Method, apparatus, and device for determining a class of power consumption characteristics of an encryption device

Abstract

The present disclosure provides a method, device and equipment for determining the power consumption feature category of an encryption device; the method comprises: obtaining side channel power consumption leakage information of a target encryption device; the target encryption device is an encryption device that needs to be subjected to power consumption feature analysis, and the side channel power consumption leakage information comprises a power consumption trajectory; inputting the side channel power consumption leakage information into a pre-trained wide learning network to obtain category parameters output by the pre-trained wide learning network, which represent the classification of power consumption features; the pre-trained wide learning network is obtained by training based on a training data set using a supervised training method. The method of the present disclosure can improve the classification accuracy to a certain extent and reduce the cost required for calculation.

Description

Technical Field

[0001] This disclosure relates to the field of chip security analysis technology, and in particular to a method, apparatus, electronic device, non-transitory computer-readable storage medium, and computer program product for determining the power consumption characteristic category of an encrypted device. Background Technology

[0002] In the field of chip security analysis, side-channel power leakage analysis, as a highly effective technique, has been widely applied to various chip attack scenarios.

[0003] Side-channel power leakage analysis schemes in related technologies are mainly based on deep learning-based power side-channel analysis techniques. These techniques analyze the power consumption information leaked by encryption devices during encryption operations and then crack the key. On one hand, deep learning technology can directly learn features from the original power consumption trajectory, eliminating the need for manually designed features and simplifying attack design. On the other hand, deep learning models can directly learn to predict intermediate values ​​of the target, further simplifying the attack process.

[0004] While deep learning-based side-channel power leakage analysis has the aforementioned advantages, it also suffers from high computational resource requirements, large data requirements, and limited model generalization ability. Summary of the Invention

[0005] This disclosure provides a method, apparatus, electronic device, non-transitory computer-readable storage medium, and computer program product for determining the power consumption characteristic category of an encryption device, in order to address the deficiencies in the prior art.

[0006] This disclosure provides a method for determining the power consumption characteristic category of an encrypted device, including: Obtain side-channel power leakage information of the target encryption device; wherein, the target encryption device is an encryption device that requires power consumption feature analysis, and the side-channel power leakage information includes power consumption trajectory; The side-channel power leakage information is input into a pre-trained width learning network to obtain category parameters representing the power consumption feature classification output by the pre-trained width learning network. The pre-trained width learning network is trained using a supervised training method based on the training dataset.

[0007] This disclosure also provides an apparatus for determining the power consumption characteristics of an encryption device, including... The data acquisition module is configured to: acquire side-channel power consumption leakage information of the target encryption device; wherein the target encryption device is an encryption device that requires power consumption feature analysis, and the side-channel power consumption leakage information includes power consumption trajectory; The prediction module is configured to: input the side channel power leakage information into a pre-trained width learning network to obtain category parameters representing the power consumption feature classification output by the pre-trained width learning network; The pre-trained width learning network is trained using a supervised training method based on the training dataset.

[0008] This disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method for determining a power consumption characteristic category of an encryption device as described above.

[0009] This disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for determining the power consumption characteristic category of an encryption device as described in any of the above.

[0010] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the method for determining a power consumption characteristic category of an encryption device as described above.

[0011] As described above, the method for determining the power consumption feature category of an encrypted device provided in this disclosure can effectively classify the side-channel power consumption leakage information of the target encrypted device through a "pre-trained wide learning network". This significantly shortens the time required for side-channel analysis, improves the classification accuracy to a certain extent, and reduces the computational cost. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is one of the flowcharts illustrating a method for determining the power consumption characteristic category of an encrypted device provided in this disclosure embodiment; Figure 2 This is a second flowchart illustrating a method for determining the power consumption characteristic category of an encrypted device, provided in an embodiment of this disclosure. Figure 3 This is a schematic diagram of the power consumption trajectory waveform provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram of the structure of the training width learning network provided in an embodiment of this disclosure; Figure 5This is a schematic diagram of the accuracy waveform of the output result of the pre-trained wide learning network provided in this embodiment of the present disclosure when performing "determining the power consumption feature category of the encrypted device" compared with the existing model; Figure 6 This is a schematic diagram of the time consumption waveform of the output result when the pre-trained width learning network provided in this embodiment performs "determining the power consumption feature category of the encryption device" compared with the existing model; Figure 7 This is a schematic diagram of the structure of an apparatus for determining the power consumption characteristic category of an encryption device, provided in an embodiment of this disclosure; Figure 8 This is a schematic diagram of the structure of the electronic device provided in this disclosure. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this disclosure clearer, the technical solutions of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0015] To address the problems of the prior art, the inventors of this disclosure, after research, propose a scheme for determining the power consumption characteristic category of an encryption device, as described in the following embodiments.

[0016] Example: The following description, in conjunction with the accompanying drawings, details the implementation of the scheme disclosed herein for determining the power consumption characteristic category of encrypted devices.

[0017] Figure 1 This is a schematic flowchart illustrating a method for determining the power consumption characteristic category of an encrypted device, provided in an exemplary embodiment of this disclosure. This embodiment can be applied to electronic devices (e.g., servers or cloud computing platforms), such as... Figure 1 As shown, the method for determining the power consumption characteristic category of an encrypted device includes the following steps: S110. Obtain the side-channel power consumption leakage information of the target encryption device.

[0018] The target encryption device is an encryption device that requires power consumption characteristic analysis.

[0019] The side-channel power leakage information includes power consumption trajectories.

[0020] Among these, power consumption leakage information is a type of side-channel leakage information from an encrypted device; for example, it can be physical characters of the encrypted device, which may include, but are not limited to, radio frequency, power consumption, and computation time. This power consumption leakage information leaks sensitive intermediate value information related to the operation of the encrypted device.

[0021] This disclosure does not limit the method of acquisition. For example, an electronic device (e.g., a server or cloud computing platform) acting as the execution entity can directly collect the side-channel power leakage information from the target encryption device; or, it can communicate with a server storing the "side-channel power leakage information" to obtain the side-channel power leakage information.

[0022] S120. Input the side channel power leakage information into the pre-trained width learning network to obtain the category parameters representing the power consumption feature classification output by the pre-trained width learning network.

[0023] The pre-trained width learning network is trained using a supervised training method based on the training dataset.

[0024] The training process of the "pre-trained width learning network" will be described below and will not be repeated here.

[0025] As described above, the method for determining the power consumption feature category of an encrypted device provided in this disclosure can effectively classify the side-channel power consumption leakage information of the target encrypted device through a "pre-trained wide learning network". This significantly shortens the time required for side-channel analysis, improves the classification accuracy to a certain extent, and reduces the computational cost.

[0026] exist Figure 1 Based on the embodiments, as an optional implementation method, refer to Figure 2 The pre-trained width learning network includes a graph feature layer (i.e., Figure 2 Mapped Features and Enhancement Layer (i.e.) Figure 2 (See the Enhancement Nodes).

[0027] Mapped Features can be divided into multiple groups, such as Group 1 to Group m; EnhancementNodes can also be divided into multiple groups, such as Group 1 to Group n.

[0028] Optionally, refer to Figure 2 The workflow of a pre-trained width learning network is summarized as follows: Step 1: Input the side-channel power consumption leakage information (i.e., input data X) of the target encryption device into the Mapped Features layer; Step 2: The Mapped Features layer can perform operations based on the input data X to construct linear mapping features. ; Step 3: Enhancement Nodes can be processed based on linear mapping features to obtain additional enhancement nodes. ; Step 4: Based on the randomly generated features, set the connection weights from the linear mapping features to the additional enhancement nodes. W c ; Step 5: Using the calculation formula By performing the calculation, the relevant output Y can be obtained. Among them, the feature classification result corresponding to "generating the relevant output Y" can be calculated through pseudo-inverse.

[0029] exist Figure 1 Based on the embodiments, as an optional implementation, the training steps of the pre-trained width learning network include: Step 1) Obtain the training dataset.

[0030] The training dataset includes multiple sets of training data, each set of training data including a subset of sample data and a corresponding subset of labels; the subset of sample data includes multiple historical power consumption trajectories of the target encryption device; the subset of labels includes multiple labels determined based on a preset leakage model, the labels being used to characterize the power consumption feature classification of the multiple historical power consumption trajectories in the subset of sample data.

[0031] As an optional example, obtaining the training dataset includes: Step 11) Sample the side-channel power leakage energy of the target encryption device to obtain multiple initial power trajectories.

[0032] Optionally, side-channel power leakage energy traces can be collected from the target encryption device. (Refer to...) Figure 3 The single initial power consumption trajectory shown uses the sampling time (i.e., points of sampling time) as the horizontal axis and the voltage (characterizing energy) as the vertical axis, recording the corresponding power consumption leakage information to form the initial power consumption trajectory.

[0033] Alternatively, analysis equipment similar to the target encryption device can be used to collect side-channel power leakage data to obtain the corresponding initial power trajectory.

[0034] Step 12) Preprocess the multiple initial power consumption trajectories to obtain multiple historical power consumption trajectories.

[0035] Optionally, preprocessing includes data normalization.

[0036] Step 13) Divide the multiple historical power consumption trajectories into multiple sample data subsets.

[0037] Step 14) Use a preset leakage model to determine the tag set corresponding to each subset of sample data.

[0038] It should be explained that the selection of the label set can be set according to the characteristics of the sample data subset. Commonly used labels include the S (Substitution-box) output median, Hamming distance, and Hamming weight.

[0039] Optionally, in this scheme, the "S-box output intermediate value model" is selected as the preset leakage model, where the S-box is the basic structure for permutation calculations performed by the symmetric key algorithm. Then, the intermediate values ​​of the S-box outputs corresponding to the sample data subsets are set as the label set. Specifically, the S-box output intermediate value model can be used to set label sets for multiple sample data subsets after data standardization.

[0040] Step 15) Based on the plurality of sample data subsets and the corresponding label sets, the training dataset is obtained.

[0041] Optionally, an input feature matrix with a bias term can be formed using multiple subsets of sample data with a set of labels, which can then be used as the training dataset.

[0042] Step 2) Input the multiple sets of training data into the width learning network to be trained, and output the power consumption feature classification prediction value corresponding to each set of training data through the width learning network to be trained.

[0043] The training-to-be-trained width learning network includes a trainable graph augmentation layer and a training-to-be-trained augmentation layer. (Refer to...) Figure 4 The training graph enhancement layers can be divided into multiple groups, such as Mapped Features 1 to Mapped Features n; the training enhancement layers are Enhancement Nodes.

[0044] Based on this, as an optional example, step 2) may include the following steps: Step 21) Using preset feature construction rules, determine the mapping features and associated nonlinear features of each set of training data in the training graph enhancement layer.

[0045] Optionally, in the case of "the input feature matrix with a bias term as the training dataset", a linear feature map is constructed for each feature value of the input, and the relevant nonlinear features are calculated through an activation function. Specifically, the mapping features for each set of training data can be determined using the following calculation formula (3): (3) in, Z Represents mapping features; The vector form of the mapping feature function is represented; X represents the training data.W Indicates the corresponding weight; b This indicates the bias term.

[0046] Furthermore, the calculation of nonlinear features is implemented using the 'ReLU' function in this scheme. It should be noted that extracting nonlinear features enables the "trainable width learning network" to fit the data more accurately when processing side-channel power leakage information, thereby improving the accuracy of prediction and classification.

[0047] Step 22) Using the preset augmentation node correspondence rules, determine the augmentation node corresponding to each of the mapped features in the augmentation layer to be trained.

[0048] Alternatively, the enhancement node corresponding to each of the mapping features can be determined using the following calculation formula (4): (4) in, Z Represents mapping features; H Indicates an enhanced node; This represents the random weights associated with the augmenting nodes; This represents the bias terms associated with the augmented node; This represents the vector form of the enhancement function.

[0049] Preferably, They are typically set to orthogonal form to reduce potential correlations between the inputs of the augmentation nodes.

[0050] Step 23) Based on the mapping features and the corresponding enhancement nodes, determine the weighted connectivity matrix between the feature layer and the enhancement layer of the graph to be trained using preset connectivity rules.

[0051] Optionally, in step 23), the weighted connectivity matrix can be determined using the following calculation formulas (1) and (2). T ; (1) (2) in, A This represents the connectivity matrix between the graph feature layer and the enhancement layer; Z This represents the mapping features in the graph feature layer; H This represents the enhancement node in the enhancement layer; T Denotes the weighted connectivity matrix; This represents the weight matrix from input to output.

[0052] Step 24) Based on the weighted connected matrix, use the pseudo-inverse calculation rule to obtain the power consumption feature classification prediction value corresponding to each group of training data.

[0053] Step 3) Using the preset error estimation rules, obtain the prediction error between the predicted value of each power consumption feature classification and the corresponding tag subset.

[0054] As an optional example, step 3) can be implemented as follows: First, based on the predicted values ​​of each group of power consumption features and the corresponding subset of labels, the accuracy or inaccuracy of the predicted values ​​of each group of power consumption features is determined; then, the prediction error is determined using the accuracy or inaccuracy.

[0055] Specifically, a prediction label set corresponding to each group of power consumption feature classification prediction values ​​can be determined. Then, the accuracy or inaccuracy rate can be calculated using the prediction label set and the corresponding label subset (i.e., the sample true values). Taking the calculation of inaccuracy rate as an example, the calculated inaccuracy rate can be directly used as the prediction error. Taking the calculation of accuracy rate as an example, the result of subtracting the accuracy rate from 1 is used as the prediction error.

[0056] Step 4) In response to the prediction error being greater than a preset prediction error threshold, update the training width learning network using a preset incremental rule.

[0057] As an optional example, step 4) may include the following steps: Step 41) Based on the learning network to be trained, add new augmenting nodes and new mapping features to obtain an updated network. Step 42) Using the current connectivity matrix of the learning network to be trained, determine the input matrix of the updated network. Step 43) Based on the input matrix, determine the connection weights and bias terms from the new mapping features to the new augmenting nodes in the updated network.

[0058] Among them, reference Figure 4 The red nodes (Additonal Enhancement Nodes) represent newly added enhancement nodes.

[0059] The input matrix of the update network can be determined using the following formula (5): (5) in, Represents the input matrix; Represents the current connected matrix; Represents the vector form of the enhanced function; This represents the mapping features in the graph feature layer; This represents the connection weights from the new mapped features to the new enhanced nodes; This represents the bias term from the new mapped features to the new enhanced nodes.

[0060] Alternatively, when dealing with power trajectories with subtle feature differences, the incremental enhancements of nodes may not be sufficient for feature classification if only a limited number of feature maps selected from the base model are relied upon. Therefore, we introduce more novel feature maps without repeatedly training the entire width learning network.

[0061] Specifically, as another optional example, step 4) may include the following steps: Step 41') Based on the learning network to be trained, add new mapping features to obtain an updated network; Step 42') Using the preset augmentation node correspondence rules, determine the new augmentation node corresponding to each new mapping feature; Step 43') Using the current connectivity matrix of the learning network to be trained, determine the input matrix of the updated network; Step 44') Based on the input matrix, determine the connection weights and bias terms from the new mapping features to the new augmentation nodes in the updated network.

[0062] The input matrix of the update network can be determined using the following formula (5): (5) in, Represents the input matrix; Represents the current connected matrix; Represents the vector form of the enhanced function; This represents the mapping features in the graph feature layer; This represents the connection weights from the new mapped features to the new enhanced nodes; This represents the bias term from the new mapped features to the new enhanced nodes.

[0063] It should be noted that in the above example, the reference is... Figure 4 , Through steps 43') to 44'), add new enhanced nodes (i.e., additional nodes, such as... Figure 4 The input matrix obtained after the red nodes (representing Additonal Enhancement Nodes) are shown. (Adding extra nodes is similar to a model optimization step.) Performing pseudo-inverse calculations can determine the relevant connection weights. and bias terms .

[0064] In this scheme, the pseudo-inverse calculations of matrices are all performed using regularization methods.

[0065] Step 5) Iteratively execute steps 2) to 4) until the prediction error is less than the preset prediction error threshold, and obtain the pre-trained width learning network from the width learning network to be trained.

[0066] Based on the above embodiments and implementation methods, as an optional implementation method, the method for determining the power consumption feature category of the encryption device further includes: a verification step of the pre-trained width learning network obtained through training.

[0067] Optionally, the verification step includes: inputting the output of the pre-trained wide learning network into the mathematical analysis module for accuracy verification, calculating the time taken for the analysis process, and outputting the accuracy and time consumption as the analysis results. Specifically, the mathematical analysis module can perform accuracy and time consumption calculations. Accuracy refers to the output result obtained by comparing the label set obtained by the model (i.e., the pre-trained wide learning network) through predictive feature classification of the validation dataset (which can be divided into training and validation datasets in an 8:2 ratio, for example) with the actual real label set. Time consumption refers to the time spent by the model analyzing the validation dataset and classifying its features. A corresponding real label set is established based on the validation dataset. The validation dataset is used as input data for classification prediction of the trained wide learning model, and the predicted label set is compared with the real label set to obtain the model output results. After the above two steps are completed, this scheme can directly output the time consumption and accuracy as the results of the model's side-channel analysis feature classification of the validation dataset.

[0068] Verification Example: The feature mapping and number of nodes were set to 10, and the batch size of the dataset was adjusted to 100 to balance computational efficiency and model representation; accuracy and time consumption were used as the analysis results.

[0069] After selecting the parameters, a scenario was implemented using the AES-128 encryption algorithm on an AT89S52 microcontroller. A leakage model was used to combine the plaintext with the key, calculate the median output of the S-box, and use it as the key label. The label and dataset were input into the model for attack analysis. Finally, the model's performance was analyzed based on the output results.

[0070] like Figure 5 and Figure 6 As shown, the "pre-trained width learning network" provided in this disclosure demonstrates good performance in the AES-128 encryption algorithm. Compared with other deep learning models, the "pre-trained width learning network" provided in this disclosure outperforms other models in terms of accuracy; it also consumes less time than LSTM and SVC models.

[0071] The apparatus for determining the power consumption characteristic category of an encrypted device provided in this disclosure is described below. The apparatus for determining the power consumption characteristic category of an encrypted device described below can be referred to in correspondence with the method for determining the power consumption characteristic category of an encrypted device described above.

[0072] Figure 7 This is a schematic diagram of a device for determining the power consumption characteristic category of an encryption device, provided in an exemplary embodiment of this disclosure. Figure 7 As shown, the apparatus for determining the power consumption characteristic category of an encryption device includes: The data acquisition module 710 is configured to: acquire side-channel power consumption leakage information of a target encryption device; wherein the target encryption device is an encryption device that requires power consumption feature analysis, and the side-channel power consumption leakage information includes power consumption trajectories; Prediction module 720 is configured to: input the side channel power leakage information into a pre-trained width learning network to obtain category parameters representing the power consumption feature classification output by the pre-trained width learning network; The pre-trained width learning network is trained using a supervised training method based on the training dataset.

[0073] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can invoke logical instructions in the memory 830 to execute a method for determining the power consumption characteristic category of an encrypted device. This method includes: acquiring side-channel power leakage information of a target encrypted device; wherein the target encrypted device is an encrypted device requiring power consumption characteristic analysis, and the side-channel power leakage information includes a power consumption trajectory; inputting the side-channel power leakage information into a pre-trained wide learning network to obtain category parameters representing the power consumption characteristic classification output by the pre-trained wide learning network; wherein the pre-trained wide learning network is trained using supervised training based on a training dataset.

[0074] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a part 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 disclosure. 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.

[0075] On the other hand, this disclosure also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the methods provided above for determining the power consumption feature category of an encryption device. The method includes: acquiring side-channel power consumption leakage information of a target encryption device; wherein the target encryption device is an encryption device that needs to be analyzed for power consumption features, and the side-channel power consumption leakage information includes a power consumption trajectory; inputting the side-channel power consumption leakage information into a pre-trained width learning network to obtain category parameters representing the power consumption feature classification output by the pre-trained width learning network; wherein the pre-trained width learning network is trained using a supervised training method based on a training dataset.

[0076] In another aspect, this disclosure also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the methods provided above for determining the power consumption feature category of an encryption device. The method includes: acquiring side-channel power consumption leakage information of a target encryption device; wherein the target encryption device is an encryption device requiring power consumption feature analysis, and the side-channel power consumption leakage information includes a power consumption trajectory; inputting the side-channel power consumption leakage information into a pre-trained width learning network to obtain category parameters representing the power consumption feature classification output by the pre-trained width learning network; wherein the pre-trained width learning network is trained using supervised training based on a training dataset.

[0077] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0078] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure.

Claims

1. A method for determining the power consumption characteristic category of an encrypted device, comprising: Obtain side-channel power leakage information of the target encryption device; wherein, the target encryption device is an encryption device that requires power consumption feature analysis, and the side-channel power leakage information includes power consumption trajectory; The side-channel power leakage information is input into a pre-trained width learning network to obtain category parameters representing the power consumption feature classification output by the pre-trained width learning network. The pre-trained width learning network is trained using a supervised training method based on a training dataset. The pre-trained width learning network includes a graph feature layer and an enhancement layer; The training steps of the pre-trained width learning network include: Step 1) Obtain the training dataset; Step 2) Input multiple sets of training data into the width learning network to be trained, and output the power consumption feature classification prediction value corresponding to each set of training data through the width learning network to be trained. Step 3) Using the preset error estimation rules, obtain the prediction error between the predicted value of each power consumption feature classification and the corresponding tag subset; Step 4) In response to the prediction error being greater than a preset prediction error threshold, update the training width learning network using a preset incremental rule; Step 5) Iteratively execute steps 2) to 4) until the prediction error is less than the preset prediction error threshold, and obtain the pre-trained width learning network from the width learning network to be trained; The width learning network to be trained includes a graph augmentation layer to be trained and an augmentation layer to be trained; the power consumption feature classification prediction value output by the width learning network to be trained for each group of training data includes: Step 21) Using preset feature construction rules, determine the mapping features and associated nonlinear features of each set of training data in the training graph enhancement layer; Step 22) Using the preset augmentation node correspondence rules, determine the augmentation node corresponding to each of the mapped features in the augmentation layer to be trained; Step 23) Based on the mapping features and the corresponding enhancement nodes, determine the weighted connectivity matrix between the feature layer and the enhancement layer of the graph to be trained using preset connectivity rules; Step 24) Based on the weighted connected matrix, use the pseudo-inverse calculation rule to obtain the power consumption feature classification prediction value corresponding to each group of training data.

2. The method according to claim 1, characterized in that, The training dataset includes multiple sets of training data, each set of training data including a subset of sample data and a corresponding subset of labels; the subset of sample data includes multiple historical power consumption trajectories of the target encryption device; the subset of labels includes multiple labels determined based on a preset leakage model, the labels being used to characterize the power consumption feature classification of the multiple historical power consumption trajectories in the subset of sample data.

3. The method according to claim 2, characterized in that, The acquisition of the training dataset includes: Step 11) Sample the side-channel power leakage energy of the target encryption device to obtain multiple initial power trajectories; Step 12) Preprocess the multiple initial power consumption trajectories to obtain multiple historical power consumption trajectories; Step 13) Divide the multiple historical power consumption trajectories into multiple sample data subsets; Step 14) Using a preset leakage model, determine the label set corresponding to each subset of sample data; Step 15) Based on the plurality of sample data subsets and the corresponding label sets, the training dataset is obtained.

4. The method according to claim 1, characterized in that, The step of updating the training width learning network using a preset incremental rule in response to the prediction error exceeding a preset prediction error threshold includes: Based on the learning network of the width to be trained, new enhancement nodes and new mapping features are added to obtain the updated network; The current connectivity matrix of the network is learned using the width to be trained, and the input matrix of the updated network is determined. Based on the input matrix, in the updated network, the connection weights and bias terms from the new mapped features to the new enhanced nodes are determined; or, Based on the network to be trained, new mapping features are added to obtain an updated network; Using preset enhanced node correspondence rules, determine the new enhanced node corresponding to each of the new mapping features; The current connectivity matrix of the network is learned using the width to be trained, and the input matrix of the updated network is determined. Based on the input matrix, the connection weights and bias terms from the new mapped features to the new enhanced nodes are determined in the updated network.

5. The method according to claim 1, characterized in that, Based on the mapping features and the corresponding enhancement nodes, a weighted connectivity matrix between the feature layer and the enhancement layer of the graph to be trained is determined using preset connectivity rules. include: The weighted connectivity matrix is ​​determined using the following formulas (1) and (2). T ; （1） （2） in, A This represents the connectivity matrix between the graph feature layer and the enhancement layer; Z This represents the mapping features in the graph feature layer; H This represents the enhancement node in the enhancement layer; T Denotes the weighted connectivity matrix; This represents the weight matrix from input to output.

6. The method according to claim 1, characterized in that, The step of obtaining the prediction error between each group of power consumption feature classification prediction values ​​and the corresponding label subset using a preset error estimation rule includes: Based on the predicted value of each power consumption feature classification and the corresponding tag subset, determine the accuracy or inaccuracy of each group of power consumption feature classification prediction values. The prediction error is determined using the accuracy or inaccuracy rate.

7. An apparatus for determining the power consumption characteristics of an encryption device, comprising: The data acquisition module is configured to: acquire side-channel power consumption leakage information of the target encryption device; wherein, The target encryption device is an encryption device that requires power consumption characteristic analysis, and the side channel power leakage information includes power consumption trajectory; The prediction module is configured to: input the side channel power leakage information into a pre-trained width learning network to obtain category parameters representing the power consumption feature classification output by the pre-trained width learning network; The pre-trained width learning network is trained using a supervised training method based on a training dataset. The pre-trained width learning network includes a graph feature layer and an enhancement layer; The training steps of the pre-trained width learning network include: Step 1) Obtain the training dataset; Step 2) Input multiple sets of training data into the width learning network to be trained, and output the power consumption feature classification prediction value corresponding to each set of training data through the width learning network to be trained. Step 3) Using the preset error estimation rules, obtain the prediction error between the predicted value of each power consumption feature classification and the corresponding tag subset; Step 4) In response to the prediction error being greater than a preset prediction error threshold, update the training width learning network using a preset incremental rule; Step 5) Iteratively execute steps 2) to 4) until the prediction error is less than the preset prediction error threshold, and obtain the pre-trained width learning network from the width learning network to be trained; The width learning network to be trained includes a graph augmentation layer to be trained and an augmentation layer to be trained; the power consumption feature classification prediction value output by the width learning network to be trained for each group of training data includes: Step 21) Using preset feature construction rules, determine the mapping features and associated nonlinear features of each set of training data in the training graph enhancement layer; Step 22) Using the preset augmentation node correspondence rules, determine the augmentation node corresponding to each of the mapped features in the augmentation layer to be trained; Step 23) Based on the mapping features and the corresponding enhancement nodes, determine the weighted connectivity matrix between the feature layer and the enhancement layer of the graph to be trained using preset connectivity rules; Step 24) Based on the weighted connected matrix, use the pseudo-inverse calculation rule to obtain the power consumption feature classification prediction value corresponding to each group of training data.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for determining the power consumption characteristics of an encryption device as described in any one of claims 1 to 6.

Images