally outputs a traffic flow prediction value.

[0017] In summary, the present invention proposes a high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics, which includes a multivariable entropy-driven interaction field module, a collaborative disturbance reconstruction module, a spatial manifold mapping partition module and a prediction module. First, the multivariable entropy-driven interaction field module maps the multi-source disturbance characteristics of vehicle speed, weather, static road conditions and dynamic events into a unified energy field, calculates the joint information entropy density and constructs a joint interaction field, and differentially modulates the original feature sequence, thereby effectively improving the system's unified expression capability for the impact of multivariable disturbances; secondly, the collaborative disturbance reconstruction module adopts a learnable mapping matrix and a multi-scale mechanism to extract the dynamic differences of features at different time scales, and generates enhanced disturbance response features through a decoupling network after fusing the global disturbance collaborative response, thereby enhancing the response to high-frequency changes and asymmetry. The robustness of modeling of emergencies; then, the spatial manifold mapping partitioning module constructs a structural tension tensor, and based on the deformation response model and dynamic scaling mechanism, groups and aggregates the representation according to spatial heterogeneity to realize the spatial expression and partition modeling of the evolution trend of traffic tension; then, the prediction module combines the disturbance amplitude factor and the weighted disturbance feature to construct an asymmetric prediction structure and adopts the mean square error, introduces the disturbance constraint term to train the model, and predicts the test set through the trained high-speed traffic flow prediction model, and outputs the final traffic flow prediction result. The present invention fully integrates the multi-source nonlinear disturbance characteristics to construct a full-process modeling framework from energy density regulation to structural response drive, with stronger generalization ability, stability and sudden adaptability, and provides solid support for traffic flow prediction and regulation in complex scenarios in intelligent transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a step-by-step diagram of a high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics.

[0019] Figure 2 This is the structure diagram of the high-speed traffic flow prediction model.

[0020] Figure 3 This is the module structure diagram of the multivariable entropy-driven interaction field.

[0021] Figure 4 Module structure diagram for collaborative perturbation reconstruction.

[0022] Figure 5 The structural diagram of the spatial manifold mapping partition module.

[0023] Figure 6 This is a diagram of the model training process.

[0024] Figure 7 The prediction model realizes the prediction evaluation fitting effect diagram of high-speed traffic flow. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] See also Figure 1-Figure 7 The present invention provides a technical solution: a high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics, including a multi-variable entropy-driven interaction field module, a collaborative disturbance reconstruction module, a spatial manifold mapping partition module and a prediction module. First, the multi-source characteristics are processed by the multi-variable entropy-driven interaction field module; secondly, the collaborative disturbance reconstruction module introduces a multi-scale mechanism to perform sliding window residual calculation, and performs a weighted sum operation to extract enhanced disturbance response characteristics; then, the spatial manifold mapping partition module constructs a structural tension tensor and introduces a deformation response model to complete the spatial partitioning and tension aggregation representation; finally, an asymmetric prediction structure is constructed, and the model is trained based on the mean square error and the disturbance constraint term is introduced to achieve high-precision prediction of traffic flow. The specific steps are as follows: Figure 1 shown.

[0027] Construct a high-speed traffic flow prediction model, the structure is as follows Figure 2 shown.

[0028] S1. Collect traffic flow data on highways and construct an original data set, which includes vehicle speed, weather, static road conditions, and dynamic event multi-source disturbance features.

[0029] Furthermore, the dataset of the present invention includes 1,000 vehicle flow data, and the dataset is divided into a training set and a test set according to a ratio of 7:3.

[0030] S2. According to the original data set, the multi-source disturbance characteristics are mapped into a unified system energy, an energy field is constructed, the energy field is used to simulate the joint information distribution and calculate the joint information entropy density, a joint interaction field is constructed based on the energy field and the joint information entropy density, and the original characteristics are processed to obtain the first data set.

[0031] Furthermore, the multivariable entropy driven interaction field module structure is as follows Figure 3 As shown, time The eigenvectors under , is the dimension of the feature vector, the value is 14, is the high-speed vehicle speed, is the weather variable, is a static road condition variable, As a dynamic event variable, the multi-source disturbance characteristics are first mapped into a unified system energy to construct an energy field , the mathematical model is: ; Where, The power energy is calculated by calculating the speed variable of high-speed vehicles The square of and then divided by 2, For environmental energy, for each weather variable, the normalized square energy is defined, and the mathematical model is: ; Where, is the dimension number of weather variables, the value is 6, representing the 6 dimensions of temperature, wind speed, rainfall, snowfall, humidity, and haze. For the Dimensional weather variables, is the global mean of the weather variable, is the global standard deviation, is the weight parameter of the weather variable, which represents the impact of different weather variables on high-speed traffic flow. The weight parameter value of rainfall is 0.26, the weight parameter value of snowfall is 0.22, the weight parameter value of haze is 0.18, the weight parameter value of wind speed is 0.14, the weight parameter value of humidity is 0.12, and the weight parameter value of temperature is 0.08.

[0032] Furthermore, in the mathematical model of energy field The energy of static road conditions is relatively fixed, but it also affects the traffic flow. The mathematical model is: ; Where, is the dimension number of the static road condition variable, with a value of 4, representing the speed limit, number of lanes, road length, and number of entrances and exits. is the weight parameter of the road condition, the weight parameter value of the number of lanes is 0.3, the weight parameter value of the speed limit is 0.26, the weight parameter value of the number of entrances and exits is 0.24, and the weight parameter value of the road length is 0.2. For the Static road condition variables of dimensions, It is a function that maps highway traffic variables to energy, where the energy function of the speed limit variable is The mathematical model is: ; Where, To adjust the parameter, the value is 0.3 to prevent the current speed limit When the speed limit is minimum When it is 0, The minimum speed limit for the entire expressway. The energy function of the maximum speed limit and number of lanes on the entire highway The energy function of the road length is calculated by adding 1 to the number of highway lanes and then taking the inverse. The energy function of the number of entrances and exits is calculated by adding 1 to the road length and then taking the inverse. It is calculated by adding 1 to the number of entrances and exits and taking the inverse.

[0033] Furthermore, in the mathematical model of energy field is the dynamic event energy, and the mathematical model is: ; Where, is the dimension number of the dynamic event variable, and its value is 3, representing the three dimensions of accidents, construction, and holidays. is the energy weight of different events, the weight of accidents is 0.37, the weight of holidays is 0.33, and the weight of construction is 0.3. It indicates 1 when the event occurs and 0 when it does not occur. For the Dynamic event variables of 2 dimensions.

[0034] Furthermore, the probability distribution of joint features is simulated, and the joint information entropy density is constructed based on the information entropy and thermodynamic distribution principles. , the mathematical model is: ; Where, is the density value of the probability distribution, and the mathematical model is: ; Where, is a normalization parameter used to ensure The sum of all density values ​​at the moment is 1, thus forming an effective joint disturbance feature distribution. The mathematical model is: ; Where, The number of samples participating in the normalization calculation at the same time, the value depends on the actual number of samples at the same time, is the kernel width, the value is 1.5, For time The j-th eigenvector under For the moment The characteristic mean of is the square of the Euclidean distance between the current feature vector and the overall center. The larger the distance, the higher the degree of feature disturbance at the time point. Then, the joint interaction field is constructed based on the cross entropy gradient principle. , which is used to reflect the spatial propagation direction and intensity of the information disturbance gradient and the dynamic disturbance gradient. The mathematical model is: ; Where, For time point The average joint information entropy density of all samples is Time point The average energy field of all samples, is the weight coefficient of energy and information coupling, and its value is 0.25.

[0035] Furthermore, the joint interaction field is used to process the original data characteristics, and the mathematical model is: ; Where, is the correction coefficient vector of the variable. The correction coefficient value of the vehicle speed variable is 0.23, the correction coefficient value of the weather variable is 0.1, the correction coefficient value of the static road condition variable is 0.2, and the correction value of the dynamic event variable is 0.18. Finally, the processed feature is obtained. as the first dataset.

[0036] S3. Capture the characteristic disturbance response in the first data set through an asymmetric disturbance channel, introduce a multi-scale mechanism to perform sliding window residual calculation on the characteristic disturbance response, perform a weighted sum operation, and use a decoupling network to obtain the second data set.

[0037] Furthermore, in step S3, the first data set is input into the collaborative disturbance reconstruction module for deep reconstruction, thereby extracting stable disturbance structure features and providing enhanced representation for high-precision prediction of high-speed traffic flow. The process is as follows: Figure 4 The specific steps are as follows: Aiming at the nonlinear and diversified trajectory behaviors presented by multi-dimensional disturbance characteristics on the time axis, the design There are asymmetric disturbance channels, each of which captures the characteristic disturbance response through a nonlinear path mapping function. The mathematical model is: ; Where, represents the asymmetric perturbation channel index, and Represents the learnable mapping matrix corresponding to the channel, represents the bias vector, For the Path at time The characteristic disturbance response of .

[0038] Furthermore, in this example, The value is 3 to describe the disturbance mechanisms of weather, static road conditions, and dynamic events in highway traffic data, ensuring that the model has the ability to model multi-source disturbance behaviors in a channel-by-channel manner. Use uniform distribution initialization, that is , according to the back-propagation algorithm through gradient descent update, the error of each layer will be passed back through the chain rule, the gradient will be calculated, and updated in each training iteration. The update formula is: , is the learning rate, set , is the cross entropy loss function; Use normal initialization, that is , the update formula is , through gradient update, and It will be gradually optimized during training, accurately mapping the perturbation path, thereby improving the model's ability to learn different perturbation channels; the bias vector In order to avoid bias in the initial stage, the network is optimized from scratch during training. The initial setting is all zero vectors, and then updated during the back propagation process. The update formula is .

[0039] Furthermore, in order to further improve the model's ability to learn external disturbance signals, a multi-scale mechanism is introduced based on the theoretical basis of trend residual modeling. By performing sliding window residual calculations on each disturbance path, the dynamic differences of each disturbance channel at different time scales are extracted. , capturing the time-varying pattern of disturbance characteristics, the mathematical model is: ; Where, It is used to capture the disturbance features of different scales in weather, static road conditions and dynamic events, helping the model to extract the differences in multi-scale disturbances, enhancing the ability to respond to disturbance signals of different scales, and improving the generalization performance of the model. For the The characteristic disturbance response of each path at time j.

[0040] Furthermore, in order to make full use of the information in the multi-scale residuals, this example introduces a cross-scale weighted fusion mechanism based on the information weighting theory. The dynamic differences at different scales are weighted and synthesized through learnable weights to form a unified global disturbance coordinated response. , used for subsequent structural decoupling, the mathematical model is: ; Where, is the learnable weight, indicating the Path and The fusion coefficient of each scale is weighted to the response of disturbances of different scales to ensure that the model can dynamically adjust the learning and fusion mode according to the information of different disturbance paths and scales. The initial value is set to 0.3. Dynamic value between.

[0041] Furthermore, in order to obtain the global disturbance coordinated response After that, a decoupling network is used to process the disturbance signal to separate the dominant component and the noise component, retaining the effective disturbance characteristics. The mathematical model is: ; Where, is the decoupling transformation matrix, used to adjust the global disturbance response, Use uniform distribution initialization, that is , the update formula is: , is the learning rate, set , is the cross entropy loss function, and regularization is used to avoid overfitting, so that the weight matrix will not be too large and the complexity of the model is maintained. The formula is: , is the regularization coefficient, and its value is ; is the bias vector, which is initially set to a zero vector and is subsequently updated during the back propagation process. The update formula is: Indicates time The decoupled disturbance signal removes noise disturbance through adaptive decoupling transformation, retains the main disturbance factors affecting high-speed traffic flow, makes the features more refined and has stronger predictive ability, and the final output feature matrix is As the second data set, it is used as input for subsequent modules.

[0042] S4. Calculate the structural tension tensor using the tension attenuation coefficient based on the second data set, construct a deformation response model based on the structural tension tensor, introduce a dynamic scaling factor to calculate the spatial grouping weight, and perform a weighted aggregation operation to obtain a third data set, where the third data set is divided into a training set and a prediction set.

[0043] Furthermore, in step S4, the second data set Input, spatial manifold mapping partition module performs spatial grouping processing on features, the process is as follows Figure 5 The specific steps are as follows: Highway traffic data has obvious disturbance coupling, time drift and strong nonlinear evolution characteristics. This example is based on tension mechanics and transforms each eigenvector into As a spatial node in the tension field, it simulates the attraction and repulsion relationship between features, drives the feature points to produce deformation response in the manifold, and then completes the dynamic grouping operation of spatial clustering. First, for any two time points , define the structural tension tensor The mathematical model for simulating the tension relationship between the cooperative disturbance and the external signal fusion feature space is: ; Where, is the tension attenuation coefficient, which is used to control the influence of long-distance tension and has a value of 0.7. Indicates the measurement of relative position relationship. The coupling amount that represents the overall strength of the features of two nodes enhances the perception of the total energy of the features.

[0044] Furthermore, based on the structural tension tensor, a deformation response model is constructed to simulate the vector deviation behavior under local tension disturbance. The mathematical model is: ; Where, Indicates time The deformation response vector of To prevent positive numbers introduced by division by zero, the value is set to The above design embodies a "tension manifold gravity", which drives the characteristic vector to produce directional deviation under the guidance of tension.

[0045] Furthermore, the deformation response vector Map to spatial grouping, introducing a dynamic scaling factor, constructing the The weight of the spatial grouping , the mathematical model is: ; Where, For the The learnable center vector of the space group, is the dynamic scaling factor, and the mathematical model is: ; Where, The preset number of partition groups is 500. For the The learnable center vector of the space group.

[0046] Furthermore, before model training, the deformation response vectors of all high-speed traffic data samples are first calculated. Perform cluster initialization to obtain the initial center vector of each space group. The mathematical model is: ; To improve stability and cluster consistency, this example introduces an exponential sliding average mechanism to dynamically update the center vector without gradient. The mathematical model is: ; Where, is a smoothing factor used to control the update amplitude, with an initial value of 0.8. Dynamic value between For the The weighted average of all deformation response vectors in a spatial grouping is calculated. Through the above design, the model can adjust its center over time to adapt to different disturbance modes.

[0047] Furthermore, weighted aggregation is performed according to the weights to generate an aggregate representation of each spatial grouping , the mathematical model is: ; Finally, the combination of all space groups is output The third dataset is divided into a training set and a test set according to a ratio of 7:3.

[0048] S5. Use a compression mechanism of feature disturbance intensity to map the training set into a prediction vector, construct an asymmetric prediction structure based on the prediction vector, use mean square error, and introduce disturbance constraint terms to train the model.

[0049] Furthermore, the training set was input into the highway traffic flow prediction model. The model uses the PyTorch deep learning framework and is accelerated by the NVIDIA V100 32GB GPU. During the training process, the batch size is set to 128. During the training phase, the training set is first mapped to a prediction vector based on the compression mechanism of the feature perturbation intensity. , to capture the changing pattern of high-speed traffic flow, the mathematical model is: ; Where, For the The aggregate representation of spatial groupings reduces the interference of small perturbations by implicitly fusing perturbation perception and feature compression.

[0050] Furthermore, an asymmetric prediction structure is constructed based on the historical traffic flow change direction to improve the response capability to drastic fluctuations in traffic flow and form an accurate prediction mechanism for traffic surges and drops in high-speed scenarios. The mathematical model is: ; Where, is the predicted value of highway traffic flow, The difference between the current and previous traffic volumes is used as a basis for trend judgment. The trainable weight vector representing the upward trend branch is initialized using a normal distribution, i.e. , the update formula is , is the learning rate, set , is the cross entropy loss function, The trainable weight vector representing the descending trend branch is initialized with uniform distribution, i.e. , the update formula is , is the bias term, which is initially set to a zero vector and is subsequently updated during the back propagation process. The update formula is: , is the bias term, the initial setting and update formula are the same as consistent.

[0051] Furthermore, to further improve the model's prediction accuracy, this example uses the mean square error (MSE) and adds a perturbation constraint to enhance stability. This measure and optimize the difference between the predicted value and the true label. The mathematical model is: ; Where, For the moment The traffic flow prediction value, For real high-speed traffic flow, is the disturbance suppression intensity factor, which is used to control the degree of trade-off between the two parts of the loss function. Its value is 0.08. During the training process, the model minimizes the above loss function through the Adam optimizer, iteratively updates the network parameters, and gradually approaches the global optimal prediction structure. The introduction of the disturbance suppression term effectively curbs the disturbance expression deviation caused by abnormal data or emergencies, and improves the prediction stability of the model under high volatility and high uncertainty road conditions. The training process is as follows: Figure 6 As shown in the figure, the number of training rounds is 70. It can be seen from the figure that with the increase of training discussion, the loss value gradually decreases from the high level in the initial stage and tends to be stable in the later stage, indicating that the model converges stably during the training process. Finally, a well-trained high-speed traffic flow prediction model is obtained.

[0052] S6. The prediction set is input into a trained high-speed traffic flow prediction model, and finally a traffic flow prediction value is output.

[0053] Furthermore, the high-speed traffic flow prediction model realizes the traffic flow prediction fitting effect diagram as shown in the figure below: Figure 7As shown in the figure, the horizontal axis is time, the vertical axis is traffic volume value, the dot solid line represents the actual traffic volume value, and the cross dotted line represents the model prediction value. It can be seen from the figure that the prediction value is roughly similar to the change trend of the actual high-speed traffic volume value. The experimental results show that the high-speed traffic volume prediction model can effectively capture the change trend of high-speed traffic volume data and can better predict the high-speed traffic volume value.

Claims

1. A high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics, characterized by: Collecting traffic flow data on highways to construct an original data set, wherein the original data set includes vehicle speed, weather, static road conditions, and dynamic event multi-source disturbance characteristics; According to the original data set, the multi-source disturbance features are mapped into a unified system energy, an energy field is constructed, the energy field is used to simulate the joint information distribution and calculate the joint information entropy density, a joint interaction field is constructed based on the energy field and the joint information entropy density, and the original features are processed to obtain a first data set; Capturing characteristic disturbance responses in the first data set through an asymmetric disturbance channel, introducing a multi-scale mechanism to perform sliding window residual calculation on the characteristic disturbance responses, performing a weighted sum operation, and obtaining a second data set using a decoupling network; Calculating a structural tension tensor using a tension attenuation coefficient according to the second data set, constructing a deformation response model based on the structural tension tensor, introducing a dynamic scaling factor to calculate spatial grouping weights, and performing a weighted aggregation operation to obtain a third data set, wherein the third data set is divided into a training set and a prediction set; The training set is mapped into a prediction vector using a compression mechanism of characteristic disturbance intensity, an asymmetric prediction structure is constructed based on the prediction vector, and a mean square error is used to train the model by introducing a disturbance constraint term; The prediction set is input into a trained high-speed traffic flow prediction model, and ultimately outputs a traffic flow prediction value.

2. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 1 is characterized in that: The original data set includes vehicle speed, weather, static road conditions, and multi-source disturbance features of dynamic events. Weather features include temperature, wind speed, rainfall, snowfall, humidity, and haze; static road condition features include road speed limit, number of lanes, road length, and number of entrances and exits; dynamic event features include traffic accidents, road construction, and holiday information.

3. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 2 is characterized in that: Mapping the multi-source disturbance characteristics into a unified system energy to construct the energy field , which is composed of power energy, environmental energy, static road condition energy and dynamic event energy. The feature mean at each moment is obtained, and the square of the Euclidean distance between the feature vector at the current moment and the feature mean is accumulated to obtain the overall deviation value. The deviation value is weighted and scaled using the preset kernel width. The scaled deviation value is negative and then exponentially converted and normalized to obtain the density value of the probability distribution. , after taking the negative value of the density value, and then performing logarithmic transformation, the joint information entropy density is obtained .

4. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 3 is characterized in that: Calculate the sample average joint information entropy density and the average energy field , the joint information entropy density and energy field at each moment are subtracted from the average joint information entropy density and average energy field, and the weight coefficient is introduced The differences are proportionally fused to obtain the joint interaction field , for the four types of variables in the original data, namely vehicle speed, weather, static road conditions and dynamic events, correction coefficients are assigned respectively, and after weighting the joint interaction field according to the corresponding coefficients, the weighted result is subtracted from the original feature vector to finally obtain the processed feature sequence As the first data set.

5. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 4 is characterized in that: The external interference feature sequence in the first data set is input into the asymmetric perturbation channel, and the learnable mapping matrix Perform linear transformation on the feature sequence and compare the transformation result with the bias vector Add, and get the intermediate activation value after the hyperbolic tangent function processing, using the learnable mapping matrix The activation value is linearly transformed again to output the characteristic disturbance response , introduce the multi-scale mechanism, and extract the dynamic differences of each disturbance channel at different time scales by performing sliding window residual calculation on each disturbance path .

6. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 5 is characterized in that: Introducing a cross-scale weighted fusion mechanism and utilizing learnable weights The dynamic differences at different time scales are weighted and summed to obtain the global disturbance coordinated response , the decoupling network is used to retain the effective disturbance characteristics, and the decoupling transformation matrix Perform linear transformation and superimpose bias vector , after activation function processing, the feature matrix is ​​obtained as the second data set.

7. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 6 is characterized in that: Each feature vector As a spatial node in the tension field, the L1 norm between the eigenvectors of any two time points and the square of the L2 norm of the sum of the two vectors are calculated, and the number of feature dimensions is used to introduce the tension attenuation coefficient. The influence degree of the feature is controlled exponentially to obtain the structural tension tensor Based on the structural tension tensor, the attenuation degree of the tension effect is regulated in the form of the inverse of the distance, and a positive number is introduced To prevent numerical anomalies caused by extremely small distances, the deformation response model is constructed to obtain the deformation response vector .

8. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 7 is characterized in that: Map the deformation response vector to The spatial grouping is performed, and the square of the L2 norm of the deformation response vector and the learnable center vector is calculated. The dynamic scaling factor is introduced to construct the The weight of the spatial grouping , weighted aggregation is performed based on grouping weights, and the aggregation representation combination of the output spatial grouping is output As the third data set, the third data set is divided into a training set and a prediction set in a ratio of 7:

3.

9. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 8 is characterized in that: Based on the compression mechanism of the feature perturbation strength, each group of aggregation representations The perturbation amplitude factor is formed with the learnable center vector, processed by the logarithmic function, and multiplied element by element with the original aggregate representation to obtain the weighted perturbation feature. Finally, the normalized weighted average of all spatial groups is performed to obtain the predicted vector , according to the trend difference The positive or negative value of is dynamically selected, and the weight branch is performed to perform linear weighted prediction output to obtain the asymmetric prediction structure.

10. The high-precision prediction method for high-speed traffic flow based on multi-source disturbance characteristics according to claim 9 is characterized in that: Based on the mean square error, the disturbance suppression strength factor is used The disturbance constraint term is constructed, the difference between the predicted value and the true label is measured and optimized, a trained high-speed traffic flow prediction model is obtained, and finally the traffic flow prediction value is output.

Images