A quantum perception intelligent key distribution method for an intelligent quantum communication network

Abstract

The application relates to a quantum perception intelligent key distribution method for an intelligent quantum communication network, which comprises the following steps: inputting quantum channel original observation data into a computer system and extracting a quantum feature vector at each step; stacking the quantum feature vector into a two-dimensional matrix and performing space-time representation through a convolutional neural network; combining a bidirectional long short-term memory network and an attention mechanism to obtain time sequence cleaning features; extracting associated quantum channel features and noise statistics, performing abnormality detection and error correction prediction; inputting both into a proximal policy optimization algorithm to adjust transmission distance and protocol parameters; combining the adjusted parameters to construct a saturated noise channel model, correcting errors, amplifying privacy, and generating a final secure key. The application aims to fully combine quantum feature perception, deep sequence modeling and reinforcement learning optimization mechanism, construct high-dimensional quantum feature mapping, and perform standardization, filtering and abnormal value processing, so that robust feature extraction of quantum channel original observation is realized.

Description

Technical Field

[0001] This invention belongs to the field of quantum information technology, specifically involving quantum key distribution (QKD) and other related technologies in quantum communication. In particular, it relates to a quantum-sensing intelligent key distribution method for intelligent quantum communication networks. Background Technology

[0002] With the development of quantum communication networks, quantum key distribution (QKD) is receiving increasing attention in scenarios such as secure communication, power anomaly detection, and cloud computing security. Achieving high-fidelity quantum state transmission and low quantum bit error rate (QBER) plays a crucial role in system reliability and key generation efficiency.

[0003] However, existing mainstream QKD protocols typically rely on two-dimensional feature representation or sequence statistical processing methods, which are highly sensitive to noise interference and anomalous patterns in complex quantum channels, resulting in limited key generation rates and insufficient system robustness. AI-enhanced quantum key distribution methods, utilizing deep feature extraction, sequence modeling, and reinforcement learning optimization techniques, can improve key generation efficiency and channel robustness, and have become one of the key areas of exploration and promotion by international quantum communication research institutions and academia.

[0004] In recent years, deep learning methods such as Convolutional Neural Networks (CNNs) for anomaly detection and Recurrent Neural Networks (RNNs) for error prediction have improved performance to some extent, but they still struggle to fully capture long-range time-dependent features and lack generalization ability under limited training data. Furthermore, existing solutions do not accurately model channel noise variations over transmission distance and lack adaptive optimization mechanisms for high-noise regions, resulting in significant performance degradation over long distances or in high-noise environments.

[0005] Therefore, there is an urgent need for an intelligent key distribution scheme that combines quantum feature perception, deep sequence modeling, and reinforcement learning optimization to achieve high-fidelity, low-QBER secure key generation and maintain stable and efficient key distribution capabilities in complex quantum channels. Summary of the Invention

[0006] The main objective of this invention is to provide a quantum-sensing intelligent key distribution method for intelligent quantum communication networks. It aims to fully integrate quantum feature sensing, deep sequence modeling, and reinforcement learning optimization mechanisms. By constructing a high-dimensional quantum feature map and performing standardization, filtering, and outlier processing, it achieves robust feature extraction of the original observations of the quantum channel.

[0007] Based on the first main aspect of the present invention, a quantum-sensing intelligent key distribution method for intelligent quantum communication networks is provided, comprising the following steps:

[0008] The raw observation data of the quantum channel is input into the computer system, the information of each quantum bit or photon is calculated and preprocessed, and the quantum feature vector of each time step is extracted.

[0009] The time-series features of the quantum feature vectors are stacked into a two-dimensional feature mapping matrix in time, and then input into a convolutional neural network to extract the local correlation between the time dimension and the feature dimension, thus obtaining a quantum feature sequence as a space-time representation.

[0010] A sequence embedding matrix is ​​constructed using the quantum feature sequence. The forward and backward hidden state representations are obtained by capturing long-range time dependencies through a bidirectional long short-term memory network. An attention mechanism is introduced to weight the features at different time steps. The convolutional features and weighted features are fused to obtain temporal cleanliness features.

[0011] Quantum channel features associated with temporal cleanliness features are extracted by jointly using convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. Noise statistics are formed by statistically analyzing the noise data and error prediction probability corresponding to these quantum channel features. Anomaly detection and error prediction tasks are then performed to obtain anomaly detection values ​​and error correction masks.

[0012] The quantum channel characteristics and noise statistics are input into the computer system, and the transmission distance and protocol parameters are dynamically adjusted through a near-end strategy optimization algorithm.

[0013] The computer system combines the adjusted transmission distance and protocol parameters, uses quantum channel characteristics and noise statistics to construct a saturated noise channel model, uses the generated anomaly detection value and error correction mask to correct the original key sequence to obtain the error correction key, and uses a general hash function to perform privacy amplification on the error correction key to generate the final security key.

[0014] As a further preferred embodiment, in the aforementioned method, the raw observation data of the quantum channel includes one or a combination of in-phase / orthogonal-homogeneous measurement sequences, reference phase or pilot signals, and detector counts.

[0015] The information of each quantum bit or photon includes coherence, phase information, and / or photon statistics;

[0016] The preprocessing involves z-score or min-max normalization for each channel; median filtering of phase noise; and gentle truncation of outlier counts.

[0017] Each time step obtained in the end quantum eigenvectors for:

[0018] ;

[0019] in, For quantum state coherence, For photon phase, The mean number of photons. The standard deviation of the photon number. Features of the photon count histogram Let F represent a real vector of length F.

[0020] As a further preferred embodiment, in the aforementioned method, the expression for the stacked two-dimensional feature mapping matrix is ​​as follows:

[0021]

[0022] The two-dimensional feature mapping matrix is ​​used as input to the convolutional neural network, and arranged in chronological order. quantum eigenvectors The quantum characteristic sequence was obtained. .

[0023] As a further preferred embodiment, in the aforementioned method, when fusing convolutional features and weighted features, the quantum feature sequence is... Perform sequence embedding processing;

[0024] Quantum eigenvectors at each time step First, a low-dimensional representation is obtained through linear projection and layer normalization:

[0025]

[0026] in, Represents the hidden state of sequence embedding. For layer normalization function, This represents the bias vector. Represents the sequence embedding dimension. Indicates the length of the sliding window;

[0027] Subsequently, the embedding vectors of all time steps are combined into a sequence embedding matrix:

[0028]

[0029] Add learnable or sinusoidal positional encoding to each embedding vector , forming input To capture temporal order relationships;

[0030] Finally, the sequence embedding matrix Input a bidirectional long short-term memory network to extract the long-range dependence of quantum features and the correlation between features.

[0031] As a further preferred embodiment, in the aforementioned method, when jointly extracting the quantum channel features associated with clean features, the two-dimensional feature mapping matrix is... Input CNN branch: Capture local correlations in time and feature dimensions;

[0032] Simultaneously, the sequence embedding matrix Inputting the Bi-LSTM branch yields the forward and backward hidden state representations:

[0033] ,

[0034] Then, the representations of each time step are combined into the overall LSTM output sequence matrix:

[0035]

[0036] in, This represents the hidden state of the forward / backward LSTM. This indicates LSTM feature concatenation. Indicates the hidden layer dimension of LSTM. This indicates LSTM feature concatenation;

[0037] Subsequently, an attention mechanism is introduced to weight the importance of each time step to obtain weighted features:

[0038] , ,

[0039] in, Attention energy score For attention context vectors, The hyperbolic tangent activation function is used. Indicates the action to be selected according to the strategy;

[0040] Finally, the convolutional features and weighted features are fused: ,in, This represents the input features after adding position encoding. This represents the local convolutional features of a CNN.

[0041] As a further preferred embodiment, in the aforementioned method, when dynamically adjusting the transmission distance and protocol parameters through the near-end strategy optimization algorithm, the system dynamically adjusts physical and protocol parameters using anomaly detection values ​​and error correction masks to achieve a balance between throughput and security across multiple nodes; including:

[0042] RL agent collection The aggregation state of QKD nodes:

[0043]

[0044] Each node adjusts local parameters according to the network coordination strategy, including transmit strength / modulation variance, coding rate / error correction level, block length, and privacy amplification parameters;

[0045] reward function The overall quantum state fidelity between nodes, network anomaly suppression effect, and throughput are as follows:

[0046]

[0047] in, To improve information coordination efficiency, Both sides exchange information, For the eavesdropper's upper realm, The aggregate state is represented by QBER, which represents the quantum bit error rate. Indicates the signal-to-noise ratio. Indicates strategy selection, Represents the reward function, This represents the discount factor.

[0048] As a further preferred embodiment, in the aforementioned method, when constructing the saturated noise channel model, the quantum signal travels a certain distance... The channel is affected by attenuation and saturation noise, and its channel model is as follows:

[0049]

[0050] The noise variance exhibits a saturation-like increase with distance:

[0051]

[0052] in, The attenuation coefficient is... For baseline noise, For maximum additional noise, For distance scale factor, Indicates the quantum characteristics at the receiving end. This represents the variance function of distance-related noise.

[0053] As a further preferred option, in the aforementioned method, after the saturated noise channel model is constructed, each node extracts local features. Clean embeddings are obtained through CNN + Bi-LSTM + attention mechanism. and receiving embedded Capture quantum noise statistics and channel distortion information;

[0054] Multi-node RL agents aggregate the embedding and noise statistics of each node, dynamically adjust the emission intensity, coding rate and privacy amplification parameters, and optimize global fidelity and throughput.

[0055] Using anomaly detection values ​​and error correction masks, users synchronize their original keys to a consistent error correction key:

[0056]

[0057] Subsequently, privacy is amplified using a generic hash function:

[0058]

[0059] in, To share a random seed.

[0060] As a further preferred option, in the aforementioned method, the final security key rate is calculated based on local error statistics and network-level optimization:

[0061]

[0062] Output final security key It can be directly provided to upper-layer cryptographic protocols to ensure high fidelity and robustness of quantum key distribution under multi-node, long-distance, and high-noise conditions.

[0063] Based on a second key aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the aforementioned quantum-sensing smart key distribution method for intelligent quantum communication networks.

[0064] Compared with existing technologies, this invention fully integrates quantum feature sensing, deep sequence modeling and reinforcement learning optimization mechanisms. By constructing a high-dimensional quantum feature map and performing standardization, filtering and outlier processing, it achieves robust feature extraction of the original observations of the quantum channel.

[0065] In this invention, a multi-layer bidirectional long short-term memory network (Bi-LSTM) combined with an attention mechanism is further utilized to effectively capture the long-range time dependence and local fluctuations of quantum states, thereby improving the ability to detect anomalies and estimate errors.

[0066] Based on this, the present invention also introduces a saturated noise channel model to characterize the distance-dependent Gaussian noise characteristics, and through an AI-driven reinforcement learning optimization strategy, adaptively selects the transmission distance and protocol parameters to achieve multi-objective optimization and balance signal fidelity and noise level.

[0067] This invention effectively improves the robustness, key generation efficiency, and security of quantum key distribution systems under complex channel conditions, providing a feasible intelligent solution for high-reliability, low-error-rate key distribution in quantum communication networks. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.

[0069] Figure 1 A flowchart illustrating the steps of a quantum-sensing intelligent key distribution framework and system for intelligent quantum communication networks provided in this embodiment of the invention;

[0070] Figure 2 This is a schematic diagram of the quantum key distribution process provided in an embodiment of the present invention;

[0071] Figure 3 This is a schematic diagram of a quantum-sensing intelligent key distribution framework and system for intelligent quantum communication networks, provided as an embodiment of the present invention. Detailed Implementation

[0072] The preferred embodiments of the present invention will be described in detail below to provide a clearer understanding of the purpose, features, and advantages of the invention. It should be understood that the following embodiments are not intended to limit the scope of the invention, but are merely illustrative of the essential spirit of the technical solution of the invention.

[0073] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known techniques associated with the invention may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0074] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0075] In the following embodiments, some of the parameters may have the following meanings:

[0076] : Represents a real vector of length F, that is, the feature vector consists of F real features. : Current time step, used to identify the time-series quantum measurement data point.

[0077] : Sliding window length, used to calculate statistics (mean, variance, etc.).

[0078] The quantum state density matrix at time t is used to describe the mixed states of a quantum system; the diagonal lines reflect the probability distribution, and the off-diagonal lines reflect coherence.

[0079] The i,j element of the density matrix at time t is used to calculate the coherence.

[0080] Quantum state coherence is calculated from the sum of the absolute values ​​of the off-diagonal elements of the density matrix.

[0081] Photon phase represents the phase information of a quantum state and can be converted into continuous features.

[0082] Phase mapping features help avoid phase wrap discontinuities.

[0083] Photon count at time t, the photon detection result from the detector.

[0084] : Mean photon number, which is the expected value within the sliding window.

[0085] : Standard deviation of photon number.

[0086] Photon count histogram features are used to statistically analyze the distribution of photon counts.

[0087] : The quantum eigenvector at time t. Characterizes the original quantum measurement features at the current moment.

[0088] F: The total number of feature dimensions, determined by the number of selected features.

[0089] : A sliding window expectation operator at time t, used to calculate the expectation of samples within the interval (tW, t].

[0090] The sliding window standard deviation operator calculates the standard deviation of samples within the interval (tW, t].

[0091] Histogram statistical functions map counted samples to probability distribution features.

[0092] Normalization: Normalization operations (such as z-score or min–max) standardize the feature amplitudes of each channel.

[0093] Median filtering operation: used to suppress phase noise.

[0094] : Input matrix structure, representing time series × feature dimension.

[0095] : A T-row, F-column real matrix.

[0096] : is the kernel similarity matrix, used to measure the similarity of quantum states across all time steps.

[0097] : The hidden state of the sequence embedding, representing the context quantum state at time t.

[0098] : Layer normalization function, used to stabilize network training.

[0099] b: Bias vector, used for translation terms in linear layers or attention calculations.

[0100] d: Sequence embedding dimension / latent feature dimension, representing the feature dimension at each time step after linear transformation or embedding.

[0101] H: Hidden state dimension, the size of the hidden layer of the LSTM.

[0102] : The position encoding vector at time step t. Encodes the sequence information of time steps using a fixed sine / cosine function.

[0103] CNN local convolution features are local time-related features extracted by one-dimensional convolution, reflecting short-term quantum state changes.

[0104] : The hidden state of the forward / backward LSTM, representing the hidden state in the bidirectional LSTM at time t, capturing bidirectional time dependencies.

[0105] LSTM feature concatenation means concatenating the outputs of the forward and backward LSTMs to obtain global temporal semantic features.

[0106] LSTM hidden layer dimension, representing the hidden dimension of a unidirectional LSTM.

[0107] The overall LSTM output sequence matrix represents the output features of the LSTM path over the entire time window.

[0108] Attention energy score: used to measure the relevance of features at time t to the global task.

[0109] Attention context vector, used to calculate attention weights, reflecting the relevance of features to the task.

[0110] The hyperbolic tangent activation function compresses the input to (-1,1), increasing its nonlinear expressive power.

[0111] : Input features after adding position encoding, output features after fusing position encoding with time step input.

[0112] : The true label of task A, indicating the true label of task A (such as security status classification).

[0113] Model prediction probability (prediction distribution), multi-class prediction results after Softmax.

[0114] : The direct bit prediction value at time step t, which is directly generated as a bit (0 / 1) based on the probability output of task A.

[0115] Cross-entropy loss for Task A measures the difference between the prediction and the true label (main task loss).

[0116] : Auxiliary loss for task B, representing the loss of auxiliary tasks (such as delayed prediction or anomaly detection).

[0117] Consistency loss forces different task feature spaces to maintain structural consistency in the shared layer.

[0118] Regularization term: Limits the size of model parameters to prevent overfitting.

[0119] Status: Contains Channel parameters, MEC load, QBER, encryption latency, etc.

[0120] Actions: Strategy selection, such as key update frequency, algorithm switching, power allocation, and slice migration.

[0121] Rewards: A multi-objective reward that combines latency, key rate, and security level.

[0122] Discount factor: Used to balance long-term and immediate returns.

[0123] Quantum characteristics of the receiver, the received signal after channel attenuation and noise.

[0124] : Distance-related noise variance function, representing the channel noise that increases with physical distance. Indicates the length of a quantum link or optical fiber.

[0125] Clean embedding at the sender end, a noise-free embedding representation optimized for Stage 4 / 5, used for key generation.

[0126] : Receiver embedding, the embedding extracted by the receiver after being affected by the channel.

[0127] Mutual information between the sender and receiver, i.e., the security information rate under channel statistics.

[0128] Eve information acquisition rate, used for security threshold assessment.

[0129] Error correction mask overhead represents the leakage caused by the amount of publicly available information.

[0130] Privacy amplification overhead, bit loss due to hash function compression.

[0131] Hash functions are used for function mapping.

[0132] like Figure 1 As shown, in one embodiment of the present invention, a quantum-sensing intelligent key distribution method for intelligent quantum communication networks includes the following steps S110-S160:

[0133] S110: Input the raw observation data of the quantum channel into the computer system, calculate the information of each quantum bit or photon and perform preprocessing, and extract the quantum feature vector of each time step;

[0134] S120, the time series features of the quantum feature vector are stacked into a two-dimensional feature mapping matrix according to time, and input into a convolutional neural network to extract the local correlation between the time dimension and the feature dimension, so as to obtain the quantum feature sequence as a space-time representation;

[0135] S130, a sequence embedding matrix is ​​constructed using the quantum feature sequence, and forward and backward hidden state representations are obtained by capturing long-range time dependencies through a bidirectional long short-term memory network. An attention mechanism is introduced to weight the features at different time steps, and convolutional features and weighted features are fused to obtain temporal cleanliness features.

[0136] S140: Quantum channel features associated with temporal cleanliness features are jointly extracted by convolutional neural network, bidirectional long short-term memory network and attention mechanism. The noise data and error prediction probability corresponding to the quantum channel features are statistically analyzed to form noise statistics. Anomaly detection and error prediction tasks are performed to obtain anomaly detection values ​​and error correction masks.

[0137] S150, input the quantum channel characteristics and noise statistics into the computer system, and dynamically adjust the transmission distance and protocol parameters through the near-end strategy optimization algorithm;

[0138] S160, the computer system combines the adjusted transmission distance and protocol parameters, uses quantum channel characteristics and noise statistics to construct a saturated noise channel model, uses the generated anomaly detection value and error correction mask to correct the original key sequence to obtain the error correction key, and uses a general hash function to perform privacy amplification on the error correction key to generate the final security key.

[0139] Combination Figure 2 and Figure 3 As shown, where, Figure 2 This diagram illustrates the quantum key distribution process provided in an embodiment of the present invention. The diagram shows the overall process of the quantum key distribution system: First, the transmitter generates a quantum state. During the data processing stage, the quantum data is standardized to reduce transmission channel loss and increase the probability of correct key sharing. Then, in the prediction-error correction stage, the system uses learning models to monitor potential attackers and perform error detection and correction. Simultaneously, it leverages AI intelligent models to enhance and optimize errors, thereby improving error correction capabilities. Finally, the corrected quantum information is converted into error information, and a final key is generated through weighted summation, forming a secure shared key that can be used for encrypted communication.

[0140] Figure 3This diagram illustrates a quantum-aware intelligent key distribution framework and system for intelligent quantum communication networks, as provided in this invention. The diagram showcases a quantum feature enhancement and dual-task processing framework. First, a quantum source generates a signal, which is modulated to form a quantum feature map and encoded as input through an embedding module. The system utilizes an AI-enhanced intelligent model to generate a key and extract tag features, combining adjustable weights to optimize noise in the receiving network, thereby improving similarity and data distribution consistency. Subsequently, the input features are processed by a quantum-aware attention module, fed into a CNN and Bi-LSTM to extract a spatiotemporal feature matrix, and branched to execute task A: anomaly detection and task B: error correction. Finally, the output is fused to obtain the enhanced prediction result and correction value, achieving intelligent enhancement of quantum features and collaborative processing of dual tasks.

[0141] In one possible implementation, steps S110-S160 above are divided into six stages, namely, quantum feature extraction stage, quantum feature mapping construction stage, quantum sequence embedding and attention mechanism fusion stage, quantum feature joint embedding and task perception stage, distance and parameter optimization stage based on reinforcement learning stage, and distillation stage.

[0142] In the quantum feature extraction stage, the computer system extracts multidimensional features from the raw observation data of the quantum channel, including the coherence measurement sequence of in-phase / quadrature components, reference phase, probe optical signal, and detector counts. First, the coherence indices of the quantum states are calculated, such as... , Extracting phase features And calculate the mean number of photons. Standard deviation and histogram distribution Subsequently, standardization, filtering, and gentle truncation are performed on all features to suppress noise interference and improve feature robustness, ultimately forming the quantum feature vector at each time step. ;

[0143] In one possible implementation, the specific execution process is as follows:

[0144] Input: Raw observation data from the quantum channel, such as in-phase / quadrature (I / Q) coherent measurement sequences, reference phase or pilot signals, detector counts, etc. Output: Each time step quantum eigenvectors .

[0145] Each quantum bit or photon typically contains the following information:

[0146] 1) Coherence. Reflecting the visibility or fidelity of interference in a quantum state, it can be calculated from the off-diagonal elements of the density matrix: , Generate scalar sequence .

[0147] 2) Phase information. Photon state phase. It can be directly used as a continuous feature, or mapped to To avoid issues with sudden changes in angle.

[0148] 3) Photon statistics. This includes photon number distribution (Poisson distribution or superposition distribution), emission intensity, and noise photon count. (In a sliding window...) Internal calculation of mean, variance, and histogram: , , .

[0149] Finally, preprocessing is performed: z-score or min-max normalization is applied to each channel (i.e., the quantum sequence at each time step t); median filtering is applied to phase noise; and gentle truncation is applied to outlier counts. The final quantum eigenvector is: .

[0150] In the quantum feature mapping construction stage, time series features are stacked temporally to form a two-dimensional feature mapping matrix. This mapping is fed as input into a convolutional neural network (CNN) layer to extract the local correlation between the time dimension and the feature dimension, thereby obtaining a higher-level spatial-temporal representation and laying the foundation for subsequent sequence modeling.

[0151] In one possible implementation, the specific execution process is as follows:

[0152] enter: Output: Feature mapping matrix / tensor At each time step Constructing quantum eigenvectors: .

[0153] Stack the sequences into a matrix:

[0154] in It is the feature dimension of each time step.

[0155] This matrix serves as the input to the CNN: 1) One-dimensional convolution (time axis), with the following shape: ;2) Two-dimensional convolution (time) Features): Shape 3) Multi-channel processing: Statistical features Considered as an independent channel, shape .

[0156] In the quantum sequence embedding and attention mechanism fusion stage, a representation matrix is ​​first constructed using low-level temporal features extracted by a convolutional network. Then, a bidirectional long short-term memory (Bi-LSTM) network is used to capture long-range temporal dependencies, yielding forward and backward hidden state representations. An attention mechanism is further introduced to weight the importance of different time steps, resulting in weighted features. Finally, the convolutional features and attention-embedded features are fused to obtain a complete temporal sequence, which serves as the clean feature input for the reinforcement learning module.

[0157] In one possible implementation, the specific execution process is as follows:

[0158] The system obtains the quantum feature sequence during the quantum feature mapping construction phase. Sequence embedding processing is performed. The quantum eigenvectors at each time step t are obtained. First, a low-dimensional representation is obtained through linear projection and layer normalization: Then, the embedding vectors of all time steps are combined into a sequence embedding matrix: In order to preserve the evolutionary information of quantum features in the time dimension.

[0159] Add learnable or sinusoidal positional encoding to each embedding vector , forming input This is done to capture temporal relationships. Ultimately, the sequence embedding matrix... It can be input into a bidirectional long short-term memory network (Bi-LSTM) to further extract the long-range dependence of quantum features and the correlation between features, providing a high-dimensional and structured quantum state representation for subsequent feature fusion, anomaly detection and error correction prediction.

[0160] In the joint embedding and task awareness stage of quantum features, the system jointly extracts quantum features through a convolutional neural network (CNN), a bidirectional long short-term memory network (Bi-LSTM), and an attention mechanism to achieve joint optimization of anomaly detection and error correction prediction. CNN captures local correlations, Bi-LSTM represents temporal dependencies, and the attention mechanism highlights key time steps. After fusing the features from both, task A (anomaly detection) uses Softmax output, while task B (error correction prediction) is based on error probability regression or bit prediction.

[0161] In one possible implementation, the specific execution process is as follows:

[0162] The system extracts low-level shared quantum feature representations by combining convolutional neural networks (CNN), bidirectional long short-term memory networks (Bi-LSTM), and attention mechanisms, and simultaneously optimizes two tasks: anomaly detection and error correction prediction.

[0163] First, the two-dimensional feature mapping matrix constructed in the quantum feature mapping construction stage is... Input CNN branch: This is to capture local correlations in both time and feature dimensions.

[0164] Simultaneously, the sequence embedding matrix in the quantum sequence embedding and attention mechanism fusion stage is... Inputting the Bi-LSTM branch yields the forward and backward hidden state representations: , Then, the representations of each time step are combined into a matrix. .

[0165] Then, an attention mechanism is introduced to weight the importance of each time step: , , .

[0166] Finally, the CNN features are fused with the attention-weighted LSTM features: .

[0167] Task A (anomaly detection) outputs via Softmax. Task B (error correction prediction) can be achieved through error probability regression. Or direct bit prediction accomplish.

[0168] The loss function during training includes anomaly detection loss. Error correction loss Consistency loss and regularization terms The total loss is: .

[0169] In the distance and parameter optimization phase based on reinforcement learning, the system inputs quantum channel characteristics and noise statistics into the reinforcement learning module, and dynamically adjusts the transmission distance and protocol parameters through a near-end policy optimization algorithm. This mechanism can adaptively select the optimal transmission strategy while balancing signal fidelity and noise levels, thereby achieving efficient and robust optimization of the quantum key distribution process.

[0170] In one possible implementation, the specific execution process is as follows:

[0171] The system utilizes quantum feature joint embedding and anomaly detection values ​​and error correction mask signals from the task awareness stage to dynamically adjust physical and protocol parameters, achieving a balance between throughput and security across multiple nodes. RL agent collection. Aggregation state of QKD nodes:

[0172]

[0173] Each node adjusts local parameters according to the network's cooperative strategy, including transmit strength / modulation variance, coding rate / error correction level, block length, and privacy amplification parameters. Reward function. The overall quantum state fidelity between nodes, network anomaly suppression effect, and throughput are as follows:

[0174]

[0175] To improve information coordination efficiency, Both sides exchange information, The upper bound for eavesdroppers is given by the anomaly / error estimation.

[0176] During the key distillation stage, the system uses the error detection value and error correction mask generated by the saturated noise channel modeling results to correct the errors in the original key sequences between users, thereby obtaining a consistent error correction key.

[0177] In this process, the established quantum key distribution channel model can simulate the attenuation and noise changes of quantum signals at different transmission distances. In particular, it can accurately reflect the physical law that noise tends to saturate as the distance increases, providing a real channel constraint for error detection and correction.

[0178] Subsequently, the system employs a general hash function (such as a Toeplitz matrix or SHA-based random compression) to perform privacy amplification on the error correction key, eliminating information potentially obtained by eavesdroppers and generating the final secure key. Simultaneously, by combining the channel parameters and protocol operation parameters optimized during the reinforcement learning phase, the system is able to estimate the final secure key rate.

[0179] The final output security key can be directly used for upper-layer quantum encryption, session authentication, or one-time key communication protocols, thereby ensuring that the system still has high quantum state fidelity and strong robustness in high-noise, multi-node, and long-distance transmission environments.

[0180] In one possible embodiment, the specific execution process is as follows: the system combines deep feature extraction, network-level optimization, and key post-processing under saturated noise multi-node channels to achieve the final secure key generation.

[0181] Quantum signals at propagation distance The channel model is affected by attenuation and saturation noise. The noise variance increases saturably with distance: , The attenuation coefficient is... For baseline noise, For maximum additional noise, This is the distance scale factor.

[0182] This model characterizes the degradation of quantum features during long-distance transmission, providing physical constraints for subsequent feature extraction and error correction. Subsequently, each node extracts local features. Clean embeddings are obtained through CNN + Bi-LSTM + attention mechanism. and receiving embedded It captures quantum noise statistics and channel distortion information. The multi-node RL agent summarizes the embeddings and statistics (QBER, SNR, anomaly score, etc.) of each node and dynamically adjusts the transmit strength, coding rate and privacy amplification parameters to optimize global fidelity and throughput.

[0183] By leveraging the joint embedding of quantum features and the anomaly detection and error correction mask obtained during the task awareness phase, users synchronize their original keys into a consistent error correction key: .

[0184] Privacy is then amplified using a common hash function (such as a Toeplitz matrix or a SHA-based random extractor): ,in To share a random seed.

[0185] The final secure key rate is calculated based on local error statistics and network-level optimization: The final output security key It can be directly provided to upper-layer cryptographic protocols (such as one-time key encryption and authentication) to ensure high fidelity and strong robustness of quantum key distribution under multi-node, long-distance and high-noise conditions.

[0186] The technical terms, principles, or means related to the technical solutions of the present invention mentioned in the above embodiments, which are not described in detail above, are all well-known technologies or common practices that are known to those skilled in the art.

[0187] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A quantum-sensing intelligent key distribution method for intelligent quantum communication networks, characterized in that, Includes the following steps: The raw observation data of the quantum channel is input into the computer system, the information of each quantum bit or photon is calculated and preprocessed, and the quantum feature vector of each time step is extracted. The time-series features of the quantum feature vectors are stacked into a two-dimensional feature mapping matrix in time, and then input into a convolutional neural network to extract the local correlation between the time dimension and the feature dimension, thus obtaining a quantum feature sequence as a space-time representation. A sequence embedding matrix is ​​constructed using the quantum feature sequence. The forward and backward hidden state representations are obtained by capturing long-range time dependencies through a bidirectional long short-term memory network. An attention mechanism is introduced to weight the features at different time steps. The convolutional features and weighted features are fused to obtain temporal cleanliness features. Quantum channel features associated with temporal cleanliness features are extracted by jointly using convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. Noise statistics are formed by statistically analyzing the noise data and error prediction probability corresponding to these quantum channel features. Anomaly detection and error prediction tasks are then performed to obtain anomaly detection values ​​and error correction masks. The quantum channel characteristics and noise statistics are input into the computer system, and the transmission distance and protocol parameters are dynamically adjusted through a near-end strategy optimization algorithm. The computer system combines the adjusted transmission distance and protocol parameters, uses quantum channel characteristics and noise statistics to construct a saturated noise channel model, uses the generated anomaly detection value and error correction mask to correct the original key sequence to obtain the error correction key, and uses a general hash function to perform privacy amplification on the error correction key to generate the final security key. When dynamically adjusting transmission distance and protocol parameters through a near-end strategy optimization algorithm, the system utilizes anomaly detection values ​​and error correction masks to dynamically adjust physical and protocol parameters, achieving a balance between throughput and security across multiple nodes; including: RL agent collection The aggregation state of QKD nodes: This represents the average coherence measure of node N at time t; It is the average qubit error rate during the quantum key distribution process at node N at time t; This represents the average signal-to-noise ratio of node N at time t; This represents the control actions performed on node N within the time window [tL, t-1]. The average calculation performed; This represents the average probability of error occurring at node N at time t, where error refers to the deviation between the model's prediction and the actual quantum state value measured at the node. This represents the sequence of actions of node N within the time window [tL, t-1]. Each node adjusts local parameters according to the network coordination strategy, including transmit strength / modulation variance, coding rate / error correction level, block length, and privacy amplification parameters; reward function The overall quantum state fidelity between nodes, network anomaly suppression effect, and throughput are as follows: in, To improve information coordination efficiency, Both sides exchange information, For the eavesdropper's upper realm, Indicates the discount factor. This represents the time delay experienced by a signal at node i from the transmitter to the receiver during quantum communication. These are weighting coefficients in the reward function, used to adjust the impact of delay on the total reward. The safety threshold violation indicator is used to characterize whether the current operating state of node i exceeds the preset safe communication conditions.

2. The quantum-sensing intelligent key distribution method for intelligent quantum communication networks according to claim 1, characterized in that, The raw observation data of the quantum channel includes one or a combination of in-phase / quadrature measurement sequences, reference phase or pilot signals, and detector counts. The information of each quantum bit or photon includes coherence, phase information, and / or photon statistics; The preprocessing involves z-score or min-max normalization for each channel; median filtering of phase noise; and gentle truncation of outlier counts. Each time step obtained in the end quantum eigenvectors for: ； in, For quantum state coherence, For photon phase, The mean number of photons. The standard deviation of the photon number. Features of the photon count histogram Let F represent a real vector of length F.

3. The quantum-sensing intelligent key distribution method for intelligent quantum communication networks according to claim 2, characterized in that, The expression for the stacked two-dimensional feature mapping matrix is ​​as follows: The two-dimensional feature mapping matrix is ​​used as input to the convolutional neural network, and arranged in chronological order. quantum eigenvectors The quantum characteristic sequence was obtained. .

4. The quantum-sensing intelligent key distribution method for intelligent quantum communication networks according to claim 3, characterized in that, When fusing convolutional features and weighted features, the quantum feature sequence Perform sequence embedding processing; Quantum eigenvectors at each time step First, a low-dimensional representation is obtained through linear projection and layer normalization: in, Represents the hidden state of sequence embedding. For layer normalization function, This represents the bias vector. Represents the sequence embedding dimension. Indicates the length of the sliding window; Subsequently, the embedding vectors of all time steps are combined into a sequence embedding matrix: Add learnable or sinusoidal positional encoding to each embedding vector , forming input To capture temporal order relationships; Finally, the sequence embedding matrix Input a bidirectional long short-term memory network to extract the long-range dependence of quantum features and the correlation between features.

5. The quantum-sensing intelligent key distribution method for intelligent quantum communication networks according to claim 4, characterized in that, When jointly extracting quantum channel features associated with clean features, the two-dimensional feature mapping matrix is ​​used. Input CNN branch: Capture local correlations in time and feature dimensions; This represents the feature dimension of the CNN branch output; Simultaneously, the sequence embedding matrix Inputting the Bi-LSTM branch yields the forward and backward hidden state representations: ， Then, the representations of each time step are combined into the overall LSTM output sequence matrix: in, This represents the hidden state of the forward / backward LSTM. This represents the hidden state of the forward LSTM. This represents the hidden state of the backward LSTM. This indicates LSTM feature concatenation. Indicates the hidden layer dimension of the LSTM; Subsequently, an attention mechanism is introduced to weight the importance of each time step to obtain weighted features: ， ， in, Attention energy score The hyperbolic tangent activation function is used. This indicates the action to be selected based on the strategy. For attention matrix, For bias terms, This represents the index of all time steps in the Softmax normalization operation. This is the transpose of the attention context vector; Finally, the convolutional features and weighted features are fused: ,in, This represents the input features after adding position encoding. This represents the local convolutional features of a CNN.

6. The quantum-sensing intelligent key distribution method for intelligent quantum communication networks according to claim 1, characterized in that, When constructing the saturated noise channel model, the quantum signal travels a certain distance. The channel is affected by attenuation and saturation noise, and its channel model is as follows: in, This represents the noise term at time t corresponding to the channel noise term, where the noise variance increases saturably with distance. in, The attenuation coefficient is... For baseline noise, For maximum additional noise, For distance scale factor, Indicates the quantum characteristics at the receiving end. This represents the variance function of distance-related noise.

7. The quantum-sensing intelligent key distribution method for intelligent quantum communication networks according to claim 6, characterized in that, After the saturated noise channel model is constructed, each node extracts local features. Clean embeddings are obtained through CNN + Bi-LSTM + attention mechanism. and receiving embedded Capture quantum noise statistics and channel distortion information; Multi-node RL agents aggregate the embedding and noise statistics of each node, dynamically adjust the emission intensity, coding rate and privacy amplification parameters, and optimize global fidelity and throughput. Using anomaly detection values ​​and error correction masks, users synchronize their original keys to a consistent error correction key: This represents the key bits after error correction at time t; Subsequently, privacy is amplified using a generic hash function: in, To share a random seed.

8. The quantum-sensing intelligent key distribution method for intelligent quantum communication networks according to claim 7, characterized in that, The final secure key rate is calculated based on local error statistics and network-level optimization: It represents the mutual information between the sender and receiver, that is, the amount of information shared between them; Indicates the upper limit of the eavesdropper; This represents the overhead of error correction during quantum key distribution; This represents the overhead of privacy amplification in the quantum key distribution process; Output final security key It can be directly provided to upper-layer cryptographic protocols to ensure high fidelity and robustness of quantum key distribution under multi-node, long-distance, and high-noise conditions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the quantum-sensing smart key distribution method for intelligent quantum communication networks as described in any one of claims 1-8.

Images