A machine learning-based method and apparatus for data anonymization in intelligent connected vehicles

Abstract

This invention discloses a machine learning-based method and apparatus for de-identifying data in intelligent connected vehicles. The method includes: collecting raw vehicle information data; identifying sensitive information in the raw data based on pre-set rules to obtain sensitive data; constructing a de-identification strategy based on the sensitive data; constructing a target dataset based on the de-identification strategy; and training an initial data de-identification model using the target dataset to obtain a target data de-identification model. The target data de-identification model is used to de-identify data in intelligent connected vehicles. This invention improves the performance and effectiveness of data de-identification and can be widely applied in the field of data de-identification technology.

Description

Technical Field

[0001] This invention relates to the field of data anonymization technology, and in particular to a method and apparatus for anonymizing data of intelligent connected vehicles based on machine learning. Background Technology

[0002] With the development of vehicle-to-everything (V2X) technology, the amount of data generated by vehicles has increased dramatically, including a large amount of data involving personal privacy, such as vehicle location information and in-vehicle camera recordings. To protect users' privacy rights, this sensitive data needs to be anonymized. Currently, there are intelligent anonymization technologies and real-time anonymization technologies, but these technologies lack a deep understanding of the data context, resulting in an inability to flexibly adjust anonymization strategies in complex scenarios. Intelligent anonymization technology can adaptively adjust anonymization strategies based on the data usage scenario and user behavior, but this method may require more manual intervention to adjust the anonymization rules. While real-time anonymization technology saves storage space, in some scenarios, the anonymized data may not meet the needs of advanced analysis. Summary of the Invention

[0003] In view of this, the main objective of the embodiments of the present invention is to provide a method and apparatus for data desensitization of intelligent connected vehicles based on machine learning, in order to solve at least one of the problems of the prior art. The present invention can improve the performance and effect of data desensitization.

[0004] To achieve the above objectives, one aspect of the present invention provides a machine learning-based method for de-identifying data of intelligent connected vehicles, comprising:

[0005] Raw data for collecting vehicle information;

[0006] Based on preset rules, sensitive information in the original data is identified to obtain sensitive data;

[0007] Based on the sensitive data, a de-identification strategy is constructed;

[0008] Based on the described desensitization strategy, construct the target dataset;

[0009] Based on the target dataset, the initial data desensitization model is trained to obtain the target data desensitization model;

[0010] The target data desensitization model is used to desensitize data from intelligent connected vehicles.

[0011] In some embodiments, a machine learning-based method for de-identifying data in intelligent connected vehicles further includes:

[0012] The target data anonymization model was evaluated, and the evaluation results were obtained.

[0013] If the evaluation result indicates that the data anonymization fails to meet the standard, then return to the step of training the initial data anonymization model based on the target dataset to obtain the target data anonymization model. If the evaluation result indicates that the data anonymization meets the standard, then deploy and monitor the target data anonymization model.

[0014] In some embodiments, the collection of raw data for vehicle information includes the following steps:

[0015] The raw data of the vehicle information is obtained through active collection by the in-vehicle device and passive collection by the cloud.

[0016] In some embodiments, the process of identifying sensitive information in the original data based on preset rules to obtain sensitive data includes the following steps:

[0017] Pre-set the data type of the sensitive data;

[0018] Pre-set the recognition rules for each of the data types;

[0019] According to the identification rules, the sensitive information of the original data is identified to obtain the sensitive data.

[0020] In some embodiments, constructing a desensitization strategy based on the sensitive data includes the following steps:

[0021] The sensitivity and application scenarios of the obtained sensitive data;

[0022] Based on the sensitivity and the application scenario, the desensitization strategy is constructed;

[0023] The desensitization strategies include data replacement, data masking, data encryption, and data perturbation.

[0024] In some embodiments, constructing the target dataset according to the desensitization strategy includes the following steps:

[0025] According to the desensitization strategy, the original data is desensitized to obtain the first desensitized text;

[0026] Based on the original data and the corresponding first de-identified text, an initial dataset is obtained;

[0027] The initial dataset is preprocessed to obtain the target dataset.

[0028] In some embodiments, training the initial data anonymization model based on the target dataset to obtain the target data anonymization model includes the following steps:

[0029] The target dataset is input into the encoder of the initial data desensitization model to obtain an intermediate vector;

[0030] The intermediate vector is input into the decoder of the initial data desensitization model to obtain the second desensitized text;

[0031] Construct a cross-entropy loss function based on the first de-identified text and the second de-identified text;

[0032] By employing a teacher-forced technique, the next word of the first de-identified text is input into the decoder. Combined with the cross-entropy loss function, the initial data de-identification model is trained to obtain the target data de-identification model.

[0033] To achieve the above objectives, another aspect of this invention proposes a machine learning-based data anonymization device for intelligent connected vehicles, the device comprising:

[0034] The first module is used to collect raw data on vehicle information;

[0035] The second module is used to identify sensitive information in the original data based on preset rules, and obtain sensitive data.

[0036] The third module is used to construct a de-identification strategy based on the sensitive data;

[0037] The fourth module is used to construct the target dataset according to the desensitization strategy;

[0038] The fifth module is used to train the initial data anonymization model based on the target dataset to obtain the target data anonymization model;

[0039] The target data desensitization model is used to desensitize data from intelligent connected vehicles.

[0040] To achieve the above objectives, another aspect of the present invention provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned machine learning-based intelligent connected vehicle data de-identification method.

[0041] To achieve the above objectives, another aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned machine learning-based intelligent connected vehicle data desensitization method.

[0042] To achieve the above objectives, another aspect of the present invention provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned machine learning-based intelligent connected vehicle data de-identification method.

[0043] The embodiments of the present invention include at least the following beneficial effects: The present invention provides a method and apparatus for de-identifying data of intelligent connected vehicles based on machine learning. This scheme involves collecting raw vehicle information data; identifying sensitive information in the raw data based on pre-set rules to obtain sensitive data; constructing a de-identification strategy based on the sensitive data; constructing a target dataset based on the de-identification strategy; and training an initial data de-identification model based on the target dataset to obtain a target data de-identification model. The target data de-identification model is used to de-identify data of intelligent connected vehicles. The present invention uses a deep learning model to train the data de-identification model, enabling the model to learn complex de-identification rules, thereby improving the performance and effectiveness of data de-identification. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0045] Figure 1 This is a flowchart of a machine learning-based data anonymization method for intelligent connected vehicles provided in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of an optional implementation process for data anonymization of intelligent connected vehicles based on machine learning, provided in an embodiment of the present invention.

[0047] Figure 3 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0049] It should be noted that although functional modules are divided in the system diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the order in the flowchart. The terms "first / S100" and "second / S200" in the specification, claims, and the foregoing drawings may be used herein to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are used only to distinguish one concept from another. For example, first information may also be referred to as second information without departing from the scope of the embodiments of the invention, and similarly, second information may also be referred to as first information. Depending on the context, the words "if" or "when" as used herein may be interpreted as "when," "in response to a determination," or "in the event of a determination."

[0050] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0052] With the development of vehicle-to-everything (V2X) technology, the amount of data generated by vehicles has increased dramatically, including a large amount of data involving personal privacy, such as vehicle location information and in-vehicle camera recordings. To protect users' privacy rights, this sensitive data needs to be anonymized. V2X systems face various security threats, including data leakage, tampering, and unauthorized access. Anonymization technology can serve as part of security protection, reducing security risks by lowering the sensitivity of data. Currently, there are intelligent anonymization technologies and real-time anonymization technologies, but these technologies lack a deep understanding of the data context, resulting in an inability to flexibly adjust anonymization strategies in complex scenarios. Intelligent anonymization technology can adaptively adjust anonymization strategies based on data usage scenarios and user behavior, but this method may require more manual intervention to adjust the anonymization rules. While real-time anonymization technology saves storage space, in some scenarios, the anonymized data may not meet the needs of advanced analysis.

[0053] In view of this, such as Figure 1 As shown, this embodiment of the invention provides a data anonymization method for intelligent connected vehicles based on machine learning. This method may include, but is not limited to, steps S100 to S500:

[0054] Step S100: Collect raw data of vehicle information;

[0055] Step S200: Based on preset rules, identify sensitive information in the original data to obtain sensitive data;

[0056] Step S300: Construct a desensitization strategy based on the sensitive data;

[0057] Step S400: Construct the target dataset according to the desensitization strategy;

[0058] Step S500: Based on the target dataset, train the initial data desensitization model to obtain the target data desensitization model;

[0059] The target data desensitization model is used to desensitize data from intelligent connected vehicles.

[0060] In step S100 of some embodiments, the raw data of the vehicle information is obtained through active collection by the in-vehicle device and passive collection by the cloud. Optionally, real-time vehicle data can be collected through the in-vehicle device to achieve real-time monitoring of the vehicle's operation, or vehicle information can be collected through a cloud server to achieve passive data collection. The vehicle information may include, but is not limited to, numerical information such as the vehicle's location, speed, driving route, fuel consumption, and shock absorption, as well as vehicle status information and fault information. Collecting the raw data of vehicle information prepares for the generation of a large amount of sensitive data required for subsequent machine learning model training.

[0061] In some embodiments, step S200 may include, but is not limited to, steps S210 to S230:

[0062] Step S210: Preset the data type of the sensitive data;

[0063] Step S220: Pre-set the recognition rules corresponding to each type of data;

[0064] Step S230: According to the identification rules, the sensitive information of the original data is identified to obtain the sensitive data.

[0065] In step S210 of some embodiments, the data types of sensitive data are defined, specifying which types of data are considered sensitive. For example, for vehicle network data, sensitive data may include: personal identification information (such as name, ID number, driver's license number); vehicle information (such as license plate number, vehicle identification number, VIN); location information (such as GPS coordinates); communication information (such as phone number, email address), and so on, but not limited to these.

[0066] In steps S220 to S230 of some embodiments, corresponding identification rules and patterns can be formulated for each type of sensitive data. Exemplarily, these identification rules can be implemented based on regular expressions (Regex), for example:

[0067] The regular expression for an ID card number can be:

[0068] \d{6}(18|19|20)? \d{2}(0[1-9]|1[0-2])(0[1-9]|

[12] \d|3

[01] )\d{3}(\d|X);

[0069] The regular expression for a license plate number can be:

[0070] [Beijing, Tianjin, Shanghai, Chongqing, Hebei, Henan, Yunnan, Liaoning, Heilongjiang, Hunan, Anhui, Shandong, Xinjiang, Jiangsu, Zhejiang, Jiangxi, Hubei, Guangxi, Gansu, Shanxi, Inner Mongolia, Shaanxi, Jilin, Fujian, Guizhou, Guangdong, Qinghai, Tibet, Sichuan, Ningxia, Hainan][AZ]{1}[A-HJ-NPQRTUWXY0-9]{5};

[0071] The regular expression for a phone number can be:

[0072] (+d{1,3}[-]?)? d{10}。

[0073] A rule-based approach is used to identify and process sensitive information in raw data. By using the identification rules mentioned above, sensitive information in the raw data can be identified, thereby obtaining sensitive data.

[0074] In some embodiments, step S300 may include, but is not limited to, steps S310 to S320:

[0075] Step S310: Obtain the sensitivity and application scenario of the sensitive data;

[0076] Step S320: Construct the desensitization strategy based on the sensitivity and the application scenario.

[0077] In steps S310 to S320 of some embodiments, a personalized de-identification strategy is formulated by acquiring the sensitivity of the sensitive data and the application scenario. This de-identification strategy may include methods such as data replacement, data masking, data encryption, and data perturbation. For highly sensitive data (such as personal identification information), encryption or data perturbation methods may be used to provide stronger protection. For application scenarios that require maintaining data format and certain characteristics (such as de-identifying credit card numbers while retaining the first six and last four digits), masking can be chosen. If the de-identified data still needs to be used for statistical analysis, data perturbation can maintain the overall distribution and trend of the data.

[0078] In some embodiments, the desensitization strategy constructed based on sensitivity and application scenario includes the following methods:

[0079] 1. Replacement: Data replacement can be used when the exact value of the data is not necessary for the analysis, or when it is necessary to completely hide the true value of the data. For example, a name can be replaced with a user ID. The specific implementation involves creating a mapping table and replacing the original data value with a non-sensitive value.

[0080] 2. Masking: Masking can be used when it is necessary to retain partial information of data, such as format or partial numbers, for easier identification or further processing. For example, a credit card number can be masked to display only the last four digits. The specific implementation involves defining masking rules, such as retaining the last four digits and replacing the rest with asterisks (*).

[0081] 3. Encryption: Encryption can be used when data needs to be transmitted or stored in a secure environment, accessible only to authorized users, or when the data is used for verification. Specifically, this involves selecting a strong encryption algorithm (such as AES, RSA, MD5, etc.) to encrypt the data. For personal information that needs to be transmitted, asymmetric encryption methods such as AES and RSA can be used; for data that only needs to be verified, such as passwords, the MD5 hash function can be used for encryption, maximizing user account security.

[0082] 4. Data Perturbation: Data perturbation can be used when it is necessary to publish datasets for research use, but individual information should not be disclosed. For example, it can be used to protect participant privacy when publishing statistical data. Random noise is introduced into the data, making it impossible to accurately recover the original data, but preserving the overall statistical properties of the data. For example, Laplace noise can be added based on differential privacy techniques. Where x is the original data; This is the data after adding Laplace noise; λ is the noise parameter, determined according to the privacy protection level.

[0083] In some embodiments, step S400 may include, but is not limited to, steps S410 to S430:

[0084] Step S410: De-identify the original data according to the de-identification strategy to obtain the first de-identified text;

[0085] Step S420: Obtain the initial dataset based on the original data and the corresponding first de-identified text;

[0086] Step S430: Preprocess the initial dataset to obtain the target dataset.

[0087] In steps S410 to S420 of some embodiments, the original data can be de-identified according to the de-identification strategy to obtain the first de-identified text. An initial dataset containing the original data and its corresponding first de-identified text is constructed. The original data contains sensitive information, while the de-identified text is the data that has undergone de-identification processing.

[0088] In step S430 of some embodiments, the data quality of the initial dataset needs to be evaluated and preprocessed. Optionally, the preprocessing may include, but is not limited to, noise removal, normalization, missing value imputation, outlier handling, and erroneous data removal. These preprocessing steps can improve data accuracy and ensure data security.

[0089] In some embodiments, the method for preprocessing the initial dataset is as follows:

[0090] 1. Noise Reduction: Moving average filtering can be used, which is a simple filtering method for smoothing data. For time series data x1, x2, ..., x... n Moving average y t The formula for calculating at time point t is:

[0091]

[0092] Where M is the window size.

[0093] 2. Normalization: Min-max normalization can be used to scale the data to the [0,1] interval. The formula is:

[0094]

[0095] 3. Outlier Handling: Common methods for handling missing values ​​include mean imputation and median imputation. Real-time vehicle data generally has few outliers or extreme values, and the data approximates a normal distribution, so mean imputation can be used. The formula is:

[0096] x′=u

[0097] Where x′ is the filled value; u is the mean of the data column.

[0098] 4. Error data removal: Faults in sensors and vehicle components or errors in data reading can produce some abnormal data. The interquartile range (IQR) method can be used to remove these erroneous data. This involves sorting the data from smallest to largest, finding the median, and the two boundaries above and below the median (interquartiles).

[0099] 1) Lower bound: Median - 1.5 × IQR (interquartile range);

[0100] 2) Upper bound: Median + 1.5 × IQR;

[0101] If a data point is below the lower bound or above the upper bound, it is considered abnormal and is removed or replaced.

[0102] In some embodiments, step S500 may include, but is not limited to, steps S510 to S540:

[0103] Step S510: Input the target dataset into the encoder of the initial data desensitization model to obtain an intermediate vector;

[0104] Step S520: Input the intermediate vector into the decoder of the initial data desensitization model to obtain the second desensitized text;

[0105] Step S530: Construct a cross-entropy loss function based on the first de-identified text and the second de-identified text;

[0106] Step S540: Using teacher-forced technology, the next word of the first de-identified text is input into the decoder, and the initial data de-identification model is trained using the cross-entropy loss function to obtain the target data de-identification model.

[0107] In steps S510 to S540 of some embodiments, a Transformer-based model (such as BERT) is selected to understand the sensitive data context, and then a Seq2Seq model architecture is used for data desensitization. Transformer models such as BERT can capture contextual information when processing text data, making them suitable for identifying and processing sensitive information in text.

[0108] In step S510 of some embodiments, the collected target dataset containing sensitive information is input into the encoder of the initial data desensitization model based on Transformer. BERT is used to understand the context of the input sensitive data. The encoder processes the input sequence step by step and outputs a vector at each time step. These vectors together constitute the intermediate representation of the input text, i.e., the intermediate vector.

[0109] In step S520 of some embodiments, the decoder uses the same Transformer architecture as the encoder, generating the second de-identified text based on the encoder's intermediate vectors. The decoder's initial state is typically set to the encoder's final hidden state, thus conveying contextual information about the input sequence. The decoder starts with an empty input sequence and progressively generates each word of the second de-identified text, with the output of each step serving as the input for the next step.

[0110] For example, the decoder is initialized with an empty input sequence or a special start marker (such as...). <start>The decoder uses its own hidden state as initial input and sets the decoder's initial state as the encoder's final hidden state to pass contextual information about the input sequence. At each time step, the decoder uses not only its own hidden state but also the encoder's intermediate vectors to generate the next word of the de-identified text. Optionally, the next word of the de-identified text can be generated using an attention mechanism: the decoder calculates the attention score between its current hidden state and each of the encoder's hidden states, then weights the encoder's hidden state according to these scores to generate a context vector; this context vector is combined with the decoder's current hidden state to form a new hidden state for predicting the next word. The decoder's output layer then predicts the probability distribution of the next word based on the hidden state combined with the context vector. Based on this probability distribution, the word with the highest probability is selected as the next word output, or the next word output is selected through a sampling method (such as greedy sampling or beam search). The decoder updates its hidden state at each step and uses the output of the previous step as the input of the current step, repeating the above process until a special end marker (such as...) is generated. <end>Once the preset maximum number of steps is reached, the generation process ends, and the sequence output by the decoder is the desensitized text, i.e., the second desensitized text.

[0111] In step S530 of some embodiments, the cross-entropy loss function is constructed from the first de-identified text and the second de-identified text. The cross-entropy loss function measures the difference between the predicted de-identified text (the second de-identified text) and the true de-identified text (the first de-identified text), thereby improving the accuracy and quality of the de-identified text.

[0112] In step S540 of some embodiments, during the initial training phase, a teacher forcing technique is used. This involves directly using the next word of the actual de-identified text (the first de-identified text) as input to the decoder to stabilize the training process. The model parameters are then updated using backpropagation and gradient descent algorithms to minimize the loss function. Teacher forcing is a training strategy used to accelerate and stabilize the training process of the Seq2Seq model. Under teacher forcing, the input to the model's decoder at each time step is not the output of the previous step, but rather the next word of the actual target output (i.e., the actual de-identified text). For example, suppose there is a real de-identified text sequence "[START] Hello "; during training, the decoder may receive a special start marker "[START]" as input at the first time step, and predict the next word based on this start marker and the encoder's intermediate vector; at the second time step, according to the teacher-mandated policy, the decoder's input is not the output of the first time step, but the next word "Hello" in the real de-identified text, and then the decoder combines the encoder's intermediate vector and the input "Hello" to predict the next word, and so on, until the decoder generates a special end marker " " or reaches the preset maximum number of steps.

[0113] In some embodiments, after obtaining the target data anonymization model, the model's performance is evaluated using an independent validation set to ensure that the model is not overfitting. The model's hyperparameters, such as the learning rate, hidden layer size, and number of attention heads, are then adjusted based on the performance on the validation set.

[0114] In some embodiments, a machine learning-based method for de-identifying data of intelligent connected vehicles may also include, but is not limited to, steps S600 to S700:

[0115] Step S600: Evaluate the target data desensitization model and obtain the evaluation result;

[0116] Step S700: If the evaluation result indicates that the data anonymization is not up to standard, then return to the step of training the initial data anonymization model based on the target dataset to obtain the target data anonymization model. If the evaluation result indicates that the data anonymization is up to standard, then deploy and monitor the target data anonymization model.

[0117] In step S600 of some embodiments, the data is de-identified using a target data de-identification model, and the de-identified data is evaluated to complete the evaluation of the target data de-identification model, thereby ensuring the usability and security of the data. The evaluation metrics used may include, but are not limited to, K-anonymity, l-diversity, and t-proximity to quantify the de-identification effect. For example, the following evaluation metrics are provided:

[0118] 1. K-anonymity

[0119] K-anonymity ensures that in the published dataset, each record is indistinguishable from fewer than K other records. This means that an attacker cannot associate an individual with any record, and at least K-1 individuals' records are identical to the target individual's record.

[0120] Calculation process:

[0121] 1) For each record, identify all records with the same attribute value (except for the identifier).

[0122] 2) If the number of records with the same attribute value is at least K, then the record satisfies K-anonymity.

[0123] 3) Repeat steps 1) and 2) until all records satisfy K-anonymity.

[0124] 2. l-diversity

[0125] l-diversity ensures that in the published dataset, the sensitive attribute value of each record appears at least l times in each equivalence class. This means that an attacker cannot determine the sensitive attribute value of an individual because each value is common enough.

[0126] Calculation process:

[0127] 1) For each equivalence class (a set of records with the same non-sensitive attribute value), identify the sensitive attribute value.

[0128] 2) If each sensitive attribute value appears at least l times in an equivalence class, then the equivalence class satisfies l-diversity.

[0129] 3) Repeat steps 1 and 2 until all equivalence classes satisfy l-diversity.

[0130] 3. t-proximity

[0131] t-proximity ensures that the distribution of sensitive attribute values ​​for each record in the published dataset is sufficiently close to the distribution in the original dataset. This means that attackers cannot infer the distribution of the original data from the published data.

[0132] Calculation process:

[0133] 1) Calculate the frequency of each sensitive attribute value in the original dataset.

[0134] 2) Calculate the frequency of each sensitive attribute value in the published dataset.

[0135] 3) Use statistical tests (such as the chi-square test) to compare the differences between the two distributions. If the differences are within an acceptable threshold (usually defined by t), the published datasets satisfy t-proximity.

[0136] These evaluation metrics and algorithms can quantify the effectiveness of data anonymization algorithms, thereby ensuring data availability and security.

[0137] In step S700 of some embodiments, if the evaluation result indicates that the de-identification is unsatisfactory, i.e., the de-identification effect is poor, the training parameters of the model are adjusted, and the process returns to the step of training the model. If the evaluation result indicates that the de-identification is satisfactory, i.e., the de-identification effect is good, the target data de-identification model is deployed and monitored. For example, the deployment strategy includes: encapsulating the trained target data de-identification model as a service, for example using the Flask framework, for easy integration into the vehicle networking platform, and designing standardized API interfaces, such as RESTful APIs, to ensure compatibility and interoperability between different vehicles and systems. Simultaneously, it ensures that all data transmission is encrypted via SSL / TLS to prevent data interception during transmission; and implements API authentication and authorization mechanisms, such as OAuth 2.0, to ensure that only authorized users and systems can access the de-identification service. The monitoring strategy includes: real-time monitoring of system performance indicators, such as response time, throughput, and error rate, to ensure service stability and response speed. Detailed access logs and system logs are recorded, including request details, response results, and exception information, to facilitate problem tracking and performance analysis. Regularly evaluate the anonymization effect to ensure that the anonymized data meets privacy protection standards, such as the aforementioned K-anonymity, l-diversity, and t-proximity metrics. During monitoring, if the anonymization effect does not meet expectations, adjust the model's training parameters and return to the model training step.

[0138] refer to Figure 2 , Figure 2 The following illustrates an optional implementation process for data anonymization of intelligent connected vehicles based on machine learning, as provided in an embodiment of the present invention:

[0139] Step 1: Collect raw vehicle information data through active collection by in-vehicle devices and passive collection by the cloud.

[0140] Step 2: Use a rule-based identification method to identify and process sensitive information in the original data to obtain sensitive data.

[0141] Step 3: Develop a personalized desensitization strategy based on the sensitivity of the sensitive data and the usage scenario.

[0142] Step 4: Train the desensitization model

[0143] 1) Construct an initial dataset containing the original data and its corresponding real de-identified text;

[0144] 2) Preprocess the initial dataset to obtain the target dataset;

[0145] 3) Input the target dataset containing sensitive information into the encoder, and use BERT to understand the context of the sensitive data in the input. The input sequence is processed by the encoder step by step. The encoder outputs a vector at each time step. These vectors together constitute the intermediate representation of the input text.

[0146] 4) The decoder and encoder use the same Transformer architecture, generating predicted de-identified text based on the encoder's intermediate representation;

[0147] 5) During training, the cross-entropy loss function is used to measure the difference between the predicted de-identified text and the real de-identified text. In the early stage of training, the teacher-forced technique is used, that is, the next word of the real de-identified text is directly used as the input of the decoder to stabilize the training process. The model parameters are updated through backpropagation and gradient descent algorithms to minimize the loss function.

[0148] 6) Evaluate the model's performance on an independent validation set to ensure that the model is not overfitting, and adjust the model's hyperparameters, such as learning rate, hidden layer size, and number of attention heads, based on the performance on the validation set.

[0149] Step 5: De-identify the data using the target data de-identification model obtained after training. Evaluate the de-identified data using evaluation metrics (including but not limited to K-anonymity, l-diversity, and t-proximity). If the evaluation shows poor de-identification effect, adjust the training parameters of the model and return to Step 4. If the evaluation shows good de-identification effect, proceed to the next step.

[0150] Step 6: Deploy and monitor the target data anonymization model. This involves encapsulating the trained target data anonymization model for easy integration into the vehicle networking platform and monitoring system performance indicators in real time. If the anonymization effect does not meet expectations, adjust the model's training parameters and return to Step 4.

[0151] It should be noted that in various specific embodiments of the present invention, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when this application embodiment needs to obtain sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to a confirmation page. Only after obtaining the user's separate permission or consent is the necessary user-related data for the normal operation of this application embodiment obtained.

[0152] This invention also provides a machine learning-based intelligent connected vehicle data anonymization device, which can implement the above-mentioned machine learning-based intelligent connected vehicle data anonymization method. The device includes:

[0153] The first module is used to collect raw data on vehicle information;

[0154] The second module is used to identify sensitive information in the original data based on preset rules, and obtain sensitive data.

[0155] The third module is used to construct a de-identification strategy based on the sensitive data;

[0156] The fourth module is used to construct the target dataset according to the desensitization strategy;

[0157] The fifth module is used to train the initial data anonymization model based on the target dataset to obtain the target data anonymization model;

[0158] The target data desensitization model is used to desensitize data from intelligent connected vehicles.

[0159] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0160] This invention also provides an electronic device, which includes a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned machine learning-based intelligent connected vehicle data anonymization method. This electronic device can be any intelligent terminal, including tablet computers, in-vehicle computers, etc.

[0161] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0162] refer to Figure 3 , Figure 3 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0163] The processor 801 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0164] The memory 802 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 802 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called and executed by the processor 801 to implement the machine learning-based intelligent connected vehicle data desensitization method of this invention.

[0165] The 803 input / output interface is used to implement information input and output.

[0166] The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0167] Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804);

[0168] The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.

[0169] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned machine learning-based intelligent connected vehicle data desensitization method.

[0170] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0171] This invention also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned machine learning-based intelligent connected vehicle data de-identification method.

[0172] In summary, the data anonymization method and apparatus for intelligent connected vehicles based on machine learning according to embodiments of the present invention have the following advantages:

[0173] 1. This invention uses a deep learning model to train a desensitization algorithm, enabling the model to learn how to apply different desensitization strategies to different types of sensitive data. This method can use a large amount of training data to teach the model complex desensitization rules, thereby improving the intelligence level of desensitization.

[0174] 2. This invention utilizes evaluation metrics such as K-anonymity, L-diversity, and t-proximity to quantify the desensitization effect. This quantitative evaluation method can provide a more objective assessment of the desensitization effect, helping to further optimize desensitization strategies.

[0175] 3. In this embodiment of the invention, the trained de-identification model is deployed to the vehicle network platform, and the performance and effectiveness of the de-identification system are monitored in real time. This real-time monitoring mechanism can ensure the continuity and consistency of data de-identification, and promptly identify and address potential problems.

[0176] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0177] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0178] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0179] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0180] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0181] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0182] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. 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, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0183] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0184] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.< / end> < / start>

Claims

1. A data anonymization method for intelligent connected vehicles based on machine learning, characterized in that, Includes the following steps: Raw data for collecting vehicle information; Based on preset rules, sensitive information in the original data is identified to obtain sensitive data; Based on the sensitive data, a de-identification strategy is constructed, including the following steps: obtaining the sensitivity and application scenarios of the sensitive data; constructing the de-identification strategy based on the sensitivity and application scenarios; the de-identification strategy includes data replacement, data masking, data encryption, and data perturbation; for highly sensitive data, encryption or data perturbation methods are used; for application scenarios requiring the preservation of data format and some features, masking is used; for sensitive data used for statistical analysis after de-identification, data perturbation is used. Based on the described desensitization strategy, construct the target dataset; Based on the target dataset, an initial data anonymization model based on Transformer is trained to obtain a target data anonymization model, including the following steps: anonymizing the original data according to the anonymization strategy to obtain a first anonymized text; inputting the target dataset into the encoder of the initial data anonymization model to obtain an intermediate vector; inputting the intermediate vector into the decoder of the initial data anonymization model to obtain a second anonymized text; constructing a cross-entropy loss function based on the first and second anonymized texts; and using a teacher-forced technique, inputting the next word of the first anonymized text into the decoder, and combining the cross-entropy loss function to train the initial data anonymization model to obtain the target data anonymization model. The target data desensitization model is used to desensitize data from intelligent connected vehicles.

2. The data anonymization method for intelligent connected vehicles based on machine learning according to claim 1, characterized in that, It also includes the following steps: The target data anonymization model was evaluated, and the evaluation results were obtained. If the evaluation result indicates that the data anonymization fails to meet the standard, the process returns to the step of training the initial data anonymization model based on the target dataset to obtain the target data anonymization model. If the evaluation result indicates that the data anonymization meets the standard, the target data anonymization model is deployed and monitored.

3. The data anonymization method for intelligent connected vehicles based on machine learning according to claim 1, characterized in that, The raw data for collecting vehicle information includes the following steps: The raw data of the vehicle information is obtained through active collection by the in-vehicle device and passive collection by the cloud.

4. The data anonymization method for intelligent connected vehicles based on machine learning according to claim 1, characterized in that, The process of identifying sensitive information from the original data based on pre-set rules to obtain sensitive data includes the following steps: Pre-set the data type of the sensitive data; Pre-set the recognition rules for each of the data types; According to the identification rules, the sensitive information of the original data is identified to obtain the sensitive data.

5. The data anonymization method for intelligent connected vehicles based on machine learning according to claim 1, characterized in that, The construction of the target dataset based on the desensitization strategy includes the following steps: According to the desensitization strategy, the original data is desensitized to obtain the first desensitized text; Based on the original data and the corresponding first de-identified text, an initial dataset is obtained; The initial dataset is preprocessed to obtain the target dataset.

6. A data anonymization device for intelligent connected vehicles based on machine learning, characterized in that, include: The first module is used to collect raw data on vehicle information; The second module is used to identify sensitive information in the original data based on preset rules, and obtain sensitive data. The third module is used to construct a de-identification strategy based on the sensitive data; The third module is specifically used for: obtaining the sensitivity and application scenario of the sensitive data; constructing the desensitization strategy based on the sensitivity and application scenario; the desensitization strategy includes data replacement, data masking, data encryption, and data perturbation; for highly sensitive data, encryption or data perturbation is used; for application scenarios that require maintaining data format and some features, masking is used. For sensitive data used for statistical analysis after desensitization, the data perturbation method is adopted; The fourth module is used to construct the target dataset according to the desensitization strategy; The fifth module is used to train the initial data anonymization model based on Transformer according to the target dataset to obtain the target data anonymization model; the fifth module is specifically used to: anonymize the original data according to the anonymization strategy to obtain the first anonymized text; input the target dataset into the encoder of the initial data anonymization model to obtain an intermediate vector; input the intermediate vector into the decoder of the initial data anonymization model to obtain the second anonymized text; Based on the first de-identified text and the second de-identified text, a cross-entropy loss function is constructed; by using a teacher-forced technique, the next word of the first de-identified text is input into the decoder, and the initial data de-identification model is trained in conjunction with the cross-entropy loss function to obtain the target data de-identification model; The target data desensitization model is used to desensitize data from intelligent connected vehicles.

7. An electronic device, characterized in that, Including the processor and memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 5.

Images