A dual-field quantum key distribution method and system

Abstract

This invention provides a dual-field quantum key distribution method and system, relating to the field of quantum information technology, comprising the following steps: two transmitting ends prepare coherent states and send them to a detection end; the detection end processes the states and sends the detection results to the first and second transmitting ends; a long short-term memory network (LSTM) is designed to track, predict, and optimize parameters at subsequent time nodes; the two transmitting ends calculate key information based on the optimized parameters predicted by the LSM and the measurement results from the detection end. This invention reduces the adjustment time for optimization parameters, simplifies the protocol steps, and improves the overall efficiency of the dual-field quantum key distribution system.

Description

Technical Field

[0001] This invention relates to the field of quantum information technology, and particularly to a dual-field quantum key distribution method and system. Background Technology

[0002] Quantum communication technology is an unconditionally secure communication system that can be used by legitimate users, based on the principles of quantum physics and combined with Shannon's "one-time pad" theorem.

[0003] Quantum key distribution is a core technology for realizing quantum communication. Relay-free quantum key distribution systems have a linear limit on the key rate (PLOB limit). In recent years, researchers have proposed a dual-field quantum protocol, TF-QKD, which can break the PLAB limit in terms of key rate.

[0004] In the quantum key distribution process, the light intensity and phase of the two signal states are inconsistent. The light intensity of the two signal states will affect the performance of TF-QKD, and the parameters need to be optimized. When optimizing the existing parameters, the exhaustive search method is used to try all the parameters. Although the optimal parameters can be found, it takes a lot of time and computing resources, resulting in a decrease in the overall efficiency of the system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the above-mentioned technologies. This invention proposes a dual-field quantum key distribution method and system, which aims to solve the technical problem that the existing technology can find the optimal parameters but consumes a lot of time and computing resources, resulting in a decrease in the overall efficiency of the system.

[0006] This invention specifically provides the following technical solution: a public two-field quantum key distribution method, comprising the following steps:

[0007] The encoding mode or the decoy mode is selected by two transmitters according to probabilities p and 1-p, and the coherent state is prepared by the encoding mode or the decoy mode.

[0008] The Long Short-Term Memory Network model is used to change the values ​​of environmental parameters to optimize parameter prediction, obtain a dataset containing the optimized parameters, and modulate the dataset of optimized parameters into a coherent state.

[0009] The modulated coherent state is sent to the detection end for measurement, and the obtained successful measurement event is sent to two sending ends. The original key is extracted through the mode selected by each of the two sending ends.

[0010] Using the two sending ends, a quantitative amount of data is selected from the decoy pattern to estimate the key rate and error rate, and the final security key is generated from the original key.

[0011] Preferably, the preparation of the coherent state through the encoding mode or the decoy mode includes the following steps:

[0012] In the encoding mode, the two transmitting ends encode bit 0 or 1 in a coherent state;

[0013] In decoy mode, each of the two transmitters prepares a coherent state with a random phase.

[0014] The two transmitting ends include a first transmitting end and a second transmitting end, and the coherent state of the encoding mode of the first transmitting end is: The coherent state of the encoding mode of the second transmitter is Where θ a1 and θ a2 ∈{0,π},φ a1 and φ a2 ∈[0,2π), where α1 and α2 are the amplitudes of the coherent states of the coding mode.

[0015] Preferably, in the encoding mode, the two transmitting ends encode bit 0 or 1 in a coherent state, including the following steps:

[0016] When θ a1 and θ a2 When θ = 0, the encoded bit is 0; when θ = 0, the encoded bit is 0. a1 and θ a2 When π = 1, the encoding bit is chosen to be 1.

[0017] Preferably, in the decoy mode, each of the two transmitters prepares a coherent state with random phase, including the following steps:

[0018] The coherent state of the decoy mode at the first transmitter is: The coherent state of the decoy mode at the second transmitter is

[0019] Where k1 and k2 ∈ (0,1,...M), and β1 and β2 represent the amplitudes of the coherent states of the decoy mode.

[0020] Preferably, before using the Long Short-Term Memory (LSTM) network model to change the values ​​of environmental parameters for optimized parameter prediction, the LTM network model needs to be trained, including the following steps:

[0021] The training dataset is obtained by changing the values ​​of each environmental parameter with a uniform step size.

[0022] The feature data of the training dataset is input into the long short-term memory network model, and the light intensity α of the encoding mode and the light intensity β of the decoy mode are used as the output label data to obtain the corresponding optimization parameter prediction results.

[0023] The feature data are optimization parameters, including: the coherent state amplitude α of the coding mode and the coherent state amplitude β of the decoy mode; the environmental parameters include: the optical error rate e and the detector dark count P. d The total number of signals is N, the number of phase divisions is M, and the transmission distance is L.

[0024] Preferably, the step of sending the modulated coherent state to the detection end for measurement includes the following steps: defining an event in the published measurement results where only one single-photon detector responds as a successful measurement event, and the others as unsuccessful measurement events, and discarding the detection results of the unsuccessful measurement events.

[0025] Preferably, the step of sending the obtained successful measurement event to two sending ends and extracting the original key through the mode selected by each of the two sending ends includes the following steps:

[0026] If both transmitters are in encoded mode, each will retain the corresponding bits as key information, and when the second single-photon detector responds, the second transmitter will flip the bits.

[0027] If both transmitters are in decoy mode, they will publish their respective selected phases.

[0028] Preferably, after estimating the key rate and error rate by selecting a quantitative amount of data from the decoy mode using the two sending ends, error correction and privacy amplification processing are performed based on the measurement results and parameter estimates published by the detection end, and the key rate is calculated.

[0029] Preferably, the calculation of the key rate is specifically expressed as follows:

[0030]

[0031] Where R is the key rate in the final key, e μ Let h(x,y) = -xlog2x - ylog2y + (x + y)log2(x + y) represent the error rate, and Q be the binary entropy function. μ This represents the probability that the first sender obtains the original key in the encoded mode, and f represents the error correction efficiency. This is the upper limit of the amount of information leakage.

[0032] Preferably, the present invention further includes a dual-field quantum key distribution system, comprising:

[0033] The first preparation unit is used to select an encoding mode or a decoy mode through two transmitters according to probabilities p and 1-p, and prepare a coherent state through the encoding mode or the decoy mode.

[0034] The second preparation unit is used to use a long short-term memory network model to change the values ​​of environmental parameters to perform optimization parameter prediction, obtain a dataset containing optimization parameters, and modulate the dataset of optimization parameters into a coherent state.

[0035] The original key extraction unit is used to send the modulated coherent state to the detection end for measurement, send the obtained successful measurement event to two sending ends, and extract the original key through the mode selected by each of the two sending ends.

[0036] The final key generation unit is used to select a quantitative amount of data from the decoy pattern using the two sending ends to estimate the key rate and error rate, and generate the final security key from the original key.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] This invention uses two transmitters to select either an encoding mode or a decoy mode based on probability, and prepares coherent states using their respective selected modes. Simultaneously, to reduce parameter optimization time, a long short-term memory network is used to change the values ​​of environmental parameters to predict and optimize parameters, outputting the optimal parameters. This solves the gradient problem and the long sequence problem, significantly reducing parameter optimization time, simplifying protocol steps, and mapping the optimal parameters to the coherent state for output. This enables dual-field quantum key distribution, reducing overall computational resources and improving the overall system efficiency. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the experimental scheme for dual-field quantum key distribution of the present invention;

[0040] Figure 2 This is a schematic diagram of the Long Short-Term Memory network of this invention;

[0041] Figure 3 This is a structural diagram of the memory unit of the Long Short-Term Memory Network of this invention;

[0042] Figure 4 This is a diagram of the dataset composition of the present invention; where (a) is the feature data content and (b) is the label data content;

[0043] Figure 5 This is the training error histogram of the LSTM network of the present invention on the dataset;

[0044] Figure 6 This is a histogram showing the prediction performance of the LSTM network of this invention on the dataset;

[0045] Figure 7 This is the overall error histogram of the LSTM network on the dataset according to the present invention;

[0046] Figure 8This is a diagram showing the prediction effect of the LSTM network of the present invention on the light intensity α of the coding pattern;

[0047] Figure 9 This is a graph showing the prediction effect of the LSTM network of the present invention on the light intensity β of the decoy pattern. Detailed Implementation

[0048] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. The following examples are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0049] Quantum key distribution (QKD) establishes an unconditionally secure communication system between legitimate users based on Shannon's "one-time pad" principle and the fundamental theorem of quantum physics. Relay-free QKD systems have a linear upper bound on the key rate, known as the PLOB limit. TF-QKD systems can break this PLOB limit. TF-QKD systems involve two signal states (encoded mode and decoy mode), with inconsistent light intensities and phases. The light intensity of these two states affects the performance of TF-QKD, so optimizing these parameters is necessary to improve system performance. A common approach is to exhaustively search the parameter range to obtain the optimal parameters, which is time-consuming and computationally resource-intensive. This invention addresses these issues by proposing a dual-field quantum key distribution system and method based on Long Short-Term Memory (LSTM) networks. Using LSM networks to predict the optimal parameters reduces optimization time, simplifies protocol steps, and improves overall system efficiency.

[0050] This invention relates to a dual-field quantum key distribution method and system based on long short-term memory networks, which involves two legitimate transmitters (a first transmitter and a second transmitter) and an untrusted detector.

[0051] See Figure 1 The TF-QKD experimental scheme, a dual-field quantum key distribution method, includes the following steps:

[0052] S1: Select the encoding mode or the decoy mode through two transmitters according to probabilities p and 1-p, and prepare the coherent state through the encoding mode or the decoy mode.

[0053] Specifically:

[0054] Preparation of coherent state: In the encoding mode, the two transmitting ends encode bits 0 or 1 in a coherent state; when θ a1 and θ a2 When θ = 0, the encoded bit is 0; when θ = 0, the encoded bit is 0. a1 and θ a2When π = 1, the coded bit is set to 1. In decoy mode, each of the two transmitters prepares a coherent state with random phase; the two transmitters include a first transmitter and a second transmitter.

[0055] The first and second lasers transmit an optical pulse. In the coding mode, the first (second) random number generator generates a random number. The two optical pulses arrive at the first (second) phase modulator along the channel and are modulated into phase φ according to the random number. a1 and φ a2 φ a1 and φ a2 ∈[0,2π); then after passing through the first (second) phase modulator, the two optical pulses are modulated at the first (second) intensity modulator to produce optical intensities (amplitudes of the coherent state of the coded mode) α1 and α2, which are provided by the LSTM module. Then, the optical pulses pass through the first (second) optical attenuator, and the first (second) transmitter obtains the coherent state. and In decoy mode, the first (second) random number generator generates random numbers k1, k2 ∈ (0, 1, ... M). Two optical pulses arrive at the first (second) phase modulator along the channel and the phase of the optical pulses is modulated according to the random numbers. and M represents the number of phase divisions. After passing through the first (second) phase modulator, the two optical pulses are modulated at the first (second) intensity modulator to obtain the light intensities (amplitudes of the coherent state in the decoy mode) β1 and β2. The light intensities β1 and β2 are provided by the LSTM module. Then, the optical pulses pass through the first (second) optical attenuator, and the first (second) transmitter obtains the coherent state. and

[0056] S2: Using a Long Short-Term Memory (LSTM) network model, the environmental parameters are modified to predict optimized parameters, resulting in a dataset containing these optimized parameters. This dataset is then modulated into a coherent state. The LTM network learns the mapping from environmental parameters to optimized parameters. The optimized parameters include: the coherent state amplitude α of the encoded mode and the coherent state amplitude β of the decoy mode; the environmental parameters include: the optical error rate e and the detector dark count P. d The total number of signals is N, the number of phase divisions is M, and the transmission distance is L.

[0057] Specifically, the steps include the following:

[0058] The training dataset is obtained by changing the values ​​of each environmental parameter with a uniform step size.

[0059] The feature data from the training dataset is input into the Long Short-Term Memory (LSTM) network model, and the light intensity α of the encoding pattern and the light intensity β of the decoy pattern are used as the output label data to obtain the corresponding optimized parameter prediction results. Here, the feature data represents the optimized parameters.

[0060] The Long Short-Term Memory (LSTM) network model is specifically as follows:

[0061] The LSTM module, or Long Short-Term Memory network model, is a time-series-based neural network model with predictive capabilities. Its basic principles can be found in [link to documentation]. Figure 2 Each index represents a time point, where t is the current time, t-1 is the previous time, and t+1 is the next time. Ct represents the conveyor belt information, which records important information from previous times and transmits it along the conveyor belt. It can be seen that the output at each time point is affected by three pieces of information: the current state, the current input, and the current conveyor belt information. These three pieces of information are filtered by memory cells to produce the final output.

[0062] See the structure of memory cells. Figure 3 Its structure contains three gated units: the input gate, the forget gate, and the output gate. The output value of each gate is calculated by multiplying the weight matrix and the input signal by the dot product, and by using the activation function (usually the sigmoid function). Finally, the output value is multiplied by the value of the memory cell to perform the final calculation.

[0063] Forget gate: This determines whether information in memory cells is forgotten; in other words, it controls whether the information stored in memory cells will disappear. Its calculation process is as follows:

[0064] f t =σ(W f ·[h t-1 ,x t ]+b f ), where Wf is the weight parameter, ht-1 is the output of the previous memory cell, xt is the input at the current time, and bf is the bias parameter.

[0065] Input gate: Determines whether the current input information is written to the memory cell; in other words, it controls the impact of the input information on the memory cell. Its calculation process is as follows:

[0066] i t =σ(W i ·[h t-1 ,x t ]+b i ), Where σ and tanh are activation functions, Wf and Wc are weight parameters, and bi and bC are bias parameters.

[0067] Output gate: Determines whether information from the memory cell is output; in other words, it controls whether the information stored in the memory cell will affect subsequent network layers. Its calculation process is as follows:

[0068] o t =σ(W o ·[h t-1 ,x t ]+b o ), h t =o t *tanh(C t ), where Wo is the weight parameter and bo is the bias parameter.

[0069] The inputs to the LSTM module are environmental parameters: optical error rate e, detector dark count rate Pd, total number of signals N, number of phase divisions M, and transmission distance L. The outputs are the modulated light intensities α and β in the coded mode and decoy mode, respectively. The implementation process is as follows:

[0070] (1) Step 1, Data Acquisition. By changing the values ​​of environmental parameters, a dataset under different parameters is obtained. The sampling points of this invention are set as follows: the optical error rate e ranges from [0.01, 0.04], and sampling is performed with a step size of 0.01; the detector dark count Pd ranges from [10...]. -10 10 -6 The sampling points are arranged according to the exponential range, i.e., {10} -10 10 -9 10 -8 10 -7 10 -6 The total number of signals ranges from

[10] . 8 10 12 ], sampling points are {10 8 10 9 10 10 10 11 10 12 The phase division number ranges from [10, 20], the sampling step size is 1, the transmission distance L ranges from [0, 400 km], and the sampling step size is 40 km. The optimal light intensities α and β were obtained through exhaustive search, ultimately yielding a total of 11,000 data samples.

[0071] See sample data structure Figure 4 The sample includes feature data for input and label data as output. Figure 4 The types of feature data in (a) include: optical error rate e, detector dark count rate Pd, total number of signals N, number of phase divisions M, and transmission distance L. Figure 4 (b) The types of tag data include: light intensity α of the encoding mode and light intensity β of the decoy mode.

[0072] (2) Step 2, data preprocessing: After randomly shuffling the sample order, the training set and test set are divided in a ratio of 8:2. The training set is used for the LSTM network to learn the mapping relationship between environmental parameters and light intensity, and the test set is used to test the training effect of the network.

[0073] (3) Step 3, Training. Train the LSTM network using the data processed in Step 2. See the training error histogram of the LSTM network for details. Figure 5 , Figure 5 The horizontal axis represents the magnitude of the error, and the vertical axis represents the number of samples. It can be seen that the errors are mostly concentrated around 0. See the error histogram for the test set. Figure 6 , Figure 6 The horizontal axis represents the magnitude of the error, and the vertical axis represents the number of samples. It can be seen that the errors are concentrated around 0.001 and 0.002. See [link to full data sample error table] for more information. Figure 7 , Figure 7 The horizontal axis represents the magnitude of the error, and the vertical axis represents the number of samples. It can be seen that the error is concentrated around the 0 point.

[0074] (4) Step 4, Prediction. The trained neural network is used to predict the optimized parameters. Specifically, the network automatically predicts the light intensity α of the encoded mode and the light intensity β of the decoy mode by inputting the corresponding environmental parameters. See [link to documentation] for the prediction effect of the light intensity α of the encoded mode. Figure 8 , Figure 8 The horizontal axis represents the sample number, and the vertical axis represents the predicted value of the sample. The dashed line represents the target value curve, and the solid line represents the predicted value curve. For the light intensity β prediction effect of the decoy mode, please refer to [link to documentation]. Figure 9 The horizontal axis represents the sample number, and the vertical axis represents the predicted value of the sample. The dashed line represents the target value curve, and the solid line represents the predicted value curve.

[0075] S3: Send the modulated coherent state to the detection end for bell measurement, send the obtained successful measurement event to the two sending ends, and extract the original key through the mode selected by each of the two sending ends.

[0076] This step is specifically as follows:

[0077] An event in the published measurement results that has only one single-photon detector response is called a successful measurement event, and the others are unsuccessful measurement events. The detection results of unsuccessful measurement events are discarded.

[0078] If both transmitters are in coded mode, they each retain the corresponding bits as key information, and when the second single-photon detector responds, the second transmitter flips the bits; if both transmitters are in decoy mode, they publish their respective selected phases.

[0079] Specific implementation process:

[0080] Detection: The first (second) transmitters publish their respective selected modes. The coherent state of the encoded mode or decoy mode prepared in step 1 by the first (second) transmitter is sent to an optical beam splitter for Bell measurement. The first (second) single-photon detector responds when it detects a single-photon signal. A successful measurement event is defined as when only one of the single-photon detectors responds.

[0081] In the encoding mode, the first (second) laser at the first (second) transmitter generates an optical pulse, and the random number generator generates a random number. The optical pulse is then sent to the first (second) phase modulator, where its phase φ1 (φ2) is modulated according to the random number. After passing through the phase modulator, the optical pulse is sent to the first (second) intensity modulator to modulate its intensity α1 (α2), and then attenuated to the quantum level by the first (second) optical attenuator. In the decoy mode, the first (second) laser at the first (second) transmitter generates an optical pulse, and the random number generator generates a random number. The optical pulse is then sent to the first (second) phase modulator, where its phase is modulated according to the random number. The light pulses are sent to the first (second) intensity modulator after passing through the phase modulator, and their light intensity β1 (β2) is modulated. The light pulses are then attenuated to the quantum level by the first (second) light attenuator.

[0082] Extracting the original key: In a successful measurement event, when both the first and second transmitters are in encoding mode, each transmitter retains its own bits, specifying the first transmitter's bit value as a1, while the second transmitter inverts its bit value a2; that is, when a2 is 1, the second transmitter's bit value is 0, and when a2 is 0, the second transmitter's bit value is 1. In a successful measurement event, when both the first and second transmitters are in decoy mode, both parties publish the phase and intensity of their respective chosen coherent states.

[0083] S4: Using two sending ends, select a quantitative amount of data from the decoy pattern to estimate the key rate and error rate, and generate the final security key from the original key.

[0084] Parameter estimation is performed based on the coherent state information of the decoy mode published by both the sending and receiving ends.

[0085] Generate the final key: The key rate is calculated from the parameter estimation and Bell measurement results. When the key rate is at a secure level, both the sending and receiving parties can perform error correction and privacy amplification on the original key to obtain the final secure key.

[0086] The formula for calculating the key rate is as follows: Where e μ Let h(x,y) = -xlog2x - ylog2y + (x + y)log2(x + y) represent the error rate, and Q be the binary entropy function. μThis represents the probability that the first sender obtains the original key in the encoded mode, and f represents the error correction efficiency. This is the upper limit of the amount of information leakage.

[0087] The present invention also includes a dual-field quantum key distribution system, comprising: a first preparation unit, a first preparation unit, a raw key extraction unit, and a final key generation unit.

[0088] The system comprises the following components: A first preparation unit selects either an encoding mode or a decoy mode from two transmitters according to probabilities p and 1-p, and prepares a coherent state using either mode. A measurement unit uses a long short-term memory network model to modify environmental parameters for prediction, obtaining a dataset including optimized parameters, mapping the environmental parameters to the optimized parameters, and modulating this mapping onto the coherent state. A raw key extraction unit sends the coherent state to a detection end for measurement, sends successful measurement events to both transmitters, and extracts the raw key using the mode selected by each transmitter. A final key generation unit uses quantitative data selected from the decoy mode by both transmitters to estimate the key rate and error rate, and generates the final security key from the raw key.

[0089] The above-described embodiments are merely preferred embodiments of the present invention, and the scope of protection of the present invention is not limited thereto. Any simple changes or equivalent substitutions of the technical solutions that can be obviously obtained by those skilled in the art within the scope of the technology disclosed in the present invention shall fall within the scope of protection of the present invention.

Claims

1. A dual-field quantum key distribution method, characterized in that, Includes the following steps: The encoding mode or the decoy mode is selected by two transmitters according to probabilities p and 1-p, and the coherent state is prepared by the encoding mode or the decoy mode. The Long Short-Term Memory Network model is used to change the values ​​of environmental parameters to optimize parameter prediction, obtain a dataset containing the optimized parameters, and modulate the dataset of optimized parameters into a coherent state. The modulated coherent state is sent to the detection end for measurement, and the obtained successful measurement event is sent to two sending ends. The original key is extracted through the mode selected by each of the two sending ends. Using the two sending ends, a quantitative amount of data is selected from the decoy pattern to estimate the key rate and error rate, and the final security key is generated from the original key.

2. The dual-field quantum key distribution method as described in claim 1, characterized in that, The preparation of coherent states through the encoding mode or decoy mode includes the following steps: In the encoding mode, the two transmitting ends encode bit 0 or 1 in a coherent state; In decoy mode, each of the two transmitters prepares a coherent state with a random phase. The two transmitting ends include a first transmitting end and a second transmitting end, and the coherent state of the encoding mode of the first transmitting end is: The coherent state of the encoding mode of the second transmitter is ,in , , and The amplitude of the coherent state of the coding mode.

3. The dual-field quantum key distribution method as described in claim 2, characterized in that, In the encoding mode, the two transmitting ends encode bit 0 or 1 in a coherent state, including the following steps: when and When, the encoding bit is 0; when and When that happens, the encoding bit is set to 1.

4. The dual-field quantum key distribution method as described in claim 2, characterized in that, In the decoy mode, each of the two transmitters prepares a coherent state with random phase, including the following steps: The coherent state of the decoy mode at the first transmitter is ( The coherent state of the decoy mode at the second transmitter is ( ; in and , and The amplitude of the coherent state in the decoy mode is represented by M, where M is the number of phase divisions.

5. The dual-field quantum key distribution method as described in claim 1, characterized in that, Before using the Long Short-Term Memory (LSTM) network model to change the values ​​of environmental parameters and optimize parameter prediction, the LTM network model needs to be trained, including the following steps: The training dataset is obtained by changing the values ​​of each environmental parameter with a uniform step size. The feature data of the training dataset is input into the long short-term memory network model, and the light intensity of the encoding pattern is used. Light intensity of decoy mode The label data is used as output to obtain the corresponding optimized parameter prediction results; The feature data are optimization parameters, which include: the amplitude of the coherent state of the coding mode. Amplitude of coherent state in decoy mode Environmental parameters include: optical error rate e, detector dark count P. d The total number of signals is N, the number of phase divisions is M, and the transmission distance is L.

6. The dual-field quantum key distribution method as described in claim 1, characterized in that, The step of sending the modulated coherent state to the detection end for measurement includes the following steps: defining an event in the published measurement results where only one single-photon detector responds as a successful measurement event, and the others as unsuccessful measurement events, and discarding the detection results of the unsuccessful measurement events.

7. The dual-field quantum key distribution method as described in claim 2, characterized in that, The process of sending the obtained successful measurement event to two sending ends and extracting the original key through the mode selected by each of the two sending ends includes the following steps: If both transmitters are in encoded mode, each will retain the corresponding bits as key information, and when the second single-photon detector responds, the second transmitter will flip the bits. If both transmitters are in decoy mode, they will publish their respective selected phases.

8. The dual-field quantum key distribution method as described in claim 1, characterized in that, After estimating the key rate and error rate by selecting a quantitative amount of data from the decoy mode using the two sending ends, error correction and privacy amplification processing are performed based on the measurement results and parameter estimates published by the detection end, and the key rate is calculated.

9. A dual-field quantum key distribution method as described in claim 8, characterized in that, The specific expression for calculating the key rate is as follows: ； Where R is the key rate in the final key. Indicates the error rate. It is a binary entropy function. This represents the probability that the first sender obtains the original key in the encoded mode, and f represents the error correction efficiency. This is the upper limit of the amount of information leakage.

10. A dual-field quantum key distribution system, characterized in that, include: The first preparation unit is used to select an encoding mode or a decoy mode through two transmitters according to probabilities p and 1-p, and prepare a coherent state through the encoding mode or the decoy mode. The second preparation unit is used to use a long short-term memory network model to change the values ​​of environmental parameters to perform optimization parameter prediction, obtain a dataset containing optimization parameters, and modulate the dataset of optimization parameters into a coherent state. The original key extraction unit is used to send the modulated coherent state to the detection end for measurement, send the obtained successful measurement event to two sending ends, and extract the original key through the mode selected by each of the two sending ends. The final key generation unit is used to select a quantitative amount of data from the decoy pattern using the two sending ends to estimate the key rate and error rate, and generate the final security key from the original key.

Images