Deep belief network-based internet of vehicles information distribution strategy generation method and system

Abstract

The application discloses a kind of based on deep belief network's vehicle networking information distribution strategy generation method and system, comprising: preprocessing vehicle networking accident information, the attribute-based distribution strategy of the vehicle networking accident information data extracted is labeled, and distribution strategy is encoded;Based on deep belief network and multi-label classification network, distribution strategy generation model is constructed, the feature of vehicle networking accident information data is extracted by deep belief network, and vehicle networking accident information feature is obtained;Introduce two-layer multi-label classification network, the distribution strategy code is used as label, and the model is trained by independent threshold loss function, and the trained distribution strategy generation model is used to generate vehicle networking information distribution strategy.The application can fully exploit vehicle networking accident and its road, environmental characteristics when facing the vehicle networking accident information to be distributed, and the distribution strategy is encoded to realize flexible, dynamic, accurate strategy generation and information distribution.

Description

Technical Field

[0001] This invention relates to the field of vehicle-to-everything (V2X) information security technology, specifically to a method and system for generating V2X information distribution strategies based on deep belief networks. Background Technology

[0002] Vehicle-to-everything (V2X) networks are considered one of the most promising technologies for the future development of intelligent vehicle networks. In V2X scenarios, the distribution of V2X accident information is crucial for realizing intelligent V2X management, intelligent vehicle-road control, and route planning. For V2X accident information to be distributed, an attribute-based access control policy (referred to as a distribution policy) is formulated. Only vehicles whose attributes meet the distribution policy can access the information, enabling secure one-to-many information distribution.

[0003] In generating distribution strategies, existing technologies primarily rely on historical data mining or pre-setting rules based on traffic dynamics knowledge. While these methods can automate strategy generation, the resulting strategies are highly subjective and limited, with simplistic structures and insufficient flexibility, making them ill-suited for the highly complex and dynamic environment of connected vehicles. In the distribution of accident information within connected vehicles, a single, static strategy struggles to cover the entire group of vehicles affected by the accident. This results in affected vehicles failing to receive timely accident information, or unaffected vehicles receiving excessive amounts of redundant information, thus impacting the quality of connected vehicle information services. Summary of the Invention

[0004] The purpose of this invention is to provide a method for generating vehicle network information distribution strategies based on deep belief networks, in order to solve the problems of strong subjectivity and limitations of the generated strategies in the prior art, the relatively simple distribution strategy structure, and the lack of sufficient flexibility.

[0005] On one hand, embodiments of the present invention provide a method for generating vehicle-to-everything (V2X) information distribution strategies based on deep belief networks, comprising: S1, acquiring V2X accident information and preprocessing the V2X accident information to obtain preprocessed V2X accident vectors, and labeling the V2X accident information data with attribute-based distribution strategies, and encoding the distribution strategies to obtain distribution strategy codes; S2, constructing a distribution strategy generation model based on deep belief networks and multi-label classification networks, introducing a Gaussian distribution function at the bottom layer of the deep belief network, and extracting features from the V2X accident information data through the deep belief network to obtain V2X accident information features; S3, introducing a two-layer multi-label classification network at the top layer of the deep belief network, using the distribution strategy codes as labels, and training the deep belief network and multi-label classification network through an independent threshold loss function to obtain a trained distribution strategy generation model; S4, collecting real-time V2X accident information and preprocessing it, inputting the preprocessed V2X accident vectors into the trained V2X information access strategy generation model to obtain distribution strategy codes, and converting the distribution strategy codes into text-based distribution strategies according to the distribution strategy coding standard.

[0006] Furthermore, the vehicle-to-everything (V2X) accident information obtained in S1 includes the severity of the accident, the road where the accident occurred, the direction of the road where the accident occurred, the length of the affected road, visibility, and weather.

[0007] Preprocessing of vehicle network accident information also includes: one-hot encoding of accident severity, road where the accident occurred, road direction and weather; normalization of affected road length and visibility, mapping them to the [0,1] value range to obtain the preprocessed vehicle network accident vector.

[0008] Furthermore, in the attribute-based distribution strategy for labeling the extracted vehicle network accident information data, the labeling distribution strategy includes attribute expressions and logical operators between attribute expressions, denoted as pol:

[0009]

[0010] Where ∧ represents the AND operation, k represents the total number of attribute expressions, and AttrExpr i This represents the i-th attribute expression; the types of attribute expressions include expressions for vehicle type, driving road, and driving direction.

[0011] Furthermore, the attribute expression also includes logical operators between attribute leaves, and the attribute expression AttrExpr i satisfy:

[0012]

[0013] Where op represents the logical AND or logical OR operation, n i Let i ∈ [1, k] represent the number of attribute leaves contained in the i-th attribute expression, where k is the number of attribute expressions. `Attrleaf` ij This represents the leaf element of the j-th attribute in the i-th attribute expression, where j∈[1,n]. i When the distribution strategy includes a certain attribute, the corresponding attribute leaf is set to 1; otherwise, it is set to 0.

[0014] Furthermore, the attribute-based dispatch strategy is encoded as follows: starting from the root node of the tree strategy, the distribution strategy is encoded from top to bottom and from left to right to obtain the dispatch strategy code (polcode). In the code, the 0th bit represents the logical operator of the root node, and the 1st to k+1th bits represent the logical operators in the k attribute expressions, using 1 to represent logical "AND" and 0 to represent logical "OR"; the last bit of the code (n1+n2+...+n) represents the logical operator in the k attribute expressions. k Each position represents an attribute leaf in the strategy, and the value of the attribute leaf is the value of the corresponding position in the encoding.

[0015] Furthermore, the deep belief network is generated by superimposing two layers of restricted Boltzmann machines. A Gaussian distribution is introduced into the bottom layer of the deep belief network to handle continuous data, and the hidden variable layer adopts a Bernoulli distribution. Its energy function is expressed as:

[0016]

[0017] Where v is the input vector of the visible layer, h is the output vector of the hidden layer, D represents the number of observable variables, and F represents the number of latent variables; v i Let v represent observable variables, each observable variable v i Obey (μ) i ,σ i Gaussian distribution of μ; i Represents the observable variable v i The mean of the distribution, σ i Represents the observable variable v i Standard deviation of the distribution; h j b represents a latent variable. j Represented as each latent variable h j The bias; w ij Represents the observable variable v i and latent variable h j The weight of the edges between them;

[0018] Furthermore, S2 also includes: taking the preprocessed vehicle-to-everything (V2X) accident vector obtained in S1 as input, and extracting the features of the V2X accident and its road and environment in an unsupervised manner through a deep belief network.

[0019] Furthermore, a two-layer multi-label classification network is introduced at the top layer of the deep belief network, and the distribution strategy encoding polcode is used as the label for model training.

[0020] The model is trained using an independent threshold loss function, and an independent threshold function is introduced for different labels, doubling the number of output neurons as the threshold output of the labels.

[0021] Furthermore, the loss function is:

[0022]

[0023]

[0024] Where m is the number of tags, Y P For a collection of related tags, A collection of irrelevant tags. This indicates the output value of the relevant label. The threshold representing the relevant labels, The output value represents the irrelevant label. This represents the threshold for irrelevant labels.

[0025] On the other hand, embodiments of the present invention also provide a vehicle-to-everything (V2X) information distribution strategy generation system based on deep belief networks, comprising: a data acquisition unit, a preprocessing unit, a model building and training unit, and a distribution strategy generation unit; the data acquisition unit is used to acquire V2X accident information, including the severity of the accident, the road where the accident occurred, the road direction where the accident occurred, the length of the affected road, visibility, and weather; the preprocessing unit preprocesses the V2X accident information acquired by the data acquisition unit to obtain a V2X accident vector; the model building and training unit constructs a V2X information access strategy generation model based on deep belief networks and multi-label classification networks and trains it; the distribution strategy generation unit can obtain a distribution strategy encoding based on the real-time acquired V2X accident information and the trained V2X information access strategy generation model, and convert the distribution strategy encoding into a text-based distribution strategy.

[0026] The beneficial effects of this invention are as follows: The proposed method for generating vehicle-to-everything (V2X) accident information distribution strategies uses an attribute-based tree structure to achieve fine-grained and flexible distribution strategy expression, and encodes the distribution strategies to achieve dynamic strategy generation. This invention improves the accuracy of V2X accident information distribution by extracting V2X accident information and its road and environmental features. Unlike traditional strategy generation methods based on historical data or traffic dynamics, this invention dynamically generates distribution strategies based on environmental and road conditions, encodes attribute expressions, supports flexible strategy expression, and sets independent thresholds for each attribute. The loss function improves the accuracy and completeness of strategy generation, enabling more accurate delivery of accident information to the required vehicle groups when distributing accident information according to the strategy. This invention achieves fine-grained distribution of accident information based on attribute distribution strategies and encodes the tree-shaped distribution strategy to support flexible and dynamic distribution strategy generation. By mining big data on vehicle-to-everything (V2X) accidents and their road and environmental information, and using deep belief networks to analyze their impact on surrounding road conditions and vehicle driving, and setting an independent threshold loss function for each attribute during model training, the accuracy and completeness of strategy generation are improved, enabling the automated generation of accurate and dynamic distribution strategies for V2X accident information distribution. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0028] Figure 1 This is a schematic diagram of the overall process of the strategy generation method in Embodiment 1 of the present invention;

[0029] Figure 2 Generate a schematic diagram of the model structure for the distribution strategy;

[0030] Figure 3 This is a schematic diagram of the overall structure of the strategy generation system in Embodiment 2 of the present invention.

[0031] Figure 4 This is a schematic diagram of the area roads and vehicles in the simulation example;

[0032] Figure 5 This is a schematic diagram of the distribution strategy in the simulation example. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0034] In the description of this specification, the terms "comprising," "including," "having," and "containing" are open-ended terms, meaning that they include but are not limited to. The terms "an embodiment," "a specific embodiment," "some embodiments," and "for example," etc., refer to specific features, structures, or characteristics described in connection with that embodiment or example that are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. The order of steps involved in the various embodiments is used to illustrate the implementation of this application, and the order of steps is not limited and can be adjusted appropriately as needed.

[0035] Example 1

[0036] This invention provides a method for generating vehicle-to-everything (V2X) information distribution strategies based on deep belief networks. Figure 1 Here is a schematic diagram of the overall process of this method, such as Figure 1 As shown, the method includes:

[0037] S1. Obtain vehicle network accident information and preprocess the vehicle network accident information to obtain vehicle network accident vectors. Label the vehicle network accident information data with attribute-based distribution strategies and encode the distribution strategies to obtain distribution strategy codes.

[0038] The vehicle-to-everything (V2X) accident information obtained includes: accident severity, road where the accident occurred, direction of the road where the accident occurred, length of the affected road, visibility, and weather.

[0039] Preprocessing of the acquired vehicle-to-everything (V2X) accident information includes:

[0040] One-hot encoding is performed on the severity of the accident, the road where the accident occurred, the direction of the road where the accident occurred, and the weather. The length and visibility of the affected road are normalized and mapped to the value range of [0,1]. The vehicle network accident vector is obtained through preprocessing.

[0041] Furthermore, an attribute-based distribution strategy for labeling vehicle-to-everything (V2X) accident information data:

[0042] The dispatch strategy for annotations includes attribute expressions and logical operators between attribute expressions. The dispatch strategy for annotations is denoted as pol.

[0043]

[0044] Where ∧ represents the AND operation, k represents the total number of attribute expressions, and AttrExpr i This represents the expression for the i-th attribute;

[0045] Furthermore, the attribute expression AttrExpr i It can be an expression for vehicle type, driving road, and driving direction. Each attribute expression consists of attribute leaves and logical operators between them. The attribute expression `AttrExpr`... i satisfy:

[0046]

[0047] Where op represents the logical AND or logical OR operation, n i Attrleaf represents the number of attribute leaves contained in the i-th (i∈[1,k]) attribute expression. ij This represents the leaf element of the j-th attribute contained in the i-th attribute expression;

[0048] Specifically, the attribute leaflets in the vehicle type expression include ambulances, police cars, taxis, private cars, construction vehicles, etc.; the attribute leaflets in the driving road expression include all road names within a certain area of ​​the vehicle network; and the attribute leaflets in the driving direction expression include the directions of vehicle travel, such as east, west, south, and north.

[0049] When a certain attribute is included in the distribution strategy, the value of the corresponding attribute leaf is set to 1; otherwise, it is set to 0.

[0050] Encoding attribute-based distribution strategies also includes:

[0051] Starting from the root node of the tree-structured strategy, the distribution strategy is encoded in a top-down, left-to-right order to obtain the distribution strategy code (polcode). In the code, bit 0 represents the logical operator of the root node, and bits 1 through k+1 represent the logical operators in the attribute expressions, using 1 for logical AND and 0 for logical OR. The last bit in the code is n1+n2+...+n... k Each position represents an attribute leaf in the strategy, and the value of the attribute leaf is the value of the corresponding position in the encoding.

[0052] S2, a distribution strategy generation model is constructed based on deep belief network and multi-label classification network. A Gaussian distribution function is introduced into the bottom layer of deep belief network. Combined with the vehicle network accident vector obtained in S1, the deep belief network is used to extract features from the vehicle network accident information data to obtain vehicle network accident information features.

[0053] like Figure 2 As shown, the Deep Belief Network (DBN) constructed in this invention is generated by superimposing two layers of Restricted Boltzmann Machines. A Gaussian distribution is introduced into the bottom layer of the DBN network to process continuous data, and the hidden variable layer is a Bernoulli distribution. Its energy function E(v,h;θ) is expressed as:

[0054]

[0055] Where v is the input vector of the visible layer and h is the output vector of the hidden layer;

[0056] D represents the number of observable variables, and F represents the number of latent variables;

[0057] v i Let v represent observable variables, each observable variable v i Obey (μ) i ,σ i Gaussian distribution;

[0058] μ i Represents the observable variable v i The mean of the distribution;

[0059] σ i Represents the observable variable v i Standard deviation of the distribution;

[0060] h j b represents a latent variable. j Represented as each latent variable h j The bias;

[0061] w ij Represents the observable variable v i and latent variable h j The weight of the edges between them;

[0062] Furthermore, the preprocessed vehicle-to-everything (V2X) accident vector obtained from S1 is used as input to the deep belief network, and the model is generated unsupervised through a distribution strategy to extract features of V2X accidents, roads, and the environment.

[0063] That is, the input vector v in the energy function corresponds to the vehicle-to-everything (V2X) accident vector, and the observable and latent variables in the energy function can be set by those skilled in the art according to the actual situation.

[0064] S3 introduces a two-layer multi-label classification network at the top layer of the deep belief network, encodes the distribution policy as a label, and trains the deep belief network and the multi-label classification network through an independent threshold loss function to obtain the trained distribution policy generation model.

[0065] Specifically, a two-layer multi-label classification network (BP-MLL) is introduced at the top layer of the deep belief network. The features of vehicle-to-everything (V2X) accidents and their roads and environment extracted by the model in S2 are used as the input of the BP-MLL network, and the corresponding distribution policy encoding polcode obtained in S1 is used as the label for model training.

[0066] The model is trained using an independent threshold loss function. This introduces independent threshold functions for different labels, and doubles the number of output neurons to serve as the threshold output for the labels. The loss function is:

[0067]

[0068]

[0069] Where m is the number of tags, Y P For a collection of related tags, A collection of irrelevant tags. This indicates the output value of the relevant label. The threshold representing the relevant labels, The output value represents the irrelevant label. This represents the threshold for irrelevant labels.

[0070] Preferably, the threshold of the loss function is preset by the technicians, and the training ends when the loss function is less than the threshold.

[0071] S4. Collect real-time vehicle network accident information and preprocess it. Input the preprocessed vehicle network accident vector into the trained vehicle network information access strategy generation model to obtain the distribution strategy code. Then, according to the standard of the distribution strategy code, convert the distribution strategy code into a text form of distribution strategy.

[0072] Example 2

[0073] like Figure 3 As shown, in Embodiment 2, this invention proposes a vehicle network information distribution strategy generation system based on deep belief networks. The vehicle network accident information distribution strategy generation method proposed in Embodiment 1 can be implemented based on this system. The system includes: a data acquisition unit, a preprocessing unit, a model building and training unit, and a distribution strategy generation unit.

[0074] Specifically, the data acquisition unit is used to obtain vehicle-to-everything (V2X) accident information, including the severity of the accident, the road where the accident occurred, the direction of the road where the accident occurred, the length of the affected road, visibility, and weather.

[0075] The preprocessing unit preprocesses the vehicle network accident information collected by the acquisition unit to obtain the vehicle network accident vector;

[0076] The model building and training unit constructs and trains a vehicle-to-everything (V2X) information access strategy generation model based on deep belief networks and multi-label classification networks.

[0077] The distribution strategy generation unit can obtain the distribution strategy code based on the real-time collected vehicle network accident information and the trained vehicle network information access strategy generation model, and convert the distribution strategy code into a text-based distribution strategy.

[0078] Simulation Example:

[0079] To further illustrate the use and effects of the method proposed in this invention in practical applications, based on the following... Figure 4 The following is an explanation of the regional road map:

[0080] The vehicle-to-everything (V2X) accident information is preprocessed. In this embodiment, one historical V2X event information is obtained as follows:

[0081] Severity 3 Road where the accident occurred Yuhangtang Road Road direction right side Affected road length 3.23 meters Visibility 10 meters weather Light rain

[0082] In this embodiment, the severity of the accident, the road where the accident occurred, the direction of the road where the accident occurred, and the weather are all one-hot encoded. The severity is encoded as 0010, the road where the accident occurred is encoded as 10000, the road where the accident occurred is encoded as 01, and the weather is encoded as 0001000000. The length of the affected road and the visibility are normalized and mapped to the value range of [0,1]. The length of the affected road is mapped to 0.32, and the visibility is mapped to 0.3. The preprocessed vehicle network accident data is then a vector data = [0,0,1,0,1,0,0,0,0,0,1,0.32,0.3,0,0,0,1,0,0,0,0,0,0];

[0083] Furthermore, such as Figure 4 As shown, the vehicle types within the area include police cars, taxis, ambulances, and private cars; the roads include Yuhangtang Road, Liangmu Road, Gaojiao Road, Jucheng Road, and Wenyi West Road; and the driving directions include east, west, south, and north, i.e., k=3, n1=4, n2=5, n3=4. The attribute leaf values ​​in the three strategy expressions are: police cars, taxis, ambulances, private cars, Yuhangtang Road, Liangmu Road, Gaojiao Road, Jucheng Road, and Wenyi West Road, and east, west, south, and north, respectively.

[0084] The distribution strategy for the aforementioned vehicle-to-everything (V2X) accident information is that "vehicles traveling west or north on Yuhangtang Road, Liangmu Road, or Gaojiao Road, and police cars or taxis can access this message," which can be tagged as follows: Figure 3 As shown.

[0085] Depend on Figure 5As shown, in this embodiment, the root node is logical AND, the logical operators of the three attribute expressions are all logical OR, and the values ​​of the attribute leaves are shown in Table 1 below. Then the above distribution strategy can be encoded as polcode = [10001100111000101].

[0086] Table 1: Attribute values ​​of leaf elements in the examples

[0087]

[0088] In practical use, the real-time accident information obtained is as follows:

[0089] Severity 4 Road where the accident occurred Jucheng Road Road direction right side Affected road length 4.7 meters Visibility 0.8 meters weather Light Snow

[0090] The preprocessed information vector is: data = [0,0,0,1,0,0,0,1,0,0,1,0.47,0.5,0,0,0,0,0,0,1,0,0];

[0091] The distribution strategy generated by the input model is: "Cars traveling on Yuhangtang Road, Jucheng Road, or Wenyi West Road, heading south, north, or east, as well as police cars, ambulances, or taxis, can access this message."

[0092] When faced with vehicle-to-everything (V2X) accident information to be distributed, this invention can fully explore the characteristics of V2X accidents and their roads and environments, encode the distribution strategy to achieve flexible and dynamic strategy generation, formulate accurate and dynamic distribution strategies, accurately cover the vehicle groups affected by V2X accidents, and achieve accurate and safe distribution of V2X information.

[0093] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0095] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0096] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0097] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating vehicle-to-everything (V2X) information distribution strategies based on deep belief networks, characterized in that, include: S1. Obtain vehicle network accident information and preprocess the vehicle network accident information to obtain the preprocessed vehicle network accident vector. Then, label the vehicle network accident information data with a distribution strategy based on tree structure attributes, encode the distribution strategy, and obtain the distribution strategy code. S2, a distribution strategy generation model is constructed based on a deep belief network and a multi-label classification network. A Gaussian distribution function is introduced at the bottom layer of the deep belief network. The feature of the vehicle network accident information data is extracted through the deep belief network to obtain the vehicle network accident information features. S3 introduces a two-layer multi-label classification network at the top layer of the deep belief network, encodes the distribution policy as a label, and trains the deep belief network and the multi-label classification network through an independent threshold loss function to obtain the trained distribution policy generation model. S4. Collect real-time vehicle network accident information and preprocess it. Input the preprocessed vehicle network accident vector into the trained vehicle network information access strategy generation model to obtain the distribution strategy code. Then, according to the standard of the distribution strategy code, convert the distribution strategy code into a text form of distribution strategy. The vehicle-to-everything (V2X) accident information obtained in S1 includes the severity of the accident, the road where the accident occurred, the direction of the road where the accident occurred, the length of the affected road, visibility, and weather. Preprocessing of vehicle network accident information also includes: one-hot encoding of accident severity, road where the accident occurred, road direction where the accident occurred and weather; normalizing the length and visibility of the affected road and mapping them to the [0,1] value range to obtain the preprocessed vehicle network accident vector. A two-layer multi-label classification network is introduced at the top layer of the deep belief network, and the distribution strategy is encoded as polcode as the label for model training; The model is trained using an independent threshold loss function, and an independent threshold function is introduced for different labels, doubling the number of output neurons as the threshold output of the labels.

2. The method for generating vehicle network information distribution strategies based on deep belief networks according to claim 1, characterized in that: In the tree-structured attribute-based distribution strategy for labeling extracted vehicle network accident information data, the labeling distribution strategy includes attribute expressions and logical operators between attribute expressions. This labeling distribution strategy is denoted as pol. Where ∧ represents the AND operation, k represents the total number of attribute expressions, and AttrExpr i This represents the i-th attribute expression; the types of attribute expressions include expressions for vehicle type, driving road, and driving direction.

3. The method for generating vehicle network information distribution strategies based on deep belief networks according to claim 2, characterized in that: The attribute expression also includes logical operators between attribute leaves, and the attribute expression AttrExpr i satisfy: Where op represents the logical AND or logical OR operation, n i Let i ∈ [1, k] represent the number of attribute leaves contained in the i-th attribute expression, where k is the number of attribute expressions. (Attrleaf) ij This represents the leaf element of the j-th attribute in the i-th attribute expression, where j∈[1,n]. i When the distribution strategy includes a certain attribute, the corresponding attribute leaf is set to 1; otherwise, it is set to 0.

4. The method for generating vehicle network information distribution strategies based on deep belief networks according to claim 3, characterized in that: The distribution strategy based on tree-structured attributes is encoded as follows: Starting from the root node of the tree-structured strategy, the distribution strategy code (polcode) is obtained by encoding from top to bottom and left to right. In the encoding, bit 0 represents the logical operator of the root node, and bits 1 to k+1 represent the logical operators in the k attribute expressions, using 1 to represent logical "AND" and 0 to represent logical "OR". The last bit in the encoding is n1+n2+...+n k Each position represents an attribute leaf in the strategy, and the value of the attribute leaf is the value of the corresponding position in the encoding.

5. The method for generating vehicle network information distribution strategies based on deep belief networks according to claim 1, characterized in that: The deep belief network is generated by superimposing two layers of restricted Boltzmann machines. A Gaussian distribution is introduced into the bottom layer of the deep belief network to handle continuous data, and the hidden variable layer adopts a Bernoulli distribution. Its energy function is expressed as: Where v is the input vector of the visible layer, h is the output vector of the hidden layer, D represents the number of observable variables, and F represents the number of latent variables; v i Let v represent observable variables, each observable variable v i Obey (μ) i ,σ i Gaussian distribution of μ; i Represents the observable variable v i The mean of the distribution, σ i Represents the observable variable v i Standard deviation of the distribution; h j b represents a latent variable. j Represented as each latent variable h j The bias; w ij Represents the observable variable v i and latent variable h j The weight of the edges between them.

6. The method for generating vehicle network information distribution strategies based on deep belief networks according to claim 1, characterized in that: S2 further includes: taking the preprocessed vehicle-to-everything (V2X) accident vector obtained in S1 as input, and extracting the features of the V2X accident and its road and environment in an unsupervised manner through a deep belief network.

7. The method for generating vehicle network information distribution strategies based on deep belief networks according to claim 1, characterized in that: The loss function is: Where m is the number of tags, Y P For a collection of related tags, A collection of irrelevant tags. This indicates the output value of the relevant label. The threshold representing the relevant labels, The output value represents the irrelevant label. This represents the threshold for irrelevant labels.

8. A vehicle-to-everything (V2X) information distribution strategy generation system based on deep belief networks, wherein the method described in any one of claims 1-7 is characterized in that, include: The system comprises: a data acquisition unit, a preprocessing unit, a model building and training unit, and a distribution strategy generation unit. The data acquisition unit is used to obtain vehicle network accident information, including the severity of the accident, the road where the accident occurred, the direction of the road where the accident occurred, the length of the affected road, visibility, and weather. The preprocessing unit preprocesses the vehicle network accident information collected by the acquisition unit to obtain the vehicle network accident vector; The model building and training unit constructs and trains a vehicle-to-everything (V2X) information access strategy generation model based on deep belief networks and multi-label classification networks. The distribution strategy generation unit can obtain the distribution strategy code based on the real-time collected vehicle network accident information and the trained vehicle network information access strategy generation model, and convert the distribution strategy code into a text-based distribution strategy.

Images