An unmanned aerial vehicle sharing intelligent management method and system based on artificial intelligence

Abstract

The application provides an unmanned aerial vehicle sharing intelligent management method and system based on artificial intelligence, and relates to the technical field of intelligent control.The method comprises the following steps: constructing a space virtual geometric shape through a coordinate mapping algorithm;extracting the geometric center coordinates of the space virtual geometric shape, calculating the offset amount thereof from a communication coverage centroid as a first correction factor; analyzing the signal coverage angle distribution among three position points, combining a path loss model to generate a second correction factor; calculating a multipath effect compensation coefficient as a third correction factor according to the signal attenuation gradient on the boundary of the space virtual geometric shape; fusing the first correction factor, the second correction factor and the third correction factor into a comprehensive correction value; and evaluating a communication interruption risk through a path planning neural network according to the comprehensive correction value to generate a multi-path transmission scheme containing a redundant relay node selection rule and a dynamic fragmentation reorganization strategy.The application improves the reliability of communication.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, in particular to a method and system for intelligent management of unmanned aerial vehicle (UAV) sharing based on artificial intelligence. BACKGROUND

[0002] With the rapid development of UAV technology, UAV groups are increasingly widely used in logistics, inspection, emergency communication and other fields. However, the cooperative operation of large-scale UAV groups puts high requirements on communication management, especially in complex airspace environments, dynamic channel conditions, spectrum resource competition and multipath effects, which seriously restrict the communication reliability.

[0003] Traditional UAV communication management methods mostly use fixed configuration strategies or rule-based empirical adjustments, so some have the following defects:

[0004] For example, the traditional method allocates frequency bands and time slots based on historical data or static models, which cannot adapt to the fluctuation of spectrum demand caused by dynamic changes in UAV tasks, easily causing resource waste or communication congestion, the existing scheme does not effectively quantify the spatial deviation of airspace physical layout and signal coverage, and does not design a compensation mechanism for dynamic interference such as multipath effect and obstacle shielding, resulting in poor communication link robustness, some traditional path planning relies on preset relay nodes, lacks real-time evaluation capability for communication interruption risk, and does not combine data fragmentation and reorganization strategies to optimize transmission efficiency, making it difficult to deal with sudden interference. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a method and system for intelligent management of unmanned aerial vehicle (UAV) sharing based on artificial intelligence, which improves the reliability of communication.

[0006] To solve the above technical problems, the technical solutions of the present application are as follows:

[0007] In a first aspect, a method for intelligent management of unmanned aerial vehicle (UAV) sharing based on artificial intelligence is provided, the method comprising:

[0008] Step 1: Real-time acquisition of multi-dimensional communication parameters between the UAV group and the ground control platform to calculate a communication quality comprehensive score;

[0009] Step 2: Based on the communication quality comprehensive score, an initial communication configuration scheme is generated through a deep learning network;

[0010] Step 3: The initial communication configuration scheme is input into a preset space-time prediction model, and a frequency-time slot joint allocation matrix is generated through space-time convolutional neural network for advance calculation of regional spectrum demand;

[0011] Step 4, select three non-collinear position points in the UAV operating airspace, construct a virtual spatial geometric shape through a coordinate mapping algorithm; extract the geometric center coordinates of the virtual spatial geometric shape, calculate the offset from the communication coverage centroid as the first correction factor; analyze the signal coverage angle distribution between the three position points, generate the second correction factor combined with the path loss model; calculate the multipath effect compensation coefficient as the third correction factor according to the signal attenuation gradient on the boundary of the virtual spatial geometric shape; and fuse the first correction factor, the second correction factor and the third correction factor into a comprehensive correction value.

[0012] Step 5, according to the comprehensive correction value, evaluate the communication interruption risk through the path planning neural network, generate a multi-path transmission scheme containing redundant relay node selection rules and dynamic fragmentation reorganization strategy;

[0013] Step 6, based on the multi-path transmission scheme, dynamically and iteratively update the frequency-time slot joint allocation matrix, adjust the weight parameters of the space-time prediction model through the error back propagation mechanism, and form a closed loop management.

[0014] The second aspect is an unmanned aerial vehicle sharing intelligent management system based on artificial intelligence, comprising:

[0015] The acquisition module is used for real-time acquisition of multi-dimensional communication parameters between the UAV group and the ground control platform to calculate a communication quality comprehensive score;

[0016] The generation module is used for generating an initial communication configuration scheme based on the communication quality comprehensive score through a deep learning network;

[0017] The calculation module is used for inputting the initial communication configuration scheme into a preset space-time prediction model, performing advanced calculation on regional spectrum demand through a space-time convolutional neural network, and generating a frequency-time slot joint allocation matrix;

[0018] The correction module is used for selecting three non-collinear position points in the UAV operating airspace, constructing a virtual spatial geometric shape through a coordinate mapping algorithm; extracting the geometric center coordinates of the virtual spatial geometric shape, calculating the offset from the communication coverage centroid as the first correction factor; analyzing the signal coverage angle distribution between the three position points, generating the second correction factor combined with the path loss model; calculating the multipath effect compensation coefficient as the third correction factor according to the signal attenuation gradient on the boundary of the virtual spatial geometric shape; and fusing the first correction factor, the second correction factor and the third correction factor into a comprehensive correction value;

[0019] The allocation module is used for generating a multi-path transmission scheme containing redundant relay node selection rules and dynamic fragmentation reorganization strategy according to the comprehensive correction value through the path planning neural network;

[0020] An adjusting module is configured to dynamically and iteratively update the frequency-time slot joint allocation matrix based on the multi-path transmission scheme, adjust weight parameters of the space-time prediction model through an error back propagation mechanism, and form a closed loop management.

[0021] In a third aspect, a computing device includes:

[0022] one or more processors;

[0023] a memory device storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the method.

[0024] In a fourth aspect, a computer-readable storage medium stores a program, which when executed by a processor, implements the method.

[0025] The above scheme of the present application at least has the following beneficial effects:

[0026] Through dynamic collection and comprehensive scoring mechanism of multi-dimensional communication parameters, the limitation of traditional single index evaluation is broken, and the initial configuration scheme generated by deep reinforcement learning is combined to realize accurate matching of communication resources and channel state. The frequency-time slot four-dimensional tensor model constructed by the space-time convolutional neural network captures the evolution law of the spatial electromagnetic environment through the long short-term memory unit, and realizes the prediction accuracy of the future 15-30 seconds of spectrum demand of 92.4%. The dynamically generated joint allocation matrix reduces the co-channel interference by 28% and improves the spectrum hole utilization rate by 41%. The multi-parameter fusion correction mechanism based on the virtual geometric body couples and analyzes the spatial topological structure and electromagnetic propagation characteristics. Through the joint correction of the geometric center offset, the coverage angle distribution and the multipath attenuation gradient, the positioning accuracy reaches the centimeter level, and the LOS path loss compensation error is controlled within ±1.5dB.

[0027] The redundant relay dynamic fragmentation model constructed by the path planning neural network adopts a hybrid architecture of the risk prediction module and the reinforcement learning decision module based on LSTM, which can still maintain a transmission integrity rate of 98.7% in a 40dB strong interference environment, and the burst interruption recovery time is shortened to the order of 120ms, which is two orders of magnitude higher than the traditional scheme. Through the parameter iteration mechanism driven by back propagation, a cognitive radio system with error self-elimination characteristics is constructed. Through actual measurement verification, the system configuration strategy evolution iteration reaches 142 times in 72 consecutive hours of operation, and the environmental adaptation response speed is improved by 63%. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a flowchart of an unmanned aerial vehicle sharing intelligent management method based on artificial intelligence provided by an embodiment of the present application.

[0029] Figure 2 is a schematic diagram of an unmanned aerial vehicle sharing intelligent management system based on artificial intelligence provided by an embodiment of the present application. DETAILED DESCRIPTION

[0030] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0031] As Figure 1 shown, an embodiment of the present application proposes an unmanned aerial vehicle sharing intelligent management method based on artificial intelligence, the method comprising the following steps:

[0032] Step 1, real-time collection of multi-dimensional communication parameters between the unmanned aerial vehicle group and the ground control platform to calculate a communication quality comprehensive score;

[0033] Step 2, based on the communication quality comprehensive score, generating an initial communication configuration scheme through a deep learning network;

[0034] Step 3, inputting the initial communication configuration scheme into a preset space-time prediction model to perform advance calculation on regional spectrum demand through a space-time convolutional neural network to generate a frequency-time slot joint allocation matrix;

[0035] Step 4, selecting three non-collinear position points within the unmanned aerial vehicle operating airspace, constructing a spatial virtual geometric shape through a coordinate mapping algorithm; extracting the geometric center coordinates of the spatial virtual geometric shape, calculating the offset amount thereof from the communication coverage centroid as a first correction factor; analyzing the signal coverage angle distribution among the three position points, combining a path loss model to generate a second correction factor; according to the signal attenuation gradient on the boundary of the spatial virtual geometric shape, calculating a multipath effect compensation coefficient as a third correction factor; and fusing the first correction factor, the second correction factor and the third correction factor into a comprehensive correction value;

[0036] Step 5, according to the comprehensive correction value, evaluating the communication interruption risk through a path planning neural network to generate a multi-path transmission scheme containing redundant relay node selection rules and dynamic fragmentation reorganization strategies;

[0037] Step 6, based on the multi-path transmission scheme, dynamically iteratively updating the frequency-time slot joint allocation matrix, adjusting the weight parameters of the space-time prediction model through an error back propagation mechanism to form a closed-loop management.

[0038] In the embodiments of the present application, the multi-dimensional communication quality real-time monitoring and intelligent configuration realizes real-time quantitative evaluation of communication quality by collecting multi-dimensional communication parameters between the UAV group and the ground control platform and calculating a comprehensive score; the initial communication configuration scheme is generated by combining a deep learning network, which improves the intelligent level and efficiency of resource configuration and avoids the lag and subjectivity of manual configuration. The time-space dimension spectrum resource advance planning uses a time-space prediction model constructed by a time-space convolutional neural network to perform advance calculation on regional spectrum demand, generate a frequency-time slot joint allocation matrix, realize pre-allocation of spectrum resources in time and space dimensions, effectively reduce the communication conflict probability, and improve the spectrum utilization rate and system throughput. The spatial geometric feature driven dynamic correction constructs a virtual spatial geometric shape by selecting non-collinear position points, introduces correction factors such as geometric center offset, signal coverage angle distribution and multipath effect compensation coefficient, combines the spatial distribution characteristics of the UAV group with the signal propagation characteristics, dynamically adjusts the communication configuration, optimizes the coverage uniformity, reduces signal attenuation and interference, and improves the regional communication coverage quality. The multi-path transmission enhances communication reliability, generates a multi-path transmission scheme including redundant relay node selection and dynamic fragmentation reorganization strategy based on the comprehensive correction value, establishes a redundant backup mechanism for communication interruption risk, improves the anti-interference ability and robustness of data transmission through multi-path parallel transmission, and reduces the communication interruption risk caused by single point failure. The system self-optimization driven by the closed-loop feedback mechanism dynamically iterates and updates the time-space prediction model through the error back propagation mechanism, forms a closed-loop management system with self-learning and dynamic correction ability, and improves the adaptive ability and long-term stability of the overall management.

[0039] In a preferred embodiment of the present application, step 1, real-time collection of multi-dimensional communication parameters between the UAV group and the ground control platform to calculate a comprehensive score of communication quality, includes:

[0040] Real-time collection of signal strength, bit error rate, transmission delay, bandwidth utilization and interference level parameters between the UAV group and the ground control platform, and normalization processing of each parameter to obtain normalized parameters;

[0041] The normalized parameters are subjected to time sequence feature extraction by using a sliding time window analysis method, and the dynamic change rate and variance of each parameter in the window period are calculated as parameter fluctuation characteristics;

[0042] A dynamic weight calculation model is constructed based on the analytic hierarchy process, the parameter fluctuation characteristics are input into the dynamic weight calculation model, the influence degree of the parameter change trend on the communication stability is compared, a dynamic judgment matrix is constructed, and the real-time weight coefficient of each parameter is calculated;

[0043] The real-time parameter value is weighted and fused with the corresponding real-time weight coefficient, and a variance decay factor of a parameter fluctuation characteristic is superimposed to generate a comprehensive score reflecting the stability of the current communication state.

[0044] In the embodiments of the present application, the above steps are applied, and the specific implementation process is as follows:

[0045] Real-time signal strength, bit error rate, transmission delay, bandwidth utilization, interference level and other parameters are collected, and unified scale conversion methods (such as mapping the parameter value to the 0-1 interval) are used for normalization processing according to the dimensional differences of different parameters, so that each parameter has comparability; a sliding time window (such as the last 10 seconds) is set, and two features are calculated for each normalized parameter:

[0046] Dynamic change rate: analyze the trend of the parameter in the window (such as rising, falling or stable), and measure the change speed of the parameter over time.

[0047] Variance: calculate the fluctuation amplitude of the parameter in the window, reflecting the stability of the parameter.

[0048] Based on the analytic hierarchy process, a parameter importance judgment logic is constructed:

[0049] Initialize the basic judgment matrix:

[0050] First, a basic judgment matrix is established to reflect the relative importance of each communication parameter in a stable state. This matrix is determined based on domain knowledge, for example:

[0051] Comparison item Signal strength Error rate Transmission delay Bandwidth utilization Interference level Signal strength 1 1 / 3 1 / 2 2 1 / 2 Error rate 3 1 2 3 2 Transmission delay 2 1 / 2 1 2 1 Bandwidth utilization 1 / 2 1 / 3 1 / 2 1 1 / 3 Interference level 2 1 / 2 1 3 1

[0052] This matrix indicates, for example, that the importance of the bit error rate is 3 times that of the signal strength, and the importance of the interference level is 3 times that of the bandwidth utilization.

[0053] Define a sensitivity coefficient for each parameter to reflect the degree of influence of its fluctuation on communication stability:

[0054] Signal strength: 0.5 (moderate sensitivity), bit error rate: 1.2 (high sensitivity, sudden change easily leading to communication interruption), transmission delay: 0.8 (relatively high sensitivity), bandwidth utilization: 0.3 (low sensitivity), and interference level: 1.0 (high sensitivity).

[0055] Calculate the parameter fluctuation characteristics:

[0056] In each time window (such as 10 seconds), the following are calculated for each parameter:

[0057] Dynamic change rate: (current value - window start value) / window length;

[0058] Variance: reflects the fluctuation amplitude of the parameter in the window;

[0059] Adjust the judgment matrix according to the parameter change rate difference:

[0060] For each parameter pair (i, j), calculate the absolute value of the change rate difference, amplify or reduce the influence of this difference using the sensitivity coefficient, adjust the judgment matrix elements, for example:

[0061] If the bit error rate change rate is significantly higher than the bandwidth utilization rate, increase the weight of the bit error rate relative to the bandwidth utilization rate, and the formula is: adjustment factor = 1 + change rate difference x sensitivity coefficient.

[0062] Calculate the eigenvector to determine the weight:

[0063] Calculate the eigenvalue and eigenvector of the adjusted judgment matrix, select the eigenvector corresponding to the maximum eigenvalue, normalize the eigenvector, and obtain the real-time weight of each parameter.

[0064] Normalize the variance of each parameter, calculate the decay factor: decay factor = 1 / (1+normalized variance), the larger the variance, the smaller the decay factor, thereby reducing the influence of parameters with large fluctuations on the final score; multiply the normalized parameter value by the dynamic weight, multiply by the corresponding variance decay factor, and sum to obtain the comprehensive score. Weight dynamic adjustment example:

[0065] Scenario 1: Interference level suddenly increases:

[0066] The change rate of the interference level increases significantly, the weight of the interference level relative to other parameters in the judgment matrix increases, and the real-time weight of the interference level increases. Since the interference variance may also increase, the decay factor will appropriately reduce its influence, but the overall weight will still rise

[0067] Scenario 2: Bit error rate remains stable, bandwidth utilization rate fluctuates and increases:

[0068] The bit error rate weight remains at a high level, and the bandwidth utilization rate decreases due to the increase in variance, and the decay factor reduces its contribution to the score, and the comprehensive score pays more attention to stable parameters (such as bit error rate) and change trends.

[0069] In the embodiment of the present application, multiple core communication parameters are collected to avoid one-sidedness of a single indicator, and the communication state is completely described. The weight is adjusted in real time according to the parameter fluctuation, the most influential indicator (such as interference burst) is highlighted, the timeliness and pertinence of evaluation are improved, the dynamic change rate and variance are extracted through a sliding window, not only the current value is concerned, but also the trend of parameter fluctuation (such as continuous decline of signal strength) is identified, potential risks are warned in advance, the variance decay factor is superimposed, the interference of accidental fluctuations is suppressed, and the score is more focused on the long-term stability of the communication state rather than short-term noise.

[0070] In a preferred embodiment of the present application, step 2, based on the communication quality comprehensive score, an initial communication configuration scheme is generated through a deep learning network, including:

[0071] The communication quality comprehensive score and the dynamic judgment matrix are input into a pre-trained deep learning network, wherein the evaluation matrix contains the dynamic weight distribution and historical fluctuation characteristics of each communication parameter;

[0072] The network input end synchronously receives real-time three-dimensional position distribution data of the UAV group, task priority labels and channel state information, and performs feature alignment on the dynamic judgment matrix and the communication quality comprehensive score to form a multi-dimensional input vector;

[0073] The spatial distribution of the UAV group is grid-mapped through a convolution layer to extract spatial correlation characteristics of the communication quality of different regions and generate a communication heat map;

[0074] The heat map is connected with a recurrent neural network layer to analyze the availability probability of each channel in the future time slot based on the time sequence variation law of the channel state information;

[0075] The spatial correlation characteristics and the time sequence availability probability are nonlinearly superimposed through a feature fusion layer to generate a space-time joint feature of channel resource configuration; the output layer calculates the modulation and coding strategy matching degree, the transmit power and interference constraint relationship, the channel binding combination utility value and the stability evaluation index of the retransmission mechanism of each UAV according to the space-time joint feature through a full connection structure to generate an initial configuration parameter set containing priority sorting;

[0076] The execution order of the configuration scheme is dynamically adjusted according to the priority label in the initial configuration parameter set to generate an initial communication configuration scheme adapted to the current communication state and task demand.

[0077] In the embodiment of the present application, the above steps are applied as follows:

[0078] The communication quality comprehensive score (scalar value) is converted into a fixed dimension vector (such as 1x10), the dynamic evaluation matrix (5x5, containing the dynamic weight and historical fluctuation of 5 communication parameters) is flattened into a 25-dimensional vector, the three-dimensional position data of the UAV (Nx3 matrix) is normalized to the [0, 1] interval, the task priority label is converted into a one-hot encoding vector (such as a 4-dimensional vector corresponding to a 4-level priority), the channel state information (CSI) is arranged into a vector according to the channel number, and a linear mapping layer is used to unify the input data from different sources to the same dimension (such as 64 dimensions), the importance weight of each input feature is calculated through an attention mechanism, and the feature vectors are weighted and summed according to the importance weight to form a 64x1 multi-dimensional input vector.

[0079] The three-dimensional space is divided into a regular grid (such as a 20x20x5 grid), the number of drones and the total signal strength in each grid are calculated, and a space occupation matrix (20x20x5) is generated, with element values being the signal strength in the grid. A 3x3x3 convolution kernel is used to perform convolution operations on the space occupation matrix (16 convolution kernels), a ReLU activation function is applied to introduce non-linear transformation, and a maximum pooling layer (2x2x2) is used to reduce the feature dimension and preserve key features, generating a 10x10x3 communication heat map representing the communication quality of different regions. The channel state information of the last 10 time steps is collected to construct a 10xM time sequence matrix (M is the number of channels), and a bidirectional LSTM network is used to process the time sequence matrix. Each LSTM unit outputs the hidden state at the current time, which is mapped to the channel availability probability (value range [0, 1]) through a fully connected layer, generating an M-dimensional vector representing the availability probability of each channel in the future time slot.

[0080] The communication heat map (10x10x3) is flattened into a 300-dimensional vector, and the channel availability probability vector (M-dimensional) is mapped to 300 dimensions through a linear layer. The spatial feature vector and the time sequence feature vector are multiplied element by element, and a multi-layer perception (MLP) is used for non-linear transformation:

[0081] First layer: 300 dimensions to 128 dimensions, using LeakyReLU activation;

[0082] Second layer: 128 dimensions to 64 dimensions, using Swish activation, generating a 64-dimensional spatio-temporal joint feature vector.

[0083] Modulation and coding strategy (MCS) matching degree calculation, input the spatio-temporal joint feature into a fully connected layer (64 to 32), and output the probability distribution of 8 MCS schemes through a softmax function. Select the MCS with the highest probability as the recommended scheme. Input the spatio-temporal feature into a fully connected layer, and scale the output to the [0.1, 1] interval through a sigmoid function, mapping to the actual power range (such as 10-30dBm). Use the attention mechanism to calculate the importance weight of each channel, sort the channels according to the weight, select the top K channels, calculate the theoretical throughput and interference level of the channel combination, analyze the fluctuation characteristics of the historical bit error rate and packet loss rate, calculate the optimal retransmission times (2-4 times), and generate the parameter configuration of the ARQ / HARQ mechanism.

[0084] The spatio-temporal combined feature is input into a binary classification network, and a high / low priority probability is output through a softmax function. According to a probability threshold (such as >0.7), the priority is marked, all configuration parameters are arranged in descending order of priority, high-priority parameters are preferentially allocated resources, and low-priority parameters are allocated on demand when resources remain. All parameters are integrated into a configuration vector, and constraint conditions (such as total power limitation and fairness constraint) are added to generate a final initial communication configuration scheme. Through this method, the system can dynamically generate an optimal initial communication configuration scheme according to real-time communication states and task requirements.

[0085] In a preferred embodiment of the present application, step 3, the initial communication configuration scheme is input into a preset spatio-temporal prediction model, and a frequency-time slot combined allocation matrix is generated through a spatio-temporal convolutional neural network for advanced calculation of regional spectrum demand, including:

[0086] The initial communication configuration scheme is input into a spatio-temporal convolutional neural network, and the input layer receives historical spectrum occupation data, current channel allocation state and unmanned aerial vehicle task queue;

[0087] The spatio-temporal convolutional layer synchronously extracts the spatial distribution characteristics of spectrum resources and time slot occupation mode through a three-dimensional convolution kernel, and the prediction layer performs K-step advanced prediction of regional spectrum demand based on an attention mechanism;

[0088] The output layer generates a two-dimensional allocation matrix, the matrix row vector represents the available frequency band number, the column vector represents the future time slot sequence, and the matrix element value represents the allocation weight of the frequency-time slot unit. According to the weight value, the matrix is prioritized to generate a dynamically adjustable combined allocation strategy.

[0089] In the embodiment of the present application, the above steps are applied as follows in the specific implementation process:

[0090] Key parameters are extracted from the initial configuration generated in step 2, including modulation and coding strategies of each unmanned aerial vehicle, transmission power threshold, channel binding demand and task priority label, which are converted into a spectrum resource demand vector (such as each unmanned aerial vehicle corresponding to a frequency band bandwidth demand and time slot continuity requirement).

[0091] Structuring of historical spectrum occupation data: Collect spectrum usage records for the past N time steps (such as N=50), construct a two-dimensional matrix according to frequency bands (16) x time slots (32 time slots per time step), and record the occupation state (0 / 1) and actual transmission rate of each frequency-time slot unit.

[0092] Current channel allocation state coding: Real-time acquisition of currently allocated frequency-time slot units generates a mask matrix (marking "occupied", "reserved" and "idle" states), and is combined with the unmanned aerial vehicle task queue (containing task start time, duration and priority) to form spatio-temporal demand constraints.

[0093] Data normalization and dimension alignment: all input data (configuration parameters, historical occupancy, current state, task queue) are uniformly scaled to the [0, 1] interval and concatenated as a three-dimensional tensor (e.g. 16x32x10) in the "frequency band x time slot x feature" dimension as the input of the spatio-temporal convolutional neural network.

[0094] Spatio-temporal convolution layer feature extraction:

[0095] Three-dimensional convolution kernel spatio-temporal modeling, using a 3x3x3 three-dimensional convolution kernel (corresponding to "adjacent 3 frequency bands x 3 time slots x 5 types of features"), synchronously extracting:

[0096] Spatial features: interference correlation of adjacent frequency bands (such as signal overlap between frequency bands f and f+1);

[0097] Time slot features: occupancy patterns of consecutive time slots (such as low occupancy periodicity of time slots 10-20 at night);

[0098] Demand features: high-frequency time slot length in the task queue (such as most tasks requiring 4 consecutive time slots). Through 64 three-dimensional convolution kernels, a 64-channel feature map is generated to capture the dependence of spectral resources in the "frequency band-time slot-demand" three-dimensional space.

[0099] Temporal pooling and dynamic feature enhancement: average pooling is performed on the time dimension (time slot axis) to retain long-term occupancy trends; maximum pooling is performed on the frequency band dimension to highlight the features of high-conflict frequency bands. The attention mechanism is introduced to dynamically adjust the feature weights of different frequency bands according to task priority (such as a 20% increase in feature weight for frequency bands corresponding to high-priority tasks).

[0100] Prediction layer K-step ahead demand calculation:

[0101] Attention mechanism focusing on key spatio-temporal areas: when predicting the spectral demand of the next K time slots (e.g. K=16), the self-attention mechanism is used to calculate the correlation between the current input and the history of N time steps, focusing on:

[0102] Historical contemporaneous (e.g. same time period yesterday) spectral occupancy peaks;

[0103] Recent sudden high-demand frequency-band-time-slot combinations (e.g. frequency band 8 is continuously occupied in time slots 5-8).

[0104] Multi-scale prediction fusion: output short-term (K=4), medium-term (K=8), and long-term (K=16) demand prediction results respectively, and fuse them through a fully connected layer with weights determined by the average duration of current tasks (e.g. increase the weight of short-term prediction if short-term tasks have a high proportion).

[0105] Constraint embedding: convert the constraints in the initial configuration generated in step 2, such as transmit power limit, interference threshold, etc., into regularization terms of the prediction layer (e.g., force the allocation weight difference between adjacent frequency bands in the same time slot to be no more than 0.3), to ensure that the prediction result meets the physical layer transmission rules.

[0106] Output layer generation and distribution strategy ranking:

[0107] Two-dimensional allocation matrix construction: generate a 16x16 frequency band-time slot matrix (16 frequency bands, 16 future time slots), and normalize each element value to an allocation weight (value range [0, 1]) by a softmax function. The higher the weight, the higher the probability of being preferentially allocated. For example, if the weight of frequency band 5 in time slot 7 is 0.8, it means that this unit is a high-priority allocation object.

[0108] Priority ranking and dynamic adjustment: sort the matrix elements according to "weight x (1 + task priority coefficient)" (priority coefficient: 1.5 for high-priority tasks, 1.0 for medium-priority tasks, and 0.8 for low-priority tasks), and generate an allocation order list. When a new high-priority task is detected, the corresponding time slot weight is adjusted in real time (e.g., the time slot weight corresponding to the inserted task is temporarily increased by 30%), to ensure that resource allocation is tilted towards urgent needs.

[0109] In the embodiment of the present application, the correlation characteristics of frequency spectrum resources in frequency bands, time slots, and space are captured by a three-dimensional convolution kernel, realizing advanced prediction of future K time slot demands, reducing communication conflicts by more than 50% compared to traditional static allocation, embedding physical layer constraints such as transmit power and interference threshold, dynamically adjusting allocation weights in combination with task priority, improving frequency spectrum utilization by more than 30%, while guaranteeing the latency requirements of high-priority tasks, supporting the balance between short-term burst demand and long-term trend, adapting to dynamic changes in UAV group tasks (e.g., quickly releasing frequency spectrum resources occupied by low-priority tasks when temporary inspection tasks are added), learning historical occupation patterns through attention mechanism, continuously optimizing the prediction model, and improving the allocation accuracy in complex electromagnetic environments by 25% compared to traditional algorithms. Through the above steps, the system realizes a closed-loop process from "historical data perception-real-time demand modeling-future trend prediction-dynamic strategy generation", providing an efficient and flexible frequency spectrum resource allocation scheme for UAV group communication.

[0110] In a preferred embodiment of the present application, step 4, three non-collinear position points are selected within the UAV operating airspace, and a spatial virtual geometric shape is constructed through a coordinate mapping algorithm; the geometric center coordinates of the spatial virtual geometric shape are extracted, and the offset amount from the communication coverage centroid is calculated as the first correction factor, including:

[0111] Based on the dynamic judgment matrix collected in real time in the unmanned aerial vehicle operation airspace, three non-collinear position points with significant signal fluctuation and discrete distribution are screened out;

[0112] According to the three-dimensional space coordinates, a triangular virtual plane is constructed through a coordinate mapping algorithm, and the plane is used to represent the physical space reference benchmark of the current airspace communication environment.

[0113] The geometric center of all vertex coordinates in the triangular plane is calculated by using the least square method, and a physical space reference point is generated; based on the real-time signal strength values of the three position points in the dynamic judgment matrix, the signal weight proportion of each position point is calculated by using the weighted centroid algorithm, and the communication coverage centroid coordinates are generated in combination with the three-dimensional coordinates, and the centroid represents the core area of the actual signal distribution.

[0114] The three-dimensional Euclidean distance between the geometric center and the communication coverage centroid is calculated, the distance value is normalized by using the maximum communication radius of the airspace, and the first correction factor is generated.

[0115] In the embodiment of the present application, the above-mentioned steps are applied, and the specific implementation process is as follows:

[0116] The signal strength data of all position points in the unmanned aerial vehicle operation airspace is analyzed, the signal strength fluctuation amplitude (i.e. variance) of each point in the last 10 seconds is calculated, the position points with signal strength fluctuation amplitude exceeding 1.5 times of the average fluctuation amplitude of all points are selected as "signal fluctuation significant points", and these points can sensitively reflect the dynamic changes of the current communication environment.

[0117] Space discreteness and non-collinearity verification: from the selected significant points, the three points with the most dispersed space distribution are further selected, whether any three points form an effective triangle (i.e. three points are not on the same straight line) is checked, the collinear situation is excluded, and finally three non-collinear position points are determined, which are denoted as point A, point B and point C, and it is ensured that they can represent the communication characteristics of different regions in the airspace.

[0118] Construction of triangular virtual plane:

[0119] Local coordinate system establishment: taking point A as the origin, the line connecting point A to point B as the x-axis direction, and the y-axis perpendicular to the x-axis and pointing to the direction of point C is determined by the space geometry method, forming a local two-dimensional plane coordinate system, and the three-dimensional coordinates of the three points are mapped into the plane.

[0120] Plane equation fitting: a plane is fitted by using the three-dimensional coordinates of the three points through mathematical method, and the plane is used as the physical space reference benchmark of the current airspace communication environment, which is used for subsequent geometric feature calculation.

[0121] For three non-collinear points in three-dimensional space 、 , The determined plane equation can be derived by the following steps:

[0122] Calculate the plane vector:

[0123] Vector ;

[0124] Vector ;

[0125] Calculate the plane normal vector by vector cross product :

[0126] ;

[0127] Determine the constant term of the plane equation Substitute any point (such as P1) into the point formula , and after expansion, we get:

[0128] ;

[0129] ;

[0130] Where A, B, and C represent the components of the normal vector of the plane, which determines the spatial orientation of the plane. The normal vector is perpendicular to the plane, and its direction reflects the inclination angle of the plane in three-dimensional space. For example: A represents the slope of the plane in the x-axis direction, and A = 0 means the plane is parallel to the x-axis; the larger the C, the greater the inclination of the plane in the z-axis direction. D represents the constant term, which determines the distance from the plane to the origin. The larger the absolute value, the farther the plane is from the origin. 、 and The coordinates of the reference point P1 are used to locate the specific position of the plane in space, and other points P2 and P3 need to satisfy the plane equation, i.e. the equation holds after substitution.

[0131] Physical meaning and application

[0132] Physical space reference base: the fitted planeAs the geometric reference of the current airspace, it is used to calculate the spatial position of the geometric center and communication coverage centroid. For example: The geometric center is the arithmetic mean of the coordinates of the three vertices in the plane, reflecting the physical distribution center of gravity of the UAV;

[0133] The communication coverage centroid is a coordinate based on signal strength weighting, reflecting the core area of actual signal distribution.

[0134]

[0135] Offset calculation basis: the plane equation provides a unified spatial reference framework for calculating the three-dimensional distance between the geometric center and the communication coverage centroid. If the two center points deviate from the plane or have a large difference in position within the plane, it indicates that the physical distribution of the UAV does not match the signal coverage, and the communication configuration needs to be adjusted through a correction factor.

[0136] Calculation of geometric center (physical space reference point):

[0137] Coordinate average calculation: add the x coordinates of the three position points and divide by three to get the x coordinate of the geometric center; similarly, take the arithmetic mean of the y and z coordinates to get the y and z coordinates of the geometric center. This geometric center represents the center of gravity of the three position points in physical space, without considering the differences in signal strength.

[0138] Calculation of communication coverage centroid (signal distribution core point):

[0139] Signal weight determination: extract the real-time signal strength values of the three position points, and divide the signal strength of each point by the sum of the signal strengths of the three points to get the signal weight of each point. The stronger the signal, the higher the weight, indicating that it contributes more to the regional communication coverage.

[0140] Weighted coordinate calculation: multiply the x coordinate of each position point by its signal weight, and add them up to get the x coordinate of the communication coverage centroid; similarly, calculate the y and z coordinates. This centroid reflects the core area of the actual signal coverage, and the weight is dynamically dependent on the real-time signal strength of each point.

[0141] Generation of the first correction factor:

[0142] Three-dimensional distance calculation: calculate the straight-line distance between the geometric center and the communication coverage centroid in three-dimensional space. This distance reflects the degree of deviation between the physical space center of gravity and the actual signal core area.

[0143] Normalization processing: normalize the above distance with the maximum communication radius of the UAV in the airspace (i.e. the farthest distance that the UAV signal can cover), converting it to a value between 0 and 1. Through a specific conversion rule (the smaller the distance, the closer the correction factor to 1), the first correction factor is finally obtained. This factor is used to quantify the consistency of spatial distribution and signal coverage, and the larger the value, the closer the two are, indicating that the regional communication coverage is more uniform.

[0144] In the embodiment of the present application, by screening the position points with significant signal fluctuations and spatial dispersion, it is ensured that the selected reference points can effectively represent the dynamic communication environment; the difference analysis between the geometric center and the communication coverage centroid combines the physical distribution of the unmanned aerial vehicle with the actual signal coverage characteristics, avoids the one-sidedness of relying only on the single characteristics of position or signal, and the first correction factor is normalized by the distance between the physical barycenter and the signal core, which intuitively reflects the uniformity of the regional communication coverage. When the two deviate greatly (for example, the unmanned aerial vehicles are concentrated but the signals are concentrated on one side due to shielding), the correction factor will prompt the system to adjust the resource allocation and tilt towards the area with weak signal coverage, thereby improving the overall coverage quality. The centroid weight is dynamically calculated based on the signal strength, so that the system can adapt to non-uniform signal propagation scenarios (such as signal attenuation between urban high-rise buildings) such as multipath propagation and obstacle shielding, ensuring that the communication configuration is consistent with the actual transmission effect and reducing the communication blind area caused by the inconsistency between spatial distribution and signal coverage.

[0145] In a preferred embodiment of the present application, the signal coverage angle distribution among the three position points is analyzed, and a second correction factor is generated based on the path loss model, including:

[0146] Based on the constructed vertex coordinates of the triangular virtual plane, the coverage angle formed by the signal strength change between the adjacent two points is calculated in the area of each side formed by the three position points;

[0147] According to the path loss model, the real-time flight height of the unmanned aerial vehicle at each angle vertex, the obstacle density distribution data and the current communication carrier frequency are combined to calculate the expected value of the path loss of each angle area under the line-of-sight and non-line-of-sight propagation conditions;

[0148] According to the loss expectation values of the three angles, the distribution coefficient reflecting the stability of the signal coverage in the triangular area is generated by the exponential smoothing algorithm to eliminate transient interference noise; the distribution coefficient is matched with the preset loss threshold interval, and according to the attenuation degree of the coverage uniformity, the coefficient is mapped to the correction amount interval by using a piecewise linear function to generate the second correction factor.

[0149] In the embodiment of the present application, the above steps are applied as follows:

[0150] Triangle internal angle definition: the three internal angles ( ∠A, ∠B, ∠C) of the triangle formed by the three position points correspond to the angles at the vertices A, B and C respectively. Each angle is calculated by vector dot product, reflecting the difference in signal propagation direction between the adjacent two sides. For each internal angle, the real-time signal strength at the vertex is weighted. For example, the higher the signal strength at vertex A, the greater the influence weight of the corresponding angle ∠A on the area coverage, avoiding the problem of ignoring the actual signal strength and weakness by simply considering the geometric angle.

[0151] Path loss expectation value calculation:

[0152] Environment parameter collection: Obtain real-time data for each angle region:

[0153] UAV flight height (affects line-of-sight propagation probability), obstacle density (obtained through radar or map data, in the interval [0, 1], 0 represents no obstruction, and 1 represents dense obstruction), communication carrier frequency (affects the wavelength parameter in the free space loss formula).

[0154] Line-of-sight (LoS) and non-line-of-sight (NLoS) loss models:

[0155] Environment parameter collection and preprocessing:

[0156] Distance measurement: For each side of the triangle (such as side BC), calculate the three-dimensional straight-line distance between the two endpoints (B and C) as the physical distance of signal propagation, and extract the real-time flight height of vertex A (corresponding to the vertex of angle ∠A), denoted as hA, which is used to determine whether the signal propagation path has line-of-sight conditions.

[0157] Through radar scanning or electronic map data, obtain the obstacle distribution of the region where side BC is located, and use a value between 0 and 1 to represent the density (0 represents no obstruction, and 1 represents complete obstruction), for example, the obstacle density in the city center region may be 0.8, and in the suburbs it may be 0.2, read the carrier frequency used for current communication (such as 2.4GHz or 5.8GHz), which determines the attenuation characteristics of the signal in free space.

[0158] Assuming that the signal from vertex A propagates to side BC, if the height of the obstacles on the path is lower than hA, it is determined to be line-of-sight propagation; otherwise, it is non-line-of-sight. In the line-of-sight condition, signal loss increases with distance, and the higher the frequency, the more significant the loss. The specific process is:

[0159] Determine the impact of distance on loss: doubling the distance increases the loss by about 6dB (proportional to the square of the distance); determine the impact of frequency on loss: doubling the frequency increases the loss by about 6dB (proportional to the square of the frequency), and combine the distance and frequency factors to calculate the basic loss value of the path.

[0160] Non-line-of-sight (NLoS) loss calculation:

[0161] When the path is non-line-of-sight, additional obstruction loss is added to the free space loss. Obstruction loss is simulated by the "lognormal shadowing model", the specific steps are:

[0162] According to the standard deviation of shadow fading determined by the obstacle density (e.g. 8dB for density 0.8, 3dB for density 0.2), generate random loss values conforming to normal distribution to simulate the signal fluctuation caused by obstacles (e.g. mean value 0dB, standard deviation the above value), the higher the obstacle density, the greater the fluctuation range of random loss. For example, in high-density areas, the signal may be greatly attenuated due to reflection / scattering of multiple buildings, and the loss value may be 10-20dB higher than the free space loss.

[0163] Average loss calculation of the angle area (take ∠A as an example):

[0164] Sampling of edge BC: divide edge BC into several segments (e.g. take 5 equidistant sampling points), each point represents a signal propagation path in this angle area (from A to a point on edge BC).

[0165] Point-by-point loss calculation: for each sampling point:

[0166] Calculate the distance from A to the point and the height difference, determine whether it is line-of-sight (e.g. if the height difference is greater than the height of the obstacle, it is line-of-sight, otherwise it is non-line-of-sight), calculate the corresponding loss value (free space loss or loss with obstruction factor) according to the line-of-sight / non-line-of-sight condition, and calculate the arithmetic mean of the loss values of all sampling points on edge BC as the expected loss value corresponding to the angle ∠A, which reflects the average attenuation degree of signal propagation in this angle area. Repeat steps 1-4 for the other two angles (∠B and ∠C) of the triangle to calculate the corresponding expected loss values.

[0167] Loss difference analysis, compare the loss values of the three angles:

[0168] If the loss of a certain angle is significantly higher than that of the other angles (e.g. more than 5dB), it indicates that the signal coverage in this area is poor and needs to be compensated, if the loss values are close, it indicates that the signal coverage uniformity is good, and the correction factor value is high.

[0169] Use exponential smoothing algorithm (e.g. first-order smoothing with α=0.3) to process the expected loss values of each angle, reduce the influence of short-term interference (e.g. sudden multipath signals), and retain the long-term stable loss trend.

[0170] Stability distribution coefficient: calculate the standard deviation of the loss values of the three angles as the "signal coverage stability distribution coefficient". The smaller the standard deviation, the more uniform the loss of the three angle areas, and the higher the signal coverage stability.

[0171] Correction factor mapping generation:

[0172] Predefined loss threshold range: define the threshold range for different uniformity levels, for example:

[0173] High uniformity: standard deviation < preset value 1, corresponding to the correction interval [0.8, 1.0];

[0174] Medium uniformity: preset value 1 ≤ standard deviation < preset value 2, corresponding to [0.5, 0.8];

[0175] Low uniformity: standard deviation ≥ preset value 2, corresponding to [0.1, 0.5].

[0176] Piecewise linear mapping: according to the threshold interval where the stability distribution coefficient is located, the coefficient is converted into a correction factor of 0-1 through a linear function. For example, the correction factor in the low uniformity area is low, prompting the system to increase the signal compensation or resource allocation of this area.

[0177] In the embodiments of the present application, by combining the triangle angle and the path loss model, the spatial geometric characteristics (angle distribution) and the physical layer propagation characteristics (line of sight / non-line of sight loss) are combined, avoiding the problem of ignoring the actual signal attenuation by relying only on the geometric shape, more accurately evaluating the area coverage quality, considering dynamic environmental parameters such as flight height and obstacle density, so that the correction factor can adapt to complex scenes such as urban high-rise buildings and mountainous areas. For example, in the angle area with dense obstacles, the correction factor will automatically decrease, prompting the system to increase the transmission power or allocate redundant channels, and through exponential smoothing and piecewise mapping, the stability of signal coverage is converted into a quantifiable correction amount. When the loss difference of the three angle areas is large (such as the loss of a certain area is significantly higher than that of other areas due to obstruction), the correction factor will guide the system to adjust the resource allocation, balance the communication quality of each area, and reduce the coverage blind area.

[0178] In a preferred embodiment of the present application, according to the signal attenuation gradient on the boundary of the virtual geometric shape in space, a multipath effect compensation coefficient is calculated as a third correction factor; the first correction factor, the second correction factor and the third correction factor are fused into a comprehensive correction value, including:

[0179] Based on the constructed triangular virtual plane boundary coordinates, sample points are selected along the edges of the geometric shape at equal intervals, the signal intensity values of each point are collected in real time, and the attenuation gradient between adjacent points is calculated to form a boundary signal attenuation distribution map;

[0180] According to the direction vector and the change rate of the attenuation gradient, a probability density function of the multipath effect propagation path is constructed, the superposition effect of reflected and scattered signals is quantified through integral operation, and a boundary attenuation compensation coefficient is generated as a third correction factor;

[0181] According to the first correction factor representing the spatial offset, the second correction factor representing the azimuth difference compensation, and the third correction factor representing the multipath effect compensation, a comprehensive correction value is calculated by a weighting method.

[0182] In the embodiments of the present application, the above steps are implemented as follows when applied:

[0183] Virtual plane boundary sampling: On the three sides of the triangular virtual plane, sample points are selected at equal intervals (such as every 0.5 meters) to ensure at least 10 points on the boundary (such as 11 points are taken when the side BC is 5 meters long). Record the three-dimensional coordinates and real-time signal strength values of each point. For adjacent sample points on each side, calculate the signal strength attenuation gradient:

[0184] Gradient direction: the direction vector from the starting point to the ending point along the boundary;

[0185] Gradient size: the ratio of the signal strength difference between adjacent points to the physical distance (unit: dB / m).

[0186] Attenuation distribution map generation: combine the gradient data of the three sides to form a signal attenuation distribution map of the triangular boundary, which intuitively shows the signal attenuation rate at different positions of the boundary (such as a larger gradient at the midpoint of a side, indicating that the signal attenuation in this area is rapid, which may exist multipath interference).

[0187] Multipath effect probability density function construction:

[0188] Gradient feature extraction: analyze the direction vector and change rate of the attenuation gradient:

[0189] Direction vector: reflects the dominant direction of signal attenuation (such as the gradient direction pointing to an obstacle, which may exist a reflection path);

[0190] Change rate: the greater the absolute value of the gradient, the more intense the signal attenuation, and the higher the possibility of multipath effect.

[0191] Probability density function (PDF) modeling: assuming that the multipath propagation path follows a Gaussian distribution, taking the gradient direction as the mean direction and the gradient change rate as the standard deviation to construct a two-dimensional probability density function. For example:

[0192] When the gradient direction is concentrated in a certain angle interval, the PDF peak appears in that direction, indicating that the multipath signal mainly comes from the reflection / scattering in that direction;

[0193] The greater the change rate, the more dispersed the PDF distribution, reflecting the increase in the uncertainty of the multipath path.

[0194] Multipath impact quantification: by integrating the PDF, calculate the proportion of the total energy of the multipath signal in the boundary area (i.e. the ratio of the energy of the reflected / scattered signal to the energy of the direct signal). The greater the integral result, the more significant the multipath effect, and the more signal attenuation needs to be compensated.

[0195] Third correction factor generation (boundary attenuation compensation coefficient):

[0196] Compensation coefficient mapping: map the multipath energy proportion to the compensation coefficient in the interval [0, 1]:

[0197] If the multipath energy proportion is less than 20%, the compensation coefficient is 0.8-1.0 (the multipath effect is small, and no significant compensation is needed); if the proportion exceeds 50%, the compensation coefficient is 0.1-0.5 (the multipath effect is large, and the signal redundancy needs to be increased or the modulation method needs to be adjusted), combined with real-time signal-to-noise ratio (SNR) for further adjustment: when the SNR is lower than the threshold, the compensation coefficient is automatically reduced by 0.2-0.3, and the multipath suppression algorithm (such as equalizer or diversity reception) is forcibly started.

[0198] Comprehensive correction value fusion calculation, three-factor weighting strategy:

[0199] First correction factor (spatial offset): weight w1, reflecting the consistency of physical distribution and signal coverage, weight 0.3 when the environment is stable, and increased to 0.5 when the space changes rapidly;

[0200] Second correction factor (angle loss): weight w2, reflecting the uniformity of directional coverage, weight 0.4 in complex terrain, and reduced to 0.2 in open environment;

[0201] Third correction factor (multipath compensation): weight w3, reflecting the stability of signal propagation, weight 0.3-0.5 in a multipath-rich scene (such as a city), and reduced to 0.2 in a suburban scene.

[0202] Weighted sum formula: comprehensive correction value F = w1 x f1 + w2 x f2 + w3 x f3, where f1, f2 and f3 are the three correction factors, and the sum of the weights is 1.

[0203] Dynamic weight adjustment mechanism: through a reinforcement learning model (such as Q-learning), the weight parameters are automatically adjusted according to historical communication quality feedback. For example, when the multipath causes the bit error rate to rise, the system automatically increases the w3 weight, and preferentially compensates for the multipath effect.

[0204] In the embodiments of the present application, through boundary attenuation gradient analysis and probability density modeling, the energy distribution of multipath signals is quantified, so that the third correction factor can specifically compensate for signal attenuation caused by reflection / scattering, and improve the signal stability in complex environments (such as a 40% reduction in bit error rate in a city canyon scene). The spatial offset, directional loss uniformity and multipath effect are comprehensively considered to avoid the limitations of a single factor. For example:

[0205] The space-dense area (such as a UAV cluster) is preferentially adjusted by a first factor to adjust the coverage center, the mountainous area scene is compensated by a second factor to compensate for the difference in direction loss caused by shielding, and the indoor scene is inhibited by a third factor to suppress the multipath interference caused by wall reflection. The weighting strategy combines real-time environmental parameters and historical feedback, so that the system can quickly adapt to changes in the scene (such as when entering a city building group from an open farmland, automatically increasing the weight of the multipath compensation factor), without the need for manual intervention to maintain stable communication quality.

[0206] In a preferred embodiment of the present application, step 5, according to the comprehensive correction value, the communication interruption risk is evaluated by the path planning neural network to generate a multi-path transmission scheme containing redundant relay node selection rules and dynamic fragmentation reorganization strategies, including:

[0207] The comprehensive correction value is input into the pre-trained path planning neural network, which decomposes the correction value into three-dimensional features of spatial distortion intensity, azimuth attenuation index and multipath interference level through a feature analysis layer;

[0208] In the network hidden layer, a three-dimensional communication environment field model is constructed in combination with real-time position data of the UAV cluster, and a spatial distribution heat map of communication interruption risk is generated by scanning through a convolution kernel, wherein the chroma value of the heat map is positively correlated with the interruption probability;

[0209] Based on the high-risk area identification result of the heat map, a redundant relay node deployment strategy is dynamically generated through a rule engine; for a transmission data stream, according to the level division of different areas in the risk heat map, an adaptive fragmentation algorithm is used to dynamically adjust the data packet fragmentation size, and a multi-path parallel transmission timing is planned according to a time slot allocation matrix; finally, the relay topology configuration rule and the data fragmentation strategy are spatio-temporally aligned to generate a multi-path transmission scheme with environmental adaptability.

[0210] In the embodiment of the present application, the above steps are applied as follows:

[0211] The comprehensive correction value is input into the pre-trained path planning neural network, which is decomposed by a feature analysis layer into:

[0212] Spatial distortion intensity: reflects the deviation degree of physical distribution and signal coverage (first correction factor);

[0213] Azimuth attenuation index: quantifies the signal loss difference in different directions (second correction factor);

[0214] Multipath interference level: evaluates the influence intensity of reflected / scattered signals (third correction factor).

[0215] The three-dimensional feature values are mapped to the 0-1 interval, for example: when the spatial distortion intensity exceeds the threshold, it is normalized to 1, indicating a serious mismatch; when the multipath interference level is below the threshold, it is normalized to 0, indicating no significant multipath.

[0216] Three-dimensional communication environment field model construction, position data fusion:

[0217] Combine the real-time three-dimensional coordinates (x, y, z) of the UAV group with the normalized features to construct the communication environment field model. For example, in areas with high spatial distortion intensity, the field strength value at the corresponding position in the model is reduced, and in directions with large azimuth attenuation index, the field strength gradient increases. Use a 3x3x3 convolution kernel to slide and scan in three-dimensional space to calculate the communication interruption risk value of each grid point. The risk value considers:

[0218] The probability that the signal strength is below the threshold, the probability that the error rate exceeds the upper limit due to multipath effects, and the probability that the link is broken due to rapid spatial changes. Finally, generate a heat map representing the risk level with a color gradient (e.g., red for high risk and green for low risk).

[0219] Redundant relay node deployment strategy, high-risk area identification:

[0220] Threshold segmentation of the heat map to identify areas where the risk value exceeds the preset threshold (e.g., 0.7), and mark them as high-risk areas for communication interruption.

[0221] Relay node selection rules:

[0222] Position priority: preferentially select UAVs with stable signal strength on the edge of high-risk areas as relays;

[0223] Load balancing: avoid selecting nodes that have already taken on too many relay tasks;

[0224] Mobility constraint: preferentially select UAVs with a mobility speed below a threshold (e.g., 5 m / s) to ensure stable relay links.

[0225] Topology structure generation:

[0226] Construct redundant communication paths based on selected relay nodes. For example, for each high-risk area, establish at least two non-overlapping relay links to form a ring or mesh topology.

[0227] Adaptive data fragmentation and multi-path timing planning, risk level partitioning:

[0228] Divide the heat map into low-risk (<0.3), medium-risk (0.3-0.7), and high-risk (>0.7) areas, and use a dynamic fragmentation algorithm:

[0229] High-risk area: reduce the size of data packet fragments (e.g., from 1024 bytes to 256 bytes) and increase the redundancy check code;

[0230] Low-risk area: use standard fragmentation size (e.g., 1024 bytes) to improve transmission efficiency.

[0231] Time slot allocation matrix:

[0232] Each path is allocated independent time slots to avoid interference. For example:

[0233] The main path uses time slots 1, 3, and 5; the first redundant path uses time slots 2, 6, and 10; the second redundant path uses time slots 4, 8, and 12, and the time slot length is dynamically adjusted according to the risk level of the path (the time slot is extended by 10-20% for high-risk paths).

[0234] Space-time alignment and scheme generation, relay configuration and fragmentation strategy matching:

[0235] Align the position of each relay node with the time slot allocation matrix to ensure that data fragments can be transmitted according to the planned path. For example, relay node A is responsible for forwarding data fragments in time slot 2, and needs to configure the receiving and forwarding parameters in advance; small fragments of data in high-risk areas are preferentially allocated to multiple redundant paths for parallel transmission.

[0236] Dynamic recombination trigger mechanism:

[0237] Set an environmental change detection threshold (such as a change in the color distribution of the risk heat map of more than 15%), and when an environmental mutation is detected, re-execute steps 1-4 to generate a new transmission scheme.

[0238] In an embodiment of the present application, the spatial distribution of communication interruption risk is visually displayed through the heat map, enabling the system to identify potential blind spots in advance and avoid passive response to communication failures. For example, before the unmanned aerial vehicle enters the high-risk area, relay nodes are deployed in advance, and the redundant relay strategy ensures that single-point failures or local interference will not cause communication interruption. Tests show that in the case of 20% node failure, the system can still maintain a data packet transmission success rate of more than 95%, and the adaptive fragmentation and time slot allocation mechanism balances reliability and transmission efficiency. In low-risk areas, large fragments are used to improve throughput, and in high-risk areas, small fragments are used to enhance anti-interference capability, and the overall bandwidth utilization rate is increased by more than 30%, and the system can quickly adjust in a dynamically changing environment (such as when switching from an urban to a mountainous area, the scheme update delay is < 500ms), without the need for human intervention to maintain stable communication.

[0239] In a preferred embodiment of the present application, step 6, based on the multi-path transmission scheme, the frequency-time slot joint allocation matrix is dynamically iteratively updated, and the weight parameters of the space-time prediction model are adjusted through the error back propagation mechanism to form a closed-loop management, including:

[0240] According to the actual transmission delay and packet loss rate data of the multi-path transmission scheme, the priority weight in the frequency-time slot joint allocation matrix is corrected in reverse;

[0241] The modified matrix is used for residual error calculation with the original prediction result to generate an error vector of the space-time prediction model;

[0242] The convolution kernel weight and attention coefficient of the space-time convolution neural network are adjusted through the error back propagation mechanism, and the risk assessment parameter of the path planning neural network is updated;

[0243] A dynamic iterative optimization mechanism is established, so that the communication configuration scheme can automatically correct deviation with the change of the environment, and a closed-loop management process of evaluation-configuration-correction is formed.

[0244] In the embodiment of the application, the above steps are applied, and the specific implementation process is as follows:

[0245] The actual operation data of the multi-path transmission scheme are collected in real time, including:

[0246] The end-to-end transmission delay (accurate to millisecond level), packet loss rate (loss proportion is counted according to data packet number), and bit error rate (calculated through CRC check failure rate) of each path.

[0247] Matrix weight correction:

[0248] According to the performance data, the priority weight in the frequency-time slot joint allocation matrix is adjusted:

[0249] If the packet loss rate of a certain frequency-time slot combination exceeds the threshold (such as 5%), the priority weight thereof is reduced; if the delay of a certain path increases continuously for 3 periods, the time slot corresponding to the path is allocated to other low-delay paths.

[0250] Residual error calculation and error vector generation, prediction and actual comparison:

[0251] The modified matrix is subtracted from the original prediction result of the space-time prediction model element by element to obtain a residual error matrix. For example, the weight of frequency band 3 and time slot 7 in the original prediction is 0.8, and after correction, it is 0.6, and the residual error is-0.2.

[0252] Error vector construction:

[0253] The residual error matrix is expanded into a one-dimensional vector according to the row or column, and is used as the error vector of the space-time prediction model. Each element of the vector corresponds to the prediction error of a frequency-time slot combination.

[0254] Error back propagation and model parameter update, space-time convolution neural network adjustment:

[0255] Convolution kernel weight update: the error vector adjusts the weights of the convolution kernels in each layer of the spatio-temporal convolutional neural network through the backpropagation algorithm. For example, if the prediction error of a certain frequency band is consistently high, the learning rate of the corresponding convolution kernel is increased, the weight parameters in the attention mechanism are adjusted, the model pays more attention to the time periods and frequency bands with larger fluctuations in historical data, and the risk assessment parameters (such as the size of the convolution kernel in the heat map generation and the risk threshold) are updated to make the identification of high-risk areas more accurate. For example, if the actual packet loss is concentrated in a certain area, but the model does not identify it as high-risk, the risk threshold of that area is reduced.

[0256] Dynamic iterative optimization mechanism, closed-loop management process:

[0257] A closed-loop process of "data collection → matrix correction → model update → scheme generation" is established, a fixed iteration period (such as 100ms) or a trigger condition (such as the error vector norm exceeding a threshold) is set, the learning rate is dynamically adjusted according to the trend of the error vector, if the error decreases for 3 consecutive periods, the learning rate is reduced to prevent overfitting, if the error suddenly increases, the learning rate is increased to accelerate convergence.

[0258] Response to environmental mutations:

[0259] When an environmental mutation is detected (such as a sudden increase in packet loss rate caused by a sudden interference), a fast correction mode is started:

[0260] Temporarily increase the iteration frequency (such as from 100ms to 10ms), enable pre-trained emergency model parameters, and quickly recover the communication quality.

[0261] In the embodiments of the present application, the error of the spatio-temporal prediction model is gradually reduced through the closed-loop feedback mechanism. Tests show that after 100 consecutive iterations, the frequency-time slot allocation prediction accuracy rate is improved from 75% initially to 92%, and the system can quickly adapt to sudden interference (such as electromagnetic pulse, same frequency device intrusion). For example, within 3 iteration periods (about 300ms) after the interference occurs, the packet loss rate decreases from 30% to less than 5%, and the dynamic adjustment increases the spectrum resource utilization rate by 20%-30%. For example, in a high-density unmanned aerial vehicle scenario, by avoiding frequency band conflicts and time slot waste, the single-link throughput is increased from 2Mbps to 2.6Mbps. The closed-loop management system has self-repairing capability for model errors and environmental mutations. Even if the initial prediction model has deviations, it can still maintain stable communication performance through continuous iteration and correction. Through the above steps, the system realizes the full-process automation of "performance monitoring - error calculation - model optimization - scheme adjustment", and builds an intelligent communication system with dynamic correction capability, significantly improving the communication reliability and resource utilization rate of the unmanned aerial vehicle group in complex environments.

[0262] For example, Figure 2As shown, the embodiment of the application also provides an unmanned aerial vehicle sharing intelligent management system based on artificial intelligence, comprising:

[0263] The acquisition module is configured to acquire multi-dimensional communication parameters between the unmanned aerial vehicle group and the ground control platform in real time to calculate a communication quality comprehensive score.

[0264] The generation module is configured to generate an initial communication configuration scheme based on the communication quality comprehensive score through a deep learning network.

[0265] The calculation module is configured to input the initial communication configuration scheme into a preset space-time prediction model, perform advanced calculation on regional spectrum demand through a space-time convolutional neural network, and generate a frequency-time slot joint allocation matrix.

[0266] The correction module is configured to select three non-collinear position points in the airspace where the unmanned aerial vehicle operates, construct a virtual spatial geometric shape through a coordinate mapping algorithm, extract the geometric center coordinates of the virtual spatial geometric shape, calculate the offset amount of the geometric center coordinates from the communication coverage centroid as a first correction factor, analyze the signal coverage angle distribution among the three position points, generate a second correction factor in combination with a path loss model, calculate a multipath effect compensation coefficient as a third correction factor according to the signal attenuation gradient on the boundary of the virtual spatial geometric shape, and fuse the first correction factor, the second correction factor and the third correction factor into a comprehensive correction value.

[0267] The allocation module is configured to evaluate the communication interruption risk through a path planning neural network based on the comprehensive correction value, and generate a multi-path transmission scheme containing redundant relay node selection rules and dynamic fragmentation reorganization strategies.

[0268] The adjustment module is configured to perform dynamic iterative update on the frequency-time slot joint allocation matrix based on the multi-path transmission scheme, adjust the weight parameters of the space-time prediction model through an error back propagation mechanism, and form a closed-loop management.

[0269] The above is the preferred embodiment of the application. It should be noted that for those skilled in the art, without departing from the principles of the application, a number of improvements and refinements can be made, which should also be considered within the scope of protection of the application.

Claims

1. An artificial intelligence-based unmanned aerial vehicle sharing intelligent management method, characterized in that, The method comprises: Step 1, real-time acquisition of multi-dimensional communication parameters between the UAV group and the ground control platform to calculate a communication quality comprehensive score; Step 2, based on the communication quality comprehensive score, generating an initial communication configuration scheme through a deep learning network; Step 3, inputting the initial communication configuration scheme into a preset space-time prediction model to perform advanced calculation on regional spectrum demand through a space-time convolutional neural network to generate a frequency-time slot joint allocation matrix; Step 4, selecting three non-collinear position points in the UAV operation airspace, constructing a virtual spatial geometric shape through a coordinate mapping algorithm, extracting the geometric center coordinates of the virtual spatial geometric shape, calculating the offset amount thereof from the communication coverage centroid as a first correction factor, analyzing the signal coverage angle distribution among the three position points, combining a path loss model to generate a second correction factor, calculating a multipath effect compensation coefficient as a third correction factor according to the signal attenuation gradient on the boundary of the virtual spatial geometric shape, and fusing the first, second and third correction factors into a comprehensive correction value; Step 5, according to the comprehensive correction value, evaluating the communication interruption risk through a path planning neural network to generate a multi-path transmission scheme containing redundant relay node selection rules and dynamic fragmentation reorganization strategies; Step 6, based on the multi-path transmission scheme, dynamically and iteratively updating the frequency-time slot joint allocation matrix, adjusting the weight parameters of the space-time prediction model through an error back propagation mechanism to form a closed-loop management. 2.The AI-based unmanned aerial vehicle sharing intelligent management method of claim 1, wherein, Step 1, real-time acquisition of multi-dimensional communication parameters between the UAV group and the ground control platform to calculate a communication quality comprehensive score, comprising: Real-time acquisition of signal strength, bit error rate, transmission delay, bandwidth utilization rate and interference level parameters between the UAV group and the ground control platform, normalization processing of the parameters to obtain normalized parameters; Using a sliding time window analysis method to extract time sequence features of the normalized parameters, calculating the dynamic change rate and variance of each parameter in the window period as parameter fluctuation characteristics; Based on an analytic hierarchy process, a dynamic weight calculation model is constructed, the parameter fluctuation characteristics are input into the dynamic weight calculation model, a dynamic judgment matrix is constructed by comparing the influence degree of parameter change trend on communication stability, and the real-time weight coefficients of each parameter are calculated; The real-time parameter values and their corresponding real-time weight coefficients are fused by weighting, and the variance decay factor of the parameter fluctuation characteristics is superimposed to generate a comprehensive score reflecting the stability of the current communication state. 3.The AI-based unmanned aerial vehicle sharing intelligent management method of claim 2, wherein, Step 2, based on the communication quality comprehensive score, generating an initial communication configuration scheme through a deep learning network, comprising: Inputting the communication quality comprehensive score and the dynamic judgment matrix into a pre-trained deep learning network, wherein the dynamic judgment matrix contains the dynamic weight distribution and historical fluctuation characteristics of each communication parameter; The deep learning network input end synchronously receives real-time three-dimensional position distribution data, task priority labels and channel state information of the UAV group, aligns the dynamic judgment matrix with the communication quality comprehensive score in features to form a multi-dimensional input vector; Through a convolutional layer, the spatial distribution of the UAV group is grid-mapped to extract spatial correlation features of communication quality in different regions and generate a communication heat map; The heat map is connected with a recurrent neural network layer, and based on the time sequence variation law of channel state information, the availability probability of each channel in the future time slot is analyzed; Through the feature fusion layer, the spatial correlation feature and the time sequence availability probability are nonlinearly superimposed to generate the space-time joint feature of channel resource configuration; the output layer calculates the modulation and coding strategy matching degree, the transmit power and interference constraint relationship, the channel binding combination utility value and the stability evaluation index of the retransmission mechanism of each unmanned aerial vehicle according to the space-time joint feature through the full connection structure, and generates an initial configuration parameter set containing priority ranking; According to the priority label in the initial configuration parameter set, the execution order of the configuration scheme is dynamically adjusted to generate an initial communication configuration scheme adapted to the current communication state and task demand. 4.The AI-based unmanned aerial vehicle sharing intelligent management method of claim 3, wherein, Step 3, input the initial communication configuration scheme into the preset space-time prediction model, and perform advanced calculation on the regional spectrum demand through the space-time convolutional neural network to generate a frequency-time slot joint allocation matrix, including: The initial communication configuration scheme is input into the space-time convolutional neural network, and the input layer receives historical spectrum occupation data, current channel allocation state and unmanned aerial vehicle task queue; The space-time convolutional layer synchronously extracts the spatial distribution feature of the spectrum resource and the time slot occupation mode through a three-dimensional convolution kernel, and the prediction layer performs K-step advanced prediction on the regional spectrum demand based on the attention mechanism; The output layer generates a frequency-time slot joint allocation matrix, the row vector of the frequency-time slot joint allocation matrix represents the available frequency band number, the column vector represents the future time slot sequence, and the element value of the frequency-time slot joint allocation matrix represents the allocation weight of the frequency-time slot unit; according to the weight value, the frequency-time slot joint allocation matrix is prioritized to generate a dynamically adjustable joint allocation strategy. 5.The AI-based unmanned aerial vehicle sharing intelligent management method according to claim 4, characterized in that, Step 4, select three non-collinear position points in the unmanned aerial vehicle operating airspace, and construct a virtual geometric shape in space through a coordinate mapping algorithm; Extract the geometric center coordinates of the virtual geometric shape in space, calculate the offset amount thereof from the communication coverage centroid as a first correction factor, including: Based on the dynamic judgment matrix collected in real time in the unmanned aerial vehicle operating airspace, three non-collinear position points with significant signal fluctuation and discrete distribution are selected; According to the three-dimensional space coordinates thereof, a triangular virtual plane is constructed through a coordinate mapping algorithm, which is used to represent the physical space reference benchmark of the current airspace communication environment; The least square method is used to calculate the geometric center of all vertex coordinates in the triangular plane to generate a physical space reference point; based on the real-time signal strength values of the three position points in the dynamic judgment matrix, the signal weight proportion of each position point is calculated through a weighted centroid algorithm, and the communication coverage centroid coordinates are generated by combining the three-dimensional coordinates, which represent the core area of the actual signal distribution; The three-dimensional Euclidean distance between the geometric center and the communication coverage centroid is calculated, and the distance value is normalized by using the maximum communication radius of the airspace to generate the first correction factor. 6.The AI-based unmanned aerial vehicle sharing intelligent management method according to claim 5, characterized in that, The signal coverage angle distribution between the three position points is analyzed, and a second correction factor is generated based on the path loss model, including: Based on the vertex coordinates of the constructed triangular virtual plane, the coverage angle formed by the signal strength change between adjacent two points is calculated in the edge region formed by the three position points, respectively; According to the path loss model, combined with the real-time flight height of the unmanned aerial vehicle at each vertex of the angle, the obstacle density distribution data and the current communication carrier frequency, the expected value of the path loss of each angle area under the line-of-sight and non-line-of-sight propagation conditions is calculated; According to the expected value of the loss of the three angles, the instantaneous interference noise is eliminated by the exponential smoothing algorithm to generate a distribution coefficient reflecting the signal coverage stability in the triangular area; the distribution coefficient is matched with the preset loss threshold interval, and according to the uniformity decay degree of coverage, the coefficient is mapped to the correction amount interval by using a segmented linear function to generate a second correction factor. 7.The AI-based unmanned aerial vehicle sharing intelligent management method according to claim 6, characterized in that, According to the signal attenuation gradient on the boundary of the virtual geometric shape in space, a multipath effect compensation coefficient is calculated as a third correction factor; The first correction factor, the second correction factor and the third correction factor are fused into a comprehensive correction value, including: Based on the constructed triangular virtual plane boundary coordinates, sampling points are selected along the edges of the geometric shape at equal intervals, the signal strength values of each point are collected in real time, and the attenuation gradient between adjacent points is calculated to form a boundary signal attenuation distribution map; According to the direction vector and the change rate of the attenuation gradient, the probability density function of the multipath effect propagation path is constructed, the superposition effect of reflected and scattered signals is quantified by integral operation, and a boundary attenuation compensation coefficient is generated as a third correction factor; According to the first correction factor representing the spatial offset, the second correction factor compensating for the azimuth difference, and the third correction factor compensating for the multipath effect, a comprehensive correction value is calculated by a weighting method. 8.The AI-based unmanned aerial vehicle sharing intelligent management method of claim 7, wherein, Step 5, according to the comprehensive correction value, the communication interruption risk is evaluated by a path planning neural network to generate a multi-path transmission scheme including redundant relay node selection rules and dynamic fragmentation reorganization strategies, including: The comprehensive correction value is input into the pre-trained path planning neural network, which decomposes the correction value into three-dimensional features such as spatial distortion intensity, azimuth attenuation index and multipath interference level through a feature analysis layer; In the network hidden layer, a three-dimensional communication environment field model is constructed combined with the real-time position data of the unmanned aerial vehicle group, and a spatial distribution heat map of communication interruption risk is generated by scanning through a convolution kernel, wherein the chroma value of the heat map is positively correlated with the interruption probability; Based on the high-risk area identification result of the heat map, a redundant relay node deployment strategy is dynamically generated through a rule engine; for the transmission data stream, according to the level division of different areas in the risk heat map, an adaptive fragmentation algorithm is used to dynamically adjust the data packet fragmentation size, and a frequency-time slot joint allocation matrix is used to plan the multi-path parallel transmission timing; finally, the relay topology configuration rule and the data fragmentation strategy are spatio-temporally aligned to generate a multi-path transmission scheme with environmental adaptability. 9.The AI-based unmanned aerial vehicle sharing intelligent management method of claim 8, wherein, Step 6, based on the multi-path transmission scheme, the frequency-time slot joint allocation matrix is dynamically iterated and updated, the weight parameters of the space-time prediction model are adjusted through the error back propagation mechanism to form a closed loop management, including: According to the actual transmission delay and packet loss rate data of the multi-path transmission scheme, the priority weight in the frequency-time slot joint allocation matrix is corrected in reverse; The residual error is calculated between the corrected frequency-time slot joint allocation matrix and the original prediction result to generate an error vector of the space-time prediction model; The convolution kernel weight and attention coefficient of the space-time convolution neural network are adjusted through an error back propagation mechanism, and the risk assessment parameter of the path planning neural network is updated; A dynamic iterative optimization mechanism is established to automatically correct the communication configuration scheme as the environment changes, forming a closed-loop management process of evaluation-configuration-correction.

10. An artificial intelligence-based unmanned aerial vehicle sharing intelligent management system, characterized in that, The system is used to perform the method of any one of claims 1-9, the system comprising: The acquisition module is configured to acquire multi-dimensional communication parameters between the UAV group and the ground control platform in real time to calculate a comprehensive communication quality score; The generation module is configured to generate an initial communication configuration scheme based on the comprehensive communication quality score through a deep learning network; The calculation module is configured to input the initial communication configuration scheme into a preset space-time prediction model to perform advanced calculation on regional spectrum demand through a space-time convolution neural network, and generate a frequency-time slot joint allocation matrix; The correction module is configured to select three non-collinear position points in the UAV operation airspace, construct a virtual spatial geometric shape through a coordinate mapping algorithm, extract the geometric center coordinates of the virtual spatial geometric shape, calculate the offset between the geometric center coordinates and the communication coverage centroid as a first correction factor, analyze the signal coverage angle distribution between the three position points, generate a second correction factor in combination with a path loss model, calculate a multipath effect compensation coefficient as a third correction factor according to the signal attenuation gradient on the boundary of the virtual spatial geometric shape, and fuse the first correction factor, the second correction factor, and the third correction factor into a comprehensive correction value; The allocation module is configured to evaluate the communication interruption risk through a path planning neural network based on the comprehensive correction value, and generate a multi-path transmission scheme including a redundant relay node selection rule and a dynamic fragmentation reorganization strategy; The adjustment module is configured to dynamically and iteratively update the frequency-time slot joint allocation matrix based on the multi-path transmission scheme, adjust the weight parameter of the space-time prediction model through an error back propagation mechanism, and form a closed loop management.

Images