Quantum Communication Method Based on Decision Tree

Abstract

The present invention discloses a quantum communication method based on a decision tree, which includes collecting communicator information to extract required feature information, digitalizing the feature information and training a decision tree prediction model to output the level of the communicator; evaluating the trained decision tree prediction model by cross-validation and resampling and optimizing the model; during communication, configuring corresponding channel particles for the communicator according to the level of the communicator output by the decision tree prediction model and starting the communication; configuring corresponding measurement operations for receivers of different levels, and after performing the measurement operations, telling the results to the receivers through a classical channel; after the receiver receives the measurement results, performing a unitary operation on the collapsed state to reconstruct the information state. The present invention uses the decision tree algorithm in machine learning to classify and rate communicators, realizes the level judgment of communicators, thereby allocating matching numbers of channel particles for communicators of corresponding levels, solves the problem of allocation of channel particles, saves quantum resources and improves communication efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of channel particle allocation in quantum communication. Specifically, it relates to a quantum communication method based on a decision tree. Background Art

[0002] Quantum communication is a new type of communication method that uses quantum superposition states and entanglement effects for information transmission. Quantum communication is mainly based on the theory of quantum entanglement states and realizes information transmission through methods such as quantum teleportation (transmission), quantum information splitting (transmission), and quantum key distribution (transmission). The process of quantum communication is as follows: First, construct a pair of particles with an entangled state (there is a long-distance physical connection between the entangled particles). Place the two particles on both sides of the communication parties respectively. Perform a joint measurement (an operation) on the particle with an unknown quantum state and the particle of the sender. Then, the particle of the receiver instantaneously collapses (changes: changes based on the physical connection between the particles). The collapse (change) is into a certain state, and this state is symmetric to the state after the collapse (change) of the particle of the sender. Then, transmit the information of the joint measurement through a classical channel to the receiver. The receiver performs a unitary transformation (equivalent to an inverse transformation) on the collapsed particle according to the received information, and thus can obtain an unknown quantum state exactly the same as that of the sender.

[0003] In actual communication, generally, a communication scheme is directly designed according to the transmitted information. Among them, it is necessary to manually judge the communication levels of communication participants, and then allocate corresponding particle positions to the communicators according to the levels. However, since the number of people participating in communication may be quite large, the conditions for judging levels are too cumbersome, and the manual workload is quite large and prone to judgment errors. It is difficult for us to judge the levels of communicators, and even some communicators cannot have particle positions and cannot participate in communication. Therefore, there is a problem of the allocation of communication channel particles in the conventional quantum communication process. Summary of the Invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a quantum communication method based on a decision tree, which introduces the method of decision tree in machine learning into quantum communication to solve the problem of judging the levels of communicators, thereby solving the problem of the allocation of communication channel particles.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A quantum communication method based on a decision tree includes the following steps:

[0007] S10. Collect the communicator information of quantum communication participants, extract multiple feature information required for the decision tree prediction model from it. After preprocessing these feature information, train the decision tree prediction model, and the output is the high-level and low-level classification of quantum communication participants;

[0008] S20. Evaluate the decision tree prediction model obtained by training using cross-validation and resampling methods, and optimize the decision tree prediction model according to the evaluation results;

[0009] S30. During communication, determine the high-level and low-level classifications of quantum communication participants according to the decision tree prediction model, allocate corresponding channel particles to quantum communication participants of different levels according to preset information, and start communication;

[0010] S40. When the receiver is a quantum communication participant of different levels, perform corresponding measurement operations according to specified rules respectively. After performing the measurement operations, tell the receiver the measurement results through the classical channel;

[0011] S50. After the receiver receives all the measurement results, perform a unitary operation on the collapsed state according to the corresponding results to restore the information sent by the sender.

[0012] Specifically, the process of step S10 includes:

[0013] S11. Preset the feature information required for the decision tree prediction model according to the target requirements of quantum communication;

[0014] S12. Extract the corresponding feature information from the collected communicator information according to preset requirements;

[0015] S13. Digitalize the extracted feature information;

[0016] S14. Use the digitalized feature information as a data set, and divide the data set into a training set and a test set according to a specified ratio;

[0017] S15. Input the data of the training set into the set decision tree prediction model for model training, output results according to the set high-level and low-level classifications of quantum communication participants, and thus determine the various parameters of the decision tree prediction model;

[0018] S16. Use the test set to test the obtained decision tree prediction model, statistically analyze and judge the error rate according to the output results of the test set, and use the loss function to prune and optimize the decision tree prediction model obtained by training.

[0019] Specifically, the feature information required for the preset decision tree prediction model is the proportion of the number of honest times in all communications, the number of honest times in the last three communications, the probability of dishonesty judged by the third-party notarization platform, and whether it is listed.

[0020] Specifically, the set decision tree prediction model uses the Gini index model, expressed as:

[0021] D 1 ={(x i) ∈ DA j f(x) = a}, 0 ≤ a ≤ x i , D 2 = D - D 1

[0022]

[0023]

[0024] Where D is the data set, A j represents the j-th feature information, x i represents the i-th value of a certain feature information, a represents the branch condition value of the corresponding feature information, and according to the value of a, the data set D is divided into D 1 and D 2 two branch subsets; K represents the number of classes of feature information in D, C k represents the sample subset belonging to the k-th class in D; Gini(D) represents the Gini index of the data set D, and Gini(D, A) represents the Gini index of the data set D when taking the feature information A;

[0025] Select the feature information corresponding to the minimum Gini index as the optimal splitting point. During model training, the samples in the training set are assigned to two branch subsets according to the feature information, and such calculations and selections are repeated until the number of samples in the branch subset is less than the predetermined threshold or there are no features to select. The respective sets of the two similar branch subsets respectively correspond to the high-level quantum communication participants and the low-level quantum communication participants in the output result.

[0026] Specifically, the loss function is expressed as:

[0027] C α (T) = C(T) + αT

[0028] Where T represents any subtree, C(T) represents the prediction error of the training data, that is, the corresponding Gini index, T represents the number of leaf nodes of the subtree, and C α (T) represents the overall loss of the subtree T when the parameter α is set. The set parameter α is used to balance the fitting degree of the training data and the model complexity.

[0029] Specifically, the process of step S20 includes:

[0030] S21. According to the cross-validation error kfodLoss function and the resampling error resubLoss function, reallocate the data set D, train and test the set decision tree prediction model, and calculate the cross-validation error;

[0031] S22. Repeat the process of step S21 multiple times, and calculate the average value of the cross-validation errors after multiple allocations as the evaluation of the decision tree prediction model for training optimization.

[0032] S23. According to the evaluation result in step S22, select to prune the decision tree prediction model optimized by training under the minimum cross-validation error, and adjust the decision tree prediction model to realize the re-optimization of the decision tree prediction model.

[0033] Specifically, in step S30, the levels of the quantum communication participants are determined as high-level quantum communication participants and low-level quantum communication participants.

[0034] Specifically, the corresponding measurement operations specified by the rules in step S40 are as follows:

[0035] When the receiver is a high-level quantum communication participant, any one of the other high-level quantum communication participants performs single-particle measurement and tells the measurement result to the receiver through the classical channel.

[0036] When the receiver is a low-level quantum communication participant, all other quantum communication participants perform single-particle measurement and tell the measurement result to the receiver through the classical channel.

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

[0038] (1) The present invention proposes a method for designing a quantum communication scheme according to quantum communication participants, skillfully using the decision tree algorithm in machine learning to classify and rate quantum communication participants, realizing the level judgment of communicators, so as to allocate matching channel particle numbers to communicators of corresponding levels, solving the problem of channel particle allocation, saving quantum resources, and improving the efficiency of quantum communication. The present invention is ingeniously designed, the process is relatively simple, the implementation is convenient and reliable, and it is suitable for application in quantum communication.

[0039] (2) The present invention collects information of quantum communication participants, extracts the required feature information therefrom, and through the processing of these feature information, more accurately realizes the classification judgment of high-level and low-level communicators, providing a reliable basis for the process of quantum communication.

[0040] (3) On the basis of designing a decision tree prediction model based on feature information, the present invention also evaluates the model by means of cross-validation and resampling, so as to modify and prune the decision tree prediction model to make it have better performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flowchart of an embodiment of the present invention.

[0042] Figure 2Schematic diagram of the decision tree prediction model obtained through training in an embodiment of the present invention.

[0043] Figure 3 Schematic diagram of the decision tree prediction model optimized through testing in an embodiment of the present invention.

[0044] Figure 4 Schematic diagram of the decision tree prediction model optimized according to the evaluation results in an embodiment of the present invention. Detailed implementation manners

[0045] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. The implementation manners of the present invention include but are not limited to the following embodiments.

[0046] Existing quantum communication methods are directly designed based on the transmitted information and mainly rely on the theory of quantum entanglement states. It uses methods such as quantum teleportation (a transmission method) and quantum information splitting (a propagation method) to achieve information transmission. The process is generally as follows: First, a pair of entangled particles a is constructed (the entangled particles have a long-distance physical connection). The two particles are placed on both sides of the communication and the particle with an unknown quantum state and the particle of the sender are jointly measured (an operation). Then, the particle of the receiver will instantaneously collapse (the change refers to the change based on the physical connection between the particles), and the collapse (change) will be a specific state, which is symmetric to the state of the collapsed (changed) particle of the transmitter. Then, the information obtained from the joint measurement is transmitted to the receiver through a classical channel. The receiver performs a unitary operation (equivalent to an inverse operation) on the collapsed particle according to the received information to obtain the same unknown quantum state as the transmitter. For example: If the transmitted information has 2 bits, then 1 bit of quantum state is required to carry the information as an information particle. To transmit this quantum state, quantum information splitting is used as the transmission method, and a 4-bit quantum entanglement state is used as the quantum channel. The sender has a particle 1 and a particle 2 with 1 bit of quantum state and 4 bits of quantum state of the channel particles. The receiver has particle 3, and a third-party quantum communication participant has particle 4. The sender jointly measures the information particle and the channel particles 1 and 2. If the sender wants the receiver to obtain the information, the sender tells the measurement result to the third-party quantum communication participant. The third-party quantum communication also measures its own particle 4. If the receiver wants to obtain the original information, it needs the measurement results of the sender and the third-party quantum communication participant and performs a unitary operation according to the measurement results, so that the receiver can obtain the original information.

[0047] According to the communication transmission efficiency calculation formula: c is the number of qubits transmitted, q is the number of bits used in the quantum channel, and t is the number of classical bits, which are the bits transmitted through the classical channel. Generally, the ciphertext to be transmitted can meet all the transmission requirements with two-particle states, so information can be transmitted with very few channel particles. For example, the channel particles are and so on. There are so many ways to communicate. How do we know which is the most suitable communication method we need? Moreover, with the same number of channel particles, the states may also be different, such as |0000> + |0101> + |1010> + |1111> and |0000> + |0101> + |1010> - |1111>. Additionally, how do we change when adding one more communicator? How to determine whether the communication scheme is optimal? Facing these problems, conventionally, only manual judgment based on experience is possible, which is very complex and prone to errors.

[0048] Against this background, the inventor proposed this quantum communication method based on decision trees. First, a quantum communication scheme is designed according to the communicators who need to participate. Then, the decision tree algorithm in machine learning is introduced to solve the problem of judging the levels of communicators, thereby solving the problem of the allocation of channel particles, and finally realizing an accurate, efficient, and low-quantum-resource-consumption quantum communication process.

[0049] As Figure 1 shown, this quantum communication method based on decision trees includes the following steps:

[0050] S10. Collect the communicator information of quantum communication participants, extract multiple feature information required for the decision tree prediction model from it. After preprocessing these feature information, train the decision tree prediction model, and the output is the high-level and low-level classification of quantum communication participants. The specific process is as follows:

[0051] S11. Preset the feature information required for the decision tree prediction model according to the target requirements of quantum communication. For example, if the target requirement values the integrity of the communicator more, the preset feature information is: the proportion of the number of times of integrity in all communications participated, the number of times of integrity in the last three communications, the possibility of dishonesty judged by a third-party notarization platform, whether it is listed, and other four features. Another example is that if the target requirement values the qualifications of the communicator more, the preset feature information can be: the company's personnel scale, the company's asset valuation, whether it is a top 100 enterprise in the world, and so on. Another example is that if the target requirement values the credit of the communicator more, the preset feature information can be: default situation, loan situation, whether the loan is repaid, and so on;

[0052] S12. Extract corresponding feature information from the collected communicator information according to preset requirements; the collected communicator information is generally relatively comprehensive various information of quantum communication participants. However, if comprehensive information is used for processing, it will lead to an overly complex algorithm. Therefore, extracting partial feature operations and processing according to the aforementioned target demand tendency can simplify the complexity of algorithm processing well while ensuring the pertinence of the results and improving data processing efficiency; in practical applications, the types of communicator information during collection can also be set based on preset target requirements and basic information, which can reduce the amount of collected data and the requirements for data storage to a certain extent;

[0053] S13. Digitalize the extracted feature information. Since the directly extracted feature information is not necessarily the data required by the algorithm, digitalizing the extracted feature information realizes standardization matching the algorithm. Among them, for the feature information whose result is a number itself, it is processed according to the specified standard, and for the feature information whose result is not a number itself, it is assigned a value according to the algorithm requirements; for example: the result of the proportion of the number of times of integrity in all communications is any decimal between 0 and 1. When digitalizing, it takes a value with one decimal place; the results of the number of times of integrity in the last three communications are 0, 1, 2, and 3. When digitalizing, they take the corresponding values of 0, 1, 2, and 3 respectively; the results of the possibility of dishonesty judged by the third-party notarization platform are dishonesty and honesty. When digitalizing, the value 1 is assigned to the honest result and the value 0 is assigned to the dishonest result; the results of whether it is listed are yes and no. When digitalizing, the value 0 is assigned to the yes result and the value 1 is assigned to the no result;

[0054] S14. Use the digitalized feature information as a data set, and divide the data set into a training set and a test set according to a specified ratio. The training set is used for model training, and the test set is used for model verification. The two can be divided in a ratio of 8:2, or other ratios can be set according to requirements;

[0055] S15. Input the data of the training set into the set decision tree prediction model for model training, and output results according to the classification of high-level and low-level quantum communication participants set, thereby determining each parameter of the decision tree prediction model. Among them, the model parameters mainly include the number of branch nodes of the decision tree, the sorting of each branch node, and the branch conditions of each branch node; the set decision tree prediction model uses the Gini index model, which is expressed as:

[0056] D 1 ={(x i )∈D|A j (x)=a},0≤a≤x i ,D 2 =D - D 1

[0057]

[0058]

[0059] In the formula, D is the data set, and during training, the training set is used for D; A j represents the j-th feature information, and x i represents the i-th value of a certain feature information, and a represents the branch condition value of the corresponding feature information. According to the value of a, the data set D is divided into D 1 and D 2 two branch subsets, corresponding to the two branches of a branch node. For example, in the aforementioned example, j ∈ [1, 4] and is an integer. The number of i values corresponding to different j is 2, 3, or more. Different j respectively correspond to the branch condition values matching their values; K represents the number of types of feature information in the data set D, which matches the maximum value of j, and C k represents the subset of samples belonging to the k-th class in the data set D. Gini(D) represents the Gini index of the data set D, and Gini(D, A) represents the Gini index of the data set D when taking the feature information A; the above formula corresponds to a branch node of the decision tree. The representations of other branch nodes are similar, except that the branch subsets divided by each branch node may be different. The aggregation of all branch subsets with the same classification corresponds to the high-level and low-level classifications of the set quantum communication participants;

[0060] The smaller the Gini index, the smaller the uncertainty of the data set, and the corresponding feature information A is the optimal feature. The feature information corresponding to the minimum Gini index is selected as the optimal splitting point. During model training, the samples in the training set are assigned to the two branch subsets according to the feature information, and this calculation and selection are repeated until the number of samples in the branch subset is less than the predetermined threshold or there are no features to select. The respective sets of the two similar branch subsets respectively correspond to the high-level quantum communication participants and low-level quantum communication participants in the output result;

[0061] S16. Use the test set to test the obtained decision tree prediction model, statistically analyze and judge the error rate according to the output result of the test set, and use the loss function to prune and optimize the decision tree prediction model obtained by training. By repeatedly calling the leaf nodes of the generated original decision tree prediction model, the pruning optimization is performed by calculating and comparing the prediction errors of the decision tree models before and after restricting the leaf nodes. The loss function is expressed as:

[0062] C α (T) = C(T) + α|T|

[0063] In the formula, T represents any subtree, C(T) represents the prediction error for the training data, that is, the corresponding Gini index, T represents the number of leaf nodes of the subtree, and C α(T) represents the overall loss of subtree T when the parameter is α. The set parameter α is used to balance the fitting degree of the training data and the model complexity. Pruning can simplify the decision tree prediction model, thereby making better predictions for unknown data and improving the generalization ability of the model.

[0064] S20. Evaluate the decision tree prediction model obtained through training optimization by using the methods of cross-validation and resampling, and optimize the decision tree prediction model according to the evaluation results;

[0065] S21. According to the cross-validation error kfodLoss function and the resampling error resubLoss function, redistribute the dataset D to obtain a new training set and a test set, then use the processes of steps S15 and S16 to obtain the decision tree prediction model again, and calculate the cross-validation error E by comparing with the preset high-level and low-level classifications of quantum communication participants;

[0066] S22. Repeat the process of step S21 multiple times, and calculate the average value of the cross-validation errors after multiple distributions as the evaluation of the decision tree prediction model obtained through training optimization; expressed as:

[0067]

[0068] In the formula, represents the average value of the cross-validation errors, E i represents the cross-validation error calculated in the i-th time, and n is the total number of times of executing step S21;

[0069] S23. According to the evaluation results of step S22, select to prune the decision tree prediction model obtained through training optimization under the minimum cross-validation error, and adjust the decision tree prediction model to realize the re-optimization of the decision tree prediction model.

[0070] S30. During communication, determine the high-level and low-level classifications of quantum communication participants according to the decision tree prediction model, and allocate corresponding channel particles to quantum communication participants of different levels according to the preset information to start communication.

[0071] S40. When the receiver is a quantum communication participant of different levels, execute the corresponding measurement operations according to the specified rules respectively. After executing the measurement operations, tell the measurement results to the receiver through the classical channel; where the specified rules for executing the corresponding measurement operations are:

[0072] When the receiver is a high-level quantum communication participant, any one of the other high-level quantum communication participants executes single-particle measurement and tells the measurement results to the receiver through the classical channel;

[0073] When the receiver is a low-level quantum communication participant, all other quantum communication participants perform single-particle measurements and tell the measurement results to the receiver through the classical channel.

[0074] S50. After the receiver receives all the measurement results, perform a unitary operation on the collapsed state according to the corresponding results to restore the information sent by the sender.

[0075] Verify and illustrate through the following example data

[0076] Set four characteristics for evaluating the communicator level, namely: the proportion of the number of honest times in all communications, the number of honest times in the last three communications, the possibility of dishonesty judged by the third-party notarization platform, and whether it is listed. The total sample size of all the collected data is 1131.

[0077] Sort the data. There are 409 cases of high-level communicators with a proportion greater than or equal to 0.5, a honesty possibility of 1 and being listed, and 722 cases of low-level communicators, thus obtaining the data set D. After dividing the data set D into a training set and a test set according to the ratio of 8:2, use the training set to train the decision tree prediction model to obtain the results as Figure 2 shown, and use the test set to optimize Figure 2 to obtain the results as Figure 3 shown. Adopt the methods of cross-validation and resampling to evaluate and optimize Figure 3 to obtain the results as Figure 4 shown. In the figure, x1 represents the proportion of the number of honest times in all communications, x2 represents the number of honest times in the last three communications, x3 represents the possibility of dishonesty judged by the third-party notarization platform, and x4 represents whether it is listed.

[0078] According to the allocation scheme of high-level and low-level communicators obtained from the above decision tree prediction model, allocate channel particles for them according to the preset allocation strategy. For example, among N communication participants, one is the sender (owning particles 1 and 2), one is the receiver (owning particles 3 and 4), the high-level communicators Bob1, Bob2,..., Bobx own particles from 5 to 5 + x, and the low-level communicators Charlie1, Charle2,..., Charliey own particles from 6 + x to 6 + x + y; then communication can be carried out.

[0079] The comparison of the communication efficiency of the communication scheme OS of the present invention with the traditional communication schemes 1, 2, 3, and 4 is shown in Table 1 below. Scheme 1 is a two-particle information three-particle channel transmission scheme, Scheme 2 is a single-particle information 4-particle transmission scheme, Scheme 3 is a single-particle information 4-particle separated transmission scheme, Scheme 4 is a two-particle information eight-particle transmission scheme, and the references are respectively:

[0080] 1. Wen Z, Liu Y M, Zhang Z J, et al. Splitting a qudit state via Greenberger–Horne–Zeilinger states of qubits[J]. Optics Communications, 2010, 283(4): 628-632.

[0081] 2. Jouguet P, Kunz-Jacques S, Leverrier A, et al. Experimental demonstration of long-distance continuous-variable quantum key distribution[J]. Nature Photonics, 2013, 7(5): 378-381.

[0082] 3. Lu Y J. A Novel Practical Quantum Secure Direct Communication Protocol[J]. International Journal of Theoretical Physics, 2021(3).

[0083] 4. Lucas L. Quantum Reinforcement Leaming with Quantum Photonics[J]. Photonics, 2021, 8(2): 33.

[0084] Compare the four aspects: quantum resource consumption QS, the number of transmitted qubits QT, transmission efficiency η%, and the number of communication participants NC. It can be seen that the results of the present invention save quantum resources and improve transmission efficiency compared with various traditional communication schemes.

[0085] Table 1 Comparison of communication efficiency

[0086] Schemes QS QT transmission efficiency(η%) NC 1 10 2 10％ 5 2 4 1 12.5％ 4 3 4 1 12.5％ 4 4 8 2 12.5％ 4 OS 8 2 14.3％ 5

[0087] Thus, it can be seen that by applying the decision tree model to the quantum communication process, the present invention can effectively reduce the complexity involved in the communication process and improve the overall working efficiency of the system, while achieving optimal resource allocation.

[0088] The above embodiments are only the preferred embodiments of the present invention, and do not limit the protection scope of the present invention. Any changes made by using the design principle of the present invention and non-creative labor based on this should fall within the protection scope of the present invention.

Claims

1. A quantum communication method based on decision tree, characterized in that, it includes the following steps: S10. Collect the communicator information of quantum communication participants, extract multiple feature information required by the decision tree prediction model from it. After preprocessing these feature information, train the decision tree prediction model, and the output is the high-level and low-level classification of quantum communication participants; S20. Evaluate the trained decision tree prediction model by using the methods of cross-validation and resampling, and optimize the decision tree prediction model according to the evaluation results; S30. During communication, determine the high-level and low-level classification of quantum communication participants according to the decision tree prediction model, allocate corresponding channel particles to quantum communication participants of different levels according to the preset information, and start communication; S40. When the receiver is a quantum communication participant of different levels, perform corresponding measurement operations according to the specified rules respectively. After performing the measurement operations, tell the measurement results to the receiver through the classical channel; S50. After the receiver receives all the measurement results, perform a unitary operation on the collapsed state according to the corresponding results to restore the information sent by the sender.

2. The quantum communication method based on decision tree according to claim 1, characterized in that, the process of step S10 includes: S11. Preset the feature information required by the decision tree prediction model according to the target requirements of quantum communication; S12. Extract the corresponding feature information from the collected communicator information according to the preset requirements; S13. Digitalize the extracted feature information; S14. Use the digitalized feature information as a data set, and divide the data set into a training set and a test set according to a specified ratio; S15. Input the data of the training set into the set decision tree prediction model for model training, and output the results according to the high-level and low-level classification of the set quantum communication participants, so as to determine each parameter of the decision tree prediction model; S16. Use the test set to test the obtained decision tree prediction model, statistically analyze and judge the error rate according to the output results of the test set, and use the loss function to prune and optimize the trained decision tree prediction model.

3. The quantum communication method based on decision tree according to claim 2, characterized in that, the preset feature information required by the decision tree prediction model is the proportion of the number of honest times in all communications, the number of honest times in the last three communications, the possibility of dishonesty judged by the third-party notarization platform, and whether it is listed.

4. The quantum communication method based on decision tree according to claim 2, characterized in that, the decision tree prediction model set in step S15 adopts the Gini index model, and the formula is expressed as: D 1 = {(x i ) ∈ D | A j (x) = a}, 0 ≤ a ≤ x i , D 2 = D - D 1 where D is the data set, and A j represents the j-th feature information, and x i represents the i-th value of a certain feature information, and a represents the branch condition value of the corresponding feature information. According to the value of a, the data set D is divided into D 1 and D 2 two branch subsets; K represents the number of classes of the feature information in D, and C k represents the sample subset in D that belongs to the k-th class; Gini(D) represents the Gini index of the data set D, and Gini(D, A) represents the Gini index of the data set D when taking the feature information A; Select the feature information corresponding to the minimum Gini index as the optimal splitting point. During model training, allocate the samples in the training set to two branch subsets according to the feature information, and calculate and select in this way repeatedly until the number of samples in the branch subset is less than the predetermined threshold or there is no feature to select. The respective sets of the two similar branch subsets respectively correspond to the high-level quantum communication participants and the low-level quantum communication participants in the output results.

5. The quantum communication method based on decision tree according to claim 2, characterized in that, The loss function in step S16 is expressed as: C α (T) = C(T) + α|T| In the formula, T represents any subtree, C(T) represents the prediction error of the training data, that is, the corresponding Gini index, |T| represents the number of leaf nodes of the subtree, and C α (T) represents the overall loss of the subtree T when the parameter α is set. The set parameter α is used to balance the fitting degree of the training data and the model complexity.

6. The quantum communication method based on decision tree according to claim 4, wherein, The process of step S20 includes: S21. According to the cross-validation error kfodLoss function and the resampling error resubLoss function, reallocate the dataset D, train and test the set decision tree prediction model, and calculate the cross-validation error; S22. Repeat the process of step S21 multiple times, and calculate the average value of the cross-validation errors after multiple allocations as the evaluation of the trained and optimized decision tree prediction model; S23. According to the evaluation result of step S22, select to prune the trained and optimized decision tree prediction model under the minimum cross-validation error, adjust the decision tree prediction model, and realize the re-optimization of the decision tree prediction model.

7. The quantum communication method based on decision tree according to claim 6, wherein, In step S30, the classifications of the quantum communication participants are determined as high-level quantum communication participants and low-level quantum communication participants.

8. The quantum communication method based on decision tree according to claim 7, wherein, The corresponding measurement operation specified by the rule in step S40 is: When the receiver is a high-level quantum communication participant, any one of the other high-level quantum communication participants performs single-particle measurement, and tells the measurement result to the receiver through the classical channel; When the receiver is a low-level quantum communication participant, all other quantum communication participants perform single-particle measurement, and tell the measurement result to the receiver through the classical channel.

Images