A method for recognizing block ciphers based on quantum self-organizing fuzzy neural network

Abstract

The application relates to a grouping cipher identification method based on a quantum self-organizing fuzzy neural network. The method comprises the following steps: acquiring ciphertext data to be identified and encoding the ciphertext data into an input quantum state; processing the ciphertext data to be identified and an activation intensity vector of a multi-expert system by using a biological heuristic gating network and a neural synapse excitation-inhibition mechanism to determine a net synapse input value of each quantum expert; adopting a Top-K sparsification strategy to select k quantum experts with the maximum net synapse input value from the multi-expert system to form an activated expert subset; analyzing the net synapse input value and the corresponding activation intensity of each quantum expert in the activated expert subset to determine the final gating weight of each quantum expert in the activated expert subset; inputting the input quantum state into each quantum expert in the activated expert subset to identify the grouping cipher type, and obtaining a final grouping cipher type identification result. Resource consumption is reduced.

Description

Technical Field

[0001] This application relates to the fields of information security and artificial intelligence technology, and in particular to a block cipher recognition method based on a quantum self-organizing fuzzy neural network. Background Technology

[0002] Information security is the cornerstone of the digital age and a guarantee for the stable operation of society as a whole. In recent years, with the rapid development of the Internet of Things, big data, and artificial intelligence technologies, data security and privacy protection have ushered in new opportunities and challenges. Block ciphers, as the pillar of modern cryptography, are widely deployed in various information systems, undertaking the core task of protecting data confidentiality. Therefore, in scenarios such as network attack and defense, security auditing, and protocol analysis, accurately and efficiently identifying the block cipher algorithms used in unknown encrypted traffic is a crucial step in ensuring cyberspace security and responding to advanced persistent threats (APTs).

[0003] Classical block cipher identification methods require significant investment from cryptographers and rely heavily on complex statistical analysis of the ciphertext. However, modern block cipher algorithms, such as the Advanced Encryption Standard (AES), are designed, in part, to produce ciphertext that is statistically indistinguishable from a truly random sequence. This poses a significant challenge to traditional identification methods that depend on finding statistical patterns, often resulting in low accuracy and method failure.

[0004] In recent years, intelligent recognition algorithms based on machine learning and deep learning have alleviated the reliance on expert knowledge to some extent. For example, algorithms like Random Forest (RF) and Support Vector Machine (SVM) classify data by extracting handcrafted features such as the NIST randomness test p-value from ciphertext. Deep learning models such as Convolutional Neural Networks (CNN) and Transformers can automatically learn features directly from the original ciphertext. However, while these intelligent algorithms achieve performance improvements, they also bring new and more serious problems. Existing deep learning models typically suffer from a huge number of parameters and high computational resource consumption. Furthermore, due to their complex nonlinear structure, the model's decision-making process often lacks interpretability, limiting its application in block cipher recognition scenarios requiring security auditing. Therefore, designing a recognition method that balances lightweightness, high accuracy, and interpretability is a pressing technical problem that needs to be solved.

[0005] With the advent of the quantum era, this field presents both challenges and opportunities. On the one hand, the potential power of quantum computing poses a threat to classical cryptography; on the other hand, quantum machine learning (QML), particularly parameterized quantum circuits, offers a novel approach to constructing parameter-efficient and highly expressive models due to its powerful parallel computing capabilities and exponential information representation space. Theoretically, QML has the potential to accelerate the handling of complex problems such as cryptography. However, QML is still under development, and the limitations of available quantum resources and the stability issues of quantum algorithms (such as the "barren plateau" phenomenon of vanishing gradients) restrict its application in solving practical problems. Therefore, it is necessary to consider how to leverage quantum advantages while addressing the "black box" and "high resource consumption" problems inherent in existing cryptographic identification methods. Summary of the Invention

[0006] Therefore, it is necessary to provide a block cipher recognition method based on quantum self-organizing fuzzy neural networks to address the aforementioned technical problems.

[0007] A block cipher recognition method based on a quantum self-organizing fuzzy neural network, the method comprising:

[0008] Acquire the ciphertext data to be identified and encode the ciphertext data to be identified into an input quantum state;

[0009] Using a bio-inspired gating network, the encrypted data to be identified and the activation intensity vector of a pre-constructed multi-expert system are processed using the neural synaptic excitation-inhibition mechanism to determine the net synaptic input value of each quantum expert in the multi-expert system. The activation intensity vector of the multi-expert system is composed of the activation intensity of the input quantum state and the antecedent prototype quantum state of the quantum fuzzy rule corresponding to all quantum experts in the multi-expert system.

[0010] A Top-K sparsity strategy is used to select the k quantum experts with the largest net synaptic input values ​​from the multi-expert system to form the subset of activated experts;

[0011] The final gating weight of each quantum expert in the activated expert subset is determined by analyzing the net synaptic input value and corresponding activation intensity of each quantum expert.

[0012] The input quantum state is input into each quantum expert in the subset of activated experts for block cipher type identification, and a weighted sum is performed in combination with the final gating weights to obtain the final block cipher type identification result.

[0013] The construction methods of the multi-expert system include:

[0014] Receive encrypted data samples and encode the encrypted data samples into input quantum state samples;

[0015] Using a quantum self-organizing fuzzy engine, quantum fuzzy rules are updated or created online based on the input quantum state samples, forming a rule base containing N quantum fuzzy rules;

[0016] Each quantum fuzzy rule in the rule base is instantiated as a quantum expert with a dynamically parameterized quantum circuit as its core, thus constructing a multi-expert system containing N quantum experts.

[0017] In one embodiment, the use of a quantum self-organizing fuzzy engine to update or create quantum fuzzy rules online based on the input quantum state samples, forming a rule base containing N quantum fuzzy rules, includes:

[0018] The activation strength of the input quantum state sample and the antecedent prototype quantum state of each existing quantum fuzzy rule in the rule base is calculated using a quantum self-organizing fuzzy engine.

[0019] A quantum search algorithm is used to search the input quantum state sample and the activation intensity of the antecedent prototype quantum state of each existing quantum fuzzy rule in the rule base, and to determine the index of the quantum fuzzy rule that has the maximum activation intensity with the input quantum state sample.

[0020] The value with the maximum activation strength of the input quantum state sample is compared with a preset new mode creation threshold. If the condition is met... If the condition is met, a new quantum fuzzy rule is created for the input quantum state sample and added to the rule base; Then, the quantum fuzzy rule with the maximum activation strength of the input quantum state sample is selected as the optimal matching rule, and a linear unitary operator combination method is used to coherently superimpose the antecedent prototype quantum state of the optimal matching rule with the input quantum state sample to update the antecedent prototype quantum state of the optimal matching rule. The value that has the maximum activation strength with respect to the input quantum state sample. Create a threshold for the preset new mode.

[0021] In one embodiment, the quantum self-organizing fuzzy engine calculation follows the fidelity square formula:

[0022] ;

[0023] in, The input quantum state sample and the currently existing first quantum state sample in the rule base. The activation strength of the prototype quantum state of the antecedent of a quantum fuzzy rule; The conjugate transpose of the input quantum state sample; Let be the prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base; The inner product of the input quantum state sample and the prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base is given.

[0024] In one embodiment, the core of the quantum search algorithm is a quantum comparison oracle, which represents a quantum fuzzy rule index. Input quantum ground state The expression for performing the phase transformation is:

[0025] ;

[0026] in, For quantum comparison oracles, Indexing quantum fuzzy rules The input quantum ground state, This represents the mapping relationship between the quantum comparison oracle and the result after performing a phase transformation on the input quantum ground state. When the input quantum state sample is related to the first... The activation strength of the antecedent prototype quantum state of a quantum fuzzy rule Greater than the dynamic threshold hour, ;otherwise, .

[0027] In one embodiment, the dynamically parameterized quantum circuit adopts a multi-stage sequential cascaded architecture, which is constructed by sequentially cascading three quantum circuits that respectively simulate the DNA double helix effect, the liquid dynamics effect, and the lattice phonon effect. The overall unitary transform operator of the dynamically parameterized quantum circuit... Represented as:

[0028] ;

[0029] in, For the global unitary transform operator of dynamically parameterized quantum circuits; symbol " "" represents the matrix multiplication of global unitary transformation operators, corresponding to the temporal execution of quantum gate operations; For capturing DNA double helix effector circuit operators used to capture long-range correlations and structural features; For capturing the fluid dynamics effects of the diffusion class of features; For lattice phonon effector circuit operators used to capture local and periodic features; , and The first in the multi-expert system The parameters that can be learned within the corresponding sub-circuit of a quantum expert.

[0030] In one embodiment, the step of processing the ciphertext data to be identified and the activation intensity vector of a pre-constructed multi-expert system using a bio-inspired gating network and the excitation-inhibition mechanism of neural synapses to determine the net synaptic input value of each quantum expert in the multi-expert system includes:

[0031] The encrypted data to be identified is concatenated with the activation intensity vector of a pre-constructed multi-expert system, and a presynaptic signal vector is generated through a projection layer. The presynaptic signal vector is represented as follows:

[0032] ;

[0033] in, Represents the presynaptic signal vector; This represents the encrypted data to be identified. This represents the activation intensity vector of a multi-expert system. This represents the concatenation of the ciphertext data to be identified and the activation strength vectors of the multi-expert system. It is a learnable projection matrix; This represents the learnable bias vector of the presynaptic signal vector. For the set of real numbers, For the dimensions of the gated network projection layer, The feature dimensions of the ciphertext data to be identified;

[0034] The presynaptic signal vector is decomposed into excitatory postsynaptic potentials and inhibitory postsynaptic potentials, and the expressions for the excitatory and inhibitory postsynaptic potentials are as follows:

[0035] ,

[0036] ;

[0037] in, and These represent excitatory and inhibitory postsynaptic potentials, respectively. This represents the Sigmoid activation function. and Let represent the learnable projection matrices of excitatory and inhibitory postsynaptic potentials, respectively. and These represent the learnable bias vectors for excitatory and inhibitory synaptic potentials, respectively.

[0038] The net synaptic input vector is calculated using the excitation-inhibition balance mechanism, and the expression for the net synaptic input vector is:

[0039] ;

[0040] in, The net synaptic input vector is represented by the net synaptic input values ​​of the quantum expert corresponding to each quantum fuzzy rule. It represents a learnable global excitability gain, simulating the regulation of overall excitability by neurotransmitters; This represents a learnable inhibition strength coefficient, enabling competition among quantum experts; This is a vector of synaptic plasticity parameters used to adjust the response strength of each quantum expert; The symbol represents the element-wise product of vectors.

[0041] In one embodiment, the step of analyzing the net synaptic input value and corresponding activation intensity of each quantum expert in the activated expert subset to determine the final gating weight of each quantum expert in the activated expert subset includes:

[0042] The net synaptic input value of each quantum expert in the activated expert subset is normalized using the Softmax function to determine the gating weight of each quantum expert in the activated expert subset.

[0043] The final gating weight of each quantum expert in the activated expert subset is determined based on the gating weights and corresponding activation strengths of all quantum experts in the activated expert subset.

[0044] In one embodiment, the activation expert subset is the first The expression for the gating weights of a quantum expert is:

[0045] ;

[0046] in, Indicates the first in the activated expert subset The gating weights of a quantum expert, For the activated expert subset, the first Net synaptic input value of a quantum expert For the activated expert subset, the first The net synaptic input value of a quantum expert.

[0047] In one embodiment, the activation expert subset is the first The expression for the final gating weights of a quantum expert is:

[0048] ;

[0049] in, For the activated expert subset, the first The gating weights of a quantum expert, For the activated expert subset, the first The gating weights of a quantum expert, For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. A preset temperature parameter is used to control the influence of gating weights and activation intensity.

[0050] In one embodiment, the step of inputting the input quantum state to each quantum expert in the activated expert subset for block cipher type identification to obtain the final block cipher type identification result includes:

[0051] The input quantum state is input to each quantum expert in the activated expert subset. The dynamically parameterized quantum circuit corresponding to each quantum expert acts on the input quantum state, and the block cipher type identification result of each quantum expert is obtained through quantum measurement. The block cipher type identification result of the p-th quantum expert in the activated expert subset is then used. The expression is:

[0052] ;

[0053] in, This indicates a quantum state measurement operation. This represents the dynamically parameterized quantum circuit unitary operator of the p-th quantum expert in the activated expert subset. For the activated expert subset, the first Internally learnable parameters of a quantum expert Indicates the input quantum state;

[0054] The block cipher type identification results of each quantum expert are weighted and summed according to their corresponding final gating weights to obtain the final block cipher type identification result. The expression for the final block cipher type identification result is as follows:

[0055] ;

[0056] in, This indicates the final block cipher type identification result; This represents the final gating weight of the p-th quantum expert in the activated expert subset.

[0057] Compared with the prior art, the beneficial effects achieved by this application are as follows:

[0058] (1) Achieving a balance between performance and resource efficiency. This application utilizes the powerful parallel processing and representation capabilities of bio-inspired dynamic quantum circuits, combined with the core working mode of a quantum self-organizing fuzzy engine, to achieve recognition accuracy comparable to or even higher than that of traditional deep learning models with a large number of parameters (hundreds of thousands to millions) at an extremely low parameter scale (approximately 14,000). This design effectively solves the problem of the strong binding between high performance and high resource consumption in existing technologies, making it possible to deploy high-performance recognition models on resource-constrained devices.

[0059] (2) This application uses a sparse gating network that simulates the excitation-inhibition mechanism of neural synapses to intelligently decompose complex classification tasks into a group of highly specialized "quantum experts" for collaborative completion. For any recognition task, the decision path, i.e., which quantum experts are activated and their respective contributions, is clear, sparse, and traceable. This makes the model formed by activating a subset of experts highly transparent and auditable in the decision-making process, fundamentally overcoming the credibility problem caused by the "black box" nature of traditional deep learning models.

[0060] (3) The core knowledge system of this application, namely the quantum fuzzy rule base, is automatically generated and evolved through a data-driven approach, completely eliminating the reliance on pre-set expert knowledge or manual rules. This application can automatically discover and extract inherent patterns from data, constructing a compact and efficient knowledge structure. This self-organizing characteristic endows this application with excellent environmental adaptability and the ability to quickly discover new patterns, making it particularly suitable for dealing with dynamic scenarios where encryption algorithms are constantly evolving. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating a block cipher recognition method based on a quantum self-organizing fuzzy neural network in one embodiment;

[0062] Figure 2 This is a schematic diagram of the overall framework based on a bio-inspired neural network in one embodiment;

[0063] Figure 3 This is a schematic diagram of the process of using Grover's algorithm to search for quantum extrema in one embodiment;

[0064] Figure 4 This is a schematic diagram of the structure of a linear unitary operator (LCU) combination method in one embodiment;

[0065] Figure 5 This is a schematic diagram of the structure of a dynamically parameterized quantum circuit in one embodiment;

[0066] Figure 6 This is a schematic diagram of the structure of a bio-inspired gating network in one embodiment. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0068] In one embodiment, such as Figure 1 As shown, a block cipher recognition method based on a quantum self-organizing fuzzy neural network is provided, which is applied to a computing device for algorithmic classification of encrypted data. The method includes the following steps:

[0069] Step S220: Obtain the ciphertext data to be identified and encode the ciphertext data to be identified into an input quantum state.

[0070] The ciphertext data to be identified can be a raw byte stream. The feature extraction module converts this byte stream into a... A feature vector S is defined. Feature extraction methods may include, but are not limited to, calculating byte frequency distributions and the P-value of the NIST randomness test. Subsequently, the feature vector is encoded into a multi-qubit input quantum state using a quantum encoding module. In order to perform subsequent quantum computing, the encoding process follows the formula:

[0071] ;

[0072] in, The input quantum state is used to encode the n qubits. For feature vectors The Euclidean norm; It is the i-th component of the eigenvector; for Computational basis vectors for a qubit system; To meet The smallest integer.

[0073] Step S240: Using a bio-inspired gating network, the encrypted data to be identified and the activation intensity vector of the pre-constructed multi-expert system are processed by the neural synaptic excitation-inhibition mechanism to determine the net synaptic input value of each quantum expert in the multi-expert system. The activation intensity vector of the multi-expert system is composed of the activation intensity of the input quantum state and the antecedent prototype quantum state of the quantum fuzzy rule corresponding to all quantum experts in the multi-expert system.

[0074] The construction methods for multi-expert systems include:

[0075] Receive encrypted data samples and encode them into input quantum state samples; use a quantum self-organizing fuzzy engine to update or create quantum fuzzy rules online based on the input quantum state samples, forming a rule base containing N quantum fuzzy rules; instantiate each quantum fuzzy rule in the rule base into a quantum expert with a dynamically parameterized quantum circuit as its core, and construct a multi-expert system containing N quantum experts.

[0076] Among them, the construction of the multi-expert system can be carried out in real time in an online mode, that is, the multi-expert system is updated in real time according to the new ciphertext data samples received, thereby enabling it to have better block cipher recognition capabilities.

[0077] The encoding method for encoding ciphertext data samples into input quantum state samples is the same as the encoding method for encoding ciphertext data to be identified into input quantum states, and will not be described again.

[0078] Step S260: Using a Top-K sparsity strategy, select the k quantum experts with the largest net synaptic input values ​​from the multi-expert system to form an activation expert subset.

[0079] Step S280: Analyze the net synaptic input value and corresponding activation intensity of each quantum expert in the activated expert subset to determine the final gating weight of each quantum expert in the activated expert subset.

[0080] Step S300: For each quantum expert in the subset of input quantum state activation experts, perform block cipher type identification, and combine the final gating weights to perform weighted summation to obtain the final block cipher type identification result.

[0081] The aforementioned block cipher recognition method based on quantum self-organizing fuzzy neural networks utilizes a quantum fuzzy rule base that is automatically generated and evolved through a data-driven approach. This completely eliminates reliance on pre-defined expert knowledge or manual rules, enabling the automatic discovery and extraction of inherent patterns from data, and the selection of compact and efficient knowledge structures (i.e., activated expert subsets). This self-organizing characteristic endows the block cipher recognition method based on quantum self-organizing fuzzy neural networks with excellent environmental adaptability and rapid discovery of new patterns, making it particularly suitable for dynamic scenarios involving the continuous evolution of encryption algorithms. Furthermore, a sparse gating network simulating the excitation-inhibition mechanism of neural synapses intelligently decomposes complex classification tasks into a collaborative process involving a group of highly specialized "quantum experts." For any given recognition task, the decision path—which quantum experts are activated and their respective contributions—is clear, sparse, and traceable. This makes the model formed by activating expert subsets highly transparent and auditable in the decision-making process, fundamentally overcoming the credibility issues caused by the "black box" nature of traditional deep learning models. Furthermore, leveraging the powerful parallel processing and representation capabilities of bio-inspired dynamic quantum circuits, and combining them with the core operating mode of a quantum self-organizing fuzzy engine, it can achieve recognition accuracy comparable to or even higher than traditional deep learning models with massive parameter counts (hundreds of thousands to millions) with an extremely low parameter scale (approximately 14,000). This design effectively solves the problem of the strong binding between high performance and high resource consumption in existing technologies, making it possible to deploy high-performance recognition models on resource-constrained devices.

[0082] In one embodiment, a quantum self-organizing fuzzy engine is used to update or create quantum fuzzy rules online based on input quantum state samples, forming a rule base containing N quantum fuzzy rules, including:

[0083] The activation strength of the input quantum state sample and the antecedent prototype quantum state of each existing quantum fuzzy rule in the rule base is calculated using a quantum self-organizing fuzzy engine.

[0084] A quantum search algorithm is used to search for the activation intensity of the prototype quantum state of the input quantum state sample and the antecedent of each existing quantum fuzzy rule in the rule base, and to determine the index of the quantum fuzzy rule that has the maximum activation intensity with the input quantum state sample.

[0085] The value with the maximum activation strength of the input quantum state sample is compared with a preset new mode creation threshold. If the condition is met... If the condition is met, a new quantum fuzzy rule is created for the input quantum state sample and added to the rule base; Then, the quantum fuzzy rule with the maximum activation strength of the input quantum state sample is selected as the optimal matching rule. A linear combination of unitary operators (LCU) method is used to coherently superimpose the antecedent prototype quantum state of the optimal matching rule with the input quantum state sample to update the antecedent prototype quantum state of the optimal matching rule. The value that has the maximum activation strength with respect to the input quantum state sample. Create a threshold for the preset new mode.

[0086] In one embodiment, the quantum self-organizing fuzzy engine computation follows the fidelity square formula:

[0087] ;

[0088] in, For input quantum state samples and the currently existing first quantum state samples in the rule base The activation strength of the prototype quantum state of the antecedent of a quantum fuzzy rule; For the conjugate transpose of the input quantum state sample; The prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base; The inner product of the input quantum state sample and the prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base.

[0089] In one embodiment, the core of the quantum search algorithm is a quantum comparison oracle, which represents a quantum fuzzy rule index. Input quantum ground state The expression for performing the phase transformation is:

[0090] ;

[0091] in, For quantum comparison oracles, Indexing quantum fuzzy rules The input quantum ground state, This represents the mapping relationship between the quantum comparison oracle and the result after performing a phase transformation on the input quantum ground state. When the input quantum state sample is the first The activation strength of the antecedent prototype quantum state of a quantum fuzzy rule Greater than the dynamic threshold hour, ;otherwise, .

[0092] In one embodiment, the multi-stage sequential cascaded architecture of dynamically parameterized quantum circuits is constructed by sequentially cascading three quantum circuits that respectively simulate the DNA double helix effect, the liquid dynamics effect, and the lattice phonon effect. The overall unitary transform operator of the dynamically parameterized quantum circuit... Represented as:

[0093] ;

[0094] in, For the global unitary transform operator of dynamically parameterized quantum circuits; symbol " "" represents the matrix multiplication of global unitary transformation operators, corresponding to the temporal execution of quantum gate operations; A DNA double helix effect sub-circuit operator for capturing long-range correlations and structured features, comprising a rotation gate simulating base rotation and a cross-qubit controlled gate simulating double-strand entanglement; A liquid dynamics effect sub-circuit operator for capturing confusion-diffusion type features, which includes rotation gates and controlled phase gates between adjacent qubits to construct local connections; For the lattice phonon effect sub-circuit operator used to capture local and periodic features, it constructs topologically constrained nearest neighbor interactions by applying regularly arranged single-qubit rotation gates to qubits and utilizing entanglement gates that connect only spatially adjacent qubits. , , The first in a multi-expert system The parameters that can be learned within the corresponding sub-circuit of a quantum expert.

[0095] In one embodiment, a bio-inspired gating network is used to process the ciphertext data to be identified and the activation intensity vector of a pre-constructed multi-expert system using a neural synaptic excitation-inhibition mechanism to determine the net synaptic input value of each quantum expert in the multi-expert system, including:

[0096] The encrypted data to be identified is concatenated with the activation intensity vector of a pre-built multi-expert system, and a presynaptic signal vector is generated through a projection layer. The presynaptic signal vector is represented as follows:

[0097] ;

[0098] in, Represents the presynaptic signal vector; This represents the encrypted data to be identified. This represents the activation intensity vector of a multi-expert system. This represents the concatenation of the ciphertext data to be identified and the activation strength vectors of the multi-expert system. It is a learnable projection matrix; This represents the learnable bias vector of the presynaptic signal vector. For the set of real numbers, For the dimensions of the gated network projection layer, The feature dimensions of the ciphertext data to be identified;

[0099] The presynaptic signal vector is decomposed into excitatory postsynaptic potentials and inhibitory postsynaptic potentials. The expressions for excitatory and inhibitory postsynaptic potentials are as follows:

[0100] ,

[0101] ;

[0102] in, and These represent excitatory and inhibitory postsynaptic potentials, respectively. This represents the Sigmoid activation function. and Let represent the learnable projection matrices of excitatory and inhibitory postsynaptic potentials, respectively. and These represent the learnable bias vectors for excitatory and inhibitory synaptic potentials, respectively.

[0103] The net synaptic input vector is calculated using the excitation-inhibition balance mechanism. The expression for the net synaptic input vector is:

[0104] ;

[0105] in, This represents the net synaptic input vector, which is composed of the net synaptic input values ​​of the quantum expert corresponding to each quantum fuzzy rule; It represents a learnable global excitability gain, simulating the regulation of overall excitability by neurotransmitters; This represents a learnable inhibition strength coefficient, enabling competition among quantum experts; This is a vector of synaptic plasticity parameters used to adjust the response strength of each quantum expert; The symbol represents the element-wise product of vectors.

[0106] In one embodiment, the final gating weight of each quantum expert in the activated expert subset is determined by analyzing the net synaptic input value and corresponding activation intensity of each quantum expert, including:

[0107] The net synaptic input value of each quantum expert in the activated expert subset is normalized using the Softmax function to determine the gating weight of each quantum expert in the activated expert subset.

[0108] The final gating weight of each quantum expert in the activated expert subset is determined based on the gating weights and corresponding activation strengths of all quantum experts in the activated expert subset.

[0109] In one embodiment, the first expert subset is activated. The expression for the gating weights of a quantum expert is:

[0110] ;

[0111] in, Indicates the activation of the expert subset. The gating weights of a quantum expert, To activate the first expert subset Net synaptic input value of a quantum expert To activate the first expert subset The net synaptic input value of a quantum expert.

[0112] In one embodiment, the first expert subset is activated. The expression for the final gating weights of a quantum expert is:

[0113] ;

[0114] in, To activate the first expert subset The gating weights of a quantum expert, To activate the first expert subset The gating weights of a quantum expert, For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. A preset temperature parameter is used to control the influence of gating weights and activation intensity.

[0115] In one embodiment, each quantum expert in the subset of input quantum state activation experts is subjected to block cipher type identification to obtain the final block cipher type identification result, including:

[0116] The input quantum state is applied to each quantum expert in the activation expert subset. The dynamically parameterized quantum circuit corresponding to each quantum expert acts on the input quantum state, and the block cipher type identification result for each quantum expert is obtained through quantum state measurement. Specifically, the block cipher type identification result for the p-th quantum expert in the activation expert subset is... The expression is:

[0117] ;

[0118] in, This indicates a quantum state measurement operation. This represents the dynamically parameterized quantum circuit unitary operator of the p-th quantum expert in the activated expert subset. To activate the first expert subset Internally learnable parameters of a quantum expert Indicates the input quantum state;

[0119] The block cipher type identification results of each quantum expert are weighted and summed according to their corresponding final gating weights to obtain the final block cipher type identification result. The expression for the final block cipher type identification result is as follows:

[0120] ;

[0121] in, This indicates the final block cipher type identification result; This represents the final gating weight of the p-th quantum expert in the activated expert subset.

[0122] Compared with the prior art, the beneficial effects achieved by this application are as follows:

[0123] (1) Achieving a balance between performance and resource efficiency. This application utilizes the powerful parallel processing and representation capabilities of bio-inspired dynamic quantum circuits, combined with the core working mode of a quantum self-organizing fuzzy engine, to achieve recognition accuracy comparable to or even higher than that of traditional deep learning models with a large number of parameters (hundreds of thousands to millions) at an extremely low parameter scale (approximately 14,000). This design effectively solves the problem of the strong binding between high performance and high resource consumption in existing technologies, making it possible to deploy high-performance recognition models on resource-constrained devices.

[0124] (2) This application uses a gating network that simulates the excitation-inhibition mechanism of neural synapses and combines it with a Top-K sparsity strategy to intelligently decompose complex classification tasks into a group of highly specialized "quantum experts" for collaborative completion. For any recognition task, the decision path, i.e., which quantum experts are activated and their respective contributions, is clear, sparse, and traceable. This makes the model (i.e., the multi-expert system) highly transparent and auditable in the decision-making process, fundamentally overcoming the credibility problem caused by the "black box" nature of traditional deep learning models.

[0125] (3) The core knowledge system of this application, namely the quantum fuzzy rule base, is automatically generated and evolved online through a data-driven approach, completely eliminating the reliance on pre-set expert knowledge or manual rules. This application can automatically discover and extract inherent patterns from data, constructing a compact and efficient knowledge structure. This self-organizing characteristic endows this application with excellent environmental adaptability and the ability to quickly discover new patterns, making it particularly suitable for dealing with dynamic scenarios where encryption algorithms are constantly evolving.

[0126] In one embodiment, such as Figure 2 As shown, a block cipher recognition method based on a quantum self-organizing fuzzy neural network is provided, and the specific steps are as follows:

[0127] Step 1: Input ciphertext data sample b (i.e., the original ciphertext) and perform feature extraction to obtain feature vector samples b1, b2, ... b d An amplitude encoding method is used to map each component of the feature vector sample to the amplitude coefficient of the quantum state, thereby encoding it into the input quantum state sample. .

[0128] Step 2: Construction of an online self-organizing quantum fuzzy rule base.

[0129] Step 2 is performed by a quantum self-organizing fuzzy engine. This engine maintains a dynamic rule base consisting of N quantum fuzzy rules, which are applied to each input quantum state sample. Perform the following operations:

[0130] 2.1 Calculate the activation strength of the input quantum state sample and the antecedent prototype quantum state of each quantum fuzzy rule in the current rule base. The quantum self-organizing fuzzy engine calculation follows the squared fidelity formula:

[0131] ;

[0132] in, For input quantum state samples and the currently existing first quantum state samples in the rule base The activation strength of the prototype quantum state of the antecedent of a quantum fuzzy rule; For the conjugate transpose of the input quantum state sample; The prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base; The inner product of the input quantum state sample and the prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base.

[0133] 2.2. Reference Figure 3 A quantum search algorithm is used to determine the activation strength. Quantum fuzzy rule index The core of the quantum search algorithm is a quantum comparison oracle, which represents a quantum fuzzy rule index. Input quantum ground state Perform the phase transformation as defined by the following equation:

[0134] ;

[0135] in, It is a quantum comparison oracle used to apply phase reversal to a quantum ground state that meets specific conditions; Indexing quantum fuzzy rules The input quantum ground state, j represents the first quantum state in the rule base. Index of quantum fuzzy rules; Defined as when the input quantum state sample is related to the first quantum state sample... The activation strength of the antecedent prototype quantum state of a quantum fuzzy rule Greater than the dynamic threshold hour, ;otherwise, The dynamic threshold y is a scalar value that increases with the number of iterations of the quantum search algorithm, used to gradually approach the maximum activation intensity during the search process.

[0136] As shown in Figure 3, the specific circuit structure of this quantum search algorithm is based on the Grover algorithm architecture. The circuit first utilizes the Hadamard gate of the input register (i.e., Figure 3 The H in the diagram initializes the initial state to a uniform superposition state. The auxiliary bits are passed through the Pauli-X gate (i.e., ...). Figure 3 The initial state is initialized to a uniform superposition state by flipping the Hadamard gate (X) of the quantum comparison oracle, which encodes all activation intensities into the input register. The quantum comparison oracle then performs a phase flip (i.e., marking) on ​​the quantum fuzzy rule index that satisfies the activation intensity condition (i.e., maximum activation intensity) based on the state of the auxiliary bits. Subsequently, the amplitude amplification process of the quantum fuzzy rule index satisfying the activation intensity condition is performed using the Grover diffusion operator. By flipping the value around the average, the probability amplitude of the target rule index is significantly increased. This process (quantum comparison oracle and diffusion operator) needs to be executed iteratively. Next, among them The total number of quantum fuzzy rules, This is the upper limit of the number of iterations required to ensure that the quantum fuzzy rule index with the maximum activation strength can be obtained with a high probability close to 1 through the measurement operation.

[0137] 2.3. Reference Figure 4 , to maximize activation intensity And the preset new mode create threshold If a comparison is made, If the input quantum state sample represents a new mode not covered by existing quantum fuzzy rules, a new quantum fuzzy rule is created for this new mode and added to the rule base; if Then, according to the quantum fuzzy rule index with the maximum activation strength... Select the corresponding best matching rule from the rule base, and use the linear unitary operator (LCU) combination method to coherently superimpose the current antecedent prototype quantum state of the best matching rule with the current input quantum state sample to update the antecedent prototype quantum state of the best matching rule.

[0138] like Figure 4 As shown, the specific quantum circuit structure and working principle of the linear unitary operator (LCU) combination method are as follows:

[0139] The circuit contains three quantum registers: a regular register ( Figure 4 The Chinese character is represented as ), input register ( Figure 4 The Chinese character is represented as ), working register ( Figure 4 The Chinese character is represented as The circuit first applies an RY rotation gate to the top auxiliary bit, the rotation angle of which is controlled by the update weight parameters in the quantum circuit, to prepare a superposition state on the auxiliary bit. Subsequently, the circuit contains a series of controlled swap gates (…). Figure 4 (Represented by lines and crosses). These controlled switching gates are controlled by the state of the auxiliary bits: when the auxiliary bits are in When the state is in the solid dot state, a swap operation is triggered, mixing the quantum state information in the input register into the regular register; when the auxiliary bit is in the solid dot state... During the state transition (hollow dot), the rule register state remains unchanged. Through this controlled coherent superposition operation (coherent superposition refers to superimposing the input quantum state sample with the preceding prototype quantum state while maintaining the coherence of the qubits, unlike classical neural networks which only fuse information through numerical weighted summation), it is possible to dynamically fuse the information of the current input quantum state sample into the preceding prototype quantum state of the optimal matching rule while maintaining quantum coherence, thereby achieving online updates of quantum fuzzy rules.

[0140] Step 3: Instantiate the quantum expert system.

[0141] Each quantum fuzzy rule in the rule base from step 2 Instantiate as a parallel quantum expert The core reasoning component of every quantum expert is a dynamically parameterized quantum circuit.

[0142] Reference Figure 5 The dynamically parameterized quantum circuit adopts a multi-stage sequential cascaded architecture, which is constructed by cascading three quantum circuits that respectively simulate the DNA double helix effect, the liquid dynamics effect, and the lattice phonon effect. Its overall unitary transform operator Represented as:

[0143] ;

[0144] in, For the global unitary transform operator of dynamically parameterized quantum circuits; symbol " "" represents the matrix multiplication of global unitary transformation operators, corresponding to the temporal execution of quantum gate operations; For capturing DNA double helix effector circuit operators used to capture long-range correlations and structural features; For capturing the fluid dynamics effects of the diffusion class of features; For lattice phonon effector circuit operators used to capture local and periodic features; , , The first in a multi-expert system The parameters that can be learned within the corresponding sub-circuit of a quantum expert.

[0145] As shown in Figure 5, the specific quantum gate structure design of the three quantum circuits is as follows:

[0146] DNA double helix effector circuits: such as Figure 5 As shown in (a), the simulated DNA double helix structure is used to capture long-range correlations and structural features. The DNA double helix effect sub-circuit includes rotation gates simulating base rotation (i.e., paired RX and RY rotation gates) to mimic the rotational properties of base pairs; and controlled gates spanning multiple qubits simulating double-strand entanglement (i.e., long-range controlled RZ gates, in...). Figure 5 (In (a), vertical lines are used to connect and RZ gates are used to simulate the entanglement and base pairing between DNA double strands across a distance, thereby capturing the long-range correlation of the input data.

[0147] Fluid dynamics effect sub-circuit: such as Figure 5 As shown in (b), the vortex and diffusion behavior of a simulated liquid is used to capture the confusion-diffusion characteristics of ciphertext. The liquid dynamics effect circuit utilizes RZ rotation gates and controlled phase gates between adjacent qubits to construct a dense local connection mode. This structure allows quantum information to flow and diffuse rapidly between adjacent bits like liquid molecules, thus effectively capturing the confusion and diffusion characteristics in cryptographic algorithms.

[0148] Lattice phonon effector circuits: such as Figure 5As shown in (c), the periodic structure of the simulated lattice and local perturbations are used to capture local and periodic features. First, "phonon vibrations" are simulated by applying regularly arranged single-qubit RX rotation gates to the qubits, exciting the local states of each lattice point in the lattice; subsequently, "nearest neighbor interactions" are simulated through two-qubit entanglement gates (CNOT). Figure 5 (c) The representation is constructed and follows strict topological constraints, i.e. Figure 5 Each entanglement gate in (c) connects only two spatially adjacent qubit lines (e.g., the i-th qubit interacts only with the (i+1)-th qubit). This topological constraint rigorously simulates the short-range interaction physical constraint in a solid lattice where atoms are only affected by the forces of directly adjacent atoms, forcing the model to focus on the tiny correlations between adjacent data bits; ultimately, a "central defect" structure (i.e., ...) is included at the very end of the line. Figure 5 (c) The local asymmetric gate operation shown on the right) This design, which introduces lattice defects to disrupt the perfect symmetry of the system, can significantly improve the sensitivity of the model (i.e., the multi-expert system) to local statistical anomalies and weak biases in block ciphers, thereby achieving high-precision classification and recognition in conjunction with the periodic structure of the front end.

[0149] Step 4: Bio-inspired expert scheduling and selection. (Refer to...) Figure 6 The overall structure of a bio-inspired gating network includes an input layer, a presynaptic signal vector processing layer, an excitation-inhibition balance layer, and a Top-K sparse selection layer. Figure 6 The "Quantum Encoding" module at the bottom center transforms the input feature vector into the input quantum state; Figure 6 The central box area illustrates the core computational process of the gating mechanism, in which... The symbol represents a linear combination of excitatory and inhibitory potentials. The symbol represents the element-wise multiplication operation with the synaptic plasticity parameter; Figure 6 The bar chart in the image visually illustrates the Top-K sparsification strategy, which retains only the net synaptic input (…). The top k quantum experts (highlighted bars in the diagram) are selected, while the remaining quantum experts are suppressed. Finally, the outputs of the selected quantum experts are fused with the calculated gating weights in the "weighted summation" module to generate the final prediction result. .

[0150] 4.1 Combine the ciphertext data to be identified with the activation strength vector of the multi-expert system The data is spliced ​​together, and a presynaptic signal vector is generated through the projection layer. , represented as:

[0151] ;

[0152] in, Represents the presynaptic signal vector; This represents the encrypted data to be identified. This represents the activation intensity vector of a multi-expert system. This represents the concatenation of the ciphertext data to be identified and the activation strength vectors of the multi-expert system. It is a learnable projection matrix; This represents the learnable bias vector of the presynaptic signal vector. For the set of real numbers, For the dimensions of the gated network projection layer, For the feature dimensions of the ciphertext data to be identified, The encrypted data to be identified and the existing rules in the rule base. The activation strength of the prototype quantum state of the antecedent of a quantum fuzzy rule.

[0153] 4.2 Transform the presynaptic signal vector Decomposed into excitatory postsynaptic potentials and inhibitory postsynaptic potentials Its calculation formula is defined as:

[0154] ,

[0155] ;

[0156] in, and These represent excitatory and inhibitory postsynaptic potentials, respectively. This represents the sigmoid activation function. and Let represent the learnable projection matrices of excitatory and inhibitory postsynaptic potentials, respectively. and represent the learnable bias vectors for excitatory and inhibitory synaptic potentials, respectively.

[0157] 4.3 Calculating the net synaptic input vector through an excitation-inhibition balance mechanism The calculation formula is as follows:

[0158] ;

[0159] in, Represents the net synaptic input vector; This indicates a learnable global excitability gain, simulating the regulation of overall excitability by neurotransmitters; This represents a learnable inhibition strength coefficient, enabling competition among quantum experts; This is a vector of synaptic plasticity parameters; The symbol represents the element-wise product of vectors.

[0160] 4.4 The net synaptic input vector calculated in step 4.3 Based on this, a Top-K sparsification strategy is adopted, namely, the net synaptic input vector The k quantum experts with the largest net synaptic input values ​​constitute the activation expert subset. These k net synaptic input values ​​are then normalized using the Softmax function to obtain the gating weight for each quantum expert in the activation expert subset. The calculation formula is shown below:

[0161] ;

[0162] in, Indicates the activation of the expert subset. The gating weights of a quantum expert, To activate the first expert subset Net synaptic input value of a quantum expert To activate the first expert subset The net synaptic input value of a quantum expert.

[0163] Step 5: Weighted Fusion Decision. This step processes the subset of activated experts selected in Step 4 to obtain the final block cipher type identification result.

[0164] 5.1 For each quantum expert in the selected subset of activated experts in step 4, the dynamically parameterized quantum circuit corresponding to each quantum expert acts on the input quantum state, and the block cipher type identification result of each quantum expert is obtained through quantum measurement. The block cipher type identification result of the p-th quantum expert in the subset of activated experts is... The expression is:

[0165] ;

[0166] in, This indicates a quantum state measurement operation. This represents the dynamically parameterized quantum circuit unitary operator of the p-th quantum expert in the activated expert subset. To activate the first expert subset Internally learnable parameters of a quantum expert This represents the input quantum state.

[0167] 5.2. Reference Figure 6 The gated network structure shown employs a dual-weighting mechanism to calculate the final prediction result. , No. The final fusion weight of each activation expert It is jointly determined by the gating weights, which quantify the relevance of quantum expert tasks, and the activation strength, which quantifies the similarity of data patterns, through collaborative fusion. The calculation formula is as follows:

[0168] ;

[0169] in, To activate the first expert subset The gating weights of a quantum expert, To activate the first expert subset The gating weights of a quantum expert, For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. A preset temperature parameter is used to control the influence of gating weights and activation intensity.

[0170] Step 5.3, Final Block Cipher Type Identification Result By analyzing the outputs of all activation experts (i.e., quantum experts within a subset of activation experts) Based on its final fusion weight The weighted summation is obtained using the following formula:

[0171] ;

[0172] in, This indicates the final block cipher type identification result; This represents the final gating weight of the p-th quantum expert in the activated expert subset.

[0173] This application provides a novel paradigm for block cipher recognition by combining a quantum self-organizing fuzzy engine with a bio-inspired multi-expert system. After model training (i.e., the construction of the multi-expert system), the resulting model (i.e., the multi-expert system) can accurately classify unknown encrypted data with low parameter count, high precision, and high interpretability. This method constructs a knowledge base through online self-organization, avoiding reliance on expert knowledge; ensures the transparency and traceability of the decision-making process through a sparsely activated expert network; and enhances the model's expressive power through bio-inspired dynamic quantum circuits. Therefore, the technical solution provided in this application effectively solves the problems of "black box" issues, parameter redundancy, and reliance on prior knowledge in existing technologies, providing a solid technical foundation for building next-generation intelligent, trustworthy, and efficient security analysis systems.

[0174] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0175] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0176] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A block cipher recognition method based on a quantum self-organizing fuzzy neural network, characterized in that, The block cipher recognition method based on quantum self-organizing fuzzy neural network includes: Acquire the ciphertext data to be identified and encode the ciphertext data to be identified into an input quantum state; Using a bio-inspired gating network, the encrypted data to be identified and the activation intensity vector of a pre-constructed multi-expert system are processed using the neural synaptic excitation-inhibition mechanism to determine the net synaptic input value of each quantum expert in the multi-expert system. The activation intensity vector of the multi-expert system is composed of the activation intensity of the input quantum state and the antecedent prototype quantum state of the quantum fuzzy rule corresponding to all quantum experts in the multi-expert system. A Top-K sparsity strategy is used to select the k quantum experts with the largest net synaptic input values ​​from the multi-expert system to form an active expert subset; The final gating weight of each quantum expert in the activated expert subset is determined by analyzing the net synaptic input value and corresponding activation intensity of each quantum expert. The input quantum state is input into each quantum expert in the subset of activated experts for block cipher type identification, and a weighted sum is performed in combination with the final gating weights to obtain the final block cipher type identification result. The construction methods of the multi-expert system include: Receive encrypted data samples and encode the encrypted data samples into input quantum state samples; Using a quantum self-organizing fuzzy engine, quantum fuzzy rules are updated or created online based on the input quantum state samples, forming a rule base containing N quantum fuzzy rules; Each quantum fuzzy rule in the rule base is instantiated as a quantum expert with a dynamically parameterized quantum circuit as its core, thus constructing a multi-expert system containing N quantum experts.

2. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 1, characterized in that, The process utilizes a quantum self-organizing fuzzy engine to update or create quantum fuzzy rules online based on the input quantum state samples, forming a rule base containing N quantum fuzzy rules, including: The activation strength of the input quantum state sample and the antecedent prototype quantum state of each existing quantum fuzzy rule in the rule base is calculated using a quantum self-organizing fuzzy engine. A quantum search algorithm is used to search the input quantum state sample and the activation intensity of the antecedent prototype quantum state of each existing quantum fuzzy rule in the rule base, and to determine the index of the quantum fuzzy rule that has the maximum activation intensity with the input quantum state sample. The value with the maximum activation strength of the input quantum state sample is compared with a preset new mode creation threshold. If the condition is met... If the condition is met, a new quantum fuzzy rule is created for the input quantum state sample and added to the rule base; Then, the quantum fuzzy rule with the maximum activation strength of the input quantum state sample is selected as the optimal matching rule, and a linear unitary operator combination method is used to coherently superimpose the antecedent prototype quantum state of the optimal matching rule with the input quantum state sample to update the antecedent prototype quantum state of the optimal matching rule. The value that has the maximum activation strength with respect to the input quantum state sample. Create a threshold for the preset new mode.

3. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 2, characterized in that, The quantum self-organizing fuzzy engine calculation follows the fidelity square formula: ； in, The input quantum state sample and the currently existing first quantum state sample in the rule base. The activation strength of the prototype quantum state of the antecedent of a quantum fuzzy rule; The conjugate transpose of the input quantum state sample; Let be the prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base; The inner product of the input quantum state sample and the prototype quantum state of the j-th quantum fuzzy rule currently existing in the rule base is given.

4. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 3, characterized in that, The core of the quantum search algorithm is a quantum comparison oracle, which represents a quantum fuzzy rule index. Input quantum ground state The expression for performing the phase transformation is: ； in, For quantum comparison oracles, Indexing quantum fuzzy rules The input quantum ground state, This represents the mapping relationship between the quantum comparison oracle and the result after performing a phase transformation on the input quantum ground state. When the input quantum state sample is related to the first... The activation strength of the antecedent prototype quantum state of a quantum fuzzy rule Greater than the dynamic threshold hour, ;otherwise, .

5. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 1, characterized in that, The dynamically parameterized quantum circuit adopts a multi-stage sequential cascaded architecture, which is constructed by sequentially cascading three quantum circuits that respectively simulate the DNA double helix effect, the liquid dynamics effect, and the lattice phonon effect. The global unitary transform operator of the dynamically parameterized quantum circuit... Represented as: ； in, For the global unitary transform operator of dynamically parameterized quantum circuits; symbol " "" represents the matrix multiplication of global unitary transformation operators, corresponding to the temporal execution of quantum gate operations; For capturing DNA double helix effector circuit operators used to capture long-range correlations and structural features; For capturing the fluid dynamics effects of the diffusion class of features; For lattice phonon effector circuit operators used to capture local and periodic features; , and The first in the multi-expert system The parameters that can be learned within the corresponding sub-circuit of a quantum expert.

6. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 1, characterized in that, The process involves using a bio-inspired gating network and the neural synaptic excitation-inhibition mechanism to process the encrypted data to be identified and the activation intensity vector of a pre-constructed multi-expert system to determine the net synaptic input value of each quantum expert in the multi-expert system, including: The encrypted data to be identified is concatenated with the pre-constructed activation intensity vector of the multi-expert system, and a presynaptic signal vector is generated through a projection layer. The presynaptic signal vector is represented as follows: ； in, Represents the presynaptic signal vector; This represents the encrypted data to be identified. This represents the activation intensity vector of a multi-expert system. This represents the concatenation of the ciphertext data to be identified and the activation strength vectors of the multi-expert system. It is a learnable projection matrix; This represents the learnable bias vector of the presynaptic signal vector. For the set of real numbers, For the dimensions of the gated network projection layer, The feature dimensions of the ciphertext data to be identified; The presynaptic signal vector is decomposed into excitatory postsynaptic potentials and inhibitory postsynaptic potentials, and the expressions for the excitatory and inhibitory postsynaptic potentials are as follows: ， ； in, and These represent excitatory and inhibitory postsynaptic potentials, respectively. This represents the Sigmoid activation function. and Let represent the learnable projection matrices of excitatory and inhibitory postsynaptic potentials, respectively. and These represent the learnable bias vectors for excitatory and inhibitory synaptic potentials, respectively. The net synaptic input vector is calculated using the excitation-inhibition balance mechanism, and the expression for the net synaptic input vector is: ； in, The net synaptic input vector is represented by the net synaptic input values ​​of the quantum expert corresponding to each quantum fuzzy rule. It represents a learnable global excitability gain, simulating the regulation of overall excitability by neurotransmitters; This represents a learnable inhibition strength coefficient, enabling competition among quantum experts; This is a vector of synaptic plasticity parameters used to adjust the response strength of each quantum expert; The symbol represents the element-wise product of vectors.

7. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 1, characterized in that, The step of analyzing the net synaptic input value and corresponding activation intensity of each quantum expert in the activated expert subset to determine the final gating weight of each quantum expert in the activated expert subset includes: The net synaptic input value of each quantum expert in the activated expert subset is normalized using the Softmax function to determine the gating weight of each quantum expert in the activated expert subset. The final gating weight of each quantum expert in the activated expert subset is determined based on the gating weights and corresponding activation strengths of all quantum experts in the activated expert subset.

8. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 7, characterized in that, The first in the activated expert subset The expression for the gating weights of a quantum expert is: ； in, Indicates the first in the activated expert subset The gating weights of a quantum expert, For the activated expert subset, the first Net synaptic input value of a quantum expert For the activated expert subset, the first The net synaptic input value of a quantum expert.

9. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 8, characterized in that, The expression for the final gating weight of the p-th quantum expert in the activated expert subset is: ； in, For the activated expert subset, the first The gating weights of a quantum expert, For the activated expert subset, the first The gating weights of a quantum expert, For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. For the input quantum state and the first The activation strength of the prototype quantum state of the antecedent of the quantum fuzzy rule corresponding to a quantum expert. A preset temperature parameter is used to control the influence of gating weights and activation intensity.

10. The block cipher recognition method based on quantum self-organizing fuzzy neural network according to claim 9, characterized in that, The step of inputting the input quantum state into each quantum expert in the activated expert subset for block cipher type identification to obtain the final block cipher type identification result includes: The input quantum state is input to each quantum expert in the activated expert subset. The dynamically parameterized quantum circuit corresponding to each quantum expert acts on the input quantum state, and the block cipher type identification result for each quantum expert is obtained through quantum state measurement. The block cipher type identification result for the p-th quantum expert in the activated expert subset is... The expression is: ； in, This indicates a quantum state measurement operation. This represents the dynamically parameterized quantum circuit unitary operator of the p-th quantum expert in the activated expert subset. For the activated expert subset, the first Internally learnable parameters of a quantum expert Indicates the input quantum state; The block cipher type identification results of each quantum expert are weighted and summed according to their corresponding final gating weights to obtain the final block cipher type identification result. The expression for the final block cipher type identification result is as follows: ； in, This indicates the final block cipher type identification result; This represents the final gating weight of the p-th quantum expert in the activated expert subset.

Images