Satellite communication resource intelligent prediction and allocation method and system based on multi-modal user behavior

Abstract

The application discloses a kind of satellite communication resource intelligent prediction and allocation method and system based on multi-modal user behavior.The method is as follows: firstly, multi-modal data is collected, and data standardization processing and missing value filling are carried out to obtain standardized data that can effectively extract features and accurately predict;Then, the user behavior mode is extracted by spatio-temporal feature fusion technology;Then, a hierarchical prediction model is used to predict resource demand for different terminal types, and optimization allocation is carried out under the multiple constraint conditions of satellite capacity constraint, frequency band compatibility constraint and service quality constraint;Finally, performance evaluation is carried out, and the scheme is optimized according to the performance evaluation result.The system includes a multi-modal data acquisition layer, a multi-modal data feature fusion layer, a prediction and allocation layer, and an evaluation layer.The application deeply integrates spatio-temporal multi-dimensional multi-source information, improves the accuracy of satellite communication resource demand prediction, and improves the utilization rate of satellite communication resources and system service quality.

Description

Technical Field

[0001] This invention relates to the field of satellite communication resource management technology, and in particular to a method and system for intelligent prediction and allocation of satellite communication resources based on multimodal user behavior. Background Technology

[0002] In early satellite communication systems, resource allocation primarily employed static pre-allocation schemes, with typical implementation methods including fixed allocation and on-demand allocation. Fixed allocation assigns satellite resources (such as frequency bands and time slots) to specific users or areas without considering fluctuations in actual demand. This method is simple to implement but has extremely low resource utilization, with statistics showing an average utilization rate of less than 40%. On-demand allocation introduces limited dynamic adjustments based on fixed allocation, temporarily allocating resources according to user requests. However, its response time can be as long as minutes, making it unable to adapt to rapidly changing user needs.

[0003] As the user base expands, predictive allocation methods based on historical statistical patterns have emerged, including time series forecasting and spatial correlation forecasting. However, these methods have a single data dimension, considering only single indicators such as traffic or number of users, ignoring multi-dimensional information such as terminal type and business characteristics, resulting in low prediction accuracy. They are usually measured in hours or days and cannot meet the needs of fine-grained scheduling at the minute level. Spatial modeling is also simple, assuming that geographical proximity means demand correlation, without considering differences in user behavior patterns.

[0004] Therefore, there is an urgent need to build an intelligent resource collaborative management framework that can deeply integrate spatiotemporal multi-dimensional and multi-source information, accurately characterize the differentiated resource demand characteristics of various terminals, dynamically adapt to changes in network topology and service load, and strictly adhere to the transmission constraints of the satellite physical layer and the service quality assurance requirements of the service layer. This framework aims to systematically break through existing technical bottlenecks and promote a fundamental paradigm shift in satellite communication resource management from "passive response allocation" to "proactive predictive optimization". Summary of the Invention

[0005] The purpose of this invention is to provide a satellite communication resource management method and system that can integrate multimodal user behavior data, accurately predict resource demand, and achieve dynamic optimization allocation, so as to solve the technical problems in the prior art such as low resource utilization, poor prediction accuracy, long response time, and inability to adapt to rapidly changing user needs.

[0006] The technical solution to achieve the purpose of this invention is: a method for intelligent prediction and allocation of satellite communication resources based on multimodal user behavior, comprising the following steps:

[0007] Step 1: Collect multimodal data, and perform data standardization and missing value imputation to obtain standardized data that can effectively extract features and make accurate predictions;

[0008] Step 2: Extract user behavior patterns using spatiotemporal feature fusion technology;

[0009] Step 3: Use a hierarchical prediction model to predict resource requirements for different terminal types, and optimize allocation under multiple constraints such as satellite capability constraints, frequency band compatibility constraints, and quality of service constraints.

[0010] Step 4: Conduct a performance evaluation and optimize the solution based on the evaluation results.

[0011] As a specific example, the multimodal data mentioned in step 1 includes network terminal data, earth station data, regional range data, number of users data, network duration data, resource utilization data, and idle resource data.

[0012] As a specific example, step 1 involves collecting multimodal data, performing data standardization and missing value imputation to obtain standardized data that can effectively extract features and make accurate predictions, as detailed below:

[0013] Step 1.1: Real-time acquisition of multimodal data is performed through terminal agent, earth station monitoring module, and signaling acquisition system;

[0014] Step 1.2: To address the issue of dimensional differences in multimodal data, the Z-score standardization method based on statistics is used for data standardization. Let the original observation data be... Its standardized data after normalization processing Defined by the following formula:

[0015]

[0016] In the formula, The arithmetic mean of the data distribution. This represents the corresponding standard deviation;

[0017] Normalization maps data of any distribution to a standard normal distribution space with a mean of 0 and a standard deviation of 1.

[0018] Step 1.3: To address data loss caused by channel interference, equipment failure, or network latency in satellite communication environments, a spatiotemporal awareness-based imputation mechanism based on the nearest neighbor algorithm is used to imput missing values. The imputation value is calculated using the following formula:

[0019]

[0020] in, This is the nearest neighbor count parameter, which is usually set to 5 depending on the data characteristics. This represents the first closest point to the missing point in the spatiotemporal dimension. One valid observation.

[0021] As a specific example, step 2, which involves extracting user behavior patterns using spatiotemporal feature fusion technology, is detailed as follows:

[0022] Step 2.1: Spatial correlation mining is performed through graph convolution to capture business relationships and geographical dependencies between earth stations;

[0023]

[0024] in , It is the original adjacency matrix, representing the Earth stations. and There is a significant connection in beam coverage or business operations; It is a normalized matrix, which ensures the numerical stability of feature propagation; It is the node feature matrix. Earth station, Dimensional features; This represents a learnable weight matrix; This indicates that it is a ReLU activation function;

[0025] Step 2.2: Extract time patterns through attention mechanisms to separate periodicity, trends, and breakthroughs at different time scales.

[0026]

[0027]

[0028] in, Indicates the first Modal features; Indicates attention weight; Represents the query vector;

[0029] Step 2.3: Achieve intelligent modal fusion through tensor decomposition, dynamically weighting different data sources according to task requirements:

[0030]

[0031] in, Represents the original three-dimensional tensor: ; Represents the core tensor; Represents the factor matrix; Operators indicate the direction along the first line. Matrix multiplication in dimension .

[0032] As a specific example, step 3 employs a hierarchical prediction model to predict resource requirements for different terminal types, and optimizes resource allocation under multiple constraints including satellite capability constraints, frequency band compatibility constraints, and quality of service constraints, as detailed below:

[0033] Step 3.1: Terminal-level demand forecasting. The weighted sum of multiple Gaussian distributions is used to approximate the true distribution, establishing a multimodal probability distribution for each type of terminal. The terminal-level demand forecasting formula is as follows:

[0034]

[0035] in, This indicates the number of mixed components, reflecting the complexity of the demand pattern; This represents the weight parameters, which learn the probability of each pattern occurring. Indicates the mean and variance;

[0036] Step 3.2: Regional-level demand aggregation to capture spatial dependencies. The formula for regional-level demand aggregation is:

[0037]

[0038]

[0039] in, Indicates the region The set of neighboring regions; Indicates the geographical distance between Earth stations; This represents the attenuation parameter, used to control the range of influence.

[0040] Step 3.3: Under multiple constraints including satellite capability constraints, frequency band compatibility constraints, and quality of service constraints, optimize the allocation. The optimization allocation model is as follows:

[0041]

[0042] The constraints are:

[0043]

[0044] in Total system capacity; For users Minimum resource requirements; For users The set of frequency bands that are allowed to be used; For users Forecasted demand; For actual allocation to users The amount of resources; For users in the previous allocation period The amount of resources; Hyperparameters for balancing matching accuracy and system stability

[0045] The optimal allocation model employs a bi-term objective function design. The goal is to minimize the gap between predicted demand and actual allocation, and to improve the accuracy of resource allocation. The aim is to reduce frequent changes in allocation schemes; hyperparameters Hyperparameters that balance matching accuracy and system stability.

[0046] As a specific example, step 4 involves performing a performance evaluation and optimizing the solution based on the evaluation results, as detailed below:

[0047] Step 4.1, Service Quality MAPE Assessment. The MAPE calculation formula is:

[0048]

[0049] in, Indicates actual resource demand; This indicates a forecast of resource demand; Indicates the number of evaluation samples; Indicates excellence; This indicates good; This indicates that improvements are still needed;

[0050] Step 4.2, Fairness Indicator Assessment, Fairness Indicators The calculation formula is:

[0051]

[0052] This indicates a waste of resources; This indicates a potential risk of overload. Represents the ideal interval;

[0053] Step 4.3, Feedback Optimization Scheme: When MAPE is too high and the prediction accuracy is not up to standard, increase training data, adjust model parameters, and optimize feature engineering; when resource utilization is abnormal and the utilization rate is lower than the set value, reduce the reserve margin and optimize the allocation strategy; when the utilization rate is higher than the set value, increase emergency reserve and implement load balancing.

[0054] A satellite communication resource intelligent prediction and allocation system based on multimodal user behavior, used to implement the aforementioned satellite communication resource intelligent prediction and allocation method based on multimodal user behavior, characterized in that it includes:

[0055] The multimodal data acquisition layer is used to acquire multimodal data and perform data standardization and missing value imputation to obtain standardized data.

[0056] The multimodal data feature fusion layer is used to reduce the extracted multidimensional features into a three-dimensional tensor of time multiplication, space multiplication, and feature multiplication. It extracts key feature patterns through Tucker decomposition, assigns adaptive weights to different feature modalities according to the current prediction task, and generates a unified feature representation.

[0057] The prediction and allocation layer is used to achieve terminal-level fine-grained prediction and regional-level spatial correlation modeling through a two-layer architecture of hybrid density network and spatial aggregation. It minimizes prediction bias, reduces frequent switching, ensures the basic needs of various users, and maximizes resource utilization through multi-objective constraint optimization balance. It also adapts to changes in user behavior patterns and network topology through a real-time update mechanism.

[0058] The evaluation layer is used to calculate the MAPE value by collecting actual observations and corresponding predicted values, summarize all allocated resources and obtain the total available resources of the system to calculate fairness indicators for performance evaluation, and optimize the scheme based on the performance evaluation results.

[0059] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior.

[0060] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior.

[0061] A computer program product includes computer instructions for causing a computer to execute the aforementioned intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior.

[0062] Compared with the prior art, the significant advantages of this invention are:

[0063] (1) By adopting a multimodal fusion mechanism, terminal data, earth station data, user behavior data and resource status data are uniformly represented and learned, which deeply integrates spatiotemporal multidimensional and multi-source information and improves the accuracy of satellite communication resource demand prediction;

[0064] (2) Terminal perception prediction is adopted, and differentiated prediction models are designed for different terminal types such as TDMA / FDMA and single / multi-beam, which accurately characterize the differentiated resource demand characteristics of various terminals.

[0065] (3) By adopting spatiotemporal joint modeling, it captures both temporal dynamics and spatial correlations, which can dynamically adapt to changes in network topology and service load, achieve dynamic optimization allocation, improve the accuracy of satellite communication resource prediction, and enhance the utilization rate of satellite communication resources and the quality of system services. Attached Figure Description

[0066] Figure 1 This is a flowchart of the intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior, as proposed in this invention.

[0067] Figure 2 This is a schematic diagram of the intelligent prediction and allocation system for satellite communication resources based on multimodal user behavior, as described in this invention.

[0068] Figure 3 This is a schematic diagram of the process for extracting multimodal data and fusing multimodal data features in this invention.

[0069] Figure 4 This is a flowchart illustrating the performance evaluation and scheme optimization process in this invention. Detailed Implementation

[0070] like Figure 1 As shown, this invention provides a method for intelligent prediction and allocation of satellite communication resources based on multimodal user behavior, comprising the following steps:

[0071] Step 1: Collect multimodal data, and perform data standardization and missing value imputation to obtain standardized data that can effectively extract features and make accurate predictions;

[0072] Step 2: Extract user behavior patterns using spatiotemporal feature fusion technology;

[0073] Step 3: Use a hierarchical prediction model to predict resource requirements for different terminal types, and optimize allocation under multiple constraints such as satellite capability constraints, frequency band compatibility constraints, and quality of service constraints.

[0074] Step 4: Conduct a performance evaluation and optimize the solution based on the evaluation results.

[0075] As a specific example, the multimodal data mentioned in step 1 includes network terminal data, earth station data, regional range data, number of users data, network duration data, resource utilization data, and idle resource data.

[0076] As a specific example, step 1 involves collecting multimodal data, performing data standardization and missing value imputation, to obtain standardized data that can effectively extract features and make accurate predictions, such as... Figure 3 As shown, the details are as follows:

[0077] Step 1.1: Real-time acquisition of multimodal data is performed through terminal agent, earth station monitoring module, and signaling acquisition system;

[0078] Step 1.2: To address the issue of dimensional differences in multimodal data, the Z-score standardization method based on statistics is used for data standardization. Let the original observation data be... Its standardized data after normalization processing Defined by the following formula:

[0079]

[0080] In the formula, The arithmetic mean of the data distribution. This represents the corresponding standard deviation;

[0081] Normalization maps data of any distribution to a standard normal distribution space with a mean of 0 and a standard deviation of 1.

[0082] Step 1.3: To address data loss caused by channel interference, equipment failure, or network latency in satellite communication environments, a spatiotemporal awareness-based imputation mechanism based on the nearest neighbor algorithm is used to imput missing values. The imputation value is calculated using the following formula:

[0083]

[0084] in, This is the nearest neighbor count parameter, which is usually set to 5 depending on the data characteristics. This represents the first closest point to the missing point in the spatiotemporal dimension. One valid observation.

[0085] As a specific example, step 2 describes extracting user behavior patterns using spatiotemporal feature fusion technology, such as... Figure 3 As shown, the details are as follows:

[0086] Step 2.1: Spatial correlation mining is performed through graph convolution to capture business relationships and geographical dependencies between earth stations;

[0087]

[0088] in , It is the original adjacency matrix, representing the Earth stations. and There is a significant connection in beam coverage or business operations; It is a normalized matrix, which ensures the numerical stability of feature propagation; It is the node feature matrix. Earth station, Dimensional features; This represents a learnable weight matrix; This indicates that it is a ReLU activation function;

[0089] Step 2.2: Extract time patterns through attention mechanisms to separate periodicity, trends, and breakthroughs at different time scales.

[0090]

[0091]

[0092] in, Indicates the first Modal features; Indicates attention weight; Represents the query vector;

[0093] Step 2.3: Achieve intelligent modal fusion through tensor decomposition, dynamically weighting different data sources according to task requirements:

[0094]

[0095] in, Represents the original three-dimensional tensor: ; Represents the core tensor; Represents the factor matrix; Operators indicate the direction along the first line. Matrix multiplication in dimension .

[0096] As a specific example, step 3 employs a hierarchical prediction model to predict resource requirements for different terminal types, and optimizes resource allocation under multiple constraints including satellite capability constraints, frequency band compatibility constraints, and quality of service constraints, as detailed below:

[0097] Step 3.1: Terminal-level demand forecasting. The weighted sum of multiple Gaussian distributions is used to approximate the true distribution, establishing a multimodal probability distribution for each type of terminal. The terminal-level demand forecasting formula is as follows:

[0098]

[0099] in, This indicates the number of mixed components, reflecting the complexity of the demand pattern; This represents the weight parameters, which learn the probability of each pattern occurring. Indicates the mean and variance;

[0100] Step 3.2: Regional-level demand aggregation to capture spatial dependencies. The formula for regional-level demand aggregation is:

[0101]

[0102]

[0103] in, Indicates the region The set of neighboring regions; Indicates the geographical distance between Earth stations; This represents the attenuation parameter, used to control the range of influence.

[0104] Step 3.3: Under multiple constraints including satellite capability constraints, frequency band compatibility constraints, and quality of service constraints, optimize the allocation. The optimization allocation model is as follows:

[0105]

[0106] The constraints are:

[0107]

[0108] in Total system capacity; For users Minimum resource requirements; For users The set of frequency bands that are allowed to be used; For users Forecasted demand; For actual allocation to users The amount of resources; For users in the previous allocation period The amount of resources; Hyperparameters for balancing matching accuracy and system stability

[0109] The optimal allocation model employs a bi-term objective function design. The goal is to minimize the gap between predicted demand and actual allocation, and to improve the accuracy of resource allocation. The aim is to reduce frequent changes in allocation schemes; hyperparameters Hyperparameters that balance matching accuracy and system stability.

[0110] As a specific example, step 4 involves performing a performance evaluation and optimizing the solution based on the evaluation results, such as... Figure 4 As shown, the details are as follows:

[0111] Step 4.1, Service Quality MAPE Assessment. The MAPE calculation formula is:

[0112]

[0113] in, Indicates actual resource demand; This indicates a forecast of resource demand; Indicates the number of evaluation samples; Indicates excellence; This indicates good; This indicates that improvements are still needed;

[0114] Step 4.2, Fairness Indicator Assessment, Fairness Indicators The calculation formula is:

[0115]

[0116] This indicates a waste of resources; This indicates a potential risk of overload. Represents the ideal interval;

[0117] Step 4.3, Feedback Optimization Scheme: When MAPE is too high and the prediction accuracy is not up to standard, increase training data, adjust model parameters, and optimize feature engineering; when resource utilization is abnormal and the utilization rate is lower than the set value, reduce the reserve margin and optimize the allocation strategy; when the utilization rate is higher than the set value, increase emergency reserve and implement load balancing.

[0118] like Figure 2 As shown, the present invention also provides a satellite communication resource intelligent prediction and allocation system based on multimodal user behavior, used to implement the aforementioned satellite communication resource intelligent prediction and allocation method based on multimodal user behavior, characterized in that it includes:

[0119] The multimodal data acquisition layer is used to acquire multimodal data and perform data standardization and missing value imputation to obtain standardized data.

[0120] The multimodal data feature fusion layer is used to reduce the extracted multidimensional features into a three-dimensional tensor of time multiplication, space multiplication, and feature multiplication. It extracts key feature patterns through Tucker decomposition, assigns adaptive weights to different feature modalities according to the current prediction task, and generates a unified feature representation.

[0121] The prediction and allocation layer is used to achieve terminal-level fine-grained prediction and regional-level spatial correlation modeling through a two-layer architecture of hybrid density network and spatial aggregation. It minimizes prediction bias, reduces frequent switching, ensures the basic needs of various users, and maximizes resource utilization through multi-objective constraint optimization balance. It also adapts to changes in user behavior patterns and network topology through a real-time update mechanism.

[0122] The evaluation layer is used to calculate the MAPE value by collecting actual observations and corresponding predicted values, summarize all allocated resources and obtain the total available resources of the system to calculate fairness indicators for performance evaluation, and optimize the scheme based on the performance evaluation results.

[0123] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior.

[0124] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior.

[0125] The present invention also provides a computer program product, including computer instructions, which are used to cause a computer to execute the aforementioned intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior.

[0126] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0127] Example

[0128] This embodiment provides a method for intelligent prediction and allocation of satellite communication resources based on multimodal user behavior, such as... Figure 1 As shown, the specific steps are as follows:

[0129] Step 1: Raw Data Acquisition and Preprocessing

[0130] Step 1.1: Multi-source data acquisition

[0131] (1) Collect real-time terminal data: record the online status, geographical location, and current bandwidth usage of each terminal;

[0132] (2) Collect earth station status: monitor earth station load, available bandwidth, bit error rate and other operating parameters;

[0133] (3) Obtain resource monitoring information: track the occupancy status, power utilization, and channel quality of each satellite frequency band in real time;

[0134] (4) Statistical analysis of user behavior data: Calculate the number of online users, average session duration, and distribution of business types.

[0135] Step 1.2: Data Cleaning Process

[0136] (1) Time alignment: unify the time granularity of data from different acquisition frequencies;

[0137] (2) Missing value imputation: For missing terminal data, the nearest neighbor algorithm is used, with K=5, to imput the missing data based on geographical location and terminal type;

[0138] (3) Outlier removal: Identify and correct outlier readings;

[0139] (4) Data standardization: Normalize data of different dimensions to eliminate differences in magnitude.

[0140] Step 1.3, Data Quality Verification

[0141] (1) Integrity check: Ensure that all data sources have complete records at each point in time;

[0142] (2) Consistency verification: Check the logical relationship between each data source, such as the total terminal bandwidth ≤ the bandwidth allocated to the earth station;

[0143] (3) Timeliness confirmation: Ensure that the data delay does not exceed 10 seconds to meet the real-time requirements.

[0144] Step 2: Feature Representation and Fusion

[0145] Step 2.1: Multimodal Feature Extraction

[0146] (1) Extracting terminal features: The terminal type is encoded using one-hot encoding, and historical usage pattern features are extracted;

[0147] (2) Constructing spatial features: Calculating the geographic distance matrix between earth stations and identifying business hotspot areas;

[0148] (3) Analyze time characteristics: extract time patterns such as hourly cycle, daily cycle, and weekly cycle;

[0149] (4) Quantify resource characteristics: Calculate the real-time utilization rate and idle resource distribution of each frequency band.

[0150] Step 2.2, Graph Structure Modeling

[0151] (1) Construct an adjacency matrix: Establish connection relationships based on the distance between earth stations and the degree of business relevance;

[0152] (2) Calculate the degree matrix: count the number of connections for each earth station for normalization processing;

[0153] (3) Feature propagation calculation: Through graph convolution algorithm, the features of each earth station are aggregated with neighbor information.

[0154] Step 2.3, Feature Fusion Processing

[0155] (1) Tensor dimensionality reduction: The extracted multidimensional features are organized into a three-dimensional tensor of time × space × features;

[0156] (2) Core feature extraction: Key feature patterns are extracted by Tucker decomposition, and the dimension is compressed from 1024 dimensions to 64 dimensions;

[0157] (3) Attention weighting: Assign adaptive weights to different feature modalities based on the current prediction task;

[0158] (4) Generate a unified feature vector: output a feature representation with unified dimensions and dense information for use by the prediction model.

[0159] Step 3: Intelligent Prediction and Optimized Allocation

[0160] Step 3.1, Layered Demand Forecasting

[0161] (1) Establish a hybrid density prediction model: adopt differentiated network structures for different terminal types;

[0162] (2) Spatial demand aggregation: Based on the weighted aggregation of terminal prediction results according to geographical relationships;

[0163] (3) Global constraint integration: Apply system-level constraints to make reasonable adjustments to the prediction;

[0164] (4) Output structured forecast: Generate a complete demand forecast report containing confidence intervals.

[0165] Step 3.2: Multi-constraint resource allocation

[0166] (1) Define multidimensional constraints: including capacity limits, quality of service, technical compatibility, and system stability;

[0167] (2) Constructing an optimization objective: minimizing prediction bias and allocating switching costs;

[0168] (3) Solving for the optimal solution: Calculate the optimal allocation that satisfies all constraints using mathematical optimization algorithms;

[0169] (4) Issue execution instructions: Convert the allocation scheme into specific configuration parameters and implement it;

[0170] (5) Pre-assessment of matching effect: Calculate the degree of consistency between the allocation plan and the predicted demand.

[0171] Step 4: Performance Evaluation and Feedback

[0172] Step 4.1, MAPE Calculation and Fairness Indicator U Calculation: Collect actual observations and corresponding predicted values ​​to calculate the MAPE value; summarize all allocated resources and obtain the total available resources of the system to calculate the fairness index;

[0173] Step 4.2, Comprehensive Evaluation: Evaluate from the dimensions of terminal type, time and space, and business.

[0174] Step 4.3, Feedback Optimization Scheme: When the prediction accuracy is not up to standard, i.e., MAPE is too high, increase training data, adjust model parameters, and optimize feature engineering; when resource utilization is abnormal, i.e., the utilization rate is too low, reduce the reserved margin and optimize the allocation strategy; when the utilization rate is too high, increase emergency reserve and implement load balancing.

[0175] This invention improves data quality by collecting multimodal data and employing Z-score normalization to eliminate dimensional differences between different data sources; it enhances the accuracy of satellite communication resource demand prediction by deeply fusing spatiotemporal multidimensional and multi-source information through graph convolution, attention mechanisms, and tensor decomposition; it improves satellite communication resource utilization and system service quality by using hierarchical prediction models and multi-constraint optimization allocation; and it achieves adaptive adjustment through performance evaluation and feedback optimization, realizing a fundamental transformation from passive response allocation to proactive predictive optimization.

[0176] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent prediction and allocation of satellite communication resources based on multimodal user behavior, characterized in that, Includes the following steps: Step 1: Collect multimodal data, and perform data standardization and missing value imputation to obtain standardized data that can effectively extract features and make accurate predictions; Step 2: Extract user behavior patterns using spatiotemporal feature fusion technology; Step 3: Use a hierarchical prediction model to predict resource requirements for different terminal types, and optimize allocation under multiple constraints such as satellite capability constraints, frequency band compatibility constraints, and quality of service constraints. Step 4: Conduct a performance evaluation and optimize the solution based on the evaluation results.

2. The intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior according to claim 1, characterized in that, The multimodal data mentioned in step 1 includes network terminal data, earth station data, regional range data, number of users data, network duration data, resource utilization data, and idle resource data.

3. The intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior according to claim 1, characterized in that, Step 1 involves collecting multimodal data, performing data standardization and missing value imputation to obtain standardized data that can effectively extract features and make accurate predictions, as detailed below: Step 1.1: Real-time acquisition of multimodal data is performed through terminal agent, earth station monitoring module, and signaling acquisition system; Step 1.2: To address the issue of dimensional differences in multimodal data, the Z-score standardization method based on statistics is used for data standardization. Let the original observation data be... Its standardized data after normalization processing Defined by the following formula: In the formula, The arithmetic mean of the data distribution. This represents the corresponding standard deviation; Normalization maps data of any distribution to a standard normal distribution space with a mean of 0 and a standard deviation of 1. Step 1.3: To address data loss caused by channel interference, equipment failure, or network latency in satellite communication environments, a spatiotemporal awareness-based imputation mechanism based on the nearest neighbor algorithm is used to imput missing values. The imputation value is calculated using the following formula: in, This is the nearest neighbor count parameter, which is usually set to 5 depending on the data characteristics. This represents the first closest point to the missing point in the spatiotemporal dimension. One valid observation.

4. The intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior according to claim 1, characterized in that, Step 2, which involves extracting user behavior patterns using spatiotemporal feature fusion technology, is detailed below: Step 2.1: Spatial correlation mining is performed through graph convolution to capture business relationships and geographical dependencies between earth stations; in , It is the original adjacency matrix, representing the Earth stations. and There is a significant connection in beam coverage or business operations; It is a normalized matrix, which ensures the numerical stability of feature propagation; It is the node feature matrix. Earth station, Dimensional features; This represents a learnable weight matrix; This indicates that it is a ReLU activation function; Step 2.2: Extract time patterns through attention mechanisms to separate periodicity, trends, and breakthroughs at different time scales. in, Indicates the first Modal features; Indicates attention weight; Represents the query vector; Step 2.3: Achieve intelligent modal fusion through tensor decomposition, dynamically weighting different data sources according to task requirements: in, Represents the original three-dimensional tensor: ; Represents the core tensor; Represents the factor matrix; Operators indicate the direction along the first line. Matrix multiplication in dimension .

5. The intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior according to claim 1, characterized in that, Step 3 describes the use of a hierarchical prediction model to predict resource requirements for different terminal types, and to optimize resource allocation under multiple constraints including satellite capability constraints, frequency band compatibility constraints, and quality of service constraints, as detailed below: Step 3.1: Terminal-level demand forecasting. The weighted sum of multiple Gaussian distributions is used to approximate the true distribution, establishing a multimodal probability distribution for each type of terminal. The terminal-level demand forecasting formula is as follows: in, This indicates the number of mixed components, reflecting the complexity of the demand pattern; This represents the weight parameters, which learn the probability of each pattern occurring. Indicates the mean and variance; Step 3.2: Regional-level demand aggregation to capture spatial dependencies. The formula for regional-level demand aggregation is: in, Indicates the area The set of neighboring regions; Indicates the geographical distance between Earth stations; This represents the attenuation parameter, used to control the range of influence. Step 3.3: Under multiple constraints including satellite capability constraints, frequency band compatibility constraints, and quality of service constraints, optimize the allocation. The optimization allocation model is as follows: The constraints are: in Total system capacity; For users Minimum resource requirements; For users The set of frequency bands that are allowed to be used; For users Forecasted demand; For actual allocation to users The amount of resources; For users in the previous allocation period The amount of resources; Hyperparameters for balancing matching accuracy and system stability The optimal allocation model employs a bi-term objective function design. The goal is to minimize the gap between predicted demand and actual allocation, and to improve the accuracy of resource allocation. The aim is to reduce frequent changes in allocation schemes; hyperparameters Hyperparameters that balance matching accuracy and system stability.

6. The intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior according to claim 5, characterized in that, Step 4 involves conducting a performance evaluation and optimizing the solution based on the evaluation results, as detailed below: Step 4.1, Service Quality MAPE Assessment. The MAPE calculation formula is: in, Indicates actual resource demand; This indicates a forecast of resource demand; Indicates the number of evaluation samples; Indicates excellence; This indicates good; This indicates that improvements are still needed; Step 4.2, Fairness Indicator Assessment, Fairness Indicators The calculation formula is: This indicates a waste of resources; This indicates a potential risk of overload. Represents the ideal interval; Step 4.3, Feedback Optimization Scheme: When MAPE is too high and the prediction accuracy is not up to standard, increase training data, adjust model parameters, and optimize feature engineering; when resource utilization is abnormal and the utilization rate is lower than the set value, reduce the reserve margin and optimize the allocation strategy; when the utilization rate is higher than the set value, increase emergency reserve and implement load balancing.

7. A satellite communication resource intelligent prediction and allocation system based on multimodal user behavior, used to implement the satellite communication resource intelligent prediction and allocation method based on multimodal user behavior as described in any one of claims 1 to 6, characterized in that, include: The multimodal data acquisition layer is used to acquire multimodal data and perform data standardization and missing value imputation to obtain standardized data. The multimodal data feature fusion layer is used to reduce the extracted multidimensional features into a three-dimensional tensor of time multiplication, space multiplication, and feature multiplication. It extracts key feature patterns through Tucker decomposition, assigns adaptive weights to different feature modalities according to the current prediction task, and generates a unified feature representation. The prediction and allocation layer is used to achieve terminal-level fine-grained prediction and regional-level spatial correlation modeling through a two-layer architecture of hybrid density network and spatial aggregation. It minimizes prediction bias, reduces frequent switching, ensures the basic needs of various users, and maximizes resource utilization through multi-objective constraint optimization balance. It also adapts to changes in user behavior patterns and network topology through a real-time update mechanism. The evaluation layer is used to calculate the MAPE value by collecting actual observations and corresponding predicted values, summarize all allocated resources and obtain the total available resources of the system to calculate fairness indicators for performance evaluation, and optimize the scheme based on the performance evaluation results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The method includes computer instructions for causing a computer to execute the intelligent prediction and allocation method for satellite communication resources based on multimodal user behavior as described in any one of claims 1 to 6.

Images