Key distribution method and system based on quantum communication

Abstract

The invention relates to the technical field of quantum communication, and discloses a key distribution method and system based on quantum communication, and the method comprises the following steps: S101, constructing a feature sequence; step S102, transmitting a reference light signal and combining with an experimental method to obtain a phase compensation value; s103, obtaining a first phase compensation predicted value and a second phase compensation predicted value; step S104, if the difference value is smaller than a first threshold value, selecting a first phase compensation predicted value for compensation; s105, when the difference value is larger than or equal to the first threshold value and smaller than the second threshold value, the two values are subjected to weighted summation compensation; s106, when the difference value is greater than or equal to a second threshold value, carrying out weighted summation compensation on the two and the phase compensation value of the current time point; according to the method, on the basis of the TF-QKD protocol, the phase compensation value is predicted through deep learning in combination with the polynomial, and the compensation decision is optimized by using the error control strategy, so that the interference contrast is improved, and the key generation rate is improved.

Description

Technical Field

[0001] The present invention relates to the field of quantum communication technology. More specifically, it relates to a key distribution method and system based on quantum communication. Background Art

[0002] Quantum Key Distribution (QKD) based on quantum communication is a technology for secure key sharing using the principles of quantum mechanics. It transmits random keys through quantum bits, uses the quantum no-cloning property (an unknown quantum state cannot be precisely replicated) to ensure that eavesdroppers cannot copy the keys, and uses quantum measurement collapse (when an eavesdropper attempts to measure the quantum state, the quantum state will change) to ensure that any eavesdropping behavior will be exposed. Taking the widely used BB84 protocol as an example, Alice (the sender) randomly generates a bit stream composed of 0 and 1, and randomly selects a basis for each bit for encoding (including 4 polarization states). After Bob (the receiver) receives the bit stream through the quantum channel, he randomly selects a basis for decoding, and then publicly discloses their respective basis information through the public channel. When the bases corresponding to the bits are the same, the bit is retained; otherwise, it is discarded. Finally, a key is generated. However, the transmission distance of the quantum channel in the BB84 protocol is limited by photon attenuation. As the transmission distance increases, the number of photons decreases sharply, resulting in a significant decrease in the secure key generation rate.

[0003] Currently, the bottleneck of long-distance communication is overcome through the Twin-Field Quantum Key Distribution (TF-QKD) protocol. That is, Alice and Bob each generate a weak coherent optical pulse encoded by a random phase. The phase encoding rule is that the phase corresponding to a 0 bit is 0 or π, and the phase corresponding to a 1 bit is π / 2 or 3π / 2. Then, the weak coherent optical pulses are sent to an intermediate station for interference measurement. The interference measurement rule is that when the phase difference is 0 or π, coherent enhancement occurs, and the detector D0 lights up. When the phase difference is π / 2 or 3π / 2, coherent attenuation occurs, and the detector D1 lights up. Finally, the interference results are announced to generate a key. Due to the setting of the intermediate station, photons do not need to traverse the entire channel, thus effectively increasing the communication distance. And because the intermediate station only performs interference measurement and announcement, even if the intermediate station is attacked, the key cannot be deciphered. Therefore, it has the advantage of measurement device independence (MDI). In addition, phase encoding is more suitable for long-distance transmission than polarization states.

[0004] However, there are phase drifts, temperature changes, fiber length changes, etc. in the optical fiber, and the TF-QKD protocol requires two optical pulses to maintain coherence, resulting in a decrease in the interference contrast and affecting the key generation rate. Summary of the Invention

[0005] The present invention provides a key distribution method and system based on quantum communication to solve the technical problems in the above background art.

[0006] The present invention provides a key distribution method based on quantum communication, including the following steps: Step S101, preprocess the historical operation data of the optical fiber to form a feature sequence as sample data for training the training samples of the first prediction model and the second prediction model; The feature sequence includes M sequence units, each sequence unit corresponding to the operation data after preprocessing at a time point, and the time intervals of each sequence unit being the same; The operation data includes: ambient temperature, bit error rate, phase drift value, and optical fiber interference factor; Step S102, obtain a phase compensation value as a sample label for training the training samples of the first prediction model and the second prediction model by transmitting a reference optical signal and combining the experimental method; Step S103, respectively perform predictions through the trained first prediction model and the second prediction model to obtain a first phase compensation prediction value and a second phase compensation prediction value; Step S104, determine that if the difference between the first phase compensation prediction value and the second phase compensation prediction value is less than the first threshold, then select the first phase compensation prediction value as the actual phase compensation value; Step S105, determine that if the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the first threshold and less than the second threshold, then perform a weighted sum calculation on the two as the actual phase compensation value; Step S106, determine that if the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the second threshold, then perform a weighted sum calculation on the two and the phase compensation value at the current time point as the actual phase compensation value.

[0007] Furthermore, the number M of sequence units, the time interval, the first threshold, the second threshold, and the third threshold of the feature sequence are all user-defined parameters.

[0008] Furthermore, the optical fiber interference factor has the following calculation formula: ; where n represents the refractive index of the optical fiber, represents the wavelength of light, L represents the length of the optical fiber, and Thermal represents the coefficient of thermal expansion.

[0009] Further, the preprocessing is specifically as follows: extracting the characteristic parameters of the operation data corresponding to each time point, concatenating the characteristic parameters with the operation data to obtain combined data, and then performing normalization processing on the combined data of each time point through the Min-max normalization method to form a characteristic sequence. The characteristic parameters include: the change rate of the environmental temperature, the change rate of the bit error rate, and the change rate of the phase drift value.

[0010] Further, the number of hidden layers of the first prediction model is the same as the number of sequence units of the characteristic sequence. The m-th hidden layer inputs the m-th sequence unit of the characteristic sequence and outputs an update vector, where 1 ≤ m ≤ M, and the dimension number of the update vector is a custom parameter; Extracting the update vector output by the M-th hidden layer and inputting it into the first classifier, and the classification space of the first classifier represents the first phase compensation prediction value; Before training the first prediction model, pre-training is added. During the pre-training process, extracting the update vector output by the M-th hidden layer and inputting it into the second classifier, and the classification space of the second classifier represents the bit error rate.

[0011] Further, the calculation formula of the m-th hidden layer includes: ; where and respectively represent the update vectors output by the m-th and the m - 1-th hidden layers, is equal to 0, represents the m-th sequence unit of the characteristic sequence input to the m-th hidden layer, and respectively represent the first control vector and the second control vector of the m-th hidden layer, and respectively represent the first weight matrix and the second weight matrix of the m-th hidden layer, tanh represents the hyperbolic tangent activation function, represents element-wise multiplication; ; where 、 and respectively represent the third weight matrix, the fourth weight matrix and the bias vector of the m-th hidden layer, sigmoid represents the sigmoid activation function; ; where MLP represents the multi-layer perceptron, and COMBINE represents the concatenation function.

[0012] Further, the phase compensation value is obtained by emitting a reference optical signal and combining with the experimental method, including the following steps: Step S201: Detect by transmitting a reference optical signal to an intermediate station to obtain a phase drift value, and perform 7 floating operations respectively to obtain 7 phase compensation values; The floating percentages are -10%, -5%, -2%, 0%, +2%, +5% and +10% respectively; Step S202: Modulate the weak coherent optical pulse according to the 7 phase compensation values respectively, and send it to the intermediate station to obtain the corresponding bit error rate, and take the phase compensation value with the minimum bit error rate; Step S203: Repeat Step S201 to Step S202 until U phase compensation values are obtained; where U is a user-defined parameter.

[0013] Further, the calculation formula of the second prediction model is as follows: ; where represents the second phase compensation prediction value output by the second prediction model, G represents the dimension number of the sequence unit, and respectively represent the g-th dimension value and the corresponding weight coefficient of the m-th sequence unit, and respectively represent the weight coefficient and the bias coefficient of the m-th sequence unit.

[0014] Further, the calculation formula for obtaining the actual phase compensation value by weighted summation calculation in Step S105 is as follows: ; where represents the first phase compensation prediction value, represents the second phase compensation prediction value; The calculation formula for obtaining the actual phase compensation value by weighted summation calculation in Step S106 is as follows: ; where Current represents the phase compensation value at the current time point.

[0015] The present invention provides a key distribution system based on quantum communication, including: A sample data construction module, which is used to preprocess the historical operation data of the optical fiber to form a feature sequence as sample data for training the first prediction model and the second prediction model; A sample label construction module, which is used to obtain a phase compensation value as a sample label for training the first prediction model and the second prediction model by transmitting a reference optical signal in combination with the experimental method; A phase compensation prediction module, which is used to respectively obtain a first phase compensation prediction value and a second phase compensation prediction value through predictions by a trained first prediction model and a second prediction model; A first judgment module, which is used to judge that the difference between the first phase compensation prediction value and the second phase compensation prediction value is less than a first threshold, and then select the first phase compensation prediction value as the actual phase compensation value; A second judgment module, which is used to judge that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the first threshold and less than a second threshold, and then perform a weighted sum calculation on the two as the actual phase compensation value; A third judgment module, which is used to judge that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the second threshold, and then perform a weighted sum calculation on the two and the phase compensation value at the current time point as the actual phase compensation value.

[0016] The beneficial effects of the present invention are as follows: On the basis of the TF-QKD protocol, the present invention comprehensively considers factors such as phase drift, temperature change, and fiber length change in the optical fiber, predicts the phase compensation value through a first prediction model (deep learning) combined with a second prediction model (polynomial equation), and uses an error control strategy to optimize the compensation decision, thereby improving the interference contrast, reducing the bit error rate, and enhancing the key generation rate. Description of the Drawings

[0017] Figure 1 is a flowchart of a key distribution method based on quantum communication according to the present invention; Figure 2 is a flowchart of obtaining a phase compensation value by transmitting a reference optical signal and combining an experimental method according to the present invention; Figure 3 is a schematic diagram of a key distribution system based on quantum communication according to the present invention.

[0018] In the figure: a sample data construction module 301, a sample label construction module 302, a phase compensation prediction module 303, a first judgment module 304, a second judgment module 305, and a third judgment module 306. Detailed Embodiments

[0019] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the protection scope of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0020] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by those of ordinary skill in the field to which the present invention pertains. In one or more embodiments of the present invention, the words such as "first", "second" and the like do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "comprising" or "including" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to represent relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0021] As Figures 1 to 3 shown, a key distribution method based on quantum communication includes the following steps: Step S101, preprocess the historical operation data of the optical fiber to form a feature sequence as sample data for training the training samples of the first prediction model and the second prediction model; The feature sequence includes M sequence units, each sequence unit corresponding to the operation data after preprocessing at a time point, and the time intervals of each sequence unit are the same; The operation data includes: ambient temperature, bit error rate, phase drift value, and optical fiber interference factor; Step S102, obtain the phase compensation value by transmitting a reference optical signal in combination with the experimental method as the sample label for training the training samples of the first prediction model and the second prediction model; Step S103, respectively perform predictions through the trained first prediction model and the second prediction model to obtain a first phase compensation prediction value and a second phase compensation prediction value; Step S104, determine that the difference between the first phase compensation prediction value and the second phase compensation prediction value is less than the first threshold, then select the first phase compensation prediction value as the actual phase compensation value; Step S105, determine that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the first threshold and less than the second threshold, then perform a weighted sum calculation on the two as the actual phase compensation value; Step S106, determine that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the second threshold, then perform a weighted sum calculation on the two and the phase compensation value at the current time point as the actual phase compensation value.

[0022] In an embodiment of the present invention, the number M of sequence units of the feature sequence, the time interval, the first threshold, the second threshold, and the third threshold are all user-defined parameters. Preferably, M is set to 12 and the time interval is set to 5 seconds. Then, the total acquisition duration for constructing a feature sequence is 12×5 = 60 seconds, and the first threshold, the second threshold, and the third threshold are respectively set to 10%, 20%, and 30% of the historical maximum phase compensation value.

[0023] In an embodiment of the present invention, the ambient temperature is acquired by a fiber optic temperature sensor; the bit error rate is represented by the ratio of the number of error bits to the total number of bits statistically counted within a user-defined preset time period. For example, the user-defined preset time period is set to 60 seconds; the phase drift value is detected by transmitting a reference optical signal to an intermediate station, where the reference optical signal is a strong coherent light wave; the fiber interference factor has the following calculation formula: ; where n represents the fiber refractive index, represents the wavelength of light, L represents the fiber length, and Thermal represents the thermal expansion coefficient.

[0024] It should be noted that the fiber refractive index, the fiber length, and the thermal expansion coefficient are all positively correlated with the fiber interference factor. The larger the fiber refractive index, the slower the optical signal propagates in the fiber; the longer the fiber length, the greater the influence of temperature and mechanical stress, and the lower the transmission efficiency; the thermal expansion coefficient is related to the fiber material and the ambient temperature. The larger the thermal expansion coefficient, the more obvious the influence of temperature on the fiber.

[0025] In an embodiment of the present invention, the preprocessing is specifically as follows: extracting the characteristic parameters of the operation data corresponding to each time point, splicing the characteristic parameters with the operation data to obtain combined data, and then performing normalization processing on the combined data of each time point through the Min-max normalization method to form a feature sequence. The characteristic parameters include: the change rate of the ambient temperature, the change rate of the bit error rate, and the change rate of the phase drift value. That is, each sequence unit is represented by a vector with a dimension of 7, namely the normalized ambient temperature, bit error rate, phase drift value, fiber interference factor, change rate of the ambient temperature, change rate of the bit error rate, and change rate of the phase drift value.

[0026] In an embodiment of the present invention, the number of hidden layers of the first prediction model is the same as the number of sequence units of the feature sequence. The m-th hidden layer inputs the m-th sequence unit of the feature sequence and outputs an update vector, where 1≤m≤M. The dimension number of the update vector is a user-defined parameter. Preferably, the dimension number of the update vector is set to 16; extracting the update vector output by the M-th hidden layer and inputting it into the first classifier, and the classification space of the first classifier represents the first phase compensation prediction value; Before training the first prediction model, pre-training is added. During the pre-training process, the updated vector output by the M-th hidden layer is extracted and input into the second classifier, and the classification space of the second classifier represents the bit error rate.

[0027] It should be noted that the sample labels of the training samples used for pre-training are relatively easy to obtain, that is, the bit error rate. The target task of pre-training (predicting the bit error rate) is similar to the target task of actual training (predicting the phase compensation value). After pre-training, fine-tuning the parameters of the first prediction model can significantly reduce the number of training samples in subsequent actual training, greatly improve the actual training speed of the model, and help the model converge better.

[0028] In an embodiment of the present invention, the calculation formula of the m-th hidden layer includes: ; where and respectively represent the updated vectors output by the m-th and m - 1-th hidden layers, is equal to 0, represents the m-th sequence unit of the feature sequence input to the m-th hidden layer, and respectively represent the first control vector and the second control vector of the m-th hidden layer, and respectively represent the first weight matrix and the second weight matrix of the m-th hidden layer, and tanh represents the hyperbolic tangent activation function, represents element-wise multiplication; ; where , and respectively represent the third weight matrix, the fourth weight matrix and the bias vector of the m-th hidden layer, and sigmoid represents the sigmoid activation function; ; where MLP represents a multi-layer perceptron and COMBINE represents a concatenation function.

[0029] It should be noted that the weight matrices and bias vectors in the first prediction model are all learnable hyperparameters. For example, if the dimension number of the updated vector is set to 16, then the dimension numbers of the first control vector and the second control vector both need to be set to 16, the dimension number of the bias vector needs to be set to 16, and each sequence unit is represented by a vector with a dimension number of 7. Then the sizes of the first weight matrix and the third weight matrix both need to be set to 7×16, and the sizes of the second weight matrix and the fourth weight matrix both need to be set to 16×16.

[0030] In an embodiment of the present invention, as Figure 2 shown, the phase compensation value is obtained by emitting a reference optical signal in combination with an experimental method, including the following steps: Step S201: Detect the phase drift value by emitting a reference optical signal to the intermediate station, and perform 7 floating operations respectively to obtain 7 phase compensation values; The floating percentages are -10%, -5%, -2%, 0%, +2%, +5% and +10% respectively; Step S202: Modulate the weak coherent optical pulse according to the 7 phase compensation values respectively, and send it to the intermediate station to obtain the corresponding bit error rate, and take the phase compensation value with the minimum bit error rate; Step S203: Repeat Step S201 to Step S202 until U phase compensation values are obtained; where U is a user-defined parameter. Preferably, U is set to 500.

[0031] In an embodiment of the present invention, phase synchronization can also be performed by an optical frequency comb (OFC) or a high-precision phase-locked loop (PLL), but their deployment costs are relatively high. Obtaining the phase compensation value by emitting a reference optical signal in combination with an experimental method can effectively reduce costs, which will not be elaborated here.

[0032] In an embodiment of the present invention, the calculation formula of the second prediction model is as follows: ; where represents the second phase compensation prediction value output by the second prediction model, G represents the number of dimensions of the sequence unit, and respectively represent the g-th dimension value and the corresponding weight coefficient of the m-th sequence unit, and respectively represent the weight coefficient and the bias coefficient of the m-th sequence unit. The weight coefficient and the bias coefficient in the second prediction model are both obtained by solving through the gradient descent algorithm, and can also be obtained by solving through the least squares method, which will not be elaborated here.

[0033] In an embodiment of the present invention, the calculation formula for obtaining the actual phase compensation value by weighted summation in step S105 is as follows: ; where represents the first phase compensation prediction value, represents the second phase compensation prediction value; The calculation formula for obtaining the actual phase compensation value by weighted summation in step S106 is as follows: ; where Current represents the phase compensation value at the current time point.

[0034] In an embodiment of the present invention, as Figure 3 shown, a quantum communication-based key distribution system includes: A sample data construction module 301, which is used to preprocess the historical operation data of the optical fiber to form a feature sequence as sample data for training the first prediction model and the second prediction model; A sample label construction module 302, which is used to obtain the phase compensation value as the sample label for training the first prediction model and the second prediction model by transmitting a reference optical signal and combining the experimental method; A phase compensation prediction module 303, which is used to respectively perform predictions through the trained first prediction model and the second prediction model to obtain a first phase compensation prediction value and a second phase compensation prediction value; A first judgment module 304, which is used to judge that if the difference between the first phase compensation prediction value and the second phase compensation prediction value is less than the first threshold, then select the first phase compensation prediction value as the actual phase compensation value; A second judgment module 305, which is used to judge that if the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the first threshold and less than the second threshold, then perform a weighted sum calculation on the two as the actual phase compensation value; A third judgment module 306, which is used to judge that if the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the second threshold, then perform a weighted sum calculation on the two and the phase compensation value at the current time point as the actual phase compensation value.

[0035] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.

Claims

1. A key distribution method based on quantum communication, characterized in that, It includes the following steps: Step S101, preprocess the historical operation data of the optical fiber to form a feature sequence as the sample data for training the training samples of the first prediction model and the second prediction model; The feature sequence includes M sequence units, each sequence unit corresponds to the operation data after preprocessing at a time point, and the time intervals of each sequence unit are the same; The operation data includes: ambient temperature, bit error rate, phase drift value, and optical fiber interference factor; Step S102, obtain the phase compensation value by transmitting a reference optical signal in combination with the experimental method as the sample label for training the training samples of the first prediction model and the second prediction model; Step S103, respectively perform predictions through the trained first prediction model and the second prediction model to obtain the first phase compensation prediction value and the second phase compensation prediction value; Step S104, determine that the difference between the first phase compensation prediction value and the second phase compensation prediction value is less than the first threshold, then select the first phase compensation prediction value as the actual phase compensation value; Step S105, determine that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the first threshold and less than the second threshold, then perform weighted summation calculation on the two as the actual phase compensation value; Step S106, determine that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the second threshold, then perform weighted summation calculation on the two and the phase compensation value at the current time point as the actual phase compensation value.

2. A quantum communication-based key distribution method according to claim 1, characterized in that, The number M of sequence units, the time interval, the first threshold, the second threshold, and the third threshold of the feature sequence are all user-defined parameters.

3. A quantum communication-based key distribution method according to claim 1, characterized in that, Optical fiber interference factor The calculation formula is as follows: ； where n represents the refractive index of the optical fiber, represents the wavelength of light, L represents the length of the optical fiber, and Thermal represents the coefficient of thermal expansion.

4. A quantum communication-based key distribution method according to claim 1, characterized in that, The preprocessing is specifically as follows: extract the characteristic parameters of the operation data corresponding to each time point, splice the characteristic parameters with the operation data to obtain combined data, and then perform normalization processing on the combined data of each time point through the Min-max normalization method to form a feature sequence. The characteristic parameters include: the change rate of ambient temperature, the change rate of bit error rate, and the change rate of phase drift value.

5. A quantum communication-based key distribution method according to claim 1, characterized in that, The number of hidden layers of the first prediction model is the same as the number of sequence units of the feature sequence. The mth hidden layer inputs the mth sequence unit of the feature sequence and outputs an update vector, where 1 ≤ m ≤ M, and the dimension number of the update vector is a user-defined parameter; Extract the update vector output by the Mth hidden layer and input it into the first classifier. The classification space of the first classifier represents the first phase compensation prediction value; Before training the first prediction model, pre-training is added. During the pre-training process, extract the update vector output by the Mth hidden layer and input it into the second classifier. The classification space of the second classifier represents the bit error rate.

6. A quantum communication-based key distribution method according to claim 5, characterized in that The calculation formula of the mth hidden layer includes: ； wherein and respectively represent the updated vectors output by the m-th and (m - 1)-th hidden layers, is equal to 0, represents the m-th sequence unit of the feature sequence input to the m-th hidden layer, and respectively represent the first control vector and the second control vector of the m-th hidden layer, and respectively represent the first weight matrix and the second weight matrix of the m-th hidden layer, and tanh represents the hyperbolic tangent activation function, represents element-wise multiplication; ； Among them , and respectively represent the third weight matrix, the fourth weight matrix and the bias vector of the m-th hidden layer, and sigmoid represents the sigmoid activation function; ； where MLP represents a multi-layer perceptron, and COMBINE represents a splicing function.

7. A quantum communication-based key distribution method according to claim 1, characterized in that, Obtain the phase compensation value by transmitting a reference optical signal in combination with the experimental method, including the following steps: Step S201, transmit a reference optical signal to the intermediate station for detection to obtain the phase drift value, and perform 7 floating operations respectively to obtain 7 phase compensation values; The floating percentages are -10%, -5%, -2%, 0%, +2%, +5%, and +10% respectively; Step S202: Modulate the weak coherent optical pulse according to 7 phase compensation values respectively, and send it to the intermediate station to obtain the corresponding bit error rate, and select the phase compensation value with the minimum bit error rate. Step S203: Repeat Step S201 to Step S202 until U phase compensation values are obtained. Where U is a user-defined parameter.

8. A quantum communication-based key distribution method according to claim 1, characterized in that The calculation formula of the second prediction model is as follows: ； Among them represents the second phase compensation prediction value output by the second prediction model, G represents the number of dimensions of the sequence unit, and respectively represent the g-th dimension value and the corresponding weight coefficient of the m-th sequence unit, and respectively represent the weight coefficient and the bias coefficient of the m-th sequence unit.

9. A quantum communication-based key distribution method according to claim 1, characterized in that The actual phase compensation value is obtained by weighted summation calculation in step S105 The calculation formula is as follows: ； Among them represents the first phase compensation prediction value, represents the second phase compensation prediction value; The calculation formula for obtaining the actual phase compensation value through weighted summation calculation in Step S106 is as follows: ； Where Current represents the phase compensation value at the current time point.

10. A key distribution system based on quantum communication, characterized in that, Implementing a quantum communication-based key distribution method according to any one of claims 1 to 9, including: A sample data construction module, which is used to preprocess the historical operation data of the optical fiber to form a feature sequence as sample data for training the first prediction model and the second prediction model. A sample label construction module, which is used to obtain the phase compensation value by transmitting a reference optical signal in combination with the experimental method as sample labels for training the first prediction model and the second prediction model. A phase compensation prediction module, which is used to respectively perform predictions through the trained first prediction model and the second prediction model to obtain a first phase compensation prediction value and a second phase compensation prediction value. A first judgment module, which is used to judge that the difference between the first phase compensation prediction value and the second phase compensation prediction value is less than the first threshold, and then select the first phase compensation prediction value as the actual phase compensation value. A second judgment module, which is used to judge that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the first threshold and less than the second threshold, and then perform a weighted summation calculation on the two as the actual phase compensation value. A third judgment module, which is used to judge that the difference between the first phase compensation prediction value and the second phase compensation prediction value is greater than or equal to the second threshold, and then perform a weighted summation calculation on the two and the phase compensation value at the current time point as the actual phase compensation value.

Images