Machine Learning-Based Quantum Key Distribution System and Method

Abstract

The present application provides a quantum key distribution system and method based on machine learning. The system includes: a network management layer configured to obtain user instructions; the user instructions include at least one of instructions for performance prediction or function setting of components in the quantum key distribution system; according to whether there is a corresponding metric instruction for the user instruction, sending the user instruction to the machine learning layer or sending the metric instruction to the function layer; a machine learning layer configured to obtain the metric instruction through a target problem metric program according to the user instruction, perform model training according to the metric instruction to obtain a function model, and send the metric instruction to the network management layer for the network management layer to store it; a function layer configured to retrieve the function model according to the metric instruction, determine a target component instruction according to the output result of the function model, and send it to the target component.

Description

Technical Field

[0001] The present application relates to the field of network system technology, and in particular to a quantum key distribution system and method based on machine learning. Background Art

[0002] Quantum key distribution (QKD) is a key generation technology based on the principles of quantum mechanics. Through the preparation, transmission, and measurement of quantum states, it generates a shared, random key between the sender and receiver that cannot be eavesdropped on. Quantum key technology requires a network system consisting of a transmitter, a channel, and a receiver. Quantum key distribution systems in related technologies primarily consist of a management layer and a functional layer. These functional layers typically include at least one of the following: a control layer, a key management layer, and a quantum layer.

[0003] With the advancement of computer technology, related technologies have proposed applying machine learning models to quantum key distribution systems to improve the system's processing and computing capabilities. These technologies typically train a machine learning model based on a specific problem and then implement it into the corresponding functional layer. However, due to the system's inability to analyze the target and problem, when encountering sudden or complex situations, the system cannot self-learn and update or upgrade, relying solely on human intervention. Consequently, these technologies still suffer from the drawbacks of being unable to cope with sudden changes and complex instructions. Summary of the Invention

[0004] In view of this, the purpose of this application is to propose a quantum key distribution system and method based on machine learning.

[0005] Based on the above objectives, one or more embodiments of the present application provide a quantum key distribution system based on machine learning, including:

[0006] The network management layer is configured to obtain a user instruction; the user instruction includes at least one instruction of performing performance prediction or function setting on a component in the quantum key distribution system; and send the user instruction to the machine learning layer or send the indicator instruction to the function layer according to whether the user instruction has a corresponding indicator instruction;

[0007] a machine learning layer configured to obtain the indicator instruction through a target problem measurement program according to the user instruction, perform model training according to the indicator instruction to obtain a functional model, and send the indicator instruction to the network management layer so that the network management layer stores it;

[0008] The functional layer is configured to call the functional model according to the indicator instruction, and determine the target component instruction according to the output result of the functional model and send it to the target component.

[0009] Optionally, the network management layer includes:

[0010] An acquisition unit, configured to acquire the user instruction;

[0011] A judgment unit, configured to confirm whether there is an index instruction corresponding to the user instruction in the storage unit;

[0012] A sending unit, configured to send the index instruction to the function layer in response to determining that there is a corresponding index instruction; and send the index instruction to the machine learning layer in response to determining that there is no corresponding index instruction;

[0013] A storage unit, configured to receive the index instruction sent by the machine learning layer and store the index instruction.

[0014] Optionally, the machine learning layer includes:

[0015] A user instruction parsing unit, configured to obtain the index instruction through a target problem measurement program according to the user instruction, and send the index instruction to the model training unit;

[0016] A model training unit, configured to obtain historical operation data of a target component according to the index instruction and perform model training, and store the model as a function model after the model passes the test;

[0017] A sending module, configured to send the index instruction to the network management layer for the network management layer to store it.

[0018] Optionally, the user instruction parsing unit includes:

[0019] An analysis module, configured to analyze the user instruction according to a target and a corresponding predetermined problem, and determine the index of the user instruction;

[0020] An orchestration module, configured to obtain the index instruction through orchestration according to the index;

[0021] A sending module, configured to send the index instruction to the model training unit.

[0022] Optionally, the analysis module is further configured to: measure the user instruction according to a measurement standard based on the target and the artificially predetermined problem in the target problem measurement program, and obtain the index of the user instruction.

[0023] Optionally, the model training unit includes:

[0024] A function management module, configured to obtain historical operation data of a target component according to the index instruction and perform model training, and send the trained function model to the test module;

[0025] A test module, configured to test the functional model, and end the training of the functional model and send it to the storage module in response to determining that the test result of the functional model meets a predetermined condition;

[0026] A storage module, configured to store the functional model;

[0027] An application sandbox module, configured to simulate a training environment and a test environment.

[0028] Optionally, the function management module includes:

[0029] A data acquisition module, configured to acquire historical operation data and current operation data of the target component as training data according to the metric instruction;

[0030] A data preprocessing module, configured to clean and augment the training data to obtain preprocessed data;

[0031] A model training module, configured to train an initial model according to the preprocessed data and the metric instruction to obtain the functional model;

[0032] A sending module, configured to send the metric instruction and the functional model to the test unit.

[0033] Optionally, the function layer includes a function application unit, and the function application unit includes:

[0034] A data acquisition module, configured to acquire historical operation data and current operation data of the target component according to the metric instruction;

[0035] A data preprocessing module, configured to clean and augment the data to obtain preprocessed data;

[0036] A model deployment module, configured to retrieve the functional model from the machine learning layer according to the metric instruction;

[0037] A model application module, configured to receive the preprocessed data and the metric instruction, and an output result calculated by the functional model;

[0038] A policy construction module, configured to generate a target component instruction according to the metric instruction and the output result, and send the target component instruction to the result distribution module;

[0039] A result distribution module, configured to send the target component instruction to the target component.

[0040] Optionally, the model deployment module is further configured to: in response to determining that the metric instruction cannot retrieve the functional model from the machine learning layer, send the metric instruction to the machine learning layer so that the machine learning layer trains the model.

[0041] Based on the same inventive concept, one or more embodiments of the present application provide a machine learning-based quantum key distribution method, including:

[0042] The network management layer obtains a user instruction; the user instruction includes at least one instruction for performance prediction or function setting of components in the quantum key distribution system; according to whether there is a corresponding metric instruction for the user instruction, send the user instruction to the machine learning layer or send the metric instruction to the function layer;

[0043] The machine learning layer obtains the metric instruction through a target problem metric program according to the user instruction, trains a model according to the metric instruction to obtain a functional model, and sends the metric instruction to the network management layer for the network management layer to store it;

[0044] The function layer retrieves the functional model according to the metric instruction, determines a target component instruction according to the output result of the functional model, and sends it to the target component.

[0045] As can be seen from the above, a quantum key distribution system and method provided by one or more embodiments of the present application decompose a user instruction into clear and definable metric instructions through a target problem metric program, and then perform model training in the machine learning layer according to the above metric instructions to obtain a functional model. Through the above system and method, the system can analyze user instructions by itself and train models by itself according to user instructions to obtain corresponding functional models. When the system mutates, it has the ability of self-regulation without manual control of the system. Thus, the operation efficiency of the system can be improved in this way. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present application or related technologies, the following will briefly introduce the drawings required for use in the embodiments or related technology descriptions. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a schematic structural diagram of a machine learning-based quantum key distribution system according to one or more embodiments of the present application;

[0048] Figure 2Schematic structural diagram of a quantum key distribution system based on machine learning in the related art of the present application;

[0049] Figure 3 Schematic structural diagram of a quantum key distribution system based on machine learning according to one or more embodiments of the present application

[0050] Figure 4 Schematic flow diagram of a quantum key distribution method based on machine learning according to one or more embodiments of the present application. Detailed implementation manners

[0051] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to specific embodiments and the accompanying drawings.

[0052] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the ordinary meanings understood by those of ordinary skill in the field to which the present application belongs. The "first", "second" and similar terms used in the embodiments of the present application do not denote any order, quantity or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" 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. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right" are only used to represent relative positional relationships, and when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0053] As described in the background art, quantum key distribution is a key generation technology based on the principles of quantum mechanics. Through the preparation, transmission, and measurement of quantum states, a shared random key that cannot be stolen is generated between the sender and the receiver. Compared with the secure communication scheme that relies on algorithmic complexity, quantum key distribution utilizes the characteristics of the single quantum no-cloning theorem of quantum to enable spatially separated users to securely share keys and resist attacks from quantum computers. A quantum key distribution network consists of multiple quantum key distribution nodes and quantum key distribution links. Among them, the main components include a transmitter and a receiver. The transmitter may include a pulsed light source, a decoy state modulation module, a quantum state modulation module, an adjustable optical attenuator, a random number generator, a synchronization signal transmission module, a negotiation signal transceiver module, a transmitter control and processing module, and a communication interface module. The receiver may include a line adaptation compensation module, a quantum state demodulation module, a single photon detector, a random number generator, a synchronization signal reception module, a negotiation signal transceiver module, a receiver control and processing module, and a communication interface module. According to different physical mechanisms, the protocol types of quantum key distribution technology can be divided into discrete variable protocols based on single photon modulation, such as the BB84 protocol, the B92 protocol, the DPS protocol, the COW protocol, etc., continuous variable protocols based on multi-photon modulation, such as the GG02 protocol, etc., and entanglement protocols based on entangled photon pairs, such as the E91 protocol, etc.

[0054] As Figure 1 shown, the quantum key distribution system in the related art mainly includes a management layer and a functional layer. The functional layer mainly includes a control layer, a key management layer, and a quantum layer. Among them, the main function of the management layer is to handle faults, configuration, billing, performance, and security management issues of other layers, as well as cross-layer management and orchestration; the main functions of the control layer are access, routing, session, configuration, policy-based control, and the management and control of controllers in the quantum key distribution network; the main functions of the key management layer are key provision, key management, and key manager control, etc.; the main function of the quantum layer is key generation. Among them, the quantum layer needs to perform performance prediction according to the system operation state, and the system and system components need to perform parameter optimization and life prediction according to the system operation state; the key management layer needs to perform key data formatting, key storage management, and detection of suspicious behaviors; other functions, including the control layer and the management layer, need to perform fault prediction.

[0055] With the development of computer technology, the related art proposes to apply machine learning in the quantum key distribution system. The main solution is to train a corresponding model according to the specific problems of quantum key distribution technology and place the model in the corresponding functional layer of the system, so as to improve the computing ability of the system through the machine learning model when solving the specific problem. For example, the machine learning model is used to complete the life prediction of the main components in the quantum key distribution system or to adjust the parameters of the system.

[0056] However, on the one hand, the above system only trains the model according to specific problems and then deploys it in the system separately. It does not really establish the relationship between the machine learning model and the components in the traditional quantum key distribution system. That is, it does not establish a unified network function architecture and lacks a general process for applying the machine learning model, resulting in a lack of intelligence and scalability. On the other hand, the above system cannot parse complex user instructions and has weak analysis and problem-solving capabilities. Therefore, the above system still has problems in coping with mutation situations and solving complex instructions. Thus, the related technology still has the disadvantages of being unable to cope with mutation situations and solve complex instructions.

[0057] In summary, this application proposes to introduce the GQM (Goal-Question-Metric) program into the system. In the process of implementing this application, the applicant found that applying GQM in quantum key distribution can improve the system's ability to analyze problems by itself. GQM is a method of data collection and analysis. Its basic idea is: first, set goals according to the actual application scenario and classify each goal into a set of quantifiable questions. Each question can be answered by several specific metrics, and the above metrics can be obtained through machine learning models learning from historical instructions and user expectations or set artificially; by asking questions and decomposing the goals into multiple executable quantitative metrics, the problem of difficult model definition is solved; according to the above clear metrics, collect data and train the initial machine learning model, and then achieve the defined goals. That is, GQM's approach from goal sorting, to question modeling, and then to the metric system can transform complex user instructions into multiple measurable actual metrics. Using the GQM program, even non-technical users can design goals and questions and use it as a collaborative tool. Then the machine learning application can establish and express an implicit model based on the above questions and their answers.

[0058] One or more embodiments of this application combine the GQM program with machine learning technology to construct a quantum key distribution system including a network management layer, a machine learning layer, and a function layer, realizing the reasonable planning of the system functions, strengthening the application of machine learning, and establishing a unified and complete system architecture. Among them, by constructing the machine learning layer, the parsing and transformation of complex instructions are realized, and the deployment of the function model is completed according to the above instructions. Thus, the quantum key distribution system proposed by one or more embodiments of this application provides a more scalable and efficient quantum key distribution system solution.

[0059] Hereinafter, the technical solutions of one or more embodiments of this specification will be described in detail through specific embodiments.

[0060] Refer to Figure 2, the machine learning-based quantum key distribution system according to one or more embodiments of the present application includes: a network management layer, a machine learning layer, and a function layer.

[0061] The network management layer 11 is configured to obtain user instructions; the user instructions include at least one of instructions for performance prediction or function setting of components in the quantum key distribution system; according to whether there is a corresponding metric instruction for the user instructions, send the user instructions to the machine learning layer or send the metric instructions to the function layer.

[0062] In this layer, user instructions are first obtained. The user instructions include at least one of instructions for performance prediction or function setting of components in the quantum key distribution system. There are a large number of distribution links, quantum key distribution components in the quantum key distribution system, such as quantum transmitters and receivers, etc., as well as key management components. The above devices can all be managed through the quantum key distribution system proposed in the present application, and the management content mainly includes performance prediction and function setting of components. In this layer, after obtaining the user instructions, it is first determined whether there is a corresponding metric instruction for the above user instructions. In some embodiments, it can be determined by judging whether there is a corresponding metric instruction stored in the storage unit of this layer. Determining that there is a corresponding metric instruction means that the system has received the same user instructions and trained the corresponding function model according to the above user instructions, so the corresponding function model can be directly retrieved and subsequent work can be carried out without training the model this time. If it is determined that there is no corresponding metric instruction for the above user instructions, it means that the system has received such instructions for the first time, and it is necessary to analyze the user instructions according to the preset goals and problems and train the corresponding function model. Thus, in response to determining that there is a corresponding metric instruction, the above metric instruction is sent to the function layer; in response to determining that there is no corresponding metric instruction, the above metric instruction is sent to the machine learning layer. At the same time, this layer also includes a storage unit for receiving the above metric instructions sent by the machine learning layer and storing the above metric instructions.

[0063] In some embodiments, the above network management layer includes: an obtaining unit configured to obtain the user instructions; a judging unit configured to confirm whether there is a metric instruction corresponding to the user instructions in the storage unit; a sending unit configured to, in response to determining that there is a corresponding metric instruction, send the metric instruction to the function layer; in response to determining that there is no corresponding metric instruction, send the metric instruction to the machine learning layer; a storage unit configured to receive the metric instructions sent by the machine learning layer and store the metric instructions.

[0064] In some embodiments, in addition to having the above functions, the network management layer also enhances the functions of the network management layer in related technologies, including managing faults, configurations, billing, performance, and security management of other layers. In some embodiments, the above functions can be enhanced in performance through a machine learning model. That is, in some embodiments, the above network management layer can also be used as a functional layer, configured to retrieve a corresponding functional model according to an index instruction, and determine a target component instruction based on the output result of the above functional model and send it to the target component.

[0065] The machine learning layer 12 is configured to obtain the index instruction through a target problem measurement program according to the user instruction, perform model training according to the index instruction to obtain a functional model, and send the index instruction to the network management layer for the network management layer to store it.

[0066] When the user instruction appears for the first time, that is, when the above network management layer does not store the corresponding index instruction, the machine learning layer will receive the above user instruction. At the initial stage of establishing the system, users or personnel in other related technical fields will set system goals according to the actual application scenarios of the system and user expectations, and put forward several questions according to the system goals, which may include what types of problems the user instruction hopes to solve, which components the user instruction hopes to act on, and what data needs to be collected to complete the user instruction, etc. Different problem settings as long as they can achieve the corresponding goals will not affect the protection scope of the present invention. At the same time, users or personnel in related technical fields set measurement criteria for the problem settings. The criteria can be set manually, or a training set including problems and index instructions can be established to train a machine learning model or other models to obtain a measurement criteria model for obtaining the measurement criteria. When the user actually uses the system, the above goals, problems, and corresponding measurement criteria have been set. When the machine learning layer receives the user instruction, it can quantify and measure it according to the above problems to obtain corresponding indexes. Further, an index instruction can be obtained according to the above indexes, and a functional model can be obtained through model training according to the above index instruction. In summary, the above user instruction, the above index instruction, and the above functional model will maintain a one-to-one correspondence relationship. When the user instruction is received next time, the index instruction and the functional model can be directly retrieved.

[0067] In some embodiments, the above-mentioned machine learning layer includes: a user instruction parsing unit configured to convert the above-mentioned index instruction through a target problem metric program according to the user instruction and send the above-mentioned index instruction to the model training unit; a model training unit configured to obtain historical operation data of a target component according to the above-mentioned index instruction and perform model training, and store the model as a functional model after the model passes the test; a sending module configured to send the above-mentioned index instruction to the network management layer for the network management layer to store it. In some embodiments, the above-mentioned user instruction parsing unit includes: a parsing module configured to parse the user instruction according to a target and a corresponding predetermined problem to determine the index of the user instruction; an orchestration module configured to obtain the above-mentioned index instruction through orchestration according to the above-mentioned index; a sending module configured to send the above-mentioned index instruction to the above-mentioned model training unit. In some embodiments, the above-mentioned parsing module is further configured to: measure the user instruction according to a metric standard based on the target and the artificially predetermined problem in the target problem metric program to obtain the index of the user instruction.

[0068] In some embodiments, the above-mentioned model training unit includes: a function management module configured to obtain historical operation data of a target component according to the above-mentioned index instruction and perform model training, and send the trained functional model to the test module; a test module configured to test the above-mentioned functional model, and in response to determining that the test result of the above-mentioned functional model meets a predetermined condition, end the training of the above-mentioned functional model and send it to the storage module; a storage module configured to store the above-mentioned functional model; an application sandbox module configured to simulate a training environment and a test environment. In some embodiments, the above-mentioned function management module includes: a data collection module configured to collect historical operation data and current operation data of the above-mentioned target component as training data according to the above-mentioned index instruction; a data preprocessing module configured to clean and enhance the above-mentioned training data to obtain preprocessed data; a model training module configured to train an initial model according to the above-mentioned preprocessed data and the above-mentioned index instruction to obtain the above-mentioned functional model; a sending module configured to send the above-mentioned index instruction and the above-mentioned functional model to the above-mentioned test unit.

[0069] The function layer 13 is configured to retrieve the functional model according to the index instruction, and determine a target component instruction according to the output result of the functional model and send it to the target component.

[0070] When the user instruction appears not for the first time, that is, when the above network management layer stores the corresponding metric instruction, the machine learning layer will receive the above metric instruction. The function layer can directly retrieve the function model from the machine learning layer and calculate the required result through this function model. In some embodiments, the function layer may include a quantum layer, a key management layer, and a control layer. In some embodiments, the function layer may further include a network management layer. This system can achieve parameter optimization and life prediction of components in the system, and at the same time, it can also achieve performance prediction of quantum channels in the quantum layer, key storage management, suspicious behavior detection, and intelligent key formatting in the key management layer, and fault prediction in the control layer and network management layer.

[0071] After collecting the historical operation data of the target component according to the above metric instruction and retrieving the corresponding function model, the output result is calculated through the above function model. The target component instruction for the target component can be obtained through the above output result.

[0072] In some embodiments, the above function layer includes a function application unit, and the above function application unit includes: a data acquisition module configured to collect the historical operation data and current operation data of the above target component according to the above metric instruction; a data preprocessing module configured to clean and enhance the above data to obtain preprocessed data; a model deployment module configured to retrieve the above function model from the above machine learning layer according to the above metric instruction; a model application module configured to receive the above preprocessed data and the above metric instruction and calculate the output result through the above function model; a policy construction module configured to generate a target component instruction according to the above metric instruction and the above output result and send the above target component instruction to the above result distribution module; a result distribution module configured to send the above target component instruction to the above target component.

[0073] During the process of retrieving the function model, there may be a phenomenon that the function model cannot be retrieved due to system failures or model data loss. In some embodiments, the above model deployment module is further configured to: in response to determining that the above metric instruction cannot retrieve the above function model from the above machine learning layer, send the above metric instruction to the above machine learning layer so that the above machine learning layer trains the model.

[0074] Reference Figure 3 , an embodiment of fault prediction for a single photon detector component in this application is used to further describe the quantum key distribution system proposed in this application. This system includes:

[0075] The network management layer 21 is configured to obtain user instructions; according to whether there is a corresponding metric instruction for the user instruction, send the user instruction to the machine learning layer or send the metric instruction to the function layer. In this embodiment, the above network management layer can also manage other layers in terms of fault, configuration, billing, performance, and security management.

[0076] In this embodiment, the above network management layer includes: an acquisition unit configured to obtain the user instruction; a judgment unit configured to confirm whether there is a metric instruction corresponding to the user instruction in the storage unit; a sending unit configured to, in response to determining that there is a corresponding metric instruction, send the metric instruction to the function layer; and in response to determining that there is no corresponding metric instruction, send the metric instruction to the machine learning layer; a storage unit configured to receive the metric instruction sent by the machine learning layer and store the metric instruction. In this embodiment, the above function layer includes a quantum layer, a key management layer, and a control layer. In response to determining that there is a corresponding metric instruction, the metric instruction can be sent to the corresponding function layer according to the target component corresponding to the above metric instruction.

[0077] The machine learning layer 22 is configured to obtain the metric instruction through a target problem measurement program according to the user instruction, perform model training according to the metric instruction to obtain a function model, and send the metric instruction to the network management layer for the network management layer to store it.

[0078] In this embodiment, the above machine learning layer includes: a user instruction parsing unit configured to transform the above metric instruction through a target problem measurement program according to the user instruction and send the above metric instruction to the model training unit; a model training unit configured to obtain the historical operation data of the target component according to the above metric instruction and perform model training, and store it as a function model after the above model passes the test; a sending module configured to send the above metric instruction to the network management layer for the network management layer to store it. The above user instruction parsing unit includes: an analysis module that measures the user instruction through a measurement standard according to the target in the target problem measurement program and a problem preset by a human to obtain the metric of the user instruction; an orchestration module configured to obtain the above metric instruction through orchestration according to the above metric; a sending module configured to send the above metric instruction to the above model training unit.

[0079] In this embodiment, the above-mentioned model training unit includes: a function management module, which is configured to obtain the historical operation data of the target element and perform model training according to the above-mentioned indicator instructions, and send the trained function model to the test module; a test module, which is configured to test the above-mentioned function model, and in response to determining that the test result of the above-mentioned function model meets the predetermined conditions, terminate the training of the above-mentioned function model and send it to the storage module; a storage module, which is configured to store the above-mentioned function model; an application sandbox module, which is configured to simulate the training environment and the test environment. The above-mentioned function management module includes: a data acquisition module, which is configured to collect the historical operation data and current operation data of the above-mentioned target element as training data according to the above-mentioned indicator instructions; a data preprocessing module, which is configured to clean and enhance the above-mentioned training data to obtain preprocessed data; a model training module, which is configured to train the initial model according to the above-mentioned preprocessed data and the above-mentioned indicator instructions to obtain the above-mentioned function model; and a sending module, which is configured to send the above-mentioned indicator instructions and the above-mentioned function model to the above-mentioned test unit.

[0080] The control layer 23 is configured to manage and control access, routing, session, configuration, policy-based control, and controllers.

[0081] The key management layer 24 is configured to provide, manage the above keys and control the key manager.

[0082] The quantum layer 25 is configured to generate a key.

[0083] In this embodiment, the control layer, quantum layer, and key management layer are all functional layers. Therefore, the three functional layers are configured to call the functional model according to the indicator instruction, determine the target component instruction based on the output of the functional model, and send it to the target component.

[0084] In some embodiments, the functional layer further includes the network management layer, which can be configured to retrieve the functional model according to the indicator instruction, determine the target component instruction according to the output result of the functional model, and send the target component instruction to the target component.

[0085] In this embodiment, the above functional layer includes a functional application unit, and the above functional application unit includes: a data acquisition module configured to acquire historical operation data and current operation data of the above target component according to the above index instruction; a data preprocessing module configured to clean and enhance the above data to obtain preprocessed data; a model deployment module configured to retrieve the above functional model from the above machine learning layer according to the above index instruction; a model application module configured to receive the above preprocessed data and the above index instruction, and obtain an output result calculated by the above functional model; a policy construction module configured to generate a target component instruction according to the above index instruction and the above output result, and send the above target component instruction to the above result distribution module; a result distribution module configured to send the above target component instruction to the above target component. The above model deployment module is further configured to: in response to determining that the above index instruction cannot retrieve the above functional model from the above machine learning layer, send the above index instruction to the above machine learning layer so that the above machine learning layer trains the model.

[0086] Based on the same inventive concept, corresponding to the above-mentioned system of any embodiment, the present application further provides a quantum key distribution method based on machine learning.

[0087] Reference Figure 4 , the method includes the following steps:

[0088] Step S101: The network management layer obtains a user instruction; the user instruction includes at least one instruction for performance prediction or function setting of components in the quantum key distribution system; according to whether there is a corresponding index instruction for the user instruction, send the user instruction to the machine learning layer or send the index instruction to the functional layer;

[0089] Step S102: The machine learning layer obtains the index instruction through a target problem metric program according to the user instruction, trains a functional model according to the index instruction, and sends the index instruction to the network management layer for the network management layer to store it;

[0090] Step S103: The functional layer retrieves the functional model according to the index instruction, and determines a target component instruction according to the output result of the functional model and sends it to the target component.

[0091] Combined Figure 3 , through an embodiment of the present application for fault prediction of a single photon detector component, the quantum key distribution method proposed by the present application is further described:

[0092] In this embodiment, the goals and problems have been set according to the user's goals and the actual application scenarios of the system. The problems include: 1) Whether the current user instruction is used to solve a prediction problem or a classification problem; 2) What are the acting components or objects of the current user instruction; 3) What data needs to be collected to complete the current user instruction and from which components these data are sourced; 4) Whether data preprocessing and data format unification are required; 5) How many data are needed to complete the current user instruction; 6) What is the expected prediction result of the current user instruction; 7) What machine learning algorithm needs to be used for the current user instruction; 8) How should the parameters of the above machine learning algorithm be set; 9) To which components should the results output by the corresponding functional model be sent; 10) Which functional layer does the above user instruction act on?

[0093] After that, according to the above metric orchestration, an indicator instruction is obtained, and this indicator instruction is sent to the corresponding functional layer for subsequent steps and to the network management layer for inclusion. The corresponding functional layer will receive the fault prediction instruction for the user's SPD (Single-Photon Detector) component. In this embodiment, since the system has not received the above instruction, that is, the corresponding indicator instruction cannot be found in the management unit of the network management layer, the above user instruction is sent to the machine learning layer.

[0094] In the machine learning layer, first, the instruction is parsed by the parsing module in the user parsing unit. According to this instruction, it is transformed into an indicator instruction through the target problem metric program. Specifically, first, according to this user instruction, it is parsed through the above nine questions. In this embodiment, the indicators (answers) obtained from the above nine questions are respectively: 1) It belongs to a prediction problem; 2) 12 SPD devices in this system; 3) Historical data of SPD detection efficiency, qubit error rate, and key generation rate, as well as historical data of SPD device aging alarms; 4) Data preprocessing is required, including data cleaning and data augmentation, removing redundant data, and unifying the data format; 5) For each determination of SPD aging fault alarm data, 100 data of historical detection efficiency, qubit error rate, and key generation rate are obtained; 6) The expected prediction result is to predict whether an SPD aging fault alarm will occur within the next 15 to 30 minutes; 7) Use the deep fully connected neural network algorithm; 8) The number of input neurons is 42, which consists of 10 maximum performance values, 10 current performance values, 10 minimum performance values, and one-hot encoding of 12-bit SPD device numbers; the number of output neurons is 2, indicating whether the alarm will occur within the next 15 to 30 minutes (predicted alarm, non-alarm); the hidden layer is 2 layers, and the number of neurons in each layer is 60 and 120 respectively. The activation function is ReLU, and the loss function is the mean square error MSE; 9) The output result is applied to the manager in the network management layer; 10) It is applied to the network management layer.

[0095] After obtaining the above 10 indicators, according to the indicators, they are arranged by the arrangement module to obtain an indicator instruction, and the indicator instruction is sent to the model training unit through the sending module.

[0096] The model training unit will construct an initial model that meets the above indicators according to the above indicator instruction and start training. In this embodiment, the network environment of the system can be simulated through the application sandbox module to improve the efficiency of model training. After the training is completed, the model can be tested through the test module. The test method can be to confirm whether the accuracy rate of the model output result reaches the threshold or whether the output result of the model converges. The evaluation content can include: the accuracy of the prediction result, the response time, the degree of computing resource occupation, etc. The test can also simulate the network environment of the system through the application sandbox module to improve the accuracy of the test. After the test passes, a functional model corresponding to the above user instruction and indicator instruction will be obtained, and the storage module will store the functional model so that when the same user instruction is received next time, it can be directly applied without having to perform model training again.

[0097] According to the above indicators, the current user instruction should act on the network management layer. Therefore, the above user instruction is sent to the network management layer. The network management layer retrieves the functional model according to the above indicator instruction and obtains the corresponding data, and the data is preprocessed according to the indicators. Then, according to the data, through the functional model, the output result is calculated, that is, whether an SPD aging failure alarm (predicted alarm, non-alarm) will occur within the next 15 to 30 minutes and the corresponding SPD device number. The policy construction module of the network management layer will generate a target component instruction according to the above result. In this embodiment, the above policy construction module generates alarm data or display data according to the above result and transmits it to the display component so that the display component issues an alarm message.

[0098] Those of ordinary skill in the art should understand that: the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of the present application (including the claims) is limited to these examples; under the idea of the present application, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations in different aspects of the embodiments of the present application as described above, and they are not provided in detail for the sake of brevity.

[0099] In addition, for the sake of simplicity of description and discussion, and in order not to make the embodiments of the present application difficult to understand, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Further, the devices may be shown in block diagram form in order to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present application are to be implemented (i.e., these details should be fully within the understanding of those skilled in the art). In cases where specific details (such as circuits) are set forth to describe exemplary embodiments of the present application, it will be apparent to those skilled in the art that the embodiments of the present application may be implemented without these specific details or with variations of these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0100] Although the present application has been described in connection with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art in light of the foregoing description. For example, other memory architectures (such as dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0101] Embodiments of the present application are intended to cover all such alternatives, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the embodiments of the present application shall be included within the protection scope of the present application.

Claims

1. A quantum key distribution system based on machine learning, characterized in that: include: The network management layer is configured to obtain a user instruction; the user instruction includes at least one instruction of performing performance prediction or function setting on a component in the quantum key distribution system; and send the user instruction to the machine learning layer or send the indicator instruction to the function layer according to whether the user instruction has a corresponding indicator instruction; a machine learning layer configured to obtain the indicator instruction through a target problem measurement program according to the user instruction, perform model training according to the indicator instruction to obtain a functional model, and send the indicator instruction to the network management layer so that the network management layer stores it; The functional layer is configured to call the functional model according to the indicator instruction, and determine the target component instruction according to the output result of the functional model and send it to the target component; The machine learning layer includes: A user instruction parsing unit is configured to obtain the indicator instruction through a target problem measurement program according to the user instruction, and send the indicator instruction to the model training unit; a model training unit configured to obtain historical operating data of a target component according to the indicator instruction and perform model training, and store the model as a functional model after the model passes the test; The sending module is configured to send the indicator instruction to the network management layer so that the network management layer can store it.

2. The machine learning-based quantum key distribution system according to claim 1, characterized in that The network management layer includes: an acquiring unit, configured to acquire the user instruction; a judgment unit configured to confirm whether there is an indicator instruction corresponding to the user instruction in the storage unit; a sending unit configured to, in response to determining that there is a corresponding indicator instruction, send the indicator instruction to the function layer; in response to determining that there is no corresponding indicator instruction, send the indicator instruction to the machine learning layer; A storage unit is configured to receive the indicator instruction sent by the machine learning layer and store the indicator instruction.

3. The machine learning-based quantum key distribution system according to claim 1, characterized in that The user instruction parsing unit includes: a parsing module configured to parse the user instruction according to the target and the corresponding predetermined question, and determine the index of the user instruction; an orchestration module, configured to construct the indicator instruction for orchestration according to the indicator; The sending module is configured to send the indicator instruction to the model training unit.

4. The machine learning-based quantum key distribution system according to claim 3, characterized in that The parsing module is further configured to measure the user instruction using a metric based on the target in the target problem measurement program and the artificially predetermined problem to obtain an index of the user instruction.

5. The machine learning-based quantum key distribution system according to claim 1, characterized in that The model training unit includes: a function management module configured to obtain historical operating data of the target component according to the indicator instruction and perform model training, and send the trained function model to the test module; a testing module configured to test the functional model, and in response to determining that the test result of the functional model meets a predetermined condition, end the training of the functional model and send it to the storage module; A storage module, configured to store the function model; The application sandbox module is configured to simulate training and testing environments.

6. The machine learning-based quantum key distribution system according to claim 5, characterized in that: The function management module includes: a data acquisition module configured to acquire historical operating data and current operating data of the target element as training data according to the indicator instruction; A data preprocessing module is configured to clean and enhance the training data to obtain preprocessed data; A model training module is configured to train the initial model according to the preprocessing data and the indicator instruction to obtain the functional model; The sending module is configured to send the indicator instruction and the functional model to the testing unit.

7. The machine learning-based quantum key distribution system according to claim 1, characterized in that The functional layer includes a functional application unit, and the functional application unit includes: a data acquisition module, configured to acquire historical operating data and current operating data of the target element according to the indicator instruction; A data preprocessing module is configured to clean and enhance the data to obtain preprocessed data; A model deployment module, configured to retrieve the functional model from the machine learning layer according to the indicator instruction; a model application module configured to receive the pre-processed data and the indicator instruction and obtain an output result calculated by the functional model; a strategy construction module, configured to generate a target component instruction according to the indicator instruction and the output result, and send the target component instruction to the result distribution module; The result distribution module is configured to send the target component instruction to the target component.

8. The machine learning-based quantum key distribution system according to claim 1, characterized in that The model deployment module is further configured to: in response to determining that the indicator instruction cannot retrieve the functional model from the machine learning layer, send the indicator instruction to the machine learning layer so that the machine learning layer trains the model.

9. A quantum key distribution method based on machine learning, characterized in that: The method is applied to a quantum key distribution network system including a network management layer, a machine learning layer, and a functional layer; The network management layer obtains a user instruction, wherein the user instruction includes at least one instruction of performing performance prediction or function setting on a component in the quantum key distribution system; and sends the user instruction to the machine learning layer or sends the indicator instruction to the function layer according to whether the user instruction has a corresponding indicator instruction; The machine learning layer obtains the indicator instruction through the target problem measurement program according to the user instruction, performs model training according to the indicator instruction to obtain a functional model, and sends the indicator instruction to the network management layer so that the network management layer stores it; The functional layer calls the functional model according to the index instruction, and determines the target component instruction according to the output result of the functional model and sends it to the target component; The machine learning layer includes: A user instruction parsing unit, which obtains the indicator instruction through a target problem measurement program according to the user instruction, and sends the indicator instruction to the model training unit; A model training unit, which acquires historical operating data of a target component according to the indicator instruction and performs model training, and stores the model as a functional model after the model passes the test; The sending module sends the indicator instruction to the network management layer so that the network management layer can store it.

Images