Emergency lane dynamic management and control method and system under highway accident scene

Abstract

The invention relates to the technical field of intelligent traffic, and provides an emergency lane dynamic management and control method and system in a highway accident scene. The emergency lane dynamic management and control method in the highway accident scene comprises the following steps: acquiring accident report information, and predicting accident handling time by adopting a deep learning network model; based on the accident handling time, a traffic wave model is adopted to predict the queuing length and the dissipation time under the influence of the accident; taking the queuing length under the influence of the accident as the total length of a controlled road section, and performing segmentation discretization; taking the dissipation time under the influence of the accident as the management and control duration, and carrying out interval processing; constructing a binary management and control strategy combining a time window and a space segment; constructing a target function by taking weighted minimization of the total travel time and the total collision exposure time as a target; and solving the target function by adopting a Kriging agent model, and obtaining an emergency lane dynamic management and control strategy according to a binary management and control strategy. And the response speed of accident treatment is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent transportation technology, and in particular to an emergency lane dynamic management and control method and system under the scene of highway accidents. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] As the core artery of modern transportation network, the traffic efficiency and safety of highway directly affect the regional economic vitality and social operation order. However, with the continuous growth of traffic flow, the problem of local congestion caused by traffic accidents is increasingly prominent. After the accident, the traditional disposal mode relies on manual experience judgment and dispersed information feedback, which often leads to congestion spread and even secondary accidents due to delayed response. How to realize accurate prediction of accident influence, dynamic diversion and active control has become a key challenge for intelligent operation of highway.

[0004] Meanwhile, in the intelligent highway traffic control technology, the emergency lane opening control technology is one of the very important key technologies. At appropriate times, especially after an accident, opening the emergency lane can effectively improve the congestion of the highway and to a certain extent improve the safety of the highway operation. In the passing scene, different emergency lane opening strategies will bring different application effects. After an accident, due to the different severity of the accident, the influence on the highway is different, such as lane congestion range, lane congestion duration, lane average speed, lane average travel time, lane occupancy rate and other indicators are different. Due to the difference in congestion length and duration under different accident levels, it has a great influence on the efficiency and safety of the highway. SUMMARY

[0005] In order to solve the technical problems existing in the background art, the present application provides an emergency lane dynamic management and control method and system under the scene of highway accidents, which realizes dynamic prediction of accident influence based on machine learning model and traffic wave model by fusing real-time accident parameters and historical travel regularity. Based on the obtained space-time influence range, the continuous emergency lane opening problem is converted into a binary decision variable of "time window x space section", forming an extensible discretized space-time strategy library. "Minimize total travel time" and "minimize total collision exposure time" are included in the objective function at the same time, realizing efficiency and safety. Finally, the Kriging surrogate model is used for the first time in the scene of highway emergency lane control, which significantly improves the accuracy of the evaluation of the control strategy by using its precise fitting ability for high-dimensional, nonlinear traffic state. The genetic algorithm with strong global search ability is coupled with the Kriging surrogate model, which not only reduces the number of simulation calls, but also jumps out of the local optimum, solving the complex discrete-continuous mixed optimization problem.

[0006] To achieve the above object, the present application adopts the following technical solutions: The first aspect of the present application provides a dynamic management and control method of emergency lane under highway accident scene.

[0007] A dynamic management and control method of emergency lane under highway accident scene, comprising: Obtaining accident reporting information, using a deep learning network model to predict accident disposal time; based on the accident disposal time, using a traffic wave model to predict the queue length and dissipation time under the influence of the accident; Taking the queue length under the influence of the accident as the total length of the management section and performing segmented discretization; taking the dissipation time under the influence of the accident as the management time length and performing intervalization; constructing a binary management strategy combining time window and spatial section; constructing an objective function with the weighted minimum of total travel time and total collision exposure time as the target; Solving the objective function using a Kriging surrogate model, and obtaining the dynamic management and control strategy of the emergency lane according to the binary management strategy.

[0008] Further, the method of obtaining accident reporting information and using a deep learning network model to predict accident disposal time comprises: Obtaining route name, number of lanes, traffic volume, accident type, source of reporting, number of vehicles involved, type of vehicles involved, number of occupied lanes, whether to occupy non-emergency lanes, and time of occurrence, and using an improved TabNet network model to obtain the accident disposal time; Wherein, the improved TabNet network model comprises: introducing a hierarchical multi-step attention mask mechanism in the attention transformer, first screening important groups from the feature groups, and then selecting key features in a fine-grained manner within the group; in the training process of the TabNet network model, a random disturbance mechanism is introduced to disturb the generated feature mask.

[0009] Further, taking the queue length under the influence of the accident as the total length of the management section and performing segmented discretization; taking the dissipation time under the influence of the accident as the management time length and performing intervalization; the method comprises: In space, the total length of the management section is discretized and divided into spatial segments with substantially the same length , the length of each segment is , and the length is within a certain range; The management time length is intervalized by taking a fixed time as a time interval , the opening and closing states of different emergency lane spatial segments within the same time interval are fixed and unchanged; Let be the A 0-1 variable indicating the open / close state of the emergency lane of the spatial segment, 1 indicates that the emergency lane of the spatial segment is open in the th time interval of the th period. 0 indicates that the emergency lane is closed. In the same time interval , the control strategy of all spatial segments is mathematically expressed as: ; The control time is divided into intervals, and the control strategy of the emergency lane in the time is mathematically expressed as: ; ; The control strategy is a two-dimensional 0-1 variable matrix in mathematical form, which describes the open / close state of each spatial segment of the emergency lane control section in different time intervals within the entire control time .

[0010] Further, based on the accident disposal time, the traffic wave model is used to predict the queue length and dissipation time under the influence of the accident; the method comprises: calculating the convergence wave speed and the start-up wave speed respectively, and calculating the queue length and dissipation time under the influence of the accident according to the relationship between the convergence wave and the start-up wave.

[0011] Further, the objective function is expressed by the following formula:

[0012]

[0013]

[0014] wherein, TTS represents the TTS value of the th vehicle in the th time step of the th period; TET represents the TET value of the vehicle in the th time step of the th period; , respectively represent the weights of the efficiency and safety indicators; N represents the number of vehicles in the network; is a 0 / 1 variable indicating whether the vehicle is in the simulation area; T represents the simulation interval time; T represents the total simulation time; represents the a time step a vehicle value. a threshold value representing TTC.

[0015] Further, in the solving process of the Kriging surrogate model, an online updating mechanism is adopted, and for candidate solution points with large prediction errors or excellent target values, part of the samples are dynamically selected for real simulation evaluation, and they are included in the training set to update the Kriging surrogate model, and through the cyclic iterative updating, the dynamic evolution of the Kriging surrogate model from global rough modeling to local high-precision fitting is realized.

[0016] The second aspect of the application provides an emergency lane dynamic management and control system in a highway accident scene.

[0017] An emergency lane dynamic management and control system in a highway accident scene, comprising: An accident space-time influence prediction module configured to: acquire accident reporting information, use a deep learning network model to predict accident disposal time, and use a traffic wave model to predict queuing length and dissipation time under the influence of the accident based on the accident disposal time; A model construction module configured to: take the queuing length under the influence of the accident as the total length of the management and control section, and perform segmentation and discretization; take the dissipation time under the influence of the accident as the management and control time length, and perform intervalization; construct a binary management and control strategy combining time windows and space sections; and construct an objective function with the weighted minimum of total travel time and total collision exposure time as the target. A solving module configured to: use a Kriging surrogate model to solve the objective function, and obtain an emergency lane dynamic management and control strategy according to the binary management and control strategy.

[0018] The third aspect of the application provides a computer device, comprising: A processor adapted to execute a computer program; A computer readable storage medium, the computer readable storage medium storing a computer program, and the computer program being executed by the processor to implement the steps of the emergency lane dynamic management and control method in a highway accident scene according to the first aspect described above.

[0019] The fourth aspect of the application provides a computer readable storage medium storing a computer program, and the computer program is adapted to be loaded and executed by the processor to implement the steps of the emergency lane dynamic management and control method in a highway accident scene according to the first aspect described above.

[0020] The fifth aspect of the application provides a computer program product or a computer program.

[0021] The application provides a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the expressway accident scene emergency lane dynamic management and control method according to the first aspect.

[0022] Compared with the prior art, the application has the beneficial effects that: 1. Accident space-time influence prediction based on disposal time length prediction and traffic wave theory: according to the received expressway accident report information, a deep learning network model of a feature selection training method fusing a hierarchical attention mechanism and a mask disturbance is proposed, the model is used to predict the accident disposal time, and the traffic wave model is used to predict the accident space-time influence range in combination with the road network flow at the time of the accident, so as to realize accident response priority division.

[0023] 2. Emergency lane opening optimization model construction based on road section discretization: in combination with the predicted accident space-time influence range, the continuous emergency lane opening problem is first converted into a binary decision variable of "time window x space section", forming an extensible discretized space-time strategy library, which is convenient for genetic algorithm chromosome coding. Considering the expressway operation efficiency and safety, the objective function of the emergency lane opening optimization model is constructed.

[0024] 3. Kriging-GA-based emergency lane dynamic opening solving algorithm: the Kriging model is first used in the expressway emergency lane management and control scene, the input and output implicit relationship of the simulation model is characterized by using the Kriging model in the solving process, the precise fitting ability of the Kriging model to high-dimensional and nonlinear traffic states is used to improve the accuracy of the management and control strategy evaluation. The genetic algorithm with strong global search ability is coupled with the Kriging proxy model, and a dynamic updating mechanism is introduced, forming a "proxy-evolution-iterative optimization" three-layer framework, which not only reduces the simulation calling times, but also jumps out of the local optimum, and solves the complex discrete-continuous mixed optimization problem. BRIEF DESCRIPTION OF DRAWINGS

[0025] The drawings accompanying the specification of the application form a part of the application and serve to further understand the application. The schematic embodiments of the application and the description thereof serve to explain the application and do not constitute an improper limitation on the application.

[0026] Figure 1 is a flowchart of the emergency lane dynamic management and control method in the expressway accident scene according to the embodiments of the application; Figure 2is a technical roadmap of the emergency lane dynamic management and control method under the expressway accident scene shown by the embodiment of the present application; Figure 3 is a schematic diagram of the attention converter shown by the embodiment of the present application; Figure 4 is a schematic diagram of the traffic wave model shown by the embodiment of the present application; Figure 5 is a structural diagram of the emergency lane dynamic management and control system under the expressway accident scene shown by the embodiment of the present application; Figure 6 is a structural diagram of the computer device shown by the embodiment of the present application. DETAILED DESCRIPTION

[0027] The present application will be further described below in conjunction with the drawings and embodiments.

[0028] It should be pointed out that the following detailed description is exemplary and is intended to provide further description of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0029] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component and / or combination thereof.

[0030] The present application aims at the congestion problem caused by expressway traffic accidents, and aims to solve the three technical bottlenecks of data fragmentation, response lag and extensive strategy in traditional emergency management. By fusing historical travel rules and real-time accident parameters, accurate prediction of accident influence range and dynamic optimization of emergency lane opening strategy are realized, forming a set of emergency lane dynamic management and control method and system under the expressway accident scene.

[0031] Figure 1 is a flowchart of the emergency lane dynamic management and control method under the expressway accident scene shown by the embodiment of the present application; refer to Figure 1 The method comprises: Obtaining accident reporting information, using a deep learning network model to predict accident disposal time; based on the accident disposal time, using a traffic wave model to predict the queuing length and dissipation time under the influence of the accident; The queue length under the influence of the accident is taken as the total length of the managed section, and is discretized in sections; the dissipation time under the influence of the accident is taken as the management time length, and is intervalized; a binary management strategy combining time windows and spatial sections is constructed; a target function is constructed with the weighted minimum of total travel time and total collision exposure time as the target; The Kriging surrogate model is used to solve the target function, and the dynamic management and control strategy of the emergency lane is obtained according to the binary management strategy.

[0032] The application takes the deep application of the ETC gantry system as the core, breaks through the limitation of single data dimension, constructs a dynamic road network influence evaluation model through full-factor data fusion (integrates structured data such as ETC path flow, historical OD selection ratio, real-time accident parameters), and predicts the accident development scale in combination with a machine learning prediction model and a traffic wave model; a space-time model of the emergency lane opening strategy is built, and the actual opening strategy is abstracted as a mathematical model problem; in combination with the running efficiency and safety of the expressway, a target function of the emergency lane opening optimization model is constructed, and a suitable solving algorithm is selected; the response surface model method is introduced to solve the problems of long solving algorithm time, complex optimization process and being not conducive to actual engineering application of the emergency lane opening optimization model.

[0033] Figure 2 It is the technical roadmap of the emergency lane dynamic management and control method in the expressway accident scene shown in the embodiment of the application; according to the technical roadmap, the emergency lane dynamic management and control method in the expressway accident scene comprises: Step one: accident space-time influence prediction.

[0034] (1) Accident reporting information access Access to detailed information of the accident actively discovered or passively reported through video patrol, manual patrol, video AI, telephone reporting and the like, including the accident occurring section, stake number, direction, accident type, number of involved accident vehicles, occupied lane condition and the like.

[0035] (2) Accident space-time influence prediction based on deep learning and traffic wave model Based on the accident reporting information, the accident influence range and duration are predicted in combination with the background flow and predicted flow of the road network and the accident section.

[0036] 1) The accident disposal time is predicted based on a machine learning model, and the input fields include route name, number of lanes, traffic flow, accident type, reporting source, number of involved vehicles, type of involved vehicles, number of occupied lanes, whether non-emergency lanes are occupied, occurrence time and the like.

[0037] Firstly, the related influencing factors are analyzed: including accident attribute factors, road attribute factors, environmental attribute factors, and traffic attribute factors; secondly, the accident type, rescue method, number of occupied lanes, month of accident occurrence, date of accident occurrence, hour of accident occurrence, whether a truck or a van is involved, accident road section management company, whether it is daytime, traffic flow saturation, road section speed limit, and road section lane number are taken as inputs, and the accident disposal time is taken as output.

[0038] Secondly, the improved TabNet network model is trained.

[0039] The core design concept of TabNet is to simulate the decision manifold of decision trees through neural network structure. Traditional decision trees achieve data classification or regression through feature division, while TabNet realizes sparse selection of features through Mask layer and Attentive Transformer, and then uses Feature Transformer to process the selected features. This design not only retains the interpretability of decision trees, but also realizes more efficient weight feature allocation through sparsity regularization. The model structure of TabNet is similar to an additive model, which is composed of multiple basic decision trees, each of which corresponds to a step of feature selection and processing. This structure enables TabNet to select features like a decision tree, while also having the powerful representation ability of a neural network.

[0040] To further improve the performance of TabNet model in feature selection accuracy and generalization ability, improvements are made from the aspects of model structure and training strategy, including the following two key innovations: First: Hierarchical multi-step attention mask mechanism: In view of the limitation that the mask generation process in the original TabNet relies on single-layer sparse attention, this embodiment introduces a "group-to-group" and "in-group" two-level attention mechanism based on the Attentive Transformer module, first filters important groups from feature groups, and then selects key features in the group, thereby constructing a more structured, stable, and semantically constrained sparse mask, improving the expression ability and generalization ability of feature selection.

[0041] The hierarchical multi-step attention mask mechanism is mainly reflected in the improvement of the attention converter, as shown in Figure 3

[0042] ​The Attention Transformer is a key component of TabNet for feature selection. This module outputs an attention mask based on the feature representation of the previous layer. It introduces a hierarchical multi-step attention masking mechanism, injecting structural priors into feature selection by introducing two levels of attention: "group-wise" and "intra-group," guiding the model to form a more stable and interpretable mask sequence.

[0043] Input features Pre-divided into G feature groups: In the first layer of attention (Attention1), weights are generated for each group:

[0044] in, For between groups, Networks for generating attention, This represents the global features of the transformation of the original input X.

[0045] Then, in the second attention layer (Attention2), for the selected group g, local attention is used to perform sparse selection of features within the group:

[0046] in, For group weights, Networks for generating attention, This is the feature of the g-th group.

[0047] The final attention mask M is:

[0048] To prevent reuse, feature usage is introduced:

[0049] In the formula, i Let γ be the number of decision steps, and γ be the relaxation parameter. M [ j ] is the first j The mask is obtained step by step. To improve the sparsity of the mask, a sparse regularization term is introduced into the loss function, and the value of each step is included in the overall loss:

[0050] In the formula, ε is passed into the model as a parameter, and it is generally a very small value. For decision-making steps, B For batch size, D For the total number of features, For the first i In the step decision, the mask value of the first b The feature of the sample j Avoids excessive averaging of the mask by constraining the sparsity of the attention mask.

[0051] Second: Mask Perturbation Training Strategy: To alleviate the overfitting risk of the model relying too much on certain feature masks during training, the embodiment introduces a random perturbation mechanism in the training stage to perturb the generated feature masks (such as mask dropout, perturbation noise or order perturbation), forcing the model to maintain prediction robustness under multiple mask combinations, effectively enhancing the generalization performance and anti-interference ability of the model.

[0052] The mask perturbation training mechanism aims to break the deterministic structure of the mask generation process, and by introducing various forms of mask perturbation, the robustness of the model under multiple feature combinations is improved. The specific method is as follows: Before each step decision, random perturbation is introduced to the original mask, forcing the model to adapt to various "feature attention patterns" and improving its generalization ability. The perturbation method is mask Dropout, and the calculation formula is:

[0053] In the formula, is the selection mask vector of the i-th generated feature, is the mask after Dropout perturbation, p is the retention probability, Bernoulli( p ) is a random vector subject to Bernoulli distribution, and each element is retained with probability p and set to zero with probability 1- p .

[0054] In each training iteration, a random perturbation strategy is applied to the current generated mask to generate a perturbed mask for subsequent feature selection operations:

[0055] Where, X [ i ] is the feature subset input to the decision module at the i-th step, X is the input feature vector.

[0056] 2) Take the accident handling time as the input of the traffic wave model to predict the queue length and dissipation time under the influence of the accident, and obtain the prediction result of the spatio-temporal influence range of the accident. The traffic wave model is shown in Figure 4 .

[0057] First, the speed of the cluster wave​ is:

[0058] where, , is the traffic flow density, flow (v·k) on the road under normal conditions, respectively. , , is the density, capacity (k) at the accident location under accident conditions, respectively. , When the accident completely blocks the road, .

[0059] The traffic characteristic parameters upstream of the accident site are obtained by using the Green Shields linear relationship model to specify the formula:

[0060] where, is the actual capacity (k) of the road under normal conditions; is the traffic congestion density (k) under accident conditions.

[0061] Let , be a dimensionless parameter between 0 and 1, reflecting the continuous deterioration of traffic conditions from free flow to congestion, and .

[0062] The wave speed of the congestion wave is:

[0063] where, is the vehicle speed (v) on the road under normal conditions.

[0064] Secondly, the wave speed of the start-up wave is:

[0065] where, is the start-up flow (k) when traffic starts to recover; .

[0066] Finally, at time , the start-up wave catches up with the congestion wave, and the distances of the two wave propagations are equal, that is:

[0067] where, is the accident disposal time, is the accident duration. Then​​​​ for:

[0068] This is the maximum queue length. for:

[0069] Step 2: Constructing an optimization model for emergency lane opening based on road segment discretization.

[0070] (1) Discretization of controlled road sections Based on the above prediction of the accident impact range, the accident control strategy consists of control time and control distance, which describes the starting point and length of the entire emergency lane that need to be opened within a certain period of time.

[0071] Implementing emergency lane control strategies on highways can improve traffic conditions on congested or accident-prone sections.

[0072] Assuming the duration of control is This value is taken as the duration of the accident in step one. The total length of the road section requiring control is: This value is taken as the maximum queue length from step one. .

[0073] Spatially, the emergency lane control section is segmented and discretized, dividing the total road segment into spatial segments of roughly equal length. The length of each segment is The length is within a certain range.

[0074] In terms of time, the duration of control Interval processing is performed, with a fixed time interval defined as the time interval. Within the same time interval, the open and closed states of different emergency lane segments remain fixed.

[0075] set up To characterize the first The 0-1 variables represent the open / closed status of the emergency lane in each spatial segment. When it is 1, it means the first... Emergency vehicle lanes in the space are open; A value of 0 indicates that the emergency lane is closed.

[0076] In the same time interval The mathematical form of the control strategy for all spatial segments is: .

[0077] Control duration By time interval Divided into the time length The control strategy of the emergency lane can be expressed in mathematical form as follows: .

[0078] The control strategy is a two-dimensional 0-1 variable matrix in mathematical form, which describes the opening and closing state of each spatial segment of the emergency lane control section in the entire control time length .

[0079] (2) Safety and efficiency-oriented emergency lane opening optimization objective Considering that some control strategies are not reasonable, in order to limit the control strategy, time and space constraints are added to the control strategy. Since the implementation of each control strategy has different effects on the traffic conditions of the expressway, the optimal control strategy that can make the expressway run most efficiently and safely under the constraint condition is an urgent optimization problem to be solved.

[0080] The optimization objective is selected from the aspects of efficiency and safety, and the optimal efficiency opening strategy is sought while ensuring safety.

[0081] The optimization objective is selected from the aspects of travel efficiency and driving safety, and the most efficient opening strategy is sought while ensuring safety. The total travel time (TTS) and the total collision exposure time (TET) are selected as the travel efficiency objective and the driving safety objective respectively, and the two objectives are combined as the total objective . Since there is no explicit mathematical relationship between the strategy and the objective , this problem belongs to the "black box" problem. The Kriging surrogate model in the response surface model is introduced to approximate the relationship between the input strategy and the output objective . Each strategy corresponds to a solution. Considering the characteristics of genetic algorithm (GA) such as strong universality and wide search area, genetic algorithm is selected as the optimization algorithm. During the process of searching for the optimal solution by optimization algorithm, Kriging surrogate model is used to accelerate the optimization process.

[0082] The total travel time of all vehicles in the road network is selected as the first objective function. When the total travel time of all vehicles is the smallest, the road utilization rate of vehicles is the highest, and the output flow per unit time is the largest. The following formula represents the expression of the total travel time TTS: ​​

[0083]

[0084] where, denotes the TTS value of the i-th vehicle at the j-th time step in the k-th cycle; denotes the number of vehicles in the network; is a 0 / 1 variable indicating whether the vehicle is in the simulation region or not; denotes the simulation interval time; denotes the total simulation time. From the safety perspective, the Time Exposed Time-to-collision (TET) of all vehicles in the network is selected as the second objective function. TET represents the total time spent in a safety-critical situation, characterized by the total time below the TTC threshold. Time-to-collision (TTC) refers to the time it takes for two vehicles traveling on the same route at their respective speeds until a collision occurs. According to relevant literature, 1.5s is taken as the TTC threshold time. When TTC is less than 1.5s, the vehicle is determined to be in a dangerous emergency state, and the TET time is counted. The TET calculation formula is as follows:

[0085]

[0086]

[0087] where, denotes the TET value of the i-th vehicle at the j-th time step in the k-th cycle; denotes the number of vehicles in the network; is a 0 / 1 variable indicating whether the vehicle is in a dangerous emergency state or not; denotes the simulation interval time; denotes the total simulation time; denotes the i-th time step in the k-th cycle; denotes the TTC threshold value.

[0088] By considering both safety and efficiency indicators, the dual objectives of safety and efficiency are combined and optimized in a weighted manner. By normalizing the safety and efficiency indicators and performing a weighted sum, the total objective function is obtained, expressed as follows: ​​​​​​​​​​

[0089] In the formula, , These represent the weights of efficiency and safety indicators, respectively.

[0090] Step 3: Algorithm for dynamic opening of emergency lanes based on improved Kriging-GA.

[0091] To accelerate the optimization model's solution speed, the Kriging meta-model is used as an auxiliary model. By using the Kriging meta-model to represent the implicit relationship between the simulation model's input and output, and combining it with the search algorithm, the optimization speed is accelerated.

[0092] The basic assumptions of the Kriging metamodel are as follows:

[0093] In the formula, These are observed values, calculated from global and local estimates. It is a regression function vector. It is a vector of regression coefficients. This represents the global estimate. It is a function with a mean of zero and a variance of . The stochastic process function. Its covariance is as follows:

[0094] In the formula, It's about variables. stochastic process function, and These are sample points. Function Take a Gaussian random process, the expression is as follows:

[0095] In the formula, , and They are , and The One component. This represents the dimension of the design space.

[0096] The principle of the Kriging metamodel is to address unknown points in the experimental data. At this point, an unbiased estimate is given, and the basic expression is as follows:

[0097] In the formula, Indicates in The estimated value of the point; is the regression coefficient vector, which is estimated by the generalized least squares estimator. Each part is represented by the following formula:

[0098]

[0099]

[0100] In the formula, represents the correlation vector between the unknown point and the training point ). is a correlation matrix with a dimension of . In the formula , it refers to the training data column vector. In the formula , it is a matrix calculated by .

[0101] The unknown parameter vector encodes the traffic control strategy into the candidate solution in the solution space, that is, which open road sections are emergency lanes and the distance of the emergency lanes. Since the simulation-based strategy evaluation has high computational cost, a Kriging meta-model is introduced as a proxy model for simulation to efficiently approximate the mapping relationship between the strategy and the effect. The prediction accuracy of the Kriging model depends on the setting of its internal parameters θ. Therefore, a genetic algorithm is used to globally optimize θ, and the goal is to maximize the likelihood of the Kriging model to the simulation data, thereby ensuring the reliability of the proxy model. Under this framework, the genetic algorithm is responsible for optimizing the Kriging model parameters, and the high-precision Kriging model is responsible for quickly evaluating the pros and cons of the control strategy, ultimately driving the optimization process to find a globally better strategy.

[0102] To solve the problem of insufficient local accuracy caused by the "static training-fixed prediction" of the traditional Kriging model, an online updating mechanism is designed in this embodiment. During the optimization process, for candidate solution points with large prediction errors or excellent target values, a part of samples are dynamically selected for real simulation evaluation, and they are included in the training set to update the Kriging model. Through cyclic iterative updating, the model realizes dynamic evolution from "global rough modeling" to "local high-precision fitting", effectively improves the representation ability of the proxy model in the key search area, and speeds up the optimization convergence speed.

[0103] The core of online updating is to select new sample points and add them to the existing sample set to retrain the Kriging model. A sampling criterion based on prediction uncertainty is introduced, such as Expected Improvement (EI):

[0104]

[0105] In the formula, is the prediction mean of the Kriging model at the initial x, is the prediction standard deviation, is the prediction value corresponding to the current optimal solution, and is the distribution function and probability density of the normal distribution, Z is the standardized improvement, used to measure the potential improvement of the candidate point relative to the current optimal solution.

[0106] In the model updating stage, let the initial training set be , select a new point , and construct a new training set:

[0107] Then, the Kriging model is trained based on the new training set. The application realizes dynamic prediction of accident influence based on machine learning model and traffic wave model by fusing real-time accident parameters and historical travel rules. Based on the obtained time and space influence range, the continuous emergency lane opening problem is converted into a binary decision variable of "time window x space section", forming an extensible discretized time and space strategy library. "Minimization of total travel time" and "minimization of total collision exposure time" are simultaneously included in the objective function, and efficiency and safety are realized. Finally, the Kriging surrogate model is used in the highway emergency lane control scene for the first time, and the precise fitting ability of the Kriging surrogate model for high-dimensional and nonlinear traffic states is utilized, which significantly improves the accuracy of the control strategy evaluation. The genetic algorithm with strong global search ability is coupled with the Kriging surrogate model, which not only reduces the number of simulation calls, but also jumps out of the local optimum, solving the complex discrete-continuous mixed optimization problem.

[0108] The above Figure 1 , Figure 2 The dynamic emergency lane control method in the highway accident scene provided by the embodiment of the application is described in detail, and next, the dynamic emergency lane control system in the highway accident scene provided by the embodiment of the application will be introduced with reference to the accompanying drawings.

[0109] Figure 5 is a structural schematic diagram of the dynamic emergency lane control system in the highway accident scene according to the embodiment of the application, and with reference to Figure 5 , the system provided by the application comprises: The accident spatiotemporal impact prediction module is configured to: acquire accident reporting information, use a deep learning network model to predict the accident handling time; and based on the accident handling time, use a traffic wave model to predict the queue length and dissipation time under the impact of the accident. The model building module for road segment discretization is configured as follows: the total length of the controlled road segment is taken as the queue length under the impact of the accident, and the segment is discretized; the control duration is taken as the dissipation time under the impact of the accident, and the interval is set; a binary control strategy combining time window and spatial segment is constructed; and an objective function is constructed with the goal of minimizing the weighted average of total travel time and total collision exposure time. The solution module is configured to: solve the objective function using the Kriging surrogate model, and obtain the dynamic control strategy for the emergency lane based on the binary control strategy.

[0110] In some embodiments, obtaining accident reporting information and using a deep learning network model to predict accident handling time includes: The system obtains the route name, number of lanes, traffic flow, accident type, reporting source, number of vehicles involved, vehicle type, number of lanes occupied, whether a non-emergency lane was occupied, and the time of occurrence. Using an improved TabNet network model, the system obtains the accident handling time. The improved TabNet network model includes: introducing a hierarchical multi-step attention masking mechanism in the attention converter, first filtering important groups from among feature groups, and then selecting key features in fine-grained manner within each group; and introducing a random perturbation mechanism during the training process of the TabNet network model to perturb the generated feature mask.

[0111] In some embodiments, the total length of the controlled road segment is taken as the queue length affected by the accident, and then segmented and discretized; the control duration is taken as the dissipation time affected by the accident, and then intervalized; the method includes: Spatially, the total length of the controlled road segment is discretized into segments of roughly equal length. The length of each segment is The length is within a certain range; Control duration in terms of time Interval processing is performed, with a fixed time interval defined as the time interval. Within the same time interval, the open and closed states of different emergency lane segments remain fixed. set up To characterize the first A 0-1 variable representing the open / closed status of the emergency lane in a spatial segment. When it is 1, it means the first... Emergency vehicle lanes in the space are open; A value of 0 indicates that the emergency lane is closed; In the same time interval The mathematical form of the control strategy for all spatial segments is: ; Control duration By time interval Divided into The segment refers to the duration. The control strategy for emergency lanes can be mathematically represented as follows: ; The control strategy is mathematically a two-dimensional... A 0-1 variable matrix describing the entire control period Emergency lane control section The open and closed states of each spatial segment at different time intervals.

[0112] In some embodiments, the step of predicting the queue length and dissipation time under the influence of an accident based on a traffic wave model, using the accident handling time, includes: calculating the rally wave velocity and the initiation wave velocity respectively, and calculating the queue length and dissipation time under the influence of the accident according to the relationship between the rally wave and the initiation wave.

[0113] In some embodiments, the objective function is expressed by the following formula:

[0114]

[0115]

[0116] in, Indicates the first The first cycle The first time step The vehicle's TTS (Traffic Time Tolerance) value; It indicates the t-th period Vehicle TET value at each time step; , These represent the weights of efficiency and safety indicators, respectively. Indicates the number of vehicles currently connected to the network; A 0 / 1 variable indicating whether the vehicle is in the simulation area; Indicates the simulation interval time; This represents the total simulation time; Indicates the first The first time step The car The time to TTC is the time required for a vehicle to collide with an obstacle in front of it. This represents the threshold for TTC.

[0117] In some embodiments, in the solving process by using the Kriging surrogate model, an online updating mechanism is used, and for candidate solution points with large prediction errors or excellent target values, part of the samples are dynamically selected for real simulation evaluation, and are included in the training set to update the Kriging surrogate model, and through cyclic iterative updating, the dynamic evolution of the Kriging surrogate model from global rough modeling to local high-precision fitting is realized.

[0118] The emergency lane dynamic management and control system under the expressway accident scene according to the embodiments of the present application can correspond to the method described in the embodiments of the present application, and the above and other operations and / or functions of each module of the emergency lane dynamic management and control system under the expressway accident scene are respectively for realizing the corresponding process of each method in Figure 1 、 Figure 2 For brevity, they will not be described here.

[0119] Referring to the structural diagram of the computer device shown in Figure 6 , the computer device includes a processor, a communication interface, and a computer readable storage medium. The processor, the communication interface, and the computer readable storage medium can be connected through a bus or other means. The communication interface is used to receive and send data. The computer readable storage medium can be stored in the memory of the computer device, and the computer readable storage medium is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer readable storage medium. The processor (or CPU (Central Processing Unit, Central Processing Unit)) is the computing core and control core of the computer device, which is suitable for implementing one or more instructions, and is particularly suitable for loading and executing one or more instructions to realize the corresponding steps in the embodiment of the emergency lane dynamic management and control method under the expressway accident scene.

[0120] The embodiment provides a computer readable storage medium (Memory), which is a memory device in a computer device, and is used to store programs and data. It can be understood that the computer readable storage medium herein can include an internal storage medium in the computer device, and of course can include an extended storage medium supported by the computer device. The computer readable storage medium provides a storage space, and the storage space stores a processing system of the computer device. Also, stored in the storage space are one or more instructions adapted to be loaded and executed by the processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0121] In an embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the method embodiment of the dynamic management and control of the emergency lane under the highway accident scenario.

[0122] The embodiment provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to enable the computer device to perform the corresponding steps in the method embodiment of the dynamic management and control of the emergency lane under the highway accident scenario.

[0123] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0124] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The function of one flow or multiple flows and / or blocks Figure 1 The device for implementing the function specified in one block or multiple blocks.

[0125] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more flow or block Figure 1 one or more blocks or blocks specified in the flow.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions executed on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 one or more flow or block Figure 1 one or more blocks or blocks specified in the flow.

[0127] Those of ordinary skill in the art can understand that all or part of the flow of the above-mentioned embodiment method can be completed by computer program instructions to instruct the relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it can include the flow of the above-mentioned embodiment of each method. Among them, the storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM) and the like.

[0128] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An emergency lane dynamic management and control method in a highway accident scene, characterized in that, The method comprises the following steps: An accident report information is acquired, and a deep learning network model is used to predict an accident disposal time; Based on the accident disposal time, a traffic wave model is used to predict the queue length and dissipation time under the influence of the accident; The queue length under the influence of the accident is taken as the total length of the control section, and is discretized by sections; the dissipation time under the influence of the accident is taken as the control time length, and is discretized by intervals; a binary control strategy combining time windows and spatial sections is constructed; A target function is constructed with the weighted minimum of the total travel time and the total collision exposure time as the target; A Kriging surrogate model is used to solve the target function, and a dynamic emergency lane control strategy is obtained according to the binary control strategy.

2. The method of claim 1, wherein, The method comprises the following steps: An accident report information is acquired, and a deep learning network model is used to predict an accident disposal time; An improved TabNet network model is used to obtain the accident disposal time by taking the route name, the number of lanes, the traffic flow, the accident type, the reporting source, the number of vehicles involved, the type of vehicles involved, the number of occupied lanes, whether the non-emergency lane is occupied, and the occurrence time as input. 3.The method of claim 1, wherein, The improved TabNet network model comprises: introducing a hierarchical multi-step attention mask mechanism in the attention transformer, first screening important groups from the feature groups, and then selecting key features in a fine-grained manner within the groups; and introducing a random disturbance mechanism in the TabNet network model training process to disturb the generated feature mask. The queue length under the influence of the accident is taken as the total length of the control section, and is discretized by sections; In space, the total length of the controlled road section is segmented and discretized, and the total length of the controlled road section is divided into spatial segments with substantially the same length The length of each segment is in a certain range. Controlling the time length Intervalization, taking a fixed time as a time interval The opening and closing states of different emergency lane space segments are fixed and unchanged within the same time interval; Let be a 0-1 variable representing the open-close status of the i-th spatial segment emergency lane, be 1 if the i-th spatial segment emergency lane is open, be 0 if the i-th spatial segment emergency lane is closed.​​ In the same time interval The mathematical representation of the control policy for all spatial segments is given by: ; Controlling the time length By time interval Divided into Segments, the time length Emergency lane control strategy in mathematical form is expressed as: ; The control strategy is a two-dimensional 0-1 variable matrix in mathematical form, which describes the opening and closing states of each spatial segment of the emergency lane control section at different time intervals within the entire control duration. The control strategy is a two-dimensional 0-1 variable matrix in mathematical form, which describes the opening and closing states of each spatial segment of the emergency lane control section at different time intervals within the entire control duration. The control strategy is a two-dimensional 0-1 variable matrix in mathematical form, which describes the opening and closing states of each spatial segment 4. The method of claim 1, wherein, The dissipation time under the influence of the accident is taken as the control time length, and is discretized by intervals; the method comprises the following steps:

5. The method of claim 1, wherein, The method comprises the following steps: wherein, represents the TTS value of the vehicle at the tth cycle and the th time step; represents the TET value of the vehicle at the tth cycle and the th time step; represents the TTC value of the vehicle at the tth cycle and the th time step; represents the TTC value of the vehicle at the tth cycle and the th time step; represents the TET value of the vehicle at the tth cycle and the th time step; represents the TET value of the vehicle at the tth cycle and the th time step; , represent the weights of efficiency and safety indicators, respectively; represents the number of vehicles in the network; is a 0 / 1 variable indicating whether the vehicle is in the simulation region or not; represents the simulation interval time; represents the total simulation time; represents the TTC value of the vehicle at the tth cycle and the th time step; represents the TTC value of the vehicle at the tth cycle and the th time step; represents the TTC value of the vehicle at the tth cycle and the th time step; represents the TTC value of the vehicle at the tth cycle and the th time step; represents the threshold value of TTC.​ 6. The method of claim 1, wherein, The queue length under the influence of the accident is taken as the total length of the control section, and is discretized by sections; the dissipation time under the influence of the accident is taken as the control time length, and is discretized by intervals; the method comprises the following steps:

7. An emergency lane dynamic management and control system in a highway accident scene, characterized in that, The target function is represented by the following formula: In the solving process of the Kriging surrogate model, an online updating mechanism is used, and for candidate solution points with large prediction errors or excellent target values, part of the samples are dynamically selected for real simulation evaluation, and are included in the training set to update the Kriging surrogate model. Through cyclic iterative updating, the Kriging surrogate model realizes dynamic evolution from global rough modeling to local high-precision fitting. The method comprises the following steps: An accident report information is acquired, and a deep learning network model is used to predict an accident disposal time; Based on the accident disposal time, a traffic wave model is used to predict the queue length and dissipation time under the influence of the accident; A binary control strategy combining time windows and spatial sections is constructed by taking the queue length under the influence of the accident as the total length of the control section, and discretizing by sections; and taking the dissipation time under the influence of the accident as the control time length, and discretizing by intervals; A target function is constructed with the weighted minimum of the total travel time and the total collision exposure time as the target; A Kriging surrogate model is used to solve the target function, and a dynamic emergency lane control strategy is obtained according to the binary control strategy. 8.A computer device, comprising: a processor adapted to execute a computer program; a computer readable storage medium having stored therein a computer program, which, when executed by the processor, implements the steps of the method for dynamic management and control of emergency lane under freeway accident scenario according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium has stored therein a computer program, which is adapted to be loaded and executed by the processor to implement the steps of the method for dynamic management and control of emergency lane under freeway accident scenario according to any one of claims 1-6.

10. A computer program product, characterised in that, The computer program product comprises a computer program, which, when executed by the processor, implements the steps of the method for dynamic management and control of emergency lane under freeway accident scenario according to any one of claims 1-6.

Images