Vehicle ecu artificial intelligence recognition method and device

Abstract

The application discloses a vehicle ECU artificial intelligence identification method and device, and the method comprises the following steps: obtaining the device fingerprints of a plurality of ECU devices, constructing a fingerprint library, and the device fingerprint is the sensor feature of the ECU device; clustering the device fingerprints in the fingerprint library to obtain a fingerprint group; obtaining the device fingerprint of the ECU device to be tested as a test fingerprint, and determining the fingerprint group to which the test fingerprint belongs; using a twin neural network model to compare the test fingerprint with the device fingerprints in the fingerprint group to which the test fingerprint belongs one by one, obtaining the similarity of the test fingerprint and each device fingerprint; comparing all the similarities to find the device fingerprint with the highest similarity to the test fingerprint, and then determining the category of the ECU device corresponding to the test fingerprint. By extracting the sensor feature of the ECU device, constructing the device fingerprint, and using the clustering algorithm and the neural network model to identify the ECU device, the application can not only prevent software-level attacks, but also prevent hardware-level attacks, and ensure the safety of vehicle communication.

Description

Technical Field

[0001] This invention relates to the field of automotive safety, and in particular to a method and apparatus for artificial intelligence identification of vehicle ECUs. Background Technology

[0002] With the rapid development of automotive electronics technology, ECUs (Electronic Control Units) have become one of the core components of modern automobiles. They are responsible for controlling and monitoring multiple systems within the vehicle, such as the engine, braking system, and air conditioning system. However, with the widespread application of ECUs in automobiles, their security issues have become increasingly prominent. Unauthorized or malfunctioning ECU devices may interfere with the normal operation of the vehicle and may even cause safety accidents. Therefore, developing a method that can accurately and quickly identify ECU devices is particularly important.

[0003] Traditional ECU device identification methods primarily rely on software-level identification technologies, such as using a mapping table between the CAN ID and the ECU. However, these methods have certain limitations. First, they are susceptible to software tampering or counterfeiting, thus reducing the accuracy of identification. Second, software-level identification methods are generally unable to effectively cope with hardware failures or malicious attacks, because these failures or attacks may alter the physical characteristics of the ECU device, not just the software-level information. Summary of the Invention

[0004] To address the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a vehicle ECU artificial intelligence identification method and device that can more accurately and efficiently identify vehicle ECUs based on hardware features.

[0005] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is to provide a vehicle ECU artificial intelligence identification method, comprising the following steps:

[0006] Acquire device fingerprints of several ECU devices and construct a fingerprint database, wherein the device fingerprints are sensor features of the ECU devices;

[0007] The device fingerprints in the fingerprint database are clustered to obtain fingerprint groups;

[0008] Obtain the device fingerprint of the ECU device under test as the test fingerprint, and determine the fingerprint group to which the test fingerprint belongs;

[0009] The test fingerprint is compared with the device fingerprints in its fingerprint group one by one using a Siamese neural network model to obtain the similarity between the test fingerprint and each device fingerprint.

[0010] By comparing all similarities, the device fingerprint with the highest similarity to the test fingerprint is found. The category of the ECU device corresponding to this device fingerprint is the same as the category of the ECU device corresponding to the test fingerprint.

[0011] Furthermore, the step of acquiring device fingerprints of several ECU devices and constructing a fingerprint database specifically includes the following sub-steps:

[0012] Select a sensor type unique to each ECU device, and each sensor type includes multiple sensors;

[0013] Obtain the features of each sensor in the selected sensor type to obtain several device fingerprints;

[0014] The fingerprints of the aforementioned devices are preprocessed, and the preprocessed device fingerprints constitute a fingerprint database.

[0015] Furthermore, the step of obtaining the features of each sensor in the selected sensor type to obtain several device fingerprints includes: obtaining the features of a sensor to form a device fingerprint;

[0016] The step of acquiring the features of a sensor to form a device fingerprint specifically includes the following sub-steps:

[0017] The time-domain characteristics of the sensor are obtained by the step response method, and the time-domain characteristics include the maximum overshoot, hysteresis time, and peak time in the step response characteristic curve.

[0018] The frequency domain characteristics of the sensor are obtained by the frequency response method, and the frequency domain characteristics include the bandwidth, time constant and natural frequency in the frequency response characteristic curve;

[0019] The time-domain features and frequency-domain features are combined to form a device fingerprint, which is a six-dimensional column vector.

[0020] Furthermore, the step of clustering device fingerprints in the fingerprint database to obtain fingerprint groups specifically includes the following sub-steps:

[0021] Initialize cluster centers by selecting K device fingerprints from the fingerprint database as cluster centers for K fingerprint groups;

[0022] Assign device fingerprints, iterate through all device fingerprints, calculate the Euclidean distance between each device fingerprint and the K cluster centers, and assign each device fingerprint to the fingerprint group containing the cluster center with the smallest Euclidean distance;

[0023] Update the cluster centers by recalculating the cluster centers of each fingerprint group and using the arithmetic mean of all device fingerprints in the fingerprint group as the new cluster centers.

[0024] The process involves iterative optimization, repeating the steps of allocating device fingerprints and updating cluster centers until the cluster center positions of all fingerprint groups converge. This means that the current calculated cluster center position of each fingerprint group is less than the previously calculated cluster center position or the preset number of iterations is reached, thus obtaining the final fingerprint group.

[0025] Furthermore, the step of determining the fingerprint group to which the test fingerprint belongs includes: calculating the Euclidean distance between the test fingerprint and the cluster center of each fingerprint group, and determining the fingerprint group to which the cluster center with the smallest Euclidean distance belongs.

[0026] The Euclidean distance between the test fingerprint and the cluster center is obtained by the following formula:

[0027]

[0028] In Equation (1), D' represents the Euclidean distance between the test fingerprint and the cluster center, and X... 1n Let X represent the nth vector of the tested fingerprint. kn Let n be the nth vector representing the cluster center of the kth fingerprint group, where n = 1, 2, 3, 4, 5, 6.

[0029] Furthermore, the step of comparing the test fingerprint with each device fingerprint in its fingerprint group using a Siamese neural network model to obtain the similarity between the test fingerprint and each device fingerprint specifically includes the following sub-steps:

[0030] Constructing a Siamese neural network model;

[0031] The twin neural network model is trained using device fingerprints from the fingerprint database;

[0032] The test fingerprint and any device fingerprint in its fingerprint group are input into the trained Siamese neural network model, and the similarity between the test fingerprint and the device fingerprint is output.

[0033] Traverse all device fingerprints in the fingerprint group to which the test fingerprint belongs, and obtain the similarity between the test fingerprint and each device fingerprint in the fingerprint group to which it belongs.

[0034] Furthermore, the Siamese neural network model includes two identical sub-networks, a feature convolutional layer, and a fully connected layer; the step of training the Siamese neural network model using device fingerprints from the fingerprint database specifically includes the following sub-steps:

[0035] The device fingerprints in the fingerprint database are randomly paired and divided into training and verification sets according to a certain ratio;

[0036] The paired device fingerprints and corresponding labels from the training set are input into the Siamese neural network model. When two device fingerprints are similar, the label is 1, and when two device fingerprints are dissimilar, the label is 0.

[0037] The features of the two input device fingerprints are extracted using the two identical sub-networks, and their features are output.

[0038] The contrastive loss function is used to calculate the distance between two output features, and the contrastive loss is calculated based on whether they are similar.

[0039] Based on the results of the loss function, the gradient of the loss with respect to the network weights is calculated using the backpropagation algorithm;

[0040] The network weights are updated using an optimization algorithm based on the calculated gradient.

[0041] Repeat the iterations until the model's performance on the validation set no longer improves, i.e., the loss function is no longer properly fitted or the preset number of iterations is reached.

[0042] Furthermore, the contrast loss is obtained using the following formula:

[0043] L=(1-y)D 2 +y·max(0,mD) 2 (two)

[0044] In Equation (II), L represents the contrast loss, y represents the binary label, when the two input samples are similar, the binary label y is 1, when the two input samples are not similar, the binary label y is 0, D represents the distance between the two feature vectors extracted by the two sub-networks, and m represents the preset second threshold.

[0045] The updated weights are obtained using the following formula:

[0046]

[0047] In equation (iii), W new W represents the updated weights. old This represents the original weights, and α represents the learning rate. This represents the gradient of the loss function.

[0048] Furthermore, the step of inputting the test fingerprint and any device fingerprint from its fingerprint group into the trained Siamese neural network model and outputting the similarity between the test fingerprint and the device fingerprint specifically includes the following sub-steps:

[0049] The two identical sub-networks are used to extract features from the test fingerprint and device fingerprint input to the Siamese neural network model, respectively;

[0050] The feature convolutional layer is used to fuse the features of the extracted test fingerprint and device fingerprint through convolution operations;

[0051] The fully connected layer is used to summarize the fused feature maps into a single output and compress it to between 0 and 1, which is used as the similarity between the test fingerprint and the device fingerprint.

[0052] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is: to provide a vehicle ECU artificial intelligence identification device, comprising:

[0053] A fingerprint database module is used to acquire device fingerprints of several ECU devices and construct a fingerprint database, wherein the device fingerprints are sensor features of the ECU devices;

[0054] The clustering module is used to cluster device fingerprints in the fingerprint database to obtain fingerprint groups;

[0055] The group determination module is used to acquire the device fingerprint of the ECU device under test as a test fingerprint, and determine the fingerprint group to which the test fingerprint belongs;

[0056] The comparison module is used to compare the test fingerprint with the device fingerprints in its fingerprint group one by one using a Siamese neural network model to obtain the similarity between the test fingerprint and each device fingerprint.

[0057] The comparison and classification module is used to compare all similarities, find the device fingerprint with the highest similarity to the test fingerprint, and determine the category of the ECU device corresponding to the device fingerprint.

[0058] The vehicle ECU artificial intelligence identification method and apparatus of the present invention have at least the following beneficial effects: The present invention extracts the hardware features of ECU devices, i.e., sensor features, to construct device fingerprints, and uses advanced machine learning algorithms to identify ECU devices. This not only prevents software-level attacks but also hardware-level attacks, ensuring vehicle communication security. The device fingerprint includes the maximum overshoot, hysteresis time, and peak time of the step response, as well as the bandwidth, time constant, and natural frequency of the frequency response. These features are not only unique and stable but also not easily affected by external interference or tampering, thus improving the accuracy of ECU device identification. By introducing a clustering algorithm, the device fingerprints of a large number of ECU devices are grouped into multiple fingerprint groups. When determining the device category of a test fingerprint, it is only necessary to find the fingerprint group to which the test fingerprint belongs in the fingerprint database. Then, a Siamese neural network model is used to compare the test fingerprint with the device fingerprints in the fingerprint group one by one to determine the ECU device category corresponding to the test fingerprint. This method not only reduces the amount of computation and improves the identification speed but also makes the identification results more accurate and reliable through group division. Attached Figure Description

[0059] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0060] Figure 1 This is a flowchart of one embodiment of the vehicle ECU artificial intelligence identification method of the present invention.

[0061] Figure 2 for Figure 1 The flowchart for step S1.

[0062] Figure 3 for Figure 2 The flowchart for step S12.

[0063] Figure 4 for Figure 1 The flowchart for step S2.

[0064] Figure 5 for Figure 1 The flowchart for step S4.

[0065] Figure 6 for Figure 5 The flowchart for step S42.

[0066] Figure 7 This is a structural block diagram of one embodiment of the vehicle ECU artificial intelligence identification device of the present invention. Detailed Implementation

[0067] The invention will now be further described with reference to the accompanying drawings.

[0068] Please see Figure 1 This is a flowchart of an embodiment of the vehicle ECU artificial intelligence identification method of the present invention. This embodiment specifically includes the following steps:

[0069] S1. Obtain device fingerprints of several ECU devices and construct a fingerprint database, wherein the device fingerprints are the sensor features of the ECU devices.

[0070] In this embodiment, device fingerprints of several ECU devices of known categories are obtained.

[0071] Please see Figure 2 Step S1 includes the following sub-steps:

[0072] S11. Select the sensor type unique to each ECU device. Each type of sensor includes multiple sensors.

[0073] When obtaining the device fingerprint for each type of ECU, the unique sensor type of each ECU should be selected. The characteristics of this type of sensor should be obtained as the device fingerprint of the ECU. Each ECU has multiple sensors connected to it, and only its unique sensors can represent this type of ECU. For example, the body control module includes temperature sensors, door status sensors, window position sensors, light sensors, wiper sensors, seat belt status sensors, etc. Among them, the sensor that best represents the body control module is the door status sensor. Moreover, there is more than one such sensor, and each sensor is different each time it is produced. Therefore, obtaining the characteristics of multiple door status sensors can reduce errors.

[0074] S12. Obtain the characteristics of each sensor in the selected sensor type to obtain several device fingerprints.

[0075] Specifically, the features of one sensor are acquired to form a device fingerprint, and the features of each sensor are acquired in the same way to obtain several device fingerprints.

[0076] Please see Figure 3 The step of acquiring the features of a sensor to form a device fingerprint includes the following sub-steps:

[0077] S121. Obtain the time-domain characteristics of the sensor using the step response method. The time-domain characteristics include the maximum overshoot, hysteresis time, and peak time in the step response characteristic curve.

[0078] Specifically, a step signal is input to the sensor, and its response is recorded to obtain a step response characteristic curve. The step response characteristic curve is analyzed to calculate the maximum overshoot, hysteresis time, and peak time. The maximum overshoot refers to the maximum deviation of the step response characteristic curve from its steady-state value, usually expressed as a percentage. The hysteresis time is the time required for the step response characteristic curve to reach 50% of its steady-state value. The peak time is the time required for the step response characteristic curve to rise from zero to its first peak value.

[0079] S122. Obtain the frequency domain characteristics of the sensor by means of the frequency response method. The frequency domain characteristics include the bandwidth, time constant and natural frequency in the frequency response characteristic curve.

[0080] Specifically, a frequency response curve can be constructed by inputting a series of sinusoidal signals of different frequencies into the sensor and measuring its output. From this curve, the bandwidth can be determined, which represents the frequency range that the sensor can respond to. The time constant represents the time required for the response to decay to 1 / e of its initial value, and the natural frequency represents the frequency of the sensor during free vibration.

[0081] S123. The time-domain features and frequency-domain features are combined to form a device fingerprint, which is a six-dimensional column vector.

[0082] S13. The fingerprints of the aforementioned devices are preprocessed, and the preprocessed device fingerprints constitute a fingerprint database.

[0083] Specifically, this includes: data cleaning, checking for missing or outlier values ​​in the data, filling in missing values ​​or deleting records containing missing values ​​to remove noise and outliers; and performing data normalization or standardization to ensure that data from different sensor types are consistent in magnitude to improve the accuracy of subsequent classification and comparison.

[0084] S2. Cluster the device fingerprints in the fingerprint database to obtain fingerprint groups.

[0085] Please see Figure 4 Step S2 includes the following sub-steps:

[0086] S21. Initialize cluster centers by selecting K device fingerprints from the fingerprint database as cluster centers for K fingerprint groups.

[0087] Specifically, based on their functions and uses, ECU devices can be divided into different major categories, and the value of K depends on the number of these categories. In this embodiment, ECU devices are divided into four major categories: powertrain control, chassis and body control, new energy vehicle-specific, and intelligent driving assistance, with K taking the value of 4.

[0088] S22. Assign device fingerprints. Iterate through all device fingerprints, calculate the Euclidean distance between each device fingerprint and the K cluster centers, and assign each device fingerprint to the fingerprint group containing the cluster center with the smallest Euclidean distance.

[0089] S23. Update the cluster centers, recalculate the cluster centers of each fingerprint group, and use the arithmetic mean of all device fingerprints in the fingerprint group as the new cluster centers.

[0090] S24. Iterative optimization: Repeat steps S22-S23 until the cluster center positions of all fingerprint groups converge. That is, the current calculated cluster center position of each fingerprint group is less than the previously calculated cluster center position or the preset number of iterations is reached, thus obtaining the final fingerprint group.

[0091] S3. Obtain the device fingerprint of the ECU device under test as the test fingerprint, and determine the fingerprint group to which the test fingerprint belongs.

[0092] Specifically, the method for determining the fingerprint group to which the test fingerprint belongs is as follows: calculate the Euclidean distance between the test fingerprint and the cluster center of each fingerprint group, and determine the fingerprint group to which the test fingerprint belongs is the cluster center with the smallest Euclidean distance.

[0093] The Euclidean distance between the test fingerprint and the cluster center is calculated using the following formula:

[0094]

[0095] Where D' represents the Euclidean distance between the test fingerprint and the cluster center, and X 1n Let X represent the nth vector of the tested fingerprint. kn Let n be the nth vector representing the cluster center of the kth fingerprint group, where n = 1, 2, 3, 4, 5, 6.

[0096] S4. Using a Siamese neural network model, the test fingerprint is compared one by one with the device fingerprints in its fingerprint group to obtain the similarity between the test fingerprint and each device fingerprint.

[0097] Please see Figure 5 Step S4 includes the following sub-steps:

[0098] S41. Construct a twin neural network model.

[0099] Specifically, the Siamese neural network model includes two identical sub-networks, a feature convolutional layer, and a fully connected layer. The two identical sub-networks are used to extract features from the test fingerprint or the device fingerprint; the feature convolutional layer is used to fuse the features extracted by the two sub-networks through convolution operations; the fully connected layer is used to summarize the fused feature map into a single output and compress it to between 0 and 1, which serves as the similarity between the test fingerprint and the device fingerprint.

[0100] The two sub-networks can be convolutional neural networks (CNNs), recurrent neural networks (RNNs), or other types of neural networks. In this embodiment, the two sub-networks are convolutional neural networks, sharing the same weights and architecture, and are used to process the input test fingerprints or device fingerprints respectively. Each sub-network includes three convolutional layers to simultaneously represent features in the time and frequency domains. The first convolutional layer outputs a feature map after the convolution operation, the second convolutional layer outputs a further extracted feature map, and the third convolutional layer outputs the final extracted feature map for subsequent feature fusion or similarity calculation. Except for the third convolutional layer, each of the other convolutional layers is followed by a batch normalization layer and a modified linear unit activation function, which can significantly improve the training efficiency and performance of the model.

[0101] S42. Train the twin neural network model using the device fingerprints in the fingerprint database.

[0102] Please see Figure 6 Step S42 includes the following sub-steps:

[0103] S421. Divide the training set and validation set by randomly pairing device fingerprints in the fingerprint database and dividing them into training set and validation set according to a certain ratio.

[0104] S422. Input the training set data. Input the paired device fingerprints and corresponding labels from the training set into the Siamese neural network model. When two device fingerprints are similar, the label is 1; when two device fingerprints are dissimilar, the label is 0.

[0105] S423, Forward propagation: Extract the features of the two input device fingerprints using the two identical sub-networks and output their features.

[0106] S424. Calculate the contrast loss. Use the contrast loss function to calculate the distance between two output features and calculate the contrast loss based on whether they are similar.

[0107] If two fingerprints are similar, the contrastive loss function will reduce the distance; if two fingerprints are dissimilar, the contrastive loss function will increase the distance to ensure that similar fingerprints are close together in the feature space, while dissimilar fingerprints are far apart. The formula for calculating the contrastive loss is as follows:

[0108] L=(1-y)D 2 +y·max(0,mD) 2

[0109] Where L represents the contrast loss, y represents the binary label, y is 1 when the two input samples are similar and y is 0 when the two input samples are dissimilar, D represents the distance between the two feature vectors extracted by the two sub-networks, and m represents the preset second threshold.

[0110] S425. Backpropagation: Based on the result of the loss function, the gradient of the loss relative to the network weights is calculated using the backpropagation algorithm.

[0111] S426. Update the weights: Use an optimization algorithm to update the network weights based on the calculated gradients.

[0112] The formula for calculating the updated weights is as follows:

[0113]

[0114] Among them, W new W represents the updated weights. old This represents the original weights, and α represents the learning rate. This represents the gradient of the loss function.

[0115] S427. Repeat the iteration until the model's performance on the validation set no longer improves, i.e., the loss function curve fits or the preset number of iterations is reached.

[0116] S43. Input the test fingerprint and any device fingerprint in its fingerprint group into the trained Siamese neural network model, and output the similarity between the test fingerprint and the device fingerprint.

[0117] Specifically, the two identical sub-networks are used to extract features of the test fingerprint and device fingerprint input to the Siamese neural network model, respectively; the feature convolutional layer is used to fuse the extracted features of the test fingerprint and device fingerprint through convolution operations; the fully connected layer is used to summarize the fused feature map into a single output and compress it to between 0 and 1, which is used as the similarity between the test fingerprint and the device fingerprint.

[0118] S44. Traverse all device fingerprints in the fingerprint group to which the test fingerprint belongs, and obtain the similarity between the test fingerprint and each device fingerprint in the fingerprint group to which it belongs.

[0119] S5. Compare all similarities and find the device fingerprint with the highest similarity to the test fingerprint. The category of the ECU device corresponding to this device fingerprint is the same as the category of the ECU device corresponding to the test fingerprint.

[0120] Please see Figure 7 This is a structural block diagram of an embodiment of the vehicle ECU artificial intelligence identification device of the present invention. The vehicle ECU artificial intelligence identification device of this embodiment is used to implement the vehicle ECU artificial intelligence identification method described in the above embodiment. Specifically, the vehicle ECU artificial intelligence identification device of this embodiment includes a fingerprint database construction module 100, a clustering module 200, a group determination module 300, a comparison module 400, and a comparison and classification module 500. Wherein:

[0121] The fingerprint database construction module 100 is used to acquire device fingerprints of several ECU devices and construct a fingerprint database. The device fingerprints are sensor features of the ECU devices. The sensor features include time-domain features and frequency-domain features. The time-domain features include the maximum overshoot, hysteresis time, and peak time of the step response. The frequency-domain features include the bandwidth, time constant, and natural frequency of the frequency response.

[0122] The clustering module 200 is used to cluster device fingerprints in the fingerprint database constructed by the fingerprint database construction module 100 to obtain fingerprint groups.

[0123] The group determination module 300 is used to acquire the device fingerprint of the ECU device under test as a test fingerprint, and determine the fingerprint group to which the test fingerprint belongs.

[0124] The comparison module 400 is used to compare the test fingerprint with the device fingerprints in the fingerprint group to which the test fingerprint belongs, as determined by the group determination module 300, using a Siamese neural network model, to obtain the similarity between the test fingerprint and each device fingerprint.

[0125] The comparison and classification module 500 is used to compare all similarities obtained by the comparison module 400, find the device fingerprint with the highest similarity to the test fingerprint, and determine the category of the ECU device corresponding to the device fingerprint as the category of the ECU device corresponding to the test fingerprint.

[0126] This invention extracts hardware features of ECU devices, namely sensor features, to construct device fingerprints. It then utilizes advanced machine learning algorithms to identify ECU devices, preventing both software and hardware attacks and ensuring vehicle communication security. The device fingerprint includes the maximum overshoot, hysteresis time, and peak time of the step response, as well as the bandwidth, time constant, and natural frequency of the frequency response. These features are unique, stable, and resistant to external interference or tampering, improving the accuracy of ECU device identification. By introducing a clustering algorithm, the device fingerprints of a large number of ECU devices are grouped into multiple fingerprint groups. When determining the device category of a test fingerprint, it is only necessary to find the fingerprint group to which the test fingerprint belongs in the fingerprint database. Then, a Siamese neural network model is used to compare the test fingerprint with each device fingerprint in that group to determine the corresponding ECU device category. This method not only reduces computation and improves recognition speed but also makes the identification results more accurate and reliable through group division.

Claims

1. A vehicle ECU artificial intelligence identification method, characterized in that, Includes the following steps: Obtain device fingerprints for several ECU devices and construct a fingerprint database. The device fingerprints are the sensor features of the ECU devices. This step includes: selecting a unique sensor type for each type of ECU device, and each type of sensor includes multiple sensors. The features of each sensor in the selected sensor type are obtained to obtain several device fingerprints. The features of the sensors include time domain features and frequency domain features. The time domain features include the maximum overshoot, hysteresis time and peak time of the step response. The frequency domain features include the bandwidth, time constant and natural frequency of the frequency response. The fingerprints of the aforementioned devices are preprocessed, and the preprocessed device fingerprints constitute a fingerprint database. The device fingerprints in the fingerprint database are clustered to obtain fingerprint groups. This step includes: selecting K device fingerprints from the fingerprint database as cluster centers for the K fingerprint groups, and dividing the ECU devices into different categories according to their functions and uses. The value of K depends on the number of categories. Obtain the device fingerprint of the ECU device under test as the test fingerprint, and determine the fingerprint group to which the test fingerprint belongs; The test fingerprint is compared with the device fingerprints in its fingerprint group one by one using a Siamese neural network model to obtain the similarity between the test fingerprint and each device fingerprint. By comparing all similarities, the device fingerprint with the highest similarity to the test fingerprint is found. The category of the ECU device corresponding to this device fingerprint is the same as the category of the ECU device corresponding to the test fingerprint.

2. The vehicle ECU artificial intelligence identification method as described in claim 1, characterized in that, The step of obtaining the features of each sensor in the selected sensor type to obtain several device fingerprints includes: obtaining the features of a sensor to form a device fingerprint; The step of acquiring the features of a sensor to form a device fingerprint specifically includes the following sub-steps: The time-domain characteristics of the sensor are obtained by the step response method, and the time-domain characteristics include the maximum overshoot, hysteresis time, and peak time in the step response characteristic curve. The frequency domain characteristics of the sensor are obtained by the frequency response method, and the frequency domain characteristics include the bandwidth, time constant and natural frequency in the frequency response characteristic curve; The time-domain features and frequency-domain features are combined to form a device fingerprint, which is a six-dimensional column vector.

3. The vehicle ECU artificial intelligence identification method as described in claim 1, characterized in that, The step of clustering device fingerprints in the fingerprint database to obtain fingerprint groups specifically includes the following sub-steps: Initialize cluster centers by selecting K device fingerprints from the fingerprint database as cluster centers for K fingerprint groups; Assign device fingerprints, iterate through all device fingerprints, calculate the Euclidean distance between each device fingerprint and the K cluster centers, and assign each device fingerprint to the fingerprint group containing the cluster center with the smallest Euclidean distance; Update the cluster centers by recalculating the cluster centers of each fingerprint group and using the arithmetic mean of all device fingerprints in the fingerprint group as the new cluster centers. The process involves iterative optimization, repeating the steps of allocating device fingerprints and updating cluster centers until the cluster center positions of all fingerprint groups converge. This means that the current calculated cluster center position of each fingerprint group is less than the previously calculated cluster center position or the preset number of iterations is reached, thus obtaining the final fingerprint group.

4. The vehicle ECU artificial intelligence identification method as described in claim 1, characterized in that, The step of determining the fingerprint group to which the test fingerprint belongs includes: calculating the Euclidean distance between the test fingerprint and the cluster center of each fingerprint group, and determining that the fingerprint group containing the cluster center with the smallest Euclidean distance is the fingerprint group to which the test fingerprint belongs. The Euclidean distance between the test fingerprint and the cluster center is obtained by the following formula: (one) In formula (1), This represents the Euclidean distance between the test fingerprint and the cluster center. Indicates the first fingerprint test A vector, Indicates the first The cluster center of the fingerprint group A vector, .

5. The vehicle ECU artificial intelligence identification method as described in claim 1, characterized in that, The step of comparing the test fingerprint with each device fingerprint in its fingerprint group using a Siamese neural network model to obtain the similarity between the test fingerprint and each device fingerprint specifically includes the following sub-steps: Constructing a Siamese neural network model; The twin neural network model is trained using device fingerprints from the fingerprint database; The test fingerprint and any device fingerprint in its fingerprint group are input into the trained Siamese neural network model, and the similarity between the test fingerprint and the device fingerprint is output. Traverse all device fingerprints in the fingerprint group to which the test fingerprint belongs, and obtain the similarity between the test fingerprint and each device fingerprint in the fingerprint group to which it belongs.

6. The vehicle ECU artificial intelligence identification method as described in claim 5, characterized in that, The Siamese neural network model includes two identical sub-networks, a feature convolutional layer, and a fully connected layer; the step of training the Siamese neural network model using device fingerprints from the fingerprint database specifically includes the following sub-steps: The device fingerprints in the fingerprint database are randomly paired and divided into training and verification sets according to a certain ratio; The paired device fingerprints and corresponding labels from the training set are input into the Siamese neural network model. When two device fingerprints are similar, the label is 1, and when two device fingerprints are dissimilar, the label is 0. The features of the two input device fingerprints are extracted using the two identical sub-networks, and their features are output. The contrastive loss function is used to calculate the distance between two output features, and the contrastive loss is calculated based on whether they are similar. Based on the results of the loss function, the gradient of the loss with respect to the network weights is calculated using the backpropagation algorithm; The network weights are updated using an optimization algorithm based on the calculated gradient. Repeat the iterations until the model's performance on the validation set no longer improves, i.e., the loss function is no longer properly fitted or the preset number of iterations is reached.

7. The vehicle ECU artificial intelligence identification method as described in claim 6, characterized in that, The contrast loss is obtained by the following formula: (two) In formula (ii), Indicates comparative loss, This represents a binary label, which is used when two input samples are similar. A value of 1 indicates that the binary label is set to 1 when the two input samples are dissimilar. =0, This represents the distance between two feature vectors extracted by the two sub-networks. This indicates the preset second threshold; The updated weights are obtained using the following formula: (three) In formula (iii), This indicates the updated weights. This represents the original weight. Indicates the learning rate. This represents the gradient of the loss function.

8. The vehicle ECU artificial intelligence identification method as described in claim 6, characterized in that, The step of inputting the test fingerprint and any device fingerprint from its fingerprint group into the trained Siamese neural network model, and outputting the similarity between the test fingerprint and the device fingerprint, specifically includes the following sub-steps: The two identical sub-networks are used to extract features from the test fingerprint and device fingerprint input to the Siamese neural network model, respectively; The feature convolutional layer is used to fuse the features of the extracted test fingerprint and device fingerprint through convolution operations; The fully connected layer is used to summarize the fused feature maps into a single output and compress it to between 0 and 1, which is used as the similarity between the test fingerprint and the device fingerprint.

9. A vehicle ECU artificial intelligence identification device, characterized in that, include: A fingerprint database construction module is used to acquire device fingerprints of several ECU devices and construct a fingerprint database. The device fingerprints are sensor features of the ECU devices. The fingerprint database construction module is used to select sensor types unique to each ECU device, and each sensor type includes multiple sensors. The features of each sensor in the selected sensor type are acquired to obtain several device fingerprints. The sensor features include time-domain features and frequency-domain features. The time-domain features include the maximum overshoot, hysteresis time, and peak time of the step response. The frequency-domain features include the bandwidth, time constant, and natural frequency of the frequency response. The several device fingerprints are preprocessed, and the preprocessed device fingerprints constitute the fingerprint database. The clustering module is used to cluster the device fingerprints in the fingerprint database to obtain fingerprint groups. The clustering module is used to select K device fingerprints from the fingerprint database as the cluster centers of the K fingerprint groups. According to different functions and uses, the ECU devices are divided into different categories, and the value of K depends on the number of categories. The group determination module is used to acquire the device fingerprint of the ECU device under test as a test fingerprint, and determine the fingerprint group to which the test fingerprint belongs; The comparison module is used to compare the test fingerprint with the device fingerprints in its fingerprint group one by one using a Siamese neural network model to obtain the similarity between the test fingerprint and each device fingerprint. The comparison and classification module is used to compare all similarities, find the device fingerprint with the highest similarity to the test fingerprint, and determine the category of the ECU device corresponding to the device fingerprint.

Images