Expressway auditing model training method, device and equipment and storage medium

Abstract

The invention provides an expressway auditing model training method, device and equipment and a storage medium, and relates to the technical field of intelligent traffic, and the method comprises the steps: obtaining vehicle passing data on an expressway as sample data, carrying out the preprocessing of the sample data, and obtaining an expressway auditing model; and inputting the preprocessed data into a pre-trained neural network model for processing to obtain word vectors, combining the word vectors into a vector feature matrix, inputting the vector feature matrix into an auditing model to train the auditing model, and performing feature extraction on the vector feature matrix by the auditing model by adopting a Transform encoder. The auditing model obtained through training of the method can automatically process expressway passing data, and intelligent upgrading of auditing work is achieved. Therefore, the burden of manual auditing is reduced, the labor cost is reduced, the objectivity and fairness of auditing work are improved, and the overall efficiency and service level of highway management are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a highway audit model training method, device, equipment and storage medium. Background Art

[0002] With the rapid development of the economy, transportation infrastructure has been greatly improved, and the highway network has expanded rapidly. As an important infrastructure, the construction and maintenance of highways require a lot of capital investment, and usually tolls are collected to recover costs and maintain operations. However, for long-distance transportation companies or individual drivers, highway tolls are a large expense. Toll evasion through certain improper means can directly reduce this part of the cost, especially in the case of frequent use of highways, the long-term accumulated highway tolls are considerable. Especially for those companies or individuals with poor operating conditions, toll evasion has become a way to increase income. Toll evasion directly leads to revenue losses for highway operators and governments, affects the maintenance and expansion plans of highway facilities, and undermines the fair competition environment. Companies and individuals who pay legally therefore bear higher costs.

[0003] However, there are currently the following difficulties in highway auditing: first, the highway network is huge, and the monitoring and law enforcement resources dedicated to combating toll evasion are limited, making it difficult to fully cover all sections and time periods; second, current auditing work relies on manual verification, but the number of auditors is limited, and personnel with law enforcement qualifications are even more scarce, making it difficult to combat toll evasion. Summary of the invention

[0004] In view of the above problems, the present application provides a highway audit model training method. The audit model trained by the highway audit model training method can be used for highway audit to realize automatic and efficient highway toll evasion audit. The method includes the following contents:

[0005] In a first aspect, the present application provides a highway audit model training method, comprising:

[0006] Obtain vehicle traffic data on the highway as sample data;

[0007] Preprocessing the sample data includes removing duplicate data in the sample data; converting time data in the sample data into a time difference value relative to the entry time; replacing missing field values ​​in the sample data with a preset replacement string; and setting an identifier for a numeric field value in the sample data, the identifier being used to indicate the meaning of the field;

[0008] The preprocessed data is input into a pre-trained neural network model for processing to obtain word vectors, and the word vectors are combined into a vector feature matrix;

[0009] The vector feature matrix is ​​input into the audit model to train the audit model, and the audit model uses a Transformer encoder to extract features from the vector feature matrix.

[0010] Optionally, the vehicle traffic data on the highway includes:

[0011] Fixed-length structured data, including vehicle license plate, license plate color and time of entering and exiting toll booths; and non-fixed-length data, including gantry flow and license plate recognition flow.

[0012] Optionally, inputting the preprocessed data into a pre-trained neural network model includes:

[0013] Grouping the preprocessed data to obtain multiple groups of data;

[0014] The data in the same group are taken as an input sequence and input into the pre-trained neural network model.

[0015] Optionally, the step of setting a flag for a numeric field value in the sample data includes:

[0016] Prefixes are set for the numerical field values ​​in the sample data respectively.

[0017] Optionally, before inputting the vector feature matrix into the audit model to train the audit model, the method further includes:

[0018] The vector feature matrix is ​​divided into a training set, a validation set and a test set, and the division ratio of the data set is 0.7:0.15:0.15.

[0019] In a second aspect, the present application provides a highway audit model training device, the device comprising:

[0020] An acquisition unit, used for acquiring vehicle traffic data on the expressway as sample data;

[0021] A data preprocessing unit, used to preprocess the sample data, wherein the preprocessing includes removing duplicate data in the sample data; converting time data in the sample data into a time difference value relative to the entry time; replacing missing field values ​​in the sample data with a preset replacement string; and setting an identifier for a numeric field value in the sample data, wherein the identifier is used to indicate the meaning of the field;

[0022] A vector generation unit, used for inputting the pre-processed data into a pre-trained neural network model for processing to obtain word vectors, and combining the word vectors into a vector feature matrix;

[0023] The model training unit is used to input the vector feature matrix into the audit model to train the audit model, and the audit model uses a Transformer encoder to extract features from the vector feature matrix.

[0024] Optionally, the vehicle traffic data on the highway includes:

[0025] Fixed-length structured data, including vehicle license plate, license plate color and time of entering and exiting toll booths; and non-fixed-length data, including gantry flow and license plate recognition flow.

[0026] Optionally, the vector generation unit inputs the preprocessed data into a pre-trained neural network model including:

[0027] Grouping the preprocessed data to obtain multiple groups of data;

[0028] The data in the same group are taken as an input sequence and input into the pre-trained neural network model.

[0029] Optionally, the data preprocessing unit setting identifiers for the numeric field values ​​in the sample data includes: setting prefixes for the numeric field values ​​in the sample data respectively.

[0030] Optionally, the training unit is further used to divide the vector feature matrix into a training set, a validation set and a test set before inputting the vector feature matrix into the audit model to train the audit model, and the division ratio of the data set is 0.7:0.15:0.15.

[0031] In a third aspect, the present application provides a device comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the highway audit model training method introduced in any implementation of the first aspect.

[0032] In a fourth aspect, the present application provides a computer-readable storage medium, in which codes are stored. When the codes are executed, the device executing the codes implements the highway audit model training method introduced in any implementation of the first aspect.

[0033] The present application provides a method for training a highway audit model. When executing the method, first obtain the vehicle traffic data on the highway as sample data, and then preprocess the sample data, the preprocessing includes removing duplicate data in the sample data; converting the time data in the sample data into a time difference value relative to the entry time; replacing the missing field values ​​in the sample data with a preset replacement string; setting an identifier for the field value of the number in the sample data, the identifier is used to represent the meaning of the field; inputting the preprocessed data into a pre-trained neural network model for processing to obtain a word vector, combining the word vector into a vector feature matrix, and inputting the vector feature matrix into the audit model to train the audit model, and the audit model uses a Transformer encoder to extract features from the vector feature matrix. Through this method, the trained audit model can be used for highway audits, wherein when training the model, by using real highway traffic record data as training samples, an audit model with a deep understanding of highway traffic data can be trained. The model can efficiently identify and process various complex traffic data, thereby significantly improving the efficiency and accuracy of audit work. Using the Transformer encoder as the core component of the audit model can make full use of its powerful self-attention mechanism to capture long-distance dependencies in the data, thereby more accurately extracting feature information that has a key impact on audit decisions. This feature extraction method is more efficient and flexible than traditional feature engineering methods, and can automatically adapt to changes in different data distributions. In addition, the audit model trained by this method can automatically process highway traffic data and realize the intelligent upgrade of audit work. This not only reduces the burden of manual audits and reduces labor costs, but also improves the objectivity and fairness of audit work, which helps to improve the overall efficiency and service level of highway management. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flowchart of a highway audit model training method provided in an embodiment of the present application;

[0036] Figure 2 A schematic diagram of the structure of a highway audit model training device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0038] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0039] Figure 1 A flowchart of a highway audit model training method provided in an embodiment of the present application.

[0040] Combination Figure 1 As shown, the highway audit model training method provided in the embodiment of the present application may include:

[0041] S101. Obtain vehicle traffic data on a highway as sample data.

[0042] When passing through toll station entrances and exits and highway gantries, vehicles on the highway will generate corresponding traffic flow data, such as transaction data, license plate recognition data, etc. The traffic flow data of each vehicle on the highway can be obtained in real time through synchronous collection tools.

[0043] These data include fixed-length structured data, such as vehicle license plate, license plate color, and time of entering and exiting the toll station; and non-fixed-length data, including gantry flow and license plate recognition flow. The vehicle license plate is a unique identifier for the vehicle, which is used to identify the vehicle. The license plate color can generally represent the type of vehicle. For example, the license plate color is usually divided into blue for ordinary small cars, yellow for large cars or agricultural vehicles, and white for police cars or military vehicles. The time of entering and exiting the toll station is used to record the specific time when the vehicle enters or exits the toll station. These structured data have a certain format and length.

[0044] In addition, there are some data with non-fixed lengths, such as gantry flow, when a vehicle passes through a high-speed gantry, the gantry will record the vehicle's passing information, such as passing speed, location, etc. This information may vary depending on the vehicle's driving conditions and the configuration of the gantry, so it is of non-fixed length. License plate recognition flow, which is vehicle information recorded by the license plate recognition system. When a vehicle passes through the license plate recognition device, the device captures the image of the license plate and extracts relevant information. This information may include the license plate number, license plate color, recognition time, etc. Similarly, due to differences in the vehicle's driving conditions and the performance of the license plate recognition equipment, the length of the license plate recognition flow is also non-fixed.

[0045] After obtaining the vehicle traffic data on the highway, it is organized into a list, which includes field names, definitions and examples of values. The list is shown in Table 1 below. Table 1 is a sample data field table provided in an embodiment of the present application.

[0046] Table 1 Sample data fields

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053] S102: Preprocess the sample data.

[0054] After obtaining the sample data, the data needs to be preprocessed, including:

[0055] Remove duplicate data from the sample data. Since the amount of high-speed traffic record data is extremely large, the number of newly added data per day can reach hundreds of thousands. Depending on the data source and data access method, there may be duplicates in the sample data used for model training. Model training uses offline data, so all sample data are first deduplicated.

[0056] The time data in the sample data is converted into a time difference value relative to the entry time. The sample data contains data in multiple time formats, and the time intervals between different samples are large. In order to process these data, the time data in each sample is converted into a difference value (in seconds) relative to the entry time of the sample. For example, if the entry time of a sample is 10:43:40 on March 20, 2023, and the exit time is 11:43:57 on March 20, 2023, then after processing, the entry time will be replaced by 0 seconds, and the exit time will be replaced by 3617 seconds.

[0057] In addition, when processing sample data, there will be a problem of missing field values. Usually, for data of numerical type, methods such as taking the mode or taking the average can be used to replace missing values. However, in the data set of this application, character type data is included, and these character type data cannot directly calculate the mode or average. In order to solve this problem, and considering the need to use a neural network model (such as a CBOW model) for training in the future, a unified replacement method is adopted. Specifically, all missing values ​​in the data set are replaced with a preset string "XXXXX". This replacement method ensures that whether it is numerical type or character type data, missing values ​​can be uniformly and clearly identified.

[0058] It should be noted that missing data may hide some important information, such as toll evasion. For example, when a driver shields the OBU device (on-board electronic tag) to evade tolls, it may lead to the loss of gantry flow data. However, in the case of missing data, it is impossible to accurately determine whether the cause of the missing data is the driver's manual operation or the failure of the high-speed toll collection system. Nevertheless, by replacing all missing values ​​with the preset string "XXXXX", a unified and easy-to-process dataset is provided for subsequent model training.

[0059] An identifier is set for the field value of the number in the sample data, and the identifier is used to indicate the meaning of the field. Since the same number may represent different meanings in the acquired sample data, for example, the number '1' represents different meanings in the inlet medium type and the driving direction. In order to increase the distinction, a specific prefix can be added to the value corresponding to the field whose original value is a number, as shown in Table 2.

[0060] Table 2 Add prefix to specific fields

[0061] Fields Raw value example Example after adding prefix ENMEDIATYPE 2 media2 DIRECTION 1 dir1 ACTUALFEECLASS 3 fee3 EXMEDIATYPE 2 media2 VEHICLEUSERTYPELIST [0,0] [user0,user0] GANTRYMEDIATYPELIST [2,2] [media2,media2] DRIVEDIRLIST [2,2,2,2] [dir2,dir2,dir2,dir2]

[0062] S103, inputting the preprocessed data into a pre-trained neural network model for processing to obtain word vectors, and combining the word vectors into a vector feature matrix.

[0063] After obtaining the processed data, it is necessary to perform word embedding on the processed data. In this application, the pre-processed data is processed by the CBOW (Continuous Bag-of-Words) model. CBOW is a neural network model for word vector generation. It is one of the two main variants of the Word2Vec model and can generate high-quality word vectors that can capture word meaning and context information for subsequent prediction tasks.

[0064] Before the preprocessed data is input into the pre-trained neural network model to obtain the word vector, it is necessary to train the CBOW model for word embedding of the sample data. The trained CBOW model is used to map the original input sequence into a continuous vector space. In this process, the vectors of the context words around the target word need to be fused.

[0065] The preprocessed data contains a variety of fields, which may be related or completely independent in meaning. For example, entry time and license plate color are two relatively independent concepts in real life, and there is no direct connection between them. However, there are obvious connections between some fields, such as entry medium type and exit medium type, which together describe the charging medium of the vehicle and are therefore closely related in meaning.

[0066] In order to use these data more effectively and enhance the model's ability to identify evasion behavior, in the embodiment of the present application, the preprocessed data is grouped based on the actual meaning of the fields before being input into the pre-trained neural network model for processing. Specifically, the relevance between the fields is considered during grouping, and those fields that are closely related in real life are grouped together.

[0067] For the data in the same group, it is used as the input sequence for training a CBOW model. The specific grouping is shown in Table 3. Each group after grouping is set with a specific model number, such as CBOW_0, CBOW_1, etc. The CBOW model can capture the potential relationship between these fields and map them into a continuous vector space. In this way, each field is converted into a fixed-length vector, which not only retains the original information of the field, but also incorporates the contextual information of other related fields.

[0068] Through this grouping and application of the CBOW model, the preprocessed data can be converted into higher-quality feature representations, which are more helpful for the audit model to capture the characteristic patterns of fee evasion behavior. This method not only improves the recognition accuracy of the model, but also enhances the generalization ability of the model, enabling it to better adapt to high-speed audit tasks in different scenarios.

[0069] Table 3 CBOW model processing fields

[0070]

[0071]

[0072]

[0073] Then, the trained CBOW model is used to map the original features of a sample in the preprocessed data into a vector. As shown in Table 4, multiple vectors are concatenated into a matrix with a dimension of [m, 256], where m represents the sum of the number of structural features and the number of flows (gate frame flows and card recognition flows) of the sample.

[0074] Table 4 Sample feature matrix

[0075]

[0076] S104, inputting the vector feature matrix into the audit model to train the audit model, and the audit model uses a Transformer encoder to extract features from the vector feature matrix.

[0077] The audit model is composed of multiple identical Transformer encoder layers stacked together. The Transformer encoder is a neural network structure based on the self-attention mechanism that can process sequence data. The output of the encoder of the audit model is averaged along dimension 1 (width or feature dimension) and then input into the linear layer. Finally, the output of the linear layer is converted into a probability distribution through Softmax. There are two categories here, corresponding to non-toll evasion and toll evasion. During the training process, the model adjusts the model parameters by comparing the toll evasion behavior in the labeled samples with the model prediction results, so as to gradually learn the characteristic patterns of toll evasion behavior. Finally, the trained audit model can receive new high-speed traffic data, and convert it into a vector feature matrix to predict and identify toll evasion behavior, thereby realizing the automation and efficiency of high-speed audits.

[0078] Before the preprocessed vector feature matrix is ​​input into the audit model for training, the dataset needs to be divided into training set, validation set and test set. This step is a key step in the model development process, which helps to evaluate the performance of the model, adjust the model parameters and prevent overfitting. The dataset is divided into training set, validation set and test set in a ratio of 0.7:0.15:0.15. This means that 70% of the data will be used to train the model, 15% of the data will be used to verify the model performance (parameter adjustment during the training process), and the other 15% of the data will be used to finally test the generalization ability of the model. When dividing the dataset, it is also necessary to ensure that the sample distribution in the training set, validation set and test set is consistent. This is to ensure that the model can show stable performance on different subsets and avoid bias caused by uneven data distribution. To achieve this, a random sampling method can be used, and it is ensured that the proportion of each category (such as evasion behavior and non-evasion behavior) is kept consistent with the original dataset during the sampling process.

[0079] When the vector feature matrix is ​​input into the audit model to train the audit model, the model input samples are padded or truncated to the dimension [batch_size, max_len, vector_size] according to the size of batchsize. Batch size refers to the batch size, which is the number of data samples processed simultaneously during each training process. In deep learning, batch training can make more efficient use of computing resources and help the generalization ability of the model; max_len is the preset maximum padding length; vector_size is the word vector length set when training the CBOW model, that is, 256. The model training uses the Focal loss loss function and the NAdam optimizer.

[0080] The above embodiment introduces the training method of the highway audit model. After the training is completed, the model can be deployed to use the model for online prediction. First, use containerization technology to load the model, set the model to evaluation mode, and ensure that the environment of model training and prediction is consistent. Integrate the prediction function into a service and create a RESTful API interface so that the client can call the prediction function through HTTP requests. Then build a real-time feature engineering pipeline to ensure that the online data has undergone the correct processing steps. Finally, set up appropriate monitoring and logging mechanisms to track the health and performance indicators of the service. When the model performance shows a significant decline, use the latest data to iteratively train the model.

[0081] The trained highway audit model in the embodiment of the present application can realize fully automatic and high-precision artificial intelligence highway toll evasion audit, which can save a lot of human resources compared with the traditional manual audit method. And the method of using multiple CBOW models to process real highway traffic records in this application can effectively convert the original data into feature vectors for model training and prediction, and the generated feature vectors have contextual semantic information. The Transformer model used in combination can effectively extract features from long sequence data, and will not have the gradient vanishing or gradient exploding problems of traditional recurrent neural networks, and has high prediction accuracy.

[0082] The above are some specific implementations of a highway audit model training method provided in the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided in the embodiment of the present application will be introduced from the perspective of functional modularization.

[0083] Figure 2 A schematic diagram of the structure of a highway audit model training device provided in an embodiment of the present application. Figure 2 As shown, the highway audit model training device 200 provided in the embodiment of the present application includes:

[0084] An acquisition unit 210 is used to acquire vehicle traffic data on the highway as sample data;

[0085] The data preprocessing unit 220 is used to preprocess the sample data, wherein the preprocessing includes removing duplicate data in the sample data; converting the time data in the sample data into a time difference value relative to the entry time; replacing the missing field values ​​in the sample data with a preset replacement string; and setting an identifier for a digital field value in the sample data, wherein the identifier is used to indicate the meaning of the field;

[0086] A vector generating unit 230 is used to input the pre-processed data into a pre-trained neural network model for processing to obtain word vectors, and combine the word vectors into a vector feature matrix;

[0087] The model training unit 240 is used to input the vector feature matrix into the audit model to train the audit model, and the audit model uses a Transformer encoder to extract features from the vector feature matrix.

[0088] In an implementation method of an embodiment of the present application, the vehicle traffic data on the highway includes:

[0089] Fixed-length structured data, including vehicle license plate, license plate color and time of entering and exiting toll booths; and non-fixed-length data, including gantry flow and license plate recognition flow.

[0090] In an implementation method of an embodiment of the present application, the vector generation unit inputs the pre-processed data into a pre-trained neural network model, including:

[0091] Grouping the preprocessed data to obtain multiple groups of data;

[0092] The data in the same group are taken as an input sequence and input into the pre-trained neural network model.

[0093] In an implementation method of an embodiment of the present application, the data preprocessing unit sets identifiers for the numerical field values ​​in the sample data, including: setting prefixes for the numerical field values ​​in the sample data respectively.

[0094] In an implementation method of an embodiment of the present application, the training unit is also used to divide the vector feature matrix into a training set, a validation set and a test set before inputting the vector feature matrix into the audit model to train the audit model, and the division ratio of the data set is 0.7:0.15:0.15.

[0095] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0096] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the method described in any embodiment of the present application.

[0097] The computer storage medium stores codes, and when the codes are executed, a device executing the codes implements the method described in any embodiment of the present application.

[0098] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0099] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0100] It should also be noted that the various embodiments in this specification are described in a progressive manner, and the same and similar parts between the various embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments. The device and apparatus embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0101] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A highway audit model training method, characterized in that: The method comprises: Obtain vehicle traffic data on the highway as sample data; Preprocessing the sample data includes removing duplicate data in the sample data; converting time data in the sample data into a time difference value relative to the entry time; replacing missing field values ​​in the sample data with a preset replacement string; and setting an identifier for a numeric field value in the sample data, the identifier being used to indicate the meaning of the field; The preprocessed data is input into a pre-trained neural network model for processing to obtain word vectors, and the word vectors are combined into a vector feature matrix; The vector feature matrix is ​​input into the audit model to train the audit model, and the audit model uses a Transformer encoder to extract features from the vector feature matrix.

2. The method according to claim 1, characterized in that: The vehicle traffic data on the highway includes: Fixed-length structured data, including vehicle license plate, license plate color and time of entering and exiting toll booths; and non-fixed-length data, including gantry flow and license plate recognition flow.

3. The method according to claim 1, characterized in that The step of inputting the pre-processed data into a pre-trained neural network model comprises: Grouping the preprocessed data to obtain multiple groups of data; The data in the same group are taken as an input sequence and input into the pre-trained neural network model.

4. The method according to claim 1, characterized in that The step of setting a flag for a field value that is a number in the sample data includes: Prefixes are set for the numerical field values ​​in the sample data respectively.

5. The method according to claim 1, characterized in that Before inputting the vector feature matrix into the audit model to train the audit model, the method further includes: The vector feature matrix is ​​divided into a training set, a validation set and a test set, and the division ratio of the data set is 0.7:0.15:0.

15.

6. A highway audit model training device, characterized in that: The device comprises: An acquisition unit, used for acquiring vehicle traffic data on the expressway as sample data; A data preprocessing unit, used to preprocess the sample data, wherein the preprocessing includes removing duplicate data in the sample data; converting time data in the sample data into a time difference value relative to the entry time; replacing missing field values ​​in the sample data with a preset replacement string; and setting an identifier for a numeric field value in the sample data, wherein the identifier is used to indicate the meaning of the field; A vector generation unit, used for inputting the pre-processed data into a pre-trained neural network model for processing to obtain word vectors, and combining the word vectors into a vector feature matrix; The model training unit is used to input the vector feature matrix into the audit model to train the audit model, and the audit model uses a Transformer encoder to extract features from the vector feature matrix.

7. The device according to claim 6, characterized in that The vehicle traffic data on the highway includes: Fixed-length structured data, including vehicle license plate, license plate color and time of entering and exiting toll booths; and non-fixed-length data, including gantry flow and license plate recognition flow.

8. The device according to claim 6, characterized in that The vector generation unit inputs the pre-processed data into a pre-trained neural network model including: Grouping the preprocessed data to obtain multiple groups of data; The data in the same group are taken as an input sequence and input into the pre-trained neural network model.

9. A computing device, characterized in that The computing device includes: a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method according to any one of claims 1 to 5 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images