Key screening method, system and device for continuous variable quantum key distribution

Abstract

The application discloses a key screening method, system and device for continuous variable quantum key distribution, and relates to the technical field of quantum communication security. The method comprises the following steps: initializing a model and saving training statistics; calculating an abnormal score for training data block by block and filling a sliding window; calculating a dynamic threshold value in a training stage; completing initial screening in the training stage; performing incremental LoOP calculation on each new key block; dynamically updating the threshold value of the sliding window and making a screening decision; and the method adopts incremental LoOP, so that each new key block only needs to perform kNN query once to obtain an abnormal score, thereby avoiding the problem of repeated training of the model and realizing real-time detection capability at the level of quantum key blocks.

Description

Technical Field

[0001] This invention relates to the field of quantum communication security technology, specifically to a key screening method, system, and apparatus for continuous variable quantum key distribution. Background Technology

[0002] Quantum key distribution (QKD) technology enables information-theoretically secure key distribution without relying on classical computational problems. Continuous-variable quantum key distribution (CV-QKD) has become a research hotspot in recent years due to its advantages such as easy integration with commercial optical communication systems, high detection efficiency, and low equipment cost.

[0003] In practical fiber optic channels and optoelectronic devices, CV-QKD systems are vulnerable to various physical side-channel attacks, including but not limited to: (1) Light source attacks: Attackers inject lasers into the transmitter or local oscillator to change the system power, thereby affecting noise estimation; (2) Saturation attacks: By making the detector operate in the nonlinear region, the true noise is underestimated; (3) Blinding attacks: By using strong light to disrupt the quantum response of the detector, the attack is hidden in the noise. These attacks share the common characteristic that they change the statistical distribution of the measurement results at the receiver, causing the over-noise estimation to be lowered, thereby disrupting the parameter estimation stage of the CV-QKD protocol and making the attack undetectable.

[0004] However, traditional CV-QKD systems only have basic filtering operations such as "measuring basis mismatch bit removal", which are difficult to identify these statistical anomaly attacks. In addition, some studies have proposed using offline machine learning models (such as LOF, IsolationForest, etc.) for key filtering, but offline models have the following problems: (1) They cannot be applied to streaming keys: CV-QKD keys are generated continuously in blocks, and offline models cannot update block by block or detect in real time; (2) The threshold is fixed and not adaptive: The fixed threshold obtained by offline training cannot cope with changes in channel noise; (3) They have poor robustness against attacks: If the training data is contaminated by attackers, the offline model will be permanently invalidated and cannot recover its detection capability; (4) The computational cost is high: The model needs to be retrained every time new data is added, which is not suitable for high-speed key generation scenarios.

[0005] In summary, existing methods rely on low-rank priors of the signal, are not adaptable to sudden interference (such as impulse noise) or time-varying channels, and are difficult to detect key block anomalies in real time, thereby reducing the security and stability of quantum key generation. Summary of the Invention

[0006] To address the shortcomings of existing technologies that rely on low-rank priors of signals, are insufficiently adaptable to sudden interference (such as impulse noise) or time-varying channels, and are difficult to detect key block anomalies in real time, thereby reducing the security and stability of quantum key generation, this invention proposes a key screening method, system, and device for continuous variable quantum key distribution, thereby solving the problems existing in the prior art.

[0007] A key selection method for continuous-variable quantum key distribution includes the following steps: Obtain the key blocks generated in real time by Bob at the continuous variable quantum key distribution (CV-QKD) receiver, and extract the statistical features in each key block to construct a key block feature vector; Based on the incremental LoOP model, a k-nearest neighbor search is performed on the feature vector of the key block to calculate the standard distance between the feature vector and its nearest neighbors; the standard distance is mapped to the probability distance by setting a saliency factor; the neighborhood expected probability distance is constructed based on the probability distance; the global normalization factor is calculated based on the neighborhood expected probability distance and the probability distance; and the anomaly probability score of each key block is calculated based on the global normalization factor. Add a fixed length to the key block anomaly probability score. W A sliding window, based on preset contamination rate parameters. c Select the first anomaly probability score from all anomaly probability scores within the current sliding window. The score is used as a dynamic threshold; abnormal key blocks are filtered out by comparing the anomaly probability score of each key block with the dynamic threshold.

[0008] Furthermore, the step of extracting statistical features from Bob's measurements to construct a key block feature vector specifically includes the following steps: Bob obtains Bob's measurement value in each key block through the CV-QKD receiver, which is represented as follows: ;in, Indicates from the first k The measurement values ​​obtained from each key block, k =1,2,3,…, n ; N This indicates the number of measurements contained in each key block; Calculate the mean of Bob's measurements: ; Calculate the variance based on the mean: ; Based on the mean and variance, the key block feature vector is constructed: .

[0009] Furthermore, the calculation process of the global normalization factor specifically includes the following steps: For feature vectors Calculate the Euclidean distance between the feature vector and its nearest neighbor: ; Based on Euclidean distance, the local standard distance is defined as: ; Indicates the first j 1 nearest neighbor sample point; By setting a significance factor Mapping local standard distance to probabilistic distance: ; Based on the probability distance, construct the expected probability distance of the neighborhood: ; Based on the expected probability distance and probability distance of the neighborhood, a probabilistic local outlier factor is defined: ; The global normalization factor is then: .

[0010] Furthermore, the anomaly probability score for each key block is expressed as: ; in, erf For the error function: Its output range is [0,1], and z is a random variable that follows a standard normal distribution. t It is an integral variable.

[0011] Furthermore, the step of adding the key block anomaly probability score to a fixed length... W A sliding window, based on preset contamination rate parameters. c Select the first anomaly probability score from all anomaly probability scores within the current sliding window. The value is used as a dynamic threshold, which specifically includes the following steps: Set a fixed length W The sliding window is: ; Add the latest key block anomaly probability score to the sliding window, and delete the oldest score when the window size reaches a preset value; Based on preset pollution rate parameters c Get the first one in the current window The score is the current threshold, expressed as: .

[0012] Furthermore, the step of filtering out normal keys by comparing the anomaly probability score of the key block with a dynamic threshold specifically includes: if If the condition is met, the key block is deemed abnormal and removed; otherwise, it is considered a normal key.

[0013] The present invention also includes a key screening system for continuous-variable quantum key distribution, comprising: The acquisition module is used to acquire the key blocks generated in real time by Bob at the continuous variable quantum key distribution (CV-QKD) receiver, and extract the statistical features in each key block to construct a key block feature vector. The anomaly score calculation module is used to perform k-nearest neighbor search on the feature vector of the key block based on the incremental LoOP model, calculate the standard distance between the feature vector and its nearest neighbors; map the standard distance to the probability distance by setting a saliency factor; construct the neighborhood expected probability distance based on the probability distance; calculate the global normalization factor based on the neighborhood expected probability distance and the probability distance; and calculate the anomaly probability score of each key block based on the global normalization factor. The filtering module is used to add the anomaly probability score of the key block to a fixed length. W A sliding window, based on preset contamination rate parameters. c Select the first anomaly probability score from all anomaly probability scores within the current sliding window. The score is used as a dynamic threshold; abnormal key blocks are filtered out by comparing the anomaly probability score of each key block with the dynamic threshold.

[0014] The present invention also includes a key screening computer device for continuous variable quantum key distribution, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the key screening method for continuous variable quantum key distribution.

[0015] The present invention also includes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform the steps of the key screening method for continuous variable quantum key distribution.

[0016] This invention provides a key selection method for continuous-variable quantum key distribution, which has the following advantages: This invention employs an incremental LoOP model, enabling each new key block to obtain an anomaly probability score with only one k-nearest neighbor search, achieving real-time online detection at the key block level. A sliding window mechanism dynamically selects a threshold based on the contamination rate parameter, allowing the screening criteria to automatically adjust with the time-varying characteristics of channel noise, solving the problem of insufficient adaptability of fixed thresholds. Simultaneously, the sliding window mechanism allows new normal samples to gradually replace contaminated old samples, and the dynamic threshold can automatically recover to normal levels, maintaining an attack recognition rate of over 98% even in adversarial scenarios, achieving self-repair capability against training set contamination. Furthermore, this method requires no hardware modifications, has low computational complexity, and can effectively identify physical side-channel attacks such as light source attacks, saturation attacks, and blinding attacks. After removing abnormal key blocks, the over-noise estimation is restored to the theoretical level, thereby improving the security key rate and maximum transmission distance. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the key selection method for continuous variable quantum key distribution in an embodiment of the present invention. Figure 2 This is the feature space map of the LOF algorithm at 100km in this embodiment of the invention; Figure 3 This is a schematic diagram illustrating the dynamic evolution of the abnormal scores and their thresholds of different algorithms under a strong attack strategy during the training phase when the communication distance is 100km in an embodiment of the present invention. Figure 4 This is a schematic diagram of the over-noise estimation results under different scenarios and data lengths in embodiments of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Existing key selection methods (such as LOF and Isolation Forest) all rely on offline training with fixed thresholds, making them unable to adapt to the time-varying noise characteristics of fiber optic channels. Furthermore, when subjected to physical side-channel attacks such as light source attacks, saturation attacks, and blinding attacks, attackers can inject anomalous samples into the training set, permanently disabling the offline model. In addition, the offline model requires retraining after each selection, failing to meet the real-time requirement of continuous key block generation in CV-QKD systems. Therefore, this invention proposes a key selection method for continuous-variable quantum key distribution. This method significantly reduces noise, restores channel parameter estimation accuracy, and improves the final key rate by identifying statistical anomalies caused by attacks in real time during key generation.

[0020] The following section describes in detail the online key screening method based on incremental LoOP proposed in this invention, using a continuous-variable quantum key distribution (CV-QKD) system and a Gaussian modulation protocol as a foundation. This invention can be integrated into CV-QKD systems of arbitrary structure, without relying on additional hardware, and utilizes only measurement statistics in the key block for real-time anomaly detection.

[0021] like Figure 1 As shown, the method specifically includes the following steps: S1. Obtain the key block generated in real time and extract its features.

[0022] Bob at the CV-QKD receiver obtains in each key block N One measurement value: ; in, Indicates from the first k indivual( k =1,2,3,…, n The measurement value obtained from the key block.

[0023] The attacker's actions in light-based attacks, saturation attacks, and blinding attacks will all alter Bob's measurement value. Based on the statistical characteristics of each block, this invention extracts two basic statistics from each block: (1) Mean: ; This quantity is particularly sensitive to saturation and blinding attacks because the mean of the output will be systematically shifted when the detector enters the nonlinear region or the response is turned off.

[0024] (2) Variance: ; This quantity is highly sensitive to light source attacks, because the injection of a light source will change the local oscillator power and the detection noise.

[0025] Feature vector construction: Due to the high separability of mean-variance features, this invention constructs a two-dimensional feature vector: ; This low-dimensional feature has more stable controllability and can capture the statistical characteristics of various side-channel attacks without additional hardware.

[0026] The eigenvector further incorporates the mean and variance of combined statistics (such as x+y, yx, xy).

[0027] ; in, Represents the local standard distance, reflecting the point. The degree of dispersion of the neighborhood distribution This represents the current feature vector to be detected (a key block); Indicates the first j 1 nearest neighbor sample point; j It is the index number immediately following the index number.

[0028] S2. During the training phase, perform offline preprocessing to initialize the incremental LoOP model and calculate the anomaly probability of the real-time key block based on incremental LoOP.

[0029] This invention employs Incremental Local Outlier Probability (LoOP) to calculate an anomaly probability value for each key block in real time. Incremental LoOP introduces an incremental update mechanism based on the original LoOP, enabling the anomaly distribution probability to be obtained for new data points without retraining the model.

[0030] (1) Local Standard Distance For feature vectors Find its in the training data k One neighbor: .gather This indicates that the feature vector in the training dataset is related to the feature vector. The closest k 1,000 neighborhood sample points, of which For the first j Nearest neighbor samples.

[0031] k-nearest neighbor search is performed based on the training dataset, and k The value of corresponds to the neighborhood parameter MinPts of the model.

[0032] According to Euclidean distance: .

[0033] Define the local standard distance as: ; This quantity reflects the point The degree of dispersion of the neighborhood distribution.

[0034] (2) Probabilistic Distance Set significance factor (Can be 1, 2, or 3, corresponding to different confidence intervals of 68%, 95%, or 99.7%), mapping the local standard distance to the probability distance: ; The distribution of this probability distance can be used to describe the degree to which a sample point is located in a local anomaly region.

[0035] (3) Probability Local Outlier Factor (PLOF) To measure the relative offset between a point and its neighborhood, construct the neighborhood expected probability distance: .

[0036] Define the probability of local outlier factor: ; A value of 0 indicates that the distribution perfectly matches the neighborhood; a value greater than 0 indicates that the point deviates outward.

[0037] (4) Global normalization factor nPLOF To stabilize local deviations at different scales, LoOP defines a global normalization factor: .

[0038] This quantity is calculated once during the training phase and used in all incremental calculations to improve numerical stability.

[0039] (5) LoOP outlier probability The final anomaly probability is: ; in erf For the error function: , is used to map the degree of local outlier after global normalization to an anomaly probability value, z is a random variable that follows a standard normal distribution, and its output value range is [0,1]. Its meaning is as follows: close to 0: the key block is completely normal; greater than a certain threshold: the key block is suspected of being attacked by a side channel and should be removed.

[0040] The key feature of incremental LoOP is that it only performs a nearest neighbor search when computing LoOP and does not require updating global statistics, making it very suitable for high-speed quantum key generation scenarios.

[0041] S3. Use a sliding window to construct a dynamic filtering threshold.

[0042] Incremental LoOP outputs probability values, but classification requires a threshold. Since channel noise in CV-QKD drifts over time, a fixed threshold can lead to serious misclassifications. Therefore, this invention designs a sliding window dynamic threshold mechanism.

[0043] (1) Sliding window construction: Let the window length be W, used to store the W most recent abnormal scores: ; The window updates with the samples: if the window is not full, add the new score; if the window is full, remove the oldest score and add the latest score.

[0044] (2) Dynamic threshold calculation: Let c (contamination) is a preset contamination rate (e.g., 0.05–0.1). The contamination rate is used to dynamically adjust the screening threshold, so that the threshold is adaptively updated according to the score distribution within the current sliding window to resist training set contamination; the threshold is taken as the value of the first value in the window. The larger score is recorded as: ; in, This represents the upper quantile of the anomaly score within the sliding window. It is used to determine the adaptive screening threshold under a given contamination rate assumption, so that the system can maintain stable screening performance and robustness even when the channel statistical characteristics change over time or when it is contaminated by anomaly samples during the training phase.

[0045] Its significance is: (1) If the contamination rate is set to 0.1, it means that the system allows 10% of the samples to be distributed “deviating from the normal mode”; (2) The threshold will be automatically adjusted over time as the statistical characteristics of the key block drift; (3) If the training phase is attacked and contaminated, as the window slides, new unattacked samples will gradually enter the window, and the threshold will automatically return to the normal level, thus achieving adversarial robustness.

[0046] S4. Perform key screening, which includes the following steps: (1) Input key block and extract statistical features Measurement input The mean and variance are used to construct the feature vector. .

[0047] (2) Incremental LoOP anomaly score calculation Based on the neighborhood structure and nPLOF saved during the training phase, calculate: , , , .

[0048] (3) Threshold update: Add a sliding window and calculate .

[0049] (4) Screening decision: like: If the condition is not met, the key block is deemed abnormal and removed; otherwise, it is considered normal and participates in: channel transmission coefficient estimation and noise reduction. Estimation and key generation. The filtered normal key blocks are used for parameter estimation, including calculating the noise upper limit. And improve the security key rate.

[0050] The method of this invention is used to identify at least one of the following CV-QKD attacks: light source attack, saturation attack, and blinding attack. The incremental LoOP model uses a single k-nearest neighbor search to update the anomaly score, thereby avoiding repeated calculation of the probability distance of historical samples and improving the processing speed to O(k\log m).

[0051] Based on the above methods, we tested their application in CV-QKD parameter estimation: Filtering out anomalous key blocks significantly improves the security of CV-QKD. (1) Over-noise recovery to theoretical level: Over-noise in CV-QKD Defined as: ; in: Measure the variance for Bob. Let T be the channel transmission coefficient, where T is the Alice modulation variance. An attacker could change this by... Messenger It is underestimated.

[0052] After filtering out abnormal blocks: ; thereby: ; To prevent attackers from hiding in the noise.

[0053] (3) Significantly improved key rate: The secret key rate of CV-QKD is as follows: ; in: For Alice-Bob information content, For Eve's information content, To improve error correction efficiency.

[0054] After filtering out abnormal blocks, the noise level decreases, making Decrease, therefore: .

[0055] Traditional LOF or isolated forests cannot be updated online and have high computational costs. This invention uses incremental LoOP, which allows each new key block to obtain anomaly scores with only one kNN query, avoiding the problem of repeated model training. It achieves real-time detection capabilities at the quantum key block level, meeting the engineering requirements of high-speed QKD systems.

[0056] like Figure 2 The image shows the feature space of the LOF algorithm at 100km, with the horizontal axis... The mean of the measured values ​​is shown on the y-axis. The measurement variance is represented by red triangles, which are the theoretical expected feature points. White points are normal block feature points, while green, purple, and yellow points correspond to light source attack, saturation attack, and blinding attack block feature points, respectively. Yellow background areas indicate abnormal scores below the threshold (judged as normal), while blue background areas indicate abnormal scores above the threshold (judged as abnormal). The darker the blue, the higher the degree of abnormality. The boundary between the yellow and blue background areas is the decision boundary of the algorithm.

[0057] like Figure 3 The figure shows the dynamic evolution of anomaly scores and their thresholds for different algorithms under a strong attack strategy during the training phase when the communication distance is 100km. Among them, (a) is the anomaly score (blue curve) and anomaly score threshold (red line) detected by the offline algorithm (LOF); (b) is the anomaly score (blue curve) and anomaly score threshold (red curve) detected by the online algorithm (incremental LoOP).

[0058] like Figure 4 The results show the over-noise estimation results under different scenarios and data lengths; where (a) and (b) are standard scenarios (attack without training set), and (c) and (d) are adversarial scenarios (strong attack on Eve training set); the data lengths correspond to N=107 and 108, respectively; the red curve, black curve, green curve, and blue curve represent the over-noise levels after being attacked without any processing, without being attacked, after being filtered by the LOF algorithm, and after being filtered by the incremental LoOP algorithm, respectively.

[0059] This invention uses a sliding window of length W, selecting the quantile of outlier scores within the window as a dynamic threshold based on the contamination rate. This adaptive threshold selection allows the model to automatically recover to an unattacked state as new normal samples are added, even when it is contaminated by attackers during the training phase. This is a crucial robustness capability that traditional offline models cannot achieve. By removing outlier key blocks, this invention restores the over-noise estimation of CV-QKD to the theoretical level, preventing attackers from using noise offsets to hide their attack behavior, thereby improving the key rate and communication distance. Compared with traditional methods such as adding optical power monitoring and hardware isolation, this invention is purely software-based, non-intrusive, and easy to deploy.

[0060] Quantum channels drift with environmental changes, and fixed threshold methods often lead to numerous false positives. This invention addresses this by implementing real-time feature statistics, incremental probability updates, and adaptive thresholding via a sliding window. All computations can be performed in real-time on a conventional CPU, requiring no FPGA, GPU, or additional monitoring modules. The incremental LoOP has a time complexity of O(k log m), and the sliding window update has constant complexity, making it ideal for high-speed continuous-variable quantum key distribution systems.

[0061] Experiments verify that this invention can recover from noise over a 100 km fiber optic link and maintain an attack detection rate of over 98% in adversarial scenarios. This performance surpasses any existing publicly available machine learning-based CV-QKD screening method. It enables the system to maintain stable screening results over long-term operation, unaffected by external factors such as temperature changes and fiber attenuation.

[0062] Based on the same inventive concept, this invention also proposes a key screening system for continuous variable quantum key distribution, comprising: The acquisition module is used to acquire the key blocks generated in real time by Bob at the continuous variable quantum key distribution (CV-QKD) receiver, and extract statistical features from each key block to construct a key block feature vector.

[0063] The anomaly score calculation module is used to perform k-nearest neighbor search on the feature vector of the key block based on the incremental LoOP model, calculate the standard distance between the feature vector and its nearest neighbors; map the standard distance to the probability distance by setting a saliency factor; construct the neighborhood expected probability distance based on the probability distance; calculate the global normalization factor based on the neighborhood expected probability distance and the probability distance; and calculate the anomaly probability score of each key block based on the global normalization factor.

[0064] The filtering module is used to add the anomaly probability score of the key block to a fixed length. W A sliding window, based on preset contamination rate parameters. c Select the first anomaly probability score from all anomaly probability scores within the current sliding window. The score is used as a dynamic threshold; abnormal key blocks are filtered out by comparing the anomaly probability score of each key block with the dynamic threshold.

[0065] The present invention also proposes a key screening computer device for continuous variable quantum key distribution, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the key screening method for continuous variable quantum key distribution.

[0066] The present invention also proposes a readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, are used to perform steps of a key screening method for continuous variable quantum key distribution.

[0067] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A key selection method for continuous-variable quantum key distribution, characterized in that, Includes the following steps: Obtain the key blocks generated in real time by Bob at the continuous variable quantum key distribution (CV-QKD) receiver, and extract the statistical features in each key block to construct a key block feature vector; Based on the incremental LoOP model, a k-nearest neighbor search is performed on the feature vector of the key block to calculate the standard distance between the feature vector and its nearest neighbors; the standard distance is mapped to the probability distance by setting a saliency factor; the expected probability distance of the neighborhood is constructed based on the probability distance; and the global normalization factor is calculated based on the expected probability distance of the neighborhood and the probability distance. The anomaly probability score for each key block is calculated based on the global normalization factor. Add a fixed length to the key block anomaly probability score. W A sliding window, based on preset contamination rate parameters. c Select the first anomaly probability score from all anomaly probability scores within the current sliding window. The score is used as a dynamic threshold; abnormal key blocks are filtered out by comparing the anomaly probability score of each key block with the dynamic threshold.

2. The key selection method for continuous-variable quantum key distribution according to claim 1, characterized in that, The extraction of statistical features from Bob's measurements to construct the key block feature vector specifically includes the following steps: Bob obtains Bob's measurement value in each key block through the CV-QKD receiver, which is represented as follows: ;in, Indicates from the first k The measurement values ​​obtained from each key block, k =1,2,3,…, n ; N This indicates the number of measurements contained in each key block; Calculate the mean of Bob's measurements: ; Calculate the variance based on the mean: ; Based on the mean and variance, the key block feature vector is constructed: .

3. The key selection method for continuous-variable quantum key distribution according to claim 2, characterized in that, The calculation process of the global normalization factor is as follows: Includes the following steps: For feature vectors Calculate the Euclidean distance between the feature vector and its nearest neighbor: ; Based on Euclidean distance, the local standard distance is defined as: ; Indicates the first j 1 nearest neighbor sample point; By setting a significance factor Mapping local standard distance to probabilistic distance: ; Based on the probability distance, construct the expected probability distance of the neighborhood: ; Based on the expected probability distance and probability distance of the neighborhood, a probabilistic local outlier factor is defined: ; The global normalization factor is then: .

4. The key selection method for continuous-variable quantum key distribution according to claim 3, characterized in that, The anomaly probability score for each key block is expressed as: ； in, erf For the error function: Its output range is [0,1], and z is a random variable that follows a standard normal distribution. t It is an integral variable.

5. A key selection method for continuous-variable quantum key distribution according to claim 4, characterized in that, The step involves adding the key block anomaly probability score to a fixed length. W A sliding window, based on preset contamination rate parameters. c Select the first anomaly probability score from all anomaly probability scores within the current sliding window. The value is used as a dynamic threshold, which specifically includes the following steps: Set a fixed length W The sliding window is: ； Add the latest key block anomaly probability score to the sliding window, and delete the oldest score when the window size reaches a preset value; Based on preset pollution rate parameters c Get the first one in the current window The score is the current threshold, expressed as: 。 6. A key selection method for continuous-variable quantum key distribution according to claim 5, characterized in that, The process of filtering out normal keys by comparing the anomaly probability score of the key block with a dynamic threshold specifically includes: if If the condition is met, the key block is deemed abnormal and removed; otherwise, it is considered a normal key.

7. A key screening system for continuous-variable quantum key distribution, characterized in that, include: The acquisition module is used to acquire the key blocks generated in real time by Bob at the continuous variable quantum key distribution (CV-QKD) receiver, and extract the statistical features in each key block to construct a key block feature vector. The anomaly score calculation module is used to perform k-nearest neighbor search on the feature vector of the key block based on the incremental LoOP model, calculate the standard distance between the feature vector and its nearest neighbors; map the standard distance to the probability distance by setting a saliency factor; construct the neighborhood expected probability distance based on the probability distance; and calculate the global normalization factor based on the neighborhood expected probability distance and the probability distance. The anomaly probability score for each key block is calculated based on the global normalization factor. The filtering module is used to add the anomaly probability score of the key block to a fixed length. W A sliding window, based on preset contamination rate parameters. c Select the first anomaly probability score from all anomaly probability scores within the current sliding window. The score is used as a dynamic threshold; abnormal key blocks are filtered out by comparing the anomaly probability score of each key block with the dynamic threshold.

8. A key screening computer device for continuous-variable quantum key distribution, characterized in that, include: A memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the key screening method for continuous variable quantum key distribution as described in any one of claims 1-6.

9. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, perform the steps of the key screening method for continuous variable quantum key distribution as described in any one of claims 1-6.

Images