Halpern type fixed point adaptive optimization side channel analysis method and system

Abstract

The invention discloses a side channel analysis method and system for adaptive optimization of a Halpern type fixed point, and belongs to the field of side channel analysis, and the method comprises the following steps: carrying out the grouping and data processing of a plurality of side channel trace data, and obtaining processed data; constructing a side channel analysis model based on a convolutional neural network, and classifying and identifying the processed data based on the side channel analysis model to obtain a predicted label value probability corresponding to each piece of side channel trace data; optimizing the side channel analysis model by adopting a Halpern type fixed point adaptive optimization algorithm based on the predicted label value probabilities corresponding to the plurality of side channel trace data to obtain a final model; and obtaining a side channel analysis result based on the final model. According to the method, the model training stability is improved through the Halpern type fixed point adaptive optimization algorithm, the model generalization ability is enhanced, and an efficient solution is provided for a complex optimization task in a side channel analysis problem.

Description

Technical Field

[0001] The present invention belongs to the technical field of side channel analysis, and in particular relates to a side channel analysis method and system for Halpern type fixed point adaptive optimization. Background Art

[0002] Supervised deep learning methods have been widely used in side channel analysis. With the exploration and application of deep learning methods by the side channel community, many open source models and data sets have been proposed. The interest points of these works are mostly focused on model construction and hyperparameter tuning, as well as improving the generalization ability of the model. Few works focus on the selection of model optimizers. Recent studies have pointed out that hyperparameter tuning not only needs to consider the parts related to the neural network architecture, but also should select the optimizer that affects the training. Common adaptive optimizers Adam and RMSProp are prone to overfitting, so small-scale neural networks and short training processes should be considered. In addition, non-adaptive methods, that is, SGD-type optimizers, although they rarely overfit, have a slow convergence speed, and the effect is only good under the premise of using momentum acceleration. Therefore, the present invention proposes a side channel analysis method and system for Halpern-type fixed point adaptive optimization. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a side channel analysis method and system for Halpern-type fixed point adaptive optimization to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above object, the present invention provides a side channel analysis method for Halpern type fixed point adaptive optimization, comprising:

[0005] Grouping and processing a number of side channel trace data to obtain processed data;

[0006] Building a side channel analysis model based on a convolutional neural network, and classifying and identifying the processed data based on the side channel analysis model to obtain a predicted label value probability corresponding to each side channel trace data;

[0007] Based on the predicted label value probabilities corresponding to the side channel trace data, the side channel analysis model is optimized using a Halpern-type fixed point adaptive optimization algorithm to obtain a final model;

[0008] A side channel analysis result is obtained based on the final model.

[0009] Optionally, the process of obtaining processed data includes:

[0010] Collecting a number of side channel trace data and label values ​​corresponding to the side channel trace data;

[0011] Rearranging and grouping a plurality of side channel trace data to obtain grouped side channel trace data;

[0012] The grouped rear channel trace data is preprocessed to obtain processed data.

[0013] Optionally, the side channel analysis model includes: a first convolutional layer, a first batch of normalization layers, a first average pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, and a softmax layer;

[0014] The process of classifying and identifying the processed data based on the side channel analysis model includes:

[0015] Inputting the processed data into the first convolution layer to obtain a first convolution feature;

[0016] The first batch of standardized layers normalizes the first convolutional features to obtain normalized features of the first batch of standardized layers;

[0017] The first average pooling layer downsamples the normalized features of the first batch of normalization layers to obtain first pooling features;

[0018] The flattening layer flattens the first pooled features to obtain flattened features;

[0019] The flattened features are sequentially passed through the first fully connected layer, the second fully connected layer, the third fully connected layer and the softmax layer to obtain the predicted label value probability corresponding to each side channel trace data.

[0020] Optionally, the process of optimizing the side channel analysis model includes:

[0021] A multi-classification cross entropy loss function model is constructed based on the predicted label value probability and label value corresponding to the channel trace data on the same side of the same batch;

[0022] The final model is obtained by training the multi-classification cross entropy loss function model based on the Halpern fixed point adaptive optimization algorithm.

[0023] Optionally, the expression of the multi-classification cross entropy loss function model is:

[0024]

[0025] Where L is the loss value, C is the total number of categories, and y i is the true label value, is the predicted probability output by the softmax layer.

[0026] Optionally, the process of optimizing the side channel analysis model based on the Halpern-type fixed point adaptive optimization algorithm includes: optimizing the parameters in the side channel analysis model based on a multi-classification cross entropy loss function model to obtain optimized model parameters;

[0027] Among them, the optimized model parameters are:

[0028]

[0029] In the formula, x n+1 represents the optimized model parameters, β 3 represents the Halpern recursion coefficient of the algorithm, x 0 represents the initial parameters of the model, x n represents the model parameters of the nth iteration, represents the learning rate of the model, represents the cumulative momentum of the model, and n represents the number of iterations of the algorithm.

[0030] The present invention also discloses a Halpern type fixed point adaptive optimization side channel analysis system, which is used to implement a Halpern type fixed point adaptive optimization side channel analysis method, wherein the side channel analysis system comprises: an acquisition module, a preprocessing module, a construction module, an optimization module and an analysis module;

[0031] The acquisition module is used to rearrange and group a plurality of side channel trace data to obtain grouped side channel trace data;

[0032] The preprocessing module is used to preprocess the grouped side channel trace data to generate processed data;

[0033] The construction module is used to establish a side channel analysis model and classify and identify the processed data to obtain the predicted label probability of each side channel trace data;

[0034] The optimization module is used to optimize the side channel analysis model based on the predicted label probability of the side channel trace data to generate a final model;

[0035] The analysis module is used to obtain a side channel analysis result based on the final model.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] First, the present invention groups and processes the side channel trace data, which can effectively remove noise and improve data quality. Secondly, the analysis model is constructed using a convolutional neural network, which can accurately extract data features and lay the foundation for classification and identification. Furthermore, the Halpern-type fixed point adaptive optimization algorithm is used to optimize the model based on the predicted label value probability, so that the model parameters are continuously approached to the optimal level, thereby enhancing the generalization ability and prediction accuracy of the model. Finally, the side channel analysis results obtained based on the optimized model are more reliable, can effectively assist in information security protection, warn of potential risks in advance, and ensure the stable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0039] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0040] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0042] Embodiment 1

[0043] "Side channel analysis task" means recovering the key used by the hardware used to generate the encrypted data from the input encrypted data. "Performing the side channel analysis task using the final model" means extracting the features of the encrypted data through the optimized model to obtain the key.

[0044] In view of the problem that the existing optimization algorithms have slow convergence speed and are prone to overfitting, the present invention proposes a side channel analysis method of Halpern type fixed point adaptive optimization. The purpose of the invention is to introduce a fixed point iteration mechanism and dynamically adjust the update direction and step size using historical information, so as to effectively deal with complex non-convex optimization problems, avoid falling into saddle points or oscillations, and promote the optimization process to approach the optimal solution more quickly. This dynamic adjustment mechanism not only improves the optimization efficiency, but also reduces the dependence on parameter adjustment, making the algorithm more adaptable and robust. In terms of reducing overfitting, the algorithm naturally produces an implicit regularization effect by introducing the stability characteristics of the fixed point method, which helps to suppress the overfitting of the model to the training data, thereby improving the generalization ability. In addition, the algorithm shows good stability and convergence when processing high-dimensional data, and can maintain robust performance even in a noisy gradient environment. Therefore, the Halpern type fixed point adaptive optimization algorithm provides a solution with a solid theoretical basis and significant practical effects in improving optimization efficiency and enhancing model generalization, which is very suitable for side channel analysis problems.

[0045] like Figure 1 As shown, this embodiment provides a side channel analysis method and system for Halpern type fixed point adaptive optimization, including the following steps:

[0046] S1. Obtain multiple side channel trace data, and rearrange and group them;

[0047] Collect some side channel trace data and their corresponding labels, randomly shuffle the order of the trace data while maintaining the corresponding relationship between the label and the data, and then evenly divide them into several batches to form batch-level side channel trace data and corresponding labels. Its mathematical expression is as follows:

[0048] Batch k =((sample k1 ,label k1 ),…,(sample kn ,label kn ))

[0049] sample ki =(t ki1 ,…,t kij ,…,t kim )

[0050] label ki =(y ki )

[0051] Among them, Batch k Represents the trace data and corresponding labels of the kth batch; sample kirepresents the i-th trace data of the k-th batch; t kij Indicates the jth power consumption data in the i-th trace data of the k-th batch; label ki Indicates the label corresponding to the i-th trace data in the k-th batch.

[0052] S2. Preprocess the grouped rear channel trace data to generate processed data.

[0053] The divided side channel trace data is normalized, that is, sample ki Each t kij Preprocessing is performed to convert each t kij Scale to (0,1):

[0054]

[0055] Among them, t max Indicates sample ki The maximum value in t min Indicates sample ki The minimum value in .

[0056] S3. Build a side channel analysis model, classify and identify the processed data, and obtain the probability of the predicted label value of each side channel trace data.

[0057] The side channel analysis model is constructed using a convolutional neural network, and its structure includes:

[0058] The constructed side channel analysis model includes:

[0059] First convolutional layer, first batch of normalization layers, first average pooling layer, flattening layer, first fully connected layer, second fully connected layer, third fully connected layer, softmax layer;

[0060] The first convolutional layer, the first batch of normalization layers, the first average pooling layer, the flattening layer, the first fully connected layer, the second fully connected layer, the third fully connected layer, and the softmax layer are connected in sequence.

[0061] The first convolution layer is used to input each batch of AES_HD trace data, extract the first convolution feature through one-dimensional convolution, and output it to the first batch of normalization layers. The specific parameters are in_channels = 1, out_channels = 2, kernel_size = 2, kernel_initializer = he_uniform, activation = selu, padding = same. Among them, in_channels represents the input dimension, out_channels represents the output dimension, kernel_size represents the size of the convolution kernel, kernel_initializer represents the way to initialize the neural network, activation represents the choice of activation function, and padding represents adding zero padding to the edge of the input data.

[0062] The first batch of normalization layers is used to input the first convolution features extracted by one-dimensional convolution, and after normalizing the features, obtain the normalized features of the first batch of normalization layers and output them to the first pooling layer. The specific parameters are in_channels=2, out_channels=2.

[0063] The first average pooling layer is used to downsample the input by inputting the normalized features of the first batch of normalization layers, obtain the first pooling features, and output them to the flattening layer. The specific parameters are pool_size=2, stride=2. Pool_size represents the pooling window size, and stride represents the downsampling factor.

[0064] The flattening layer is used to input the first pooled features, flatten the two-dimensional features into a one-dimensional vector, and output it to the first fully connected layer.

[0065] The first fully connected layer is used to input the flattened features of the flattened layer, extract nonlinear features through the fully connected operation, obtain the first fully connected layer features, and output them to the second fully connected layer. The specific parameters are units = 2, kernel_initializer = he_uniform, activation = selu, where units represents the number of neurons in the fully connected layer.

[0066] The second fully connected layer is used to input the features of the first fully connected layer, extract nonlinear features through the fully connected operation, obtain the features of the second fully connected layer, and output them to the third fully connected layer. The specific parameter is units = 256.

[0067] The third fully connected layer is used to input the features of the second fully connected layer, extract nonlinear features through the fully connected operation, obtain the features of the third fully connected layer, and output them to the softmax layer. The specific parameter is units = 256.

[0068] The softmax layer is used to input the features of the third fully connected layer and calculate the probability of the predicted label value of each trace data in each batch through the softmax function.

[0069] The processed data is input into the side channel analysis model for classification and identification to obtain the probability of the predicted label value for each trace data in each batch.

[0070] S4. Predict the probability of the label value based on the acquired side channel trace data, optimize the side channel analysis model, and obtain the final model.

[0071] S401. Construct a multi-classification cross entropy loss function model using the probability of the predicted label value of each trace data in each batch and the true label value of each trace data in each batch.

[0072] The constructed multi-classification cross entropy loss function model is defined as follows:

[0073]

[0074] Where C is the total number of categories, y i is the true label value, is the predicted probability output by the softmax layer in step 3.

[0075] S402. Train the loss function model through the Halpern-type fixed point adaptive optimization algorithm to obtain the final model.

[0076] Among them, the Halpern type fixed point adaptive optimization algorithm definition includes:

[0077] Algorithm: HAdam

[0078] Input: Model parameters x 0 ∈R d , the gradient of the loss function L with respect to the model parameters Algorithm first-order moment coefficient β 1 =0.9, algorithm second-order moment coefficient β 2 =0.999, algorithm learning rate α = 1e-3, algorithm Halpern recursion coefficient β 3 =0.5, constant∈=1e-8.

[0079] Output: Optimized model parameters x n+1

[0080]

[0081] S5. Use the final model to complete the side channel analysis.

[0082] A comparative experiment was conducted on the side channel analysis models optimized by different optimizers. The side channel analysis model optimized by the Halpern fixed point adaptive optimization algorithm HAdam significantly improved the recognition accuracy of the label compared with the model optimized by the existing optimization algorithms (Nadam, Adam, RMSProp, NAG).

[0083] Embodiment 2

[0084] The present embodiment also provides a Halpern-type fixed point adaptively optimized side channel analysis system, comprising: an acquisition module, a preprocessing module, a construction module, an optimization module and an analysis module; the acquisition module is used to acquire a plurality of side channel trace data, and rearrange and group them; the preprocessing module is used to preprocess the grouped side channel trace data to generate processed data; the construction module is used to establish a side channel analysis model, and use the processed data for classification and identification, so as to obtain a predicted label probability of each side channel trace data;

[0085] The optimization module is used to adjust and optimize the side channel analysis model according to the predicted label probability of the side channel trace data to generate a final model; the analysis module is used to complete the side channel analysis using the final model.

[0086] Firstly, the acquisition module is used to obtain several side channel trace data and reorganize and divide them.

[0087] A plurality of side channel trace data and corresponding labels are obtained, and the obtained side channel trace data are randomly shuffled in order while maintaining the corresponding labels, and then evenly divided into a plurality of batches of side channel trace data and corresponding labels; wherein the expressions of the plurality of batches of side channel trace data include:

[0088] Batch k =((sample k1 ,label k1 ),…,(sample kn ,label kn ))

[0089] sample ki =(t ki1 ,…,t kij ,…,t kim )

[0090] label ki =(y ki )

[0091] Among them, Batch k Represents the trace data and corresponding labels of the kth batch; sample kirepresents the i-th trace data of the k-th batch; t kij Indicates the jth power consumption data in the i-th trace data of the k-th batch; label ki Indicates the label corresponding to the i-th trace data in the k-th batch.

[0092] The processing module pre-processes the divided side channel trace data to obtain processed data.

[0093] The divided side channel trace data is normalized, that is, sample ki Each t kij Preprocessing is performed to convert each t kij Scale to (0,1):

[0094]

[0095] Among them, t max Indicates sample ki The maximum value in t min Indicates sample ki The minimum value in .

[0096] The side channel analysis model is constructed using the building module to classify and identify the processed data and obtain the probability of the predicted label value of each side channel trace data.

[0097] The constructed side channel analysis model includes:

[0098] First convolutional layer, first batch of normalization layers, first average pooling layer, flattening layer, first fully connected layer, second fully connected layer, third fully connected layer, softmax layer;

[0099] The first convolutional layer, the first batch of normalization layers, the first average pooling layer, the flattening layer, the first fully connected layer, the second fully connected layer, the third fully connected layer, and the softmax layer are connected in sequence.

[0100] The first convolution layer is used to input each batch of AES_HD trace data, extract the first convolution feature through one-dimensional convolution, and output it to the first batch of normalization layers. The specific parameters are in_channels = 1, out_channels = 2, kernel_size = 2, kernel_initializer = he_uniform, activation = selu, padding = same. Among them, in_channels represents the input dimension, out_channels represents the output dimension, kernel_size represents the size of the convolution kernel, kernel_initializer represents the way to initialize the neural network, activation represents the choice of activation function, and padding represents adding zero padding to the edge of the input data.

[0101] The first batch of normalization layers is used to input the first convolution features extracted by one-dimensional convolution, and after normalizing the features, obtain the normalized features of the first batch of normalization layers and output them to the first pooling layer. The specific parameters are in_channels=2, out_channels=2.

[0102] The first pooling layer is used to downsample the input by inputting the normalized features of the first batch of normalization layers, obtain the first pooling features, and output them to the flattening layer. The specific parameters are pool_size=2, stride=2. Pool_size represents the pooling window size, and stride represents the downsampling factor.

[0103] The flattening layer is used to input the first pooled features, flatten the two-dimensional features into a one-dimensional vector, and output it to the first fully connected layer.

[0104] The first fully connected layer is used to input the flattened features of the flattened layer, extract nonlinear features through the fully connected operation, obtain the first fully connected layer features, and output them to the second fully connected layer. The specific parameters are units = 2, kernel_initializer = he_uniform, activation = selu, where units represents the number of neurons in the fully connected layer.

[0105] The second fully connected layer is used to input the features of the first fully connected layer, extract nonlinear features through the fully connected operation, obtain the features of the second fully connected layer, and output them to the third fully connected layer. The specific parameter is units = 256.

[0106] The third fully connected layer is used to input the features of the second fully connected layer, extract nonlinear features through the fully connected operation, obtain the features of the third fully connected layer, and output them to the softmax layer. The specific parameter is units = 256.

[0107] The softmax layer is used to input the features of the third fully connected layer and calculate the probability of the predicted label value of each trace data in each batch through the softmax function.

[0108] The processed data is input into the side channel analysis model for classification and identification to obtain the probability of the predicted label value for each trace data in each batch.

[0109] The optimization module predicts the probability of the label value based on the acquired side channel trace data, optimizes the side channel analysis model, and obtains the final model.

[0110] S401. Construct a multi-classification cross entropy loss function model using the probability of the predicted label value of each trace data in each batch and the true label value of each trace data in each batch.

[0111] The constructed cross entropy loss function model is defined as follows:

[0112]

[0113] Where C is the total number of categories, y i is the true label value, is the predicted probability output by the softmax layer in step 3.

[0114] S402. Train the loss function model through the Halpern-type fixed point adaptive optimization algorithm to obtain the final model.

[0115] Among them, the Halpern type fixed point adaptive optimization algorithm definition includes:

[0116] Algorithm: HAdam

[0117] Input: Model parameters x 0 ∈R d , the gradient of the loss function L with respect to the model parameters Algorithm first-order moment coefficient β 1 =0.9, algorithm second-order moment coefficient β 2 =0.999, algorithm learning rate α = 1e-3, algorithm Halpern recursion coefficient β 3 =0.5, constant∈=1e-8.

[0118] Output: Optimized model parameters x n+1

[0119]

[0120] Finally, the analysis module uses the final model to complete the side channel analysis.

[0121] A comparative experiment was conducted on the side channel analysis models optimized by different optimizers. The side channel analysis model optimized by the Halpern fixed point adaptive optimization algorithm HAdam significantly improved the recognition accuracy of the label compared with the model optimized by the existing optimization algorithms (Nadam, Adam, RMSProp, NAG).

[0122] As a specific implementation of this embodiment, the present invention accelerates the convergence speed of the algorithm: the Halpern-type fixed point adaptive optimization algorithm can significantly accelerate the convergence speed of the optimization process by introducing a fixed point iteration mechanism and an adaptive adjustment strategy. The algorithm combines historical gradient information and dynamic step size adjustment, and can intelligently select the optimal optimization direction in each update to reduce redundant calculations in the iterative process. In addition, it is particularly effective for optimizing complex non-convex objective functions, can avoid falling into local optimality or saddle points, and quickly approach the global optimal solution. This feature not only improves training efficiency, but also reduces dependence on hyperparameter adjustments such as learning rate, thereby reducing the time cost and complexity of manual parameter adjustment.

[0123] As a specific implementation of this embodiment, the present invention enhances the stability of model training: the Halpern-type algorithm suppresses the divergence problems that may be caused by gradient oscillation and excessive update step size through the stability support provided by the fixed point theory. Even in the case of unfavorable conditions such as high noise or sparse gradients during training, the algorithm can maintain stable convergence performance. In addition, the dynamic step size adjustment mechanism enables it to adapt to the needs of different training stages, accelerate convergence in the early stage, and refine the update in the later stage, thereby further ensuring the robustness of the training process and the reliability of the optimization results.

[0124] As a specific implementation of this embodiment, the present invention improves the generalization ability of the model: the Halpern-type fixed point adaptive optimization algorithm can effectively reduce the overfitting tendency of the model to the training data through its inherent regularization characteristics. The smoothness and stability characteristics in the fixed point method reduce the risk of overfitting to a certain extent, making the performance of the model on the test data more robust. In addition, the algorithm can gradually approach a smoother and more extensive solution during the optimization process, thereby improving the generalization performance of the model.

[0125] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A side channel analysis method with Halpern-type fixed point adaptive optimization, characterized in that: The following steps are involved: Grouping and processing a number of side channel trace data to obtain processed data; Building a side channel analysis model based on a convolutional neural network, and classifying and identifying the processed data based on the side channel analysis model to obtain a predicted label value probability corresponding to each side channel trace data; Based on the predicted label value probabilities corresponding to the side channel trace data, the side channel analysis model is optimized using a Halpern-type fixed point adaptive optimization algorithm to obtain a final model; A side channel analysis result is obtained based on the final model.

2. The method according to claim 1, characterized in that: The process of obtaining processed data includes: Collecting a number of side channel trace data and label values ​​corresponding to the side channel trace data; Rearranging and grouping a plurality of side channel trace data to obtain grouped side channel trace data; The grouped rear channel trace data is preprocessed to obtain processed data.

3. The method according to claim 1, characterized in that The side channel analysis model includes: a first convolutional layer, a first batch of normalization layers, a first average pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, a third fully connected layer, and a softmax layer; The process of classifying and identifying the processed data based on the side channel analysis model includes: Inputting the processed data into the first convolution layer to obtain a first convolution feature; The first batch of standardized layers normalizes the first convolutional features to obtain normalized features of the first batch of standardized layers; The first average pooling layer downsamples the normalized features of the first batch of normalization layers to obtain first pooling features; The flattening layer flattens the first pooled features to obtain flattened features; The flattened features are sequentially passed through the first fully connected layer, the second fully connected layer, the third fully connected layer and the softmax layer to obtain the predicted label value probability corresponding to each side channel trace data.

4. The method according to claim 2, characterized in that: The process of optimizing the side channel analysis model includes: A multi-classification cross entropy loss function model is constructed based on the predicted label value probability and label value corresponding to the same batch of channel trace data on the same side; The final model is obtained by training the multi-classification cross entropy loss function model based on the Halpern fixed point adaptive optimization algorithm.

5. The method according to claim 4, characterized in that The expression of the multi-classification cross entropy loss function model is: Where L is the loss value, C is the total number of categories, and y i is the true label value, is the predicted probability output by the softmax layer.

6. The method according to claim 5, characterized in that The process of optimizing the side channel analysis model based on the Halpern type fixed point adaptive optimization algorithm includes: optimizing the parameters in the side channel analysis model based on a multi-classification cross entropy loss function model to obtain optimized model parameters; Among them, the optimized model parameters are: In the formula, x n+1 represents the optimized model parameters, β3 represents the algorithm Halpern recursive coefficient, x0 represents the initial parameters of the model, and x n represents the model parameters of the nth iteration, represents the learning rate of the model, represents the cumulative momentum of the model, and n represents the number of iterations of the algorithm.

7. A Halpern-type fixed point adaptive optimization side channel analysis system, characterized in that: A side channel analysis method for implementing the Halpern-type fixed point adaptive optimization described in any one of claims 1 to 6, wherein the side channel analysis system comprises: an acquisition module, a preprocessing module, a construction module, an optimization module and an analysis module; The acquisition module is used to rearrange and group a plurality of side channel trace data to obtain grouped side channel trace data; The preprocessing module is used to preprocess the grouped side channel trace data to generate processed data; The construction module is used to establish a side channel analysis model and classify and identify the processed data to obtain the predicted label probability of each side channel trace data; The optimization module is used to optimize the side channel analysis model based on the predicted label probability of the side channel trace data to generate a final model; The analysis module is used to obtain a side channel analysis result based on the final model.

Images