AI-based quantum communication optimization method and system

Abstract

The invention discloses an AI-based quantum communication optimization method and system, and relates to the technical field of quantum communication optimizing.A quantum element reinforcement learning strategy library is locally deployed at a terminal, a self-adaptive spectrum feature extraction mechanism is combined, the channel mutation response rate is improved, and compared with centralized AI needing a cloud end to return a decision and avoid a network delay bottleneck, the AI-based quantum communication optimization method and system have the advantages that the reliability is high; the instantaneous anti-interference requirement of battlefield communication is met; under the federated learning framework, the terminal only uploads the model gradient, and the original data does not leave the local; the strategy migration unit is used for enabling a new terminal or a strange environment to quickly obtain a benchmark decision-making capability by using a pre-stored feature vector of an interference pattern library, and cracking the cold start defect of traditional federal learning; in addition, the quantum state parameters and the electromagnetic spectrum are fused through gating weighting, and the feature discrimination is enhanced while the original physical characteristics are reserved. The three-channel tensor structure provides high information density input for the lightweight CNN, and the misjudgment rate rise caused by model compression is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quantum communication optimization, and in particular to an AI-based quantum communication optimization method and system. BACKGROUND

[0002] The application demand of quantum communication in mobile scenarios such as military command and emergency communication is rapidly growing. Mobile terminals need to maintain quantum key distribution capability in complex environments such as urban areas and the wild. These environments have dense non-cooperative signal sources such as civilian electromagnetic radiation and directional interference equipment, which cause the quantum channel parameters to present high-frequency mutation characteristics. Terminal mobility further aggravates the spatial instability of the channel state.

[0003] The current new quantum communication optimization system adopts a centralized AI decision-making architecture, such as the National Quantum Science and Technology 2025 program. This system relies on the cloud to aggregate channel data from each terminal and generates an anti-interference strategy through a deep learning model. However, in actual electromagnetic countermeasure scenarios, the communication delay between the central node and the mobile terminal significantly increases, which cannot meet the real-time response requirement. At the same time, terminals in sensitive areas are subject to data security regulations and cannot transmit raw quantum state information to the cloud. Some solutions attempt to deploy lightweight models locally on terminals, but the cross-scene generalization ability is insufficient due to isolated training data.

[0004] The recently proposed federated learning framework (such as Google's 2024 prototype) allows terminals to train models locally and only share parameters to the central node. This method alleviates the data privacy problem, but does not solve the model cold start defect: new terminals or unfamiliar environments need to accumulate local data again, and the initial decision error rate is high. The edge computing solution reduces the delay, but still cannot dynamically adapt to the changing interference mode combination. Existing optimization systems have not formed an effective collaboration mechanism under the triple constraints of real-time, generalization, and privacy of mobile terminals. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] The present application provides an AI-based quantum communication optimization method and system to solve the three limitations of real-time response lag, data privacy constraints, and cross-scene generalization defects of mobile quantum terminals in complex electromagnetic environments.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application embodiment provides an AI-based quantum communication optimization method, which comprises,

[0009] Step S1, the mobile terminal collects local channel feature data in real time, and the channel feature data includes quantum state tomography parameters and electromagnetic spectrum interference modes;

[0010] Step S2, calling a pre-trained quantum element reinforcement learning strategy library to dynamically generate anti-interference protocol parameters according to the channel feature data, wherein the strategy library stores a mapping relationship between multiple groups of interference feature vectors and protocol parameters;

[0011] The construction of the quantum element reinforcement learning strategy library includes:

[0012] The generation mode of the interference feature vector: the electromagnetic spectrum interference mode is converted into a frequency domain energy distribution vector through Fourier transform;

[0013] The updating mechanism of the strategy network: when the terminal detects an unrecorded interference mode, the local strategy network is triggered for incremental training, and the encrypted feature vector is uploaded to the federal layer;

[0014] Step S3, performing a quantum key distribution operation based on the generated protocol parameters, and recording communication performance index feedback to the local model.

[0015] As a preferred scheme of the AI-based quantum communication optimization method, in the generation process of the interference feature vector in step S2:

[0016] The mobile terminal uniformly samples the original electromagnetic interference waveform in a single frame observation window to obtain a discrete sequence:

[0017]

[0018] wherein, X (n) represents the amplitude of the nth sampling point, f (t) represents a continuous time signal, t represents a sampling time, f represents a sampling frequency, Hz, n represents a sampling serial number, and N represents the total number of single-frame sampling points.

[0019] The adaptive Kaiser window is applied to X (n) :

[0020]

[0021] wherein, a (n) represents a window function coefficient, b (n) represents a zero-order first-order modified Bessel function, represents a constant of pi, and a represents a window shape control coefficient.

[0022] With the current signal-to-noise ratio ​​​​​​​​​​​​​​​​​Adaptive change, adaptive formula is:

[0023] ,

[0024] Wherein, and are the lower and upper limits of the window shape coefficient, is the slope adjustment coefficient, is the center offset constant, is the base of natural logarithm, is the signal-to-noise ratio, unit: dB;

[0025] After windowing, we get , wherein, is the windowed sequence;

[0026] Discrete Fourier transform is performed on :

[0027] , ,

[0028] Wherein, is the complex amplitude of the th frequency component, is the frequency index, is the imaginary unit;

[0029] Calculate and normalize the power spectrum to form the energy vector:

[0030] , ,

[0031] Wherein, is the energy proportion of the th frequency point, is the summation index, is the normalized energy distribution vector with length , denotes matrix transpose;

[0032] Real-time interference detection threshold determination:

[0033] ,

[0034] Wherein, is the length of the continuous detection window, is the starting frequency point index, is the indicator function, which takes 1 if the condition is met, otherwise 0, is the energy threshold, is the determination proportion threshold; if the condition is met, and the timestamp are packaged into a feature vector and uploaded to the local strategy network to trigger incremental training.

[0035] As a preferred solution of the AI-based quantum communication optimization method described in the present invention, the construction of the quantum element reinforcement learning strategy library in step S2 includes:

[0036] Use quantum generative adversarial networks to simulate electromagnetic interference scenarios and generate interference feature vectors;

[0037] Use the proximal policy optimization algorithm to train the policy network and output the optimal protocol parameter combination;

[0038] The simulation process of the quantum generative adversarial network includes:

[0039] Generator input: a random noise vector that follows a Rayleigh distribution; the Rayleigh distribution is used to simulate the random fading characteristics of the electromagnetic interference amplitude;

[0040] Discriminator structure: Contains quantum convolution layers to identify the distribution differences between interference patterns and real battlefield data.

[0041] As a preferred solution of the AI-based quantum communication optimization method described in the present invention, before generating the anti-interference protocol parameters in step S2, a feature fusion operation is first performed on the quantum state parameters and the electromagnetic parameters:

[0042] In the feature fusion operation, the step of splicing the quantum state parameters and the electromagnetic parameters into a mixed tensor includes:

[0043] Arrange the quantum state parameters into a column vector:

[0044] , , ,

[0045] in, For the Quantum state tomography characteristics, is the quantum characteristic dimension, is the original quantum vector, for it norm, is the normalized component;

[0046] Arrange the electromagnetic spectrum amplitude parameters into row vectors:

[0047] , , ,

[0048] in, For the Frequency domain amplitude features, is the electromagnetic characteristic dimension, is the original electromagnetic vector, is the norm, is the normalized component;

[0049] The matching dimension matrix is generated by row and column replication:

[0050] , , , ,

[0051] wherein, is the quantum feature row replication matrix, is the electromagnetic feature column replication matrix, and are the quantum and electromagnetic indices, respectively;

[0052] The row and column gating weights are introduced:

[0053] , ,

[0054] wherein, and are the gating coefficients, is the trainable scalar weight, is the bias term, is the Sigmoid activation function;

[0055] The complementary matrix is constructed using the gating coefficients:

[0056] ,

[0057] wherein, is the complementary fusion matrix, is the element;

[0058] The three-channel hybrid tensor is formed by channel dimension splicing:

[0059] , ,

[0060] wherein, is the tensor element of the channel , is the channel index, represents the concatenation operation in the channel dimension.

[0061] As a preferred scheme of the AI-based quantum communication optimization method provided by the present application, the feedback in step S3 triggers the following operations:

[0062] uploading local model gradient to the federal aggregation server when the communication error rate exceeds a threshold value;

[0063] downloading noise-injected global model parameters from the server, updating the local decision network.

[0064] In a second aspect, the present application provides an AI-based quantum communication optimization system, comprising,

[0065] Terminal layer: a local decision module deployed in a mobile terminal, comprising:

[0066] Channel monitoring unit, real-time acquisition of quantum state tomography parameters and electromagnetic spectrum data;

[0067] Lightweight CNN model, outputting protocol parameters according to input data;

[0068] The lightweight CNN model comprises:

[0069] Input layer: receiving Stokes parameters in quantum state tomography parameters and frequency domain amplitude of electromagnetic spectrum;

[0070] Feature fusion layer: concatenating quantum state parameters and electromagnetic parameters into a hybrid tensor;

[0071] Output layer: generating protocol type identifier and light source pulse frequency adjustment amount;

[0072] Federal layer: central server, comprising:

[0073] Model aggregation unit, adding Gaussian noise to the model gradient uploaded by each terminal;

[0074] Interference mode library, storing a set of feature vectors of typical electromagnetic interference scenarios.

[0075] As a preferred scheme of the AI-based quantum communication optimization system, the terminal layer further comprises:

[0076] Strategy migration unit, when the terminal enters a new electromagnetic environment, matching a similar feature vector from the interference mode library to initialize the local CNN model parameters.

[0077] As a preferred scheme of the AI-based quantum communication optimization system, the model aggregation unit of the federal layer performs:

[0078] Weighted average of the received gradient parameters;

[0079] Superimposing Laplace noise meeting the differential privacy requirement to generate a global model;

[0080] The weighted average operation comprises:

[0081] Assign gradient weights based on the terminal's historical communication success rate;

[0082] The gradient of low-weight terminals is clipped to suppress the influence of outliers.

[0083] As a preferred solution of the AI-based quantum communication optimization system described in the present invention, the interference pattern library includes:

[0084] Quantum channel response characteristics of dense urban electromagnetic radiation, directional interference equipment, and multipath fading in mobile scenarios;

[0085] The protocol parameter optimization history associated with each eigenvector;

[0086] The interference pattern library stores historical bit error rate threshold records.

[0087] As a preferred solution of the AI-based quantum communication optimization system described in the present invention, the local decision module interacts with the federal layer through an encrypted channel, and the original channel data of the terminal does not leave the local area.

[0088] The beneficial effects of the present invention are as follows: the present invention deploys a quantum element reinforcement learning strategy library locally on the terminal, combined with an adaptive spectrum feature extraction mechanism, to improve the channel mutation response rate. Compared with centralized AI that requires cloud-based decision-making, it avoids network delay bottlenecks and meets the instantaneous anti-interference requirements of battlefield communications. Under the federated learning framework, the terminal only uploads model gradients, and the original data does not leave the local area. The strategy migration unit uses the pre-stored feature vectors of the interference pattern library to enable new terminals or unfamiliar environments to quickly obtain benchmark decision-making capabilities, breaking the cold start defects of traditional federated learning. In addition, the quantum state parameters of the present invention are fused with the electromagnetic spectrum through gated weighting to enhance feature discriminability while retaining the original physical properties. The three-channel tensor structure provides high information density input for lightweight CNN, avoiding the increase in misjudgment rate due to model compression. Unrecorded interference triggers incremental training and feature upload, allowing the policy library to continuously expand its coverage. Combined with the global model aggregation of the federal layer, a closed-loop learning ecosystem of terminal perception-local optimization-federated evolution is formed to adapt to the continuous upgrading of electromagnetic confrontation scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0090] Figure 1 This is a flow chart of the AI-based quantum communication optimization method in Example 1.

[0091] Figure 2 A framework diagram of the AI-based quantum communication optimization system in Embodiment 1. DETAILED DESCRIPTION

[0092] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0093] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0094] Secondly, "one embodiment" or "embodiment" referred to herein means that a specific feature, structure or characteristic can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is separate or alternative to other embodiments.

[0095] Embodiment 1, with reference to Figure 1 and Figure 2 The embodiment provides an AI-based quantum communication optimization method, comprising the following steps:

[0096] Step S1, the mobile terminal collects local channel characteristic data in real time, and the channel characteristic data includes quantum state tomography parameters and electromagnetic spectrum interference patterns;

[0097] Step S2, calling a pre-trained quantum element reinforcement learning strategy library, dynamically generating anti-interference protocol parameters according to the channel characteristic data, wherein the strategy library stores the mapping relationship between multiple groups of interference characteristic vectors and protocol parameters;

[0098] The construction of the quantum element reinforcement learning strategy library includes:

[0099] The generation mode of the interference characteristic vector: the electromagnetic spectrum interference pattern is converted into a frequency energy distribution vector through Fourier transform;

[0100] The updating mechanism of the strategy network: when the terminal detects an unrecorded interference pattern, the local strategy network incremental training is triggered, and the encrypted characteristic vector is uploaded to the federal layer;

[0101] The construction of the quantum element reinforcement learning strategy library in step S2 includes:

[0102] Simulate the electromagnetic interference scene through the quantum generative adversarial network to generate the interference characteristic vector;

[0103] Use the proximal policy optimization algorithm to train the policy network and output the optimal protocol parameter combination;

[0104] During the generation of the interference feature vector in step S2:

[0105] The mobile terminal observes the original electromagnetic interference waveform in a single frame observation window Perform uniform sampling to obtain a discrete sequence:

[0106] , ,

[0107] in, Indicates the The amplitude of the sampling points, is a continuous-time signal, is the sampling time, is the sampling frequency in Hz, is the sampling sequence number, is the total number of sampling points in a single frame;

[0108] right Apply an adaptive Kaiser window to reduce spectral leakage:

[0109] , ,

[0110] in, is the window function coefficient, is the zero-order modified Bessel function of the first kind, is the circumference constant, is the window shape control coefficient;

[0111] With the current signal-to-noise ratio Adaptive changes, the adaptive formula is:

[0112] ,

[0113] in, and are the lower and upper limits of the window coefficient, is the slope adjustment coefficient, is the center bias constant, is the base of natural logarithms, is the signal-to-noise ratio, in dB;

[0114] After windowing, we get ,in, is the windowed sequence;

[0115] right Perform a discrete Fourier transform:

[0116] , ,

[0117] wherein, is the th frequency component complex amplitude, is the frequency index, is the imaginary unit;

[0118] The power spectrum is calculated and normalized to form an energy vector:

[0119] , ,

[0120] wherein, is the th frequency bin energy ratio, is the summation index, is the normalized energy distribution vector of length denotes the matrix transpose; Real-time interference detection threshold determination:

[0121]

[0122] , wherein,

[0123] is the continuous detection window length, is the starting frequency index, is the indicator function, taking 1 if the condition is met, otherwise 0, is the energy threshold, is the determination proportion threshold; if the condition is met, the and the timestamp are packaged as a feature vector and uploaded to the local strategy network to trigger incremental training; Specifically, this process suppresses sidelobes with adaptive Kaiser window, improves narrowband feature recognition, and makes window shape coefficients smoothly transition under different signal-to-noise ratios through logical S-shaped mapping, balancing weak signal amplification and strong interference suppression; Normalized power spectrum provides a power-independent feature scale, which is beneficial for cross-device model sharing; The determination formula uses a combination of sliding window averaging and proportion threshold to maintain detection stability when the spectrum fluctuates rapidly, and triggers incremental learning updates to enhance the model's online adaptation ability;

[0124] The simulation process of the quantum generative adversarial network includes:

[0125] Generator input: random noise vector subject to Rayleigh distribution (simulate electromagnetic environment randomness);

[0126]

[0127] ​Before generating the anti-interference protocol parameters in step S2, a feature fusion operation is performed on the quantum state parameters and the electromagnetic parameters:

[0128] In the feature fusion operation, the steps of concatenating quantum state parameters and electromagnetic parameters into a mixed tensor include:

[0129] Arrange the quantum state parameters into a column vector:

[0130] , , ,

[0131] in, For the Quantum state tomography characteristics, is the quantum characteristic dimension, is the original quantum vector, for it norm, is the normalized component;

[0132] Arrange the electromagnetic spectrum amplitude parameters into row vectors:

[0133] , , ,

[0134] in, For the Frequency domain amplitude features, is the electromagnetic characteristic dimension, is the original electromagnetic vector, for it norm, is the normalized component;

[0135] Generate a matrix of matching dimensions by copying rows and columns:

[0136] , , , ,

[0137] in, is the quantum eigenrow replication matrix, Replicate the matrix for the electromagnetic eigenvalue columns, and respectively, quantum and electromagnetic indexing;

[0138] Introducing row and column gating weights (highlighting complementarity):

[0139] , ,

[0140] in, with being a gating coefficient, being a trainable scalar weight, being a bias term, being a Sigmoid activation function;

[0141] Constructing complementary matrix with gating coefficient:

[0142] ,

[0143] wherein, is a complementary fusion matrix, is the element of the th row and the th column;

[0144] Concatenating along channel dimension to form a three-channel mixed tensor:

[0145] , ,

[0146] wherein, is the tensor element of the channel , is the channel index, denotes the concatenation operation along the channel dimension;

[0147] Specifically, this step eliminates the amplitude difference of different dimensions through vector normalization, so that the subsequent convolution kernel focuses on the relative feature distribution. Row and column replication expands the one-dimensional sequence into a two-dimensional plane that can be convolved, and the two types of features are explicitly aligned in the same coordinate system. The gating weight introduces adaptive channel attention for each row and each column, enhances the response sensitivity of the model to key quantum state features and high-energy spectral lines, and provides nonlinear coupling perspectives through the complementary matrix The three-channel design not only retains the original modal information but also injects coupling semantics, achieving high discriminability under a lightweight structure and providing rich and compressed feature representations for subsequent convolution layers.

[0148] Discriminator structure: contains a quantum convolution layer to identify the distribution difference between interference patterns and real battlefield data.

[0149] Step S3: Perform quantum key distribution operations based on the generated protocol parameters, and record communication performance index feedback to the local model;

[0150] The feedback in step S3 triggers the following operations:

[0151] When the communication error rate exceeds the threshold, upload the local model gradient to the federated aggregation server;

[0152] Download the noise-injected global model parameters from the server and update the local decision network.

[0153] The embodiment also provides an AI-based quantum communication optimization system, comprising:

[0154] Terminal layer: a local decision module deployed in a mobile terminal, comprising:

[0155] A channel monitoring unit acquires quantum state tomography parameters and electromagnetic spectrum data in real time;

[0156] A lightweight CNN model outputs protocol parameters according to input data;

[0157] The terminal layer further comprises:

[0158] A strategy migration unit matches a similar feature vector from an interference mode library and initializes local CNN model parameters when the terminal enters a new electromagnetic environment;

[0159] The lightweight CNN model comprises:

[0160] An input layer: receiving Stokes parameters in quantum state tomography parameters and frequency domain amplitudes of electromagnetic spectrum;

[0161] A feature fusion layer: concatenating quantum state parameters and electromagnetic parameters into a hybrid tensor;

[0162] An output layer: generating a protocol type identifier and a light source pulse frequency adjustment amount;

[0163] Federal layer: a central server, comprising:

[0164] A model aggregation unit adds Gaussian noise to model gradients uploaded by each terminal;

[0165] An interference mode library: storing a feature vector set of typical electromagnetic interference scenarios;

[0166] The model aggregation unit of the federal layer performs:

[0167] Weighted averaging on received gradient parameters;

[0168] Superimposing Laplace noise meeting differential privacy requirements to generate a global model;

[0169] The weighted averaging operation comprises:

[0170] Assigning gradient weights according to terminal historical communication success rates;

[0171] Amplitude clipping of gradients of low-weight terminals to suppress the influence of outliers;

[0172] The interference mode library contains:

[0173] Quantum channel response characteristics of urban dense electromagnetic radiation, directional interference equipment and multipath fading in mobile scenarios;

[0174] a protocol parameter optimization history record associated with each feature vector;

[0175] The local decision module interacts with the federation layer through an encrypted channel, and terminal raw channel data does not go out of the local.

[0176] It should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. An AI-based quantum communication optimization method, characterized in that: include, Step S1: The mobile terminal collects local channel characteristic data in real time, wherein the channel characteristic data includes quantum state tomography parameters and electromagnetic spectrum interference patterns; Step S2, calling a pre-trained quantum element reinforcement learning strategy library to dynamically generate anti-interference protocol parameters according to the channel feature data, wherein the strategy library stores a mapping relationship between multiple sets of interference feature vectors and protocol parameters; The construction of the quantum element reinforcement learning strategy library includes: Interference characteristic vector generation method: convert the electromagnetic spectrum interference pattern into frequency domain energy distribution vector through Fourier transform; Policy network update mechanism: When a terminal detects an unrecorded interference pattern, it triggers incremental training of the local policy network and uploads the encrypted feature vector to the federation layer; Step S3: Perform quantum key distribution operations based on the generated protocol parameters, and record the communication performance indicators and feed them back to the local model.

2. The AI-based quantum communication optimization method according to claim 1, wherein: During the generation of the interference feature vector in step S2: The mobile terminal observes the original electromagnetic interference waveform in a single frame observation window Perform uniform sampling to obtain a discrete sequence: ， ， in, Indicates the The amplitude of the sampling points, is a continuous-time signal, is the sampling time, is the sampling frequency in Hz, is the sampling sequence number, is the total number of sampling points in a single frame; right Apply an adaptive Kaiser window: ， ， in, is the window function coefficient, is the zero-order modified Bessel function of the first kind, is the circumference constant, is the window shape control coefficient; With the current signal-to-noise ratio Adaptive changes, the adaptive formula is: ， in, and are the lower and upper limits of the window coefficient, is the slope adjustment coefficient, is the center bias constant, is the base of natural logarithms, is the signal-to-noise ratio, in dB; After windowing, we get ,in, is the windowed sequence; right Perform a discrete Fourier transform: ， ， in, For the The complex amplitude of the frequency component, is the frequency index, is an imaginary unit; Compute and normalize the power spectrum to form the energy vector: ， ， in, For the The energy proportion of each frequency point, To sum the index, is the length The normalized energy distribution vector of Represents matrix transpose; Perform real-time interference detection threshold determination: ， in, is the continuous detection window length, is the starting frequency index, is an indicator function, which takes 1 if the condition is met, otherwise it takes 0. is the energy threshold, To determine the ratio threshold; if the condition is met, The data is packaged with the timestamp as a feature vector and uploaded to the local policy network to trigger incremental training.

3. The AI-based quantum communication optimization method according to claim 1, wherein: The construction of the quantum element reinforcement learning strategy library in step S2 includes: Use quantum generative adversarial networks to simulate electromagnetic interference scenarios and generate interference feature vectors; Use the proximal policy optimization algorithm to train the policy network and output the optimal protocol parameter combination; The simulation process of the quantum generative adversarial network includes: Generator input: a random noise vector that follows a Rayleigh distribution; the Rayleigh distribution is used to simulate the random fading characteristics of the electromagnetic interference amplitude; Discriminator structure: Contains quantum convolution layers to identify the distribution differences between interference patterns and real battlefield data.

4. The AI-based quantum communication optimization method according to claim 3, wherein: Before generating the anti-interference protocol parameters in step S2, a feature fusion operation is performed on the quantum state parameters and the electromagnetic parameters: In the feature fusion operation, the step of splicing the quantum state parameters and the electromagnetic parameters into a mixed tensor includes: Arrange the quantum state parameters into a column vector: ， ， ， in, For the Quantum state tomography characteristics, is the quantum characteristic dimension, is the original quantum vector, for it norm, is the normalized component; Arrange the electromagnetic spectrum amplitude parameters into row vectors: ， ， ， in, For the Frequency domain amplitude features, is the electromagnetic characteristic dimension, is the original electromagnetic vector, for it norm, is the normalized component; Generate a matrix of matching dimensions by copying rows and columns: ， ， ， ， in, is the quantum eigenrow replication matrix, is the electromagnetic eigenvalue column replication matrix, and respectively, quantum and electromagnetic indexing; Introducing row and column gating weights: ， ， in, and is the gating coefficient, is a trainable scalar weight, is the bias term, is the Sigmoid activation function; Construct a complementary matrix using the gating coefficients: ， in, is the complementary fusion matrix, For the element; Concatenate in the channel dimension to form a three-channel mixed tensor: ， ， in, For channel The tensor elements of is the channel index, Indicates cascading operations along the channel dimension.

5. The AI-based quantum communication optimization method according to claim 1, wherein: The feedback in step S3 triggers the following actions: When the communication bit error rate exceeds the threshold, the local model gradient is uploaded to the federated aggregation server; Download the noise-injected global model parameters from the server and update the local decision network.

6. An AI-based quantum communication optimization system, based on the AI-based quantum communication optimization method according to any one of claims 1 to 5, characterized in that: include: Terminal layer: A local decision-making module deployed on mobile terminals, including: Channel monitoring unit, which obtains quantum state tomography parameters and electromagnetic spectrum data in real time; Lightweight CNN model that outputs protocol parameters based on input data; The lightweight CNN model includes: Input layer: receives the Stokes parameters in the quantum state tomography parameters and the frequency domain amplitude of the electromagnetic spectrum; Feature fusion layer: concatenates quantum state parameters and electromagnetic parameters into a mixed tensor; Output layer: Generates protocol type identifier and light source pulse frequency adjustment; Federation layer: Central server, including: Model aggregation unit, which adds Gaussian noise to the model gradients uploaded by each terminal; Interference pattern library, which stores feature vector sets of typical electromagnetic interference scenarios.

7. The AI-based quantum communication optimization system according to claim 6, characterized in that: The terminal layer also includes: The strategy migration unit, when the terminal enters a new electromagnetic environment, matches similar feature vectors from the interference pattern library and initializes the local CNN model parameters.

8. The AI-based quantum communication optimization system according to claim 6, characterized in that: The model aggregation unit of the federation layer performs: Perform weighted averaging on the received gradient parameters; Superimpose Laplace noise that meets differential privacy requirements to generate a global model; The weighted averaging operation includes: Assign gradient weights based on the terminal's historical communication success rate; The gradient of low-weight terminals is clipped to suppress the influence of outliers.

9. The AI-based quantum communication optimization system according to claim 6, characterized in that: The interference pattern library includes: Quantum channel response characteristics of dense urban electromagnetic radiation, directional interference equipment, and multipath fading in mobile scenarios; The protocol parameter optimization history associated with each eigenvector; The interference pattern library stores historical bit error rate threshold records.

10. An AI-based quantum communication optimization system according to any one of claims 6 to 9, characterized in that: The local decision module interacts with the federal layer through an encrypted channel, and the original channel data of the terminal does not leave the local area.

Images