Traffic control method based on vehicle networking accident early warning and dynamic resource scheduling

Abstract

The application discloses a traffic control method based on vehicle networking accident early warning and dynamic resource scheduling, and steps include: collecting and preprocessing data, forming a unified format feature vector dataset of each sub-range, transforming the feature vector dataset and extracting potential variables, then outputting an accident risk prediction value through a dynamic accident prediction module, then combining the current prediction value with resource distribution data at the last moment to calculate the accident influence of each region, and finally dynamically allocating emergency resources according to the accident influence. The application can integrate multi-source data in real time through vehicle networking and generate a unified feature vector, solve the problem of data dispersion in traditional methods, and has the advantages of high early warning accuracy, stable model, prevention of catastrophic forgetting, more reasonable emergency resource allocation and the like.

Description

Technical Field

[0001] This invention belongs to the field of traffic control technology, and specifically relates to a traffic control method. Background Technology

[0002] With the acceleration of urbanization and the rapid increase in the number of motor vehicles, urban traffic management systems have become an indispensable and important component of modern urban infrastructure. Frequent traffic accidents not only cause numerous casualties and property losses, but also seriously affect urban traffic flow and residents' daily lives. Therefore, establishing an effective accident early warning and resource allocation system is of great significance for improving traffic management.

[0003] Traditional traffic accident prediction methods primarily rely on post-accident reports and static historical data, lacking diversity and real-time capabilities. In recent years, emerging technologies such as Vehicle-to-Everything (V2X) have been increasingly applied in traffic management, providing new opportunities for accident early warning and resource scheduling systems. V2X technology enables efficient communication between vehicles (V2V), between vehicles and infrastructure (V2I), and between vehicles and pedestrians (V2P), allowing for the real-time collection and transmission of multi-source traffic data. This data not only greatly enriches the input information for accident prediction but also provides strong support for improving the accuracy and real-time performance of prediction models.

[0004] However, current V2X-based solutions for incident early warning and resource scheduling have the following drawbacks:

[0005] First, the lack of an effective integration mechanism for data from different devices and sensors, such as vehicle dynamics, road conditions, and environmental parameters, makes it difficult for the model to perform comprehensive analysis.

[0006] Secondly, the distribution of different feature data usually varies greatly, and problems such as uneven data distribution and noise interference also affect the convergence speed and stability of the model.

[0007] Furthermore, existing accident prediction systems lack flexible dynamic adjustment capabilities, especially in resource scheduling. As traffic data is updated, traditional static models struggle to adapt to new data characteristics in a constantly changing environment, requiring model retraining. New data not only improves model performance in new scenarios but also leads to performance degradation in older scenarios, resulting in catastrophic forgetting and failing to provide effective support for resource scheduling. Summary of the Invention

[0008] This invention proposes a traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling. Its purpose is to solve the problems of data ineffective integration, uneven data distribution, and catastrophic forgetting after model retraining in existing methods.

[0009] The technical solution of this invention is as follows:

[0010] A traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling includes the following steps:

[0011] Step 1: Collect data and preprocess the data to form a feature vector dataset in a unified format;

[0012] Traffic control areas are divided into multiple regions, each region contains multiple sub-regions, and each sub-region obtains a feature vector dataset at the same time.

[0013] Step 2: Perform feature transformation on the feature vector dataset to obtain the transformed feature vectors;

[0014] Step 3: Use a variational autoencoder to encode and map the transformed feature vectors to obtain the corresponding latent variables;

[0015] Step 4: Input the potential variables into the dynamic accident prediction module to obtain the accident risk prediction value;

[0016] Step 5: Combine the predicted accident risk values ​​of each sub-region in each region at the current moment with the resource distribution data of each region at the previous moment to obtain the current accident impact of each region.

[0017] Step 6: Allocate emergency resources according to the impact of the accident in each region.

[0018] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling: In step 1, multi-source traffic data is collected in real time through the V2X network. The collected data includes vehicle dynamic information, road condition data, environmental parameters, and road network characteristic data.

[0019] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling: In step 1, the collected data is cleaned, denoised, and time-synchronized to ensure that data from different sensors and devices are aligned at the same point in time, thereby forming a feature vector dataset in a unified format; the feature vector dataset obtained at a certain moment is represented as:

[0020] ;

[0021] in, Represents the first eigenvector in the dataset. One feature;

[0022] In step 2, the feature vector dataset obtained in step 1 is transformed to unify the distribution of each feature, improve the skewness and kurtosis of the data, and obtain the transformed feature vector. ,in:

[0023] ;

[0024] in, It is the transformed eigenvector The first in 1 eigenvalue, These are transformation parameters.

[0025] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling: In step 3, the transformed feature vector... The input is fed into the encoder of the variational autoencoder to obtain the latent variable output in this step. ;

[0026] The training process of the variational autoencoder is as follows:

[0027] Construct training samples, each containing a feature vector. ;

[0028] eigenvectors The inputs are fed into the encoder, and the mean vector of the latent space is calculated through multiple hidden layers. and standard deviation vector And assume the corresponding latent variables Follows Gaussian distribution :

[0029] ;

[0030] in, Representing variables Follows the mean vector The covariance matrix is ​​a diagonal matrix. The normal distribution;

[0031] Generate latent variables through reparameterization Specific samples, latent variable elements in each dimension Described by the mean and standard deviation of each latent dimension:

[0032] ;

[0033] in, It is noise sampled from the standard normal distribution, ensuring the latent variable elements Differentiability, and The mean vector obtained through the encoder and standard deviation vector The corresponding element;

[0034] The latent variables are decoded through a variational autoencoder. Mapping back to the original data space yields :

[0035] ;

[0036] in, , The latent variables obtained by the decoder are respectively The mean and variance;

[0037] Through each group and The reconstruction error is obtained, and the network parameters of the variational autoencoder are updated by backpropagation with the goal of minimizing the reconstruction error.

[0038] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling: in step 4, the dynamic accident prediction module includes an MLP model.

[0039] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling, the specific processing procedure of step 4 is as follows:

[0040] Step 4-1: Combine the results obtained in Step 3 with... Latent variables at time 1 Inputting the data into an MLP model yields preliminary predictions of accident risk. ;

[0041] Specifically, the processing procedure of the MLP model is as follows:

[0042] ;

[0043] in, and These are the Sigmoid activation function and the ReLU activation function, respectively. and It is the weight matrix of the MLP model. and It is the bias term of the MLP model;

[0044] Step 4-2: Introduce the current real-time resource scheduling information and calculate the resource feedback factor of the previous moment;

[0045] Step 4-3: Based on the resource feedback factor from the previous time step Preliminary prediction of accident risk Make corrections to obtain Accident risk prediction value for the sub-range corresponding to the time. :

[0046] ;

[0047] in, It is the resource impact coefficient, used to control the extent to which resource scheduling corrects for accident risks; yes Preliminary prediction of accident risk for the sub-range corresponding to the given time.

[0048] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling: In step 4-2, based on the current potential variables... The corresponding area is The amount of emergency resources allocated at any given time ,calculate Momentary resource feedback factor The resource feedback factor reflects the inhibitory effect of previous resource scheduling on reducing accident risk.

[0049] ;

[0050] in, As a preset baseline resource level, This is for adjusting the coefficient.

[0051] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling: the MLP model is periodically updated using new training samples, and the update method is as follows:

[0052] Step T-1: Establish the Fisher information corresponding to each parameter in the MLP model after the last model update;

[0053] For the MLP model, the first Parameters The corresponding Fisher information for:

[0054] ;

[0055] in, This represents the number of training samples during the last model update. The first time since the last model update Loss per training sample Parameters after the last model update final value gradient, loss function , and These are the numbers from the last model update. The predicted and actual values ​​corresponding to each training sample;

[0056] Step T-2: Iteratively train the MLP model using new training samples. The total loss function for each iteration is:

[0057] ;

[0058] Among them, average loss The calculation method is as follows:

[0059] ;

[0060] in, The number of new training samples. and These are the current iterations at the 1st and 2nd iterations respectively. The predicted and actual values ​​corresponding to each training sample;

[0061] Regularization term The calculation method is as follows:

[0062] ;

[0063] in, For parameters The current value, Parameters after the last model update The value, It is the regularization coefficient, used to control the strength of regularization.

[0064] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling, in step 5, any region in... The method for calculating the impact of an accident at a given moment is as follows:

[0065] ;

[0066] in, Indicates the number of the current region Accident risk prediction values ​​for individual sub-ranges; This represents the number of sub-ranges in the current region. Indicates the first The distance from the center of each sub-range to the center of the region is used to calculate the Gaussian function that reduces the accident risk with spatial distance. The parameter is used to control the degree of diffusion of the Gaussian function.

[0067] As a further improvement to the traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling, step 6 employs a normalization strategy to adjust the overall available resources. Dispatch will be based on the impact of accidents in each region:

[0068] ;

[0069] in, Indicates in Time assigned to the first Emergency resource availability in each region It is the first The impact of accidents in each region It is a moderating factor used to control the system's weighting of risk sensitivity.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. This invention collects multi-source data such as vehicle dynamics, road conditions, and environmental parameters in real time through vehicle networking, and forms a feature vector dataset in a unified format through cleaning, noise reduction, and time synchronization processing, thereby solving the analysis difficulties caused by the scattered data sources and lack of integration mechanism in traditional methods.

[0072] 2. This invention performs nonlinear mapping on multiple types of traffic features, including negative, zero, and positive values, and unifies the data distribution. This not only overcomes the problems of data skewness and noise interference in traditional methods, but also provides a consistent input for the subsequent extraction of latent variables, significantly improving the stability of the model.

[0073] 3. This invention constrains the update magnitude of MLP model parameters through Fisher information, enabling the model to adapt to new scene features while retaining the memory of key parameters of old tasks when receiving new data for training. This effectively solves the problems of catastrophic forgetting and performance degradation after retraining traditional static models.

[0074] 4. This invention allocates emergency resources based on the accident-resource influence field, quantifies the spatial diffusion effect of accident risks and the resource suppression effect, realizes the coordinated and adaptive allocation of resources between regions, and optimizes emergency response efficiency.

[0075] 5. This invention dynamically corrects the accident risk prediction value through resource feedback factors, integrates real-time resource scheduling information into the prediction process, and forms a closed-loop system of "prediction-scheduling-feedback-update". This enables the prediction results to have both static feature analysis and dynamic resource modulation capabilities, thereby improving the accuracy of early warning. Detailed Implementation

[0076] The technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0077] A traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling includes the following steps:

[0078] Step 1: Collect data and preprocess the data to form a feature vector dataset in a unified format. The traffic control area is divided into multiple regions, each region contains multiple sub-regions, and at the same time, each sub-region obtains a separate feature vector dataset.

[0079] If the area under traffic control is a city, then the city center, suburbs, and highway sections can be designated as regions, and each region can be further divided into multiple sub-regions.

[0080] Specifically, multi-source traffic data is collected in real time through vehicle-to-everything (V2X) networks. The collected data includes vehicle dynamic information (speed, acceleration, etc.), road condition data (road segment level, congestion, etc.), environmental parameters (weather conditions, visibility, etc.), and road network characteristic data.

[0081] The collected data undergoes cleaning, noise reduction, and time synchronization to ensure that data from different sensors and devices are aligned at the same point in time, thereby forming a feature vector dataset in a unified format. The feature vector dataset obtained at a certain moment is represented as follows:

[0082] ;

[0083] in, Represents the first eigenvector in the dataset. One characteristic.

[0084] Step 2: Perform feature transformation on the feature vector dataset to obtain the transformed feature vectors.

[0085] Specifically, the feature vector dataset obtained in step 1 is transformed to unify the distribution of each feature, improve the skewness and kurtosis of the data, and obtain the transformed feature vector. .in:

[0086] ;

[0087] in, It is the transformed eigenvector The first in 1 eigenvalue, These are transformation parameters, which are optimized and determined by maximum likelihood estimation.

[0088] Step 3: Use a variational autoencoder to encode and map the transformed feature vectors to obtain the corresponding latent variables.

[0089] Variational autoencoders (VAEs) aim to map input data to a latent space using neural networks and then map latent variables back to the original data space using a decoder. They can not only compress the dimensionality of data but also extract latent features from the data.

[0090] Transformed feature vector The input is fed into the encoder of the variational autoencoder to obtain the latent variable output in this step. .

[0091] The training process of the variational autoencoder is as follows:

[0092] Construct training samples, each containing a feature vector. .

[0093] eigenvectors The inputs are fed into the encoder, and the mean vector of the latent space is calculated through multiple hidden layers. and standard deviation vector And assume the corresponding latent variables Follows Gaussian distribution :

[0094] ;

[0095] in, Representing variables Follows the mean vector The covariance matrix is ​​a diagonal matrix. It follows a normal distribution.

[0096] Generate latent variables through reparameterization Specific samples, latent variable elements in each dimension Described by the mean and standard deviation of each latent dimension:

[0097] ;

[0098] in, It is noise sampled from the standard normal distribution, ensuring the latent variable elements Differentiability, and The mean vector obtained through the encoder and standard deviation vector The corresponding element.

[0099] The decoder will latent variables Mapping back to the original data space :

[0100] ;

[0101] in, , The latent variables obtained by the decoder are respectively The mean and variance of.

[0102] The decoder represents the latent space through this process. Decoding dimensional data space obtained Through each group and The reconstruction error is obtained, and the network parameters of the variational autoencoder are updated by backpropagation with the goal of minimizing the reconstruction error, so that the variational autoencoder can better learn the latent structure of the data.

[0103] Step 4: Input the potential variables into the dynamic accident prediction module to obtain the accident risk prediction value.

[0104] The dynamic accident prediction module includes an MLP model, and the specific processing procedure is as follows:

[0105] Step 4-1: Combine the results obtained in Step 3 with... Latent variables at time 1 Inputting the data into an MLP model yields preliminary predictions of accident risk. .

[0106] Specifically, the processing procedure of the MLP model is as follows:

[0107] ;

[0108] in, and These are the Sigmoid activation function and the ReLU activation function, respectively. and It is the weight matrix of the MLP model. and It is the bias term of the MLP model.

[0109] Step 4-2: Introduce the current real-time resource scheduling information and calculate the resource feedback factor of the previous moment.

[0110] Specifically, to make the prediction results closer to the actual situation, current real-time resource scheduling information is introduced. This is based on the current latent variables. The corresponding area is The amount of emergency resources allocated at any given time ,calculate Momentary resource feedback factor The resource feedback factor reflects the inhibitory effect of previous resource scheduling on reducing accident risk.

[0111] ;

[0112] in, As a preset baseline resource level, The adjustment coefficients were obtained by calibration using historical data.

[0113] Step 4-3: Based on the resource feedback factor from the previous time step Preliminary prediction of accident risk Make corrections to obtain Accident risk prediction value for the sub-range corresponding to the time. :

[0114] ;

[0115] in, It is the resource impact coefficient, used to control the extent to which resource scheduling corrects for accident risks; yes Preliminary prediction of accident risk for the sub-range corresponding to the given time.

[0116] The final accident risk prediction value not only takes into account the static characteristics of traffic conditions, but also incorporates the effects of real-time resource scheduling, making the prediction results more accurate.

[0117] The MLP model is periodically updated using new training samples, and the update method is as follows:

[0118] Step T-1: Establish the Fisher information corresponding to each parameter in the MLP model after the last model update.

[0119] Specifically, for the MLP model, the first Parameters The corresponding Fisher information for:

[0120] ;

[0121] in, This represents the number of training samples during the last model update. The first time since the last model update Loss per training sample Parameters after the last model update final value gradient, loss function , and These are the numbers from the last model update. The predicted and actual values ​​corresponding to each training sample.

[0122] Fisher information is obtained based on old task data from the previous model update. It is a fixed value during this update and reflects the importance of a certain parameter in the loss function. The larger the value, the greater the impact of the parameter on the model performance.

[0123] Step T-2: Iteratively train the MLP model using new training samples. The total loss function for each iteration is:

[0124] ;

[0125] Among them, average loss The calculation method is as follows:

[0126] ;

[0127] in, The number of new training samples. and These are the current iterations at the 1st and 2nd iterations respectively. The predicted and actual values ​​corresponding to each training sample.

[0128] Regularization term To prevent catastrophic forgetting, a penalty is used to avoid over-adjustment of learned parameters. The calculation method is as follows:

[0129] ;

[0130] in, For parameters The current value, The parameters after the last model update (when the old task training ended). The value of . Note that with each training iteration, It is dynamic and changing, and It remains unchanged. It is the regularization coefficient, used to control the strength of regularization.

[0131] This method can retain the memory of old tasks while learning new data, avoiding drastic changes in key parameters, and also ensuring that the model can efficiently process new task data.

[0132] Step 5: Combine the predicted accident risk values ​​of each sub-region in each region at the current moment with the resource distribution data of each region at the previous moment to obtain the current accident impact of each region.

[0133] Any region in The method for calculating the impact of an accident at a given moment is as follows:

[0134] ;

[0135] in, Indicates the number of the current region Accident risk prediction values ​​for individual sub-ranges; This represents the number of sub-ranges in the current region. Indicates the first The distance from the center of each sub-range to the center of the region is used to calculate the Gaussian function that reduces the accident risk with spatial distance. The parameter used to control the spread of the Gaussian function is determined by hyperparameter optimization based on the statistical regularities of historical data.

[0136] Step 6: Allocate emergency resources according to the impact of the accident in each region.

[0137] In urban traffic management, each area may have different accident risks, requiring corresponding emergency resource allocation. Emergency resources are calculated in units, which include ambulances, tow trucks, and traffic police patrol teams.

[0138] In order to allocate total available resources (Provided by the emergency management department platform) Resources will be allocated rationally according to the degree of accident impact in each region, using a normalization strategy to ensure overall availability. Dispatch will be based on the impact of accidents in each region:

[0139] ;

[0140] in, Indicates in Time assigned to the first Emergency resource availability in each region It is the first The impact of accidents in each region It is a moderating factor used to control the system's weighting of risk sensitivity.

[0141] The above allocation method ensures that the overall available resources are reasonably allocated among the regions according to the level of accident risk, while enabling the resource scheduling strategy to be continuously optimized based on real-time feedback from the accident impact field.

[0142] This invention uses the resource allocation results The feedback is fed into the accident risk prediction of the area at the next moment (step 4), and the system forms a closed loop of "prediction-scheduling-feedback-update", which continuously improves the overall emergency response efficiency.

[0143] It should be noted that, as will be apparent to those skilled in the art, the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics thereof. The scope of the present invention is defined by the claims rather than the foregoing description.

Claims

1. A traffic control method based on vehicle-to-everything accident early warning and dynamic resource scheduling, characterized by the steps of include: Step 1: Collect data and preprocess the data to form a feature vector dataset in a unified format; Traffic control areas are divided into multiple regions, each region contains multiple sub-regions, and each sub-region obtains a feature vector dataset at the same time. Step 2: Perform feature transformation on the feature vector dataset to obtain the transformed feature vectors; Step 3: Use a variational autoencoder to encode and map the transformed feature vectors to obtain the corresponding latent variables; Step 4: Input the potential variables into the dynamic accident prediction module to obtain the accident risk prediction value; In step 4, the dynamic accident prediction module includes an MLP model; the specific processing procedure in step 4 is as follows: Step 4-1, the step 3 obtained and latent variables corresponding to the time input to the MLP model to obtain the preliminary prediction value of the accident risk ; Specifically, the processing procedure of the MLP model is as follows: ； wherein, and are a Sigmoid activation function and a ReLU activation function, respectively, and are weight matrices of the MLP model, and are bias terms of the MLP model; Step 4-2: Introduce the current real-time resource scheduling information and calculate the resource feedback factor of the previous moment; In step 4-2, according to the current potential variable corresponding to the region at the moment of the emergency resource amount allocated , the resource feedback factor at the moment is calculated ; the resource feedback factor reflects the inhibitory effect of previous resource scheduling on reducing the risk of accidents: ； wherein, is a preset reference resource level, is an adjustment coefficient; Step 4-3, based on the resource feedback factor of the last time Preliminary prediction value of the accident risk Amended to obtain The accident risk prediction value of the corresponding sub-range of the time : ； wherein, is a resource influence coefficient, used to control the amplitude of the accident risk modification by resource scheduling; is the preliminary accident risk prediction value for the sub-range corresponding to the time instant; Step 5: Combine the predicted accident risk values ​​of each sub-region in each region at the current moment with the resource distribution data of each region at the previous moment to obtain the current accident impact of each region. Step 6: Allocate emergency resources according to the impact of the accident in each region. 2.The traffic control method based on the vehicle networking accident early warning and dynamic resource scheduling according to claim 1, wherein: In step 1, multi-source traffic data is collected in real time through the Internet of Vehicles. The collected data includes vehicle dynamic information, road condition data, environmental parameters, and road network characteristic data. 3.The traffic control method based on the accident early warning and dynamic resource scheduling of the vehicle networking according to claim 2, characterized in that: In step 1, the collected data is cleaned, denoised, and time-synchronized to ensure that the data from different sensors and devices are aligned at the same point in time, thereby forming a feature vector dataset in a unified format. The feature vector dataset obtained at a certain moment is represented as follows: ； wherein, represents the i-th feature in the feature vector dataset; represents the i-th feature in the feature vector dataset; In step 2, the feature vector dataset obtained in step 1 is transformed to unify the distribution form of each feature, improve the skewness and kurtosis of the data, and obtain the transformed feature vector wherein: ； wherein, is the transformed feature vector is the th eigenvalue in the transformed feature vector, is the transformation parameter.

4. The traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling as described in claim 1, characterized in that: In step 3, the transformed feature vector is input to the encoder of the variational autoencoder, resulting in the latent variable of the output of this step; The training process of the variational autoencoder is as follows: Construct training samples, each containing a feature vector. ; eigenvectors The inputs are fed into the encoder, and the mean vector of the latent space is calculated through multiple hidden layers. and standard deviation vector And assume the corresponding latent variables Follows Gaussian distribution : ； in, Representing variables Follows the mean vector The covariance matrix is ​​a diagonal matrix. The normal distribution; Generate latent variables through reparameterization Specific samples, latent variable elements in each dimension Described by the mean and standard deviation of each latent dimension: ； in, It is noise sampled from the standard normal distribution, ensuring the latent variable elements Differentiability, and The mean vector obtained through the encoder and standard deviation vector The corresponding element; The latent variables are decoded through a variational autoencoder. Mapping back to the original data space yields : ； in, , The latent variables obtained by the decoder are respectively The mean and variance; Through each group and The reconstruction error is obtained, and the network parameters of the variational autoencoder are updated by backpropagation with the goal of minimizing the reconstruction error.

5. The traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling as described in claim 1, characterized in that: The MLP model is periodically updated using new training samples, and the update method is as follows: Step T-1: Establish the Fisher information corresponding to each parameter in the MLP model after the last model update; For the MLP model, the first Parameters The corresponding Fisher information for: ； in, This represents the number of training samples during the last model update. The first time since the last model update Loss per training sample Parameters after the last model update final value gradient, loss function , and These are the numbers from the last model update. The predicted and actual values ​​corresponding to each training sample; Step T-2: Iteratively train the MLP model using new training samples. The total loss function for each iteration is: ； Among them, average loss The calculation method is as follows: ； in, The number of new training samples. and These are the current iterations at the 1st and 2nd iterations respectively. The predicted and actual values ​​corresponding to each training sample; Regularization term The calculation method is as follows: ； in, For parameters The current value, Parameters after the last model update The value, It is the regularization coefficient, used to control the strength of regularization.

6. The traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling as described in claim 1, characterized in that, In step 5, any region is in The method for calculating the impact of an accident at a given moment is as follows: ； in, Indicates the number of the current region Accident risk prediction values ​​for individual sub-ranges; This represents the number of sub-ranges in the current region. Indicates the first The distance from the center of each sub-range to the center of the region is used to calculate the Gaussian function that reduces the accident risk with spatial distance. The parameter is used to control the degree of diffusion of the Gaussian function.

7. The traffic control method based on vehicle-to-everything (V2X) accident early warning and dynamic resource scheduling as described in claim 1, characterized in that, In step 6, a normalization strategy is adopted to normalize the total available resources. Dispatch will be based on the impact of accidents in each region: ； in, Indicates in Time assigned to the first Emergency resource availability in each region It is the first The impact of accidents in each region It is a moderating factor used to control the system's weighting of risk sensitivity.

Images