An artificial intelligence-based wireless communication resource allocation method and system

Abstract

The application discloses a wireless communication resource allocation method and system based on artificial intelligence, and the method comprises the following steps: obtaining fusion wireless communication data; constructing a DRL local model at each base station, and training the DRL local model by using an improved PPO algorithm; periodically aggregating the DRL local model parameters of each base station by using a central server, adding noise by using a DP differential privacy technology, inputting the fusion wireless communication data into a DP-DRL global model, and outputting an initial joint allocation strategy; allocating wireless communication resources based on the initial joint allocation strategy, statistically analyzing system performance indexes by using a sliding window, feeding back the system performance indexes to a fuzzy logic controller, adjusting the weight of a reward function, and obtaining a target joint allocation strategy. The method can train the local characteristics of each base station, the global model integrates the experience of each base station, and the global optimization of the resource allocation strategy is realized.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a wireless communication resource allocation method and system based on artificial intelligence. Background Technology

[0002] The scale of wireless communication networks is expanding rapidly, the number of terminal devices is growing explosively, and service types are becoming more diversified and complex. Traditional wireless communication resource allocation methods are mostly based on static rules or simple heuristic algorithms, which are difficult to adapt to the dynamically changing network environment and diverse user needs. Furthermore, in a multi-base station collaborative network architecture, how to efficiently integrate the resources of each base station to achieve globally optimal resource allocation while ensuring user data privacy and security has become a critical issue that urgently needs to be addressed in the current wireless communication field. Traditional methods suffer from limitations such as low resource utilization, poor user experience, and insufficient privacy protection. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a wireless communication resource allocation method based on artificial intelligence.

[0004] The technical solution of the present invention to achieve the above objectives is as follows: Further, in the above-mentioned method for allocating wireless communication resources based on artificial intelligence, the method includes the following steps:

[0005] Terminal link status data, user behavior data, and environmental interference signal data from each base station are acquired to obtain multimodal wireless communication data. Attention mechanism is used to dynamically weight the different data sources in the multimodal wireless communication data to obtain fused wireless communication data.

[0006] A local DRL model is constructed at each base station, and the local DRL model is trained using the improved PPO algorithm.

[0007] The DRL local model parameters of each base station are periodically aggregated by the central server, and noise is added using DP differential privacy technology to generate a DP-DRL global model.

[0008] The fused wireless communication data is input into the DP-DRL global model, and an initial joint allocation strategy is output.

[0009] The wireless communication resources are allocated based on the initial joint allocation strategy. The system performance indicators are statistically analyzed through a sliding window and fed back to the fuzzy logic controller to adjust the weight of the reward function, thereby obtaining the target joint allocation strategy.

[0010] Furthermore, in the aforementioned artificial intelligence-based wireless communication resource allocation method, the step of acquiring terminal link status data, user behavior data, and environmental interference signal data from each base station to obtain multimodal wireless communication data includes:

[0011] Real-time acquisition of signal strength, signal-to-noise ratio, bit error rate, and transmission delay parameters between the terminal and the base station to obtain terminal link status data;

[0012] By integrating user behavior monitoring software into terminal devices, user behavior data is obtained by collecting data on user service types, service usage periods, data transmission rate requirements, and service quality requirements.

[0013] The spectrum analyzer and electromagnetic interference detector deployed around the base station are used to detect the surrounding electromagnetic interference signals in real time, including at least co-channel interference, adjacent channel interference, industrial noise and natural noise, to obtain environmental interference signal data.

[0014] The terminal link status data, user behavior data, and environmental interference signal data are integrated to obtain multimodal wireless communication data.

[0015] Furthermore, in the aforementioned AI-based wireless communication resource allocation method, the step of dynamically weighting different data sources in the multimodal wireless communication data using an attention mechanism to obtain fused wireless communication data includes:

[0016] A multimodal attention network is constructed based on an LSTM network. The multimodal attention network contains multiple attention sub-modules, which correspond to terminal link status data, user behavior data, and environmental interference signal data, respectively.

[0017] The attention submodule is used to calculate the attention weights of different data sources in different time and space dimensions, dynamically adjust the importance of multimodal wireless communication data, and obtain initial wireless communication data.

[0018] The initial wireless communication data, after being weighted by the attention mechanism, is fused together, and the fused wireless communication data is generated by weighted summation. The weight coefficients are dynamically adjusted according to the attention weights calculated in real time to obtain the fused wireless communication data.

[0019] Furthermore, in the aforementioned AI-based wireless communication resource allocation method, the step of constructing a local DRL model at each base station and training the local DRL model using an improved PPO algorithm includes:

[0020] The feature data from the fused wireless communication data is used as the state input of the DRL local model. The dimension of the state space is dynamically adjusted according to the actual data features collected, thus defining the state space of the DRL local model.

[0021] The action space includes the allocation strategy for wireless communication resources. Each action corresponds to a specific resource allocation method, defining the model action space.

[0022] For voice call services with high real-time requirements, increase the weight of transmission latency and packet loss rate in the reward function; for video streaming services with high bandwidth requirements, increase the weight of user throughput.

[0023] Furthermore, in the aforementioned AI-based wireless communication resource allocation method, the step of periodically aggregating the DRL local model parameters of each base station using a central server, adding noise using DP differential privacy technology, and generating a DP-DRL global model includes:

[0024] Collect the local DRL model parameters of each base station and perform periodic aggregation, setting the aggregation period according to the system's operational requirements and network conditions;

[0025] The federated averaging algorithm is used as the model parameter aggregation algorithm to perform a weighted average of the local DRL model parameters of each base station, and the weights are set according to the scale of the base station.

[0026] After the DRL local model parameters are aggregated, Laplacian noise is added to the aggregated DP-DRL global model parameters using DP differential privacy technology.

[0027] By adjusting the privacy budget value when adding noise, a DP-DRL global model is generated to balance the strength of privacy protection and model performance.

[0028] Furthermore, in the aforementioned AI-based wireless communication resource allocation method, the step of inputting the fused wireless communication data into the DP-DRL global model and outputting an initial joint allocation strategy includes:

[0029] The fused wireless communication data is input into the DP-DRL global model, and the DP-DRL global model outputs the probability distribution of the action space.

[0030] Based on the probability distribution, a greedy strategy is used to select the action with the highest probability as the allocation strategy.

[0031] The allocation strategy is converted into a specific wireless communication resource allocation strategy format to obtain the initial joint allocation strategy.

[0032] Furthermore, in the aforementioned AI-based wireless communication resource allocation method, the allocation of wireless communication resources based on the initial joint allocation strategy, and the statistical analysis of system performance indicators through a sliding window, include:

[0033] The allocation of wireless communication resources based on the initial joint allocation strategy includes: setting the size and step size of the sliding window, and collecting statistics including spectrum utilization, average user throughput, QoS satisfaction, energy efficiency, and system fairness indicators.

[0034] An artificial intelligence-based wireless communication resource allocation system, comprising the following modules:

[0035] The data acquisition module is used to acquire terminal link status data, user behavior data and environmental interference signal data from each base station to obtain multimodal wireless communication data. The different data sources in the multimodal wireless communication data are dynamically weighted using an attention mechanism to obtain fused wireless communication data.

[0036] The local training module is used to build a local DRL model at each base station and train the local DRL model using the improved PPO algorithm.

[0037] The federated aggregation module is used to periodically aggregate the DRL local model parameters of each base station using the central server, add noise using DP differential privacy technology, and generate a DP-DRL global model.

[0038] The resource allocation module is used to input the converged wireless communication data into the DP-DRL global model and output an initial joint allocation strategy;

[0039] The online adaptive module is used to allocate wireless communication resources based on the initial joint allocation strategy, and to feed back the system performance indicators through a sliding window to the fuzzy logic controller to adjust the reward function weights and obtain the target joint allocation strategy.

[0040] Furthermore, in the system implementing the aforementioned AI-based wireless communication resource allocation method, the data acquisition module includes the following sub-modules:

[0041] A submodule is constructed to build a multimodal attention network based on an LSTM network. The multimodal attention network contains multiple attention submodules, which correspond to terminal link status data, user behavior data, and environmental interference signal data, respectively.

[0042] The calculation submodule is used to calculate the attention weights of different data sources in different time and space dimensions using the attention submodule, dynamically adjust the importance of multimodal wireless communication data, and obtain initial wireless communication data.

[0043] The fusion submodule is used to fuse the initial wireless communication data after it has been weighted by the attention mechanism. It generates fused wireless communication data by using a weighted summation method. The weight coefficients are dynamically adjusted according to the attention weights calculated in real time to obtain the fused wireless communication data.

[0044] Furthermore, in the system implementing the aforementioned AI-based wireless communication resource allocation method, the resource allocation module includes the following sub-modules:

[0045] The input submodule is used to input the fused wireless communication data into the DP-DRL global model, and the DP-DRL global model outputs the probability distribution of the action space.

[0046] The selection submodule is used to select the action with the highest probability as the allocation strategy based on the probability distribution using a greedy strategy.

[0047] A submodule is obtained to convert the allocation strategy into a specific wireless communication resource allocation strategy format to obtain an initial joint allocation strategy.

[0048] Its beneficial effects lie in obtaining multimodal wireless communication data by acquiring terminal link status data, user behavior data, and environmental interference signal data from each base station. Different data sources within this multimodal wireless communication data are dynamically weighted using an attention mechanism to obtain fused wireless communication data. A local DRL model is constructed at each base station, and the improved PPO algorithm is used to train this local DRL model. The parameters of the local DRL models from each base station are periodically aggregated using a central server, and noise is added using DP differential privacy technology to generate a DP-DRL global model. The fused wireless communication data is input into the DP-DRL global model, and an initial joint allocation strategy is output. Wireless communication resources are allocated based on this initial joint allocation strategy. System performance indicators are statistically analyzed using a sliding window, and feedback is given to the fuzzy logic controller to adjust the reward function weights, resulting in a target joint allocation strategy. 1. It can adapt to complex and ever-changing network environments through continuous learning and optimization. The local model is trained based on the local characteristics of each base station, while the global model integrates the experience of each base station to achieve global optimization of the resource allocation strategy. Compared with traditional methods, this significantly improves the efficiency of resource allocation and system performance, such as increasing spectrum utilization and user throughput. 2. By adding noise to protect user data and model parameters, the privacy and security of user data during resource allocation are ensured, meeting data security regulations. At the same time, by reasonably adjusting the privacy budget, the impact on model performance is controlled within an acceptable range while protecting privacy, thus achieving a balance between privacy protection and model performance. Attached Figure Description

[0049] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0050] Figure 1 This is a schematic diagram of the first embodiment of a wireless communication resource allocation method based on artificial intelligence in this invention.

[0051] Figure 2This is a schematic diagram of a second embodiment of a wireless communication resource allocation method based on artificial intelligence in this invention.

[0052] Figure 3 This is a schematic diagram of the first embodiment of an artificial intelligence-based wireless communication resource allocation system according to the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0054] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0055] The present invention will now be described in detail with reference to the accompanying drawings, such as... Figure 1 As shown, an artificial intelligence-based wireless communication resource allocation method includes the following steps:

[0056] Step 101: Obtain terminal link status data, user behavior data and environmental interference signal data from each base station to obtain multimodal wireless communication data. Use an attention mechanism to dynamically weight the different data sources in the multimodal wireless communication data to obtain fused wireless communication data.

[0057] Specifically, in this embodiment, the signal strength, signal-to-noise ratio, bit error rate, and transmission delay parameters between the terminal and the base station are collected in real time to obtain terminal link status data;

[0058] By integrating user behavior monitoring software into terminal devices, user behavior data is obtained by collecting data on user service types, service usage periods, data transmission rate requirements, and service quality requirements.

[0059] The spectrum analyzer and electromagnetic interference detector deployed around the base station are used to detect the surrounding electromagnetic interference signals in real time, including at least co-channel interference, adjacent channel interference, industrial noise and natural noise, to obtain environmental interference signal data.

[0060] By integrating terminal link status data, user behavior data, and environmental interference signal data, multimodal wireless communication data is obtained.

[0061] A multimodal attention network is constructed based on an LSTM network. The multimodal attention network contains multiple attention sub-modules, which correspond to terminal link status data, user behavior data, and environmental interference signal data, respectively.

[0062] The attention submodule is used to calculate the attention weights of different data sources in different time and space dimensions, dynamically adjust the importance of multimodal wireless communication data, and obtain the initial wireless communication data.

[0063] The initial wireless communication data, after being weighted by the attention mechanism, is fused together, and the fused wireless communication data is generated by weighted summation. The weight coefficients are dynamically adjusted according to the attention weights calculated in real time to obtain the fused wireless communication data.

[0064] Step 102: Construct a local DRL model at each base station and train the local DRL model using the improved PPO algorithm;

[0065] Specifically, in this embodiment, the feature data from the fused wireless communication data is used as the state input of the DRL local model, and the dimension of the state space is dynamically adjusted according to the actual collected data features to define the state space of the DRL local model.

[0066] The action space includes the allocation strategy for wireless communication resources. Each action corresponds to a specific resource allocation method, defining the model action space.

[0067] For voice call services with high real-time requirements, increase the weight of transmission latency and packet loss rate in the reward function; for video streaming services with high bandwidth requirements, increase the weight of user throughput.

[0068] Step 103: Periodically aggregate the DRL local model parameters of each base station using the central server, add noise using DP differential privacy technology, and generate a DP-DRL global model.

[0069] Specifically, in this embodiment, the DRL local model parameters of each base station are collected and periodically aggregated, and the aggregation period is set according to the system's operating requirements and network conditions.

[0070] The federated averaging algorithm is used as the model parameter aggregation algorithm to perform a weighted average of the local DRL model parameters of each base station, and the weights are set according to the scale of the base station.

[0071] After the DRL local model parameters are aggregated, Laplacian noise is added to the aggregated DP-DRL global model parameters using DP differential privacy technology.

[0072] By adjusting the privacy budget value when adding noise, a DP-DRL global model is generated to balance the strength of privacy protection and model performance.

[0073] Step 104: Input the fused wireless communication data into the DP-DRL global model and output the initial joint allocation strategy;

[0074] Specifically, in this embodiment, wireless communication data will be fused and input into the DP-DRL global model, and the DP-DRL global model will output the probability distribution of the action space.

[0075] Based on probability distribution, a greedy strategy is used to select the action with the highest probability as the allocation strategy;

[0076] The allocation strategy is converted into a specific wireless communication resource allocation strategy format to obtain the initial joint allocation strategy.

[0077] Step 105: Allocate wireless communication resources based on the initial joint allocation strategy, collect system performance indicators through a sliding window, feed them back to the fuzzy logic controller, adjust the reward function weights, and obtain the target joint allocation strategy.

[0078] Specifically, in this embodiment, wireless communication resources are allocated based on an initial joint allocation strategy, including: setting the size and step size of the sliding window, and collecting statistics including spectrum utilization, average user throughput, QoS satisfaction, energy efficiency, and system fairness indicators.

[0079] Its beneficial effects are as follows: 1. It can adapt to complex and ever-changing network environments through continuous learning and optimization. The local model is trained based on the local characteristics of each base station, while the global model integrates the experience of each base station to achieve global optimization of resource allocation strategies. Compared with traditional methods, this significantly improves the efficiency of resource allocation and system performance, such as increasing spectrum utilization and user throughput. 2. By adding noise to protect user data and model parameters, it ensures the privacy and security of user data during resource allocation, meeting data security regulations. At the same time, by reasonably adjusting the privacy budget, the impact on model performance is controlled within an acceptable range while ensuring privacy, achieving a balance between privacy protection and model performance.

[0080] Please see Figure 2 In an artificial intelligence-based wireless communication resource allocation method, obtaining multimodal wireless communication data by acquiring terminal link status data, user behavior data, and environmental interference signal data from each base station includes the following steps:

[0081] Step 201: Collect signal strength, signal-to-noise ratio, bit error rate and transmission delay parameters between the terminal and the base station in real time to obtain terminal link status data;

[0082] Step 202: By integrating user behavior monitoring software into the terminal device, collect user data on service type, service usage time, data transmission rate requirements, and service quality requirements to obtain user behavior data;

[0083] Step 203: Use spectrum analyzers and electromagnetic interference detectors deployed around the base station to detect electromagnetic interference signals in the surrounding area in real time, including at least co-channel interference, adjacent channel interference, industrial noise and natural noise, to obtain environmental interference signal data.

[0084] Step 204: Integrate the terminal link status data, user behavior data, and environmental interference signal data to obtain multimodal wireless communication data.

[0085] Please see Figure 3 In an artificial intelligence-based wireless communication resource allocation system, the system includes the following modules:

[0086] The data acquisition module is used to acquire terminal link status data, user behavior data and environmental interference signal data from each base station to obtain multimodal wireless communication data. It uses an attention mechanism to dynamically weight the different data sources in the multimodal wireless communication data to obtain fused wireless communication data.

[0087] The local training module is used to build local DRL models at each base station and train the local DRL models using the improved PPO algorithm.

[0088] The federated aggregation module is used to periodically aggregate the DRL local model parameters of each base station using the central server, add noise using DP differential privacy technology, and generate a DP-DRL global model.

[0089] The resource allocation module is used to input the converged wireless communication data into the DP-DRL global model and output the initial joint allocation strategy.

[0090] The online adaptive module is used to allocate wireless communication resources based on the initial joint allocation strategy. It uses a sliding window to collect system performance indicators and feeds them back to the fuzzy logic controller to adjust the weights of the reward function and obtain the target joint allocation strategy.

[0091] Specifically, this embodiment can also be implemented in the following ways:

[0092] Step 1: Multimodal data acquisition and fusion;

[0093] Data Acquisition: Terminal Link Status Data: High-precision sensors and monitoring modules are deployed at each base station to collect real-time link status parameters between the terminal and the base station, such as signal strength, signal-to-noise ratio, bit error rate, and transmission delay. Differentiated data acquisition frequencies and accuracy standards are designed for different types of terminals (such as mobile phones, IoT devices, and vehicle-mounted terminals). For example, for high-speed moving vehicle-mounted terminals, the acquisition frequency of signal strength and transmission delay is increased to 10 times per second to accurately capture dynamic changes in the link.

[0094] User Behavior Data: By integrating user behavior monitoring software into terminal devices, data is collected on user activity types (such as voice calls, video streaming, file transfers, web browsing, etc.), usage times, data transmission rate requirements, and Quality of Service (QoS) requirements. Simultaneously, combined with user location information (obtained through base station positioning or terminal GPS modules), user behavior patterns in different scenarios (such as indoor, outdoor, and high-speed mobile scenarios) are analyzed. For example, users in office areas may heavily utilize file transfers and video conferencing during weekdays, while users in residential areas may concentrate on video streaming and web browsing at night. Environmental Interference Signal Data: Using environmental monitoring equipment deployed around the base station, such as spectrum analyzers and electromagnetic interference detectors, surrounding electromagnetic interference signals are detected in real time, including co-channel interference, adjacent-channel interference, industrial noise, and natural noise. The frequency range, intensity, duration, and occurrence patterns of interference signals are recorded. For example, in some industrial areas, strong industrial noise interference may occur at specific times, affecting wireless communication quality.

[0095] Data Fusion: Attention Mechanism Design: A multimodal attention network is constructed, comprising multiple attention sub-modules corresponding to terminal link status data, user behavior data, and environmental interference signal data, respectively. Each sub-module dynamically adjusts the importance of each data source by calculating the attention weights of different data sources in different time and spatial dimensions. Specifically, for terminal link status data, the attention weights are mainly based on indicators reflecting link quality, such as signal strength and bit error rate; for user behavior data, the weights are based on indicators reflecting user needs, such as service type and QoS requirements; and for environmental interference signal data, the weights are based on indicators reflecting the impact of interference, such as interference intensity and duration.

[0096] Dynamic weighted fusion: Data from various data sources, weighted by an attention mechanism, are fused together, generating fused wireless communication data through a weighted summation. The weighting coefficients are dynamically adjusted based on real-time calculated attention weights to ensure the impact of key data sources is highlighted in different communication scenarios. For example, in scenarios with severe interference, the weight of environmental interference signal data is automatically increased to more accurately reflect the current communication environment.

[0097] Step 2: DRL local model building and training;

[0098] Local Model Construction: State Space Definition: Key features from the fused wireless communication data are used as the state inputs to the DRL local model, including terminal link state parameters (signal strength, signal-to-noise ratio, bit error rate, transmission delay, etc.), user behavior characteristics (service type, data rate requirements, QoS requirements, etc.), environmental interference characteristics (interference intensity, frequency range, etc.), and the current resource usage of the base station (such as available bandwidth, time slot allocation, etc.). The dimensions of the state space are dynamically adjusted according to the actual collected data characteristics to ensure that the model can fully perceive the operating status of the base station. Action Space Definition: The action space includes wireless communication resource allocation strategies, such as spectrum resource allocation (frequency band selection, bandwidth allocation), time slot allocation (time slot division, time slot allocation to different terminals), power control (adjusting the transmit power of the base station and terminals), etc. Each action corresponds to a specific resource allocation method, such as allocating a certain bandwidth of a frequency band to a certain terminal, or allocating a certain time slot to a specific service type. Reward Function Design: The initial reward function comprehensively considers system performance indicators, such as spectrum utilization, user throughput, service quality satisfaction, energy efficiency, etc. Different reward weights are set for different service types and user needs. For example, for voice call services with high real-time requirements, the weight of transmission latency and packet loss rate in the reward function is increased; for video streaming services with high bandwidth requirements, the weight of user throughput is increased.

[0099] Improve PPO algorithm training;

[0100] Policy and Value Network Structure: Deep neural networks are used to construct the policy and value networks. The policy network takes the state space as input and outputs the probability distribution of the action space, determining which resource allocation strategy to choose. The value network takes the state space as input and outputs a value estimate of the current state, used to evaluate the merits of the policy. The network structure uses a multilayer perceptron, with the number of hidden layers and neurons optimized according to the dimensions of the state and action spaces. For example, three hidden layers, each containing 128 neurons, are used to balance the model's expressive power and computational efficiency.

[0101] Training process optimization: Based on the traditional PPO algorithm, an experience replay mechanism and gradient pruning technique are introduced. Experience replay stores the experience samples generated during training in an experience pool, and randomly samples them for training, improving data utilization and training stability; gradient pruning limits the magnitude of gradients to avoid gradient explosion and ensure the convergence of the training process. During training, the parameters of the policy network and value network are continuously updated based on the real-time operating data of the base stations, enabling the model to gradually learn the optimal resource allocation strategy.

[0102] Step 3: Global model aggregation and differential privacy processing;

[0103] Model parameter aggregation;

[0104] Central server architecture: The central server possesses powerful computing and storage capabilities, responsible for collecting DRL local model parameters from each base station and performing periodic aggregation. The aggregation cycle is set according to the system's operational needs and network conditions, for example, aggregation every 10 minutes, ensuring that the global model can promptly reflect the local optimal strategies of each base station.

[0105] Aggregation Algorithm Design: The FedAvg algorithm is used as the basic algorithm for aggregating model parameters, which performs a weighted average of the local DRL model parameters of each base station. The weights can be set according to the scale of the base station (such as the number of terminals in the coverage area, traffic volume, etc.). For example, base stations with high traffic volume are given higher weights during aggregation to highlight the impact of their local strategies.

[0106] The noise addition method for differential privacy processing involves adding Laplacian or Gaussian noise to the aggregated DP-DRL global model parameters after DRL local model parameter aggregation. The noise intensity is adjusted based on the privacy budget ϵ and sensitivity Δ to ensure that the differential privacy definition ϵ is satisfied. For example, for each model parameter, Gaussian noise with a mean of 0 and a standard deviation of σ is added.

[0107] Balancing Privacy Protection and Model Performance: When adding noise, the privacy budget ϵ is adjusted to balance the strength of privacy protection and model performance. A smaller ϵ value provides stronger privacy protection but may have some impact on model performance; a larger ϵ value has the opposite effect. A suitable ϵ value, such as ϵ=0.5, is determined through experiments and simulations.

[0108] Step 4: Generate the initial joint allocation strategy;

[0109] Model input processing;

[0110] Data preprocessing: The fused wireless communication data obtained in step 1 is preprocessed, including data normalization and standardization, to meet the input requirements of the DP-DRL global model. The normalization method uses min-max normalization to map the data to the [0,1] interval; the standardization method uses Z-score standardization to give the data zero mean and unit variance. The preprocessed data is used as the model input to ensure that the model can correctly identify and process features of different dimensions.

[0111] Input Dimension Adaptation: Based on the input layer structure of the DP-DRL global model, the dimensions of the fused wireless communication data are adjusted to ensure that the data can be successfully input into the model. If the model input layer requires a fixed-dimensional vector, the multimodal data is converted into a fixed-dimensional input vector through feature selection or feature fusion.

[0112] Policy output parsing;

[0113] Action probability distribution processing: The DP-DRL global model outputs the probability distribution of the action space, and the optimal initial joint assignment strategy is selected based on the probability distribution. A greedy strategy can be used, selecting the action with the highest probability as the assignment strategy; alternatively, a random strategy can be used, randomly selecting actions based on the probability distribution to increase strategy diversity. In practical applications, the appropriate strategy selection method should be chosen based on the system's real-time requirements and stability needs.

[0114] Policy format conversion: Convert the actions output by the model into specific wireless communication resource allocation policy formats, including spectrum allocation schemes, time slot allocation schemes, power control schemes, etc., to ensure that each base station can accurately understand and execute the policy.

[0115] Step 5: Strategy adjustment and joint target allocation strategy generation;

[0116] System performance statistics;

[0117] Sliding window settings: Set the size and step size of the sliding window, for example, a window size of 100 time units and a step size of 10 time units, used to collect system performance metrics within the window. Metrics to be collected include spectrum utilization, average user throughput, QoS satisfaction (such as latency compliance rate, packet loss rate compliance rate, etc.), energy efficiency, and system fairness metrics (such as variance of user throughput, etc.).

[0118] Real-time statistics and storage: The real-time data processing module performs real-time statistics and storage on the performance indicators within the sliding window, providing accurate input data for the fuzzy logic controller.

[0119] Fuzzy logic controller design;

[0120] Input and output variable definitions: The input of the fuzzy logic controller is the deviation and rate of change of the system performance indicators statistically obtained by the sliding window from the preset target values, such as the deviation of spectrum utilization, the deviation of average user throughput and its rate of change, etc.; the output is the adjustment of the reward function weights, including the adjustment of spectrum utilization weight, user throughput weight, QoS satisfaction weight, energy efficiency weight, etc.

[0121] Fuzzy rule design: A fuzzy rule base is established, and the fuzzy values ​​of the output variables are determined based on the fuzzification results of the input variables. For example, when the spectral utilization deviation is "high" and the rate of change of deviation is "positive," the adjustment amount of the spectral utilization weight is increased to guide the model to focus more on improving spectral utilization. The fuzzy rules adopt an "if-then" form and are optimized through expert experience and experimental data.

[0122] Fuzzy reasoning and defuzzification: The Mamdani fuzzy reasoning method is used for fuzzy reasoning, and the centroid method is used for defuzzification to obtain the specific reward function weight adjustment.

[0123] Reward function weight adjustment and strategy update;

[0124] Weight adjustment strategy: Based on the weight adjustment amount output by the fuzzy logic controller, the weight coefficients of each indicator in the reward function are dynamically adjusted. The adjustment process follows the principle of smooth adjustment to avoid adverse effects on model training caused by drastic weight changes. For example, the adjustment magnitude does not exceed 10% of the initial weights.

[0125] Target joint allocation strategy generation: The adjusted reward function is fed back to the DRL local model in step 2, the model is retrained, and a new joint allocation strategy is generated through steps 3 and 4, forming a closed-loop optimization process, and finally obtaining a target joint allocation strategy that adapts to the current system state.

[0126] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based wireless communication resource allocation method, characterized by, The wireless communication resource allocation method based on artificial intelligence comprises the following steps: Obtain terminal link state data, user behavior data and environmental interference signal data in each base station to obtain multi-modal wireless communication data, dynamically weight different data sources in the multi-modal wireless communication data using an attention mechanism to obtain fused wireless communication data; Construct a DRL local model in each base station, introduce an experience replay mechanism and gradient clipping technology on the basis of a traditional PPO algorithm to obtain an improved PPO algorithm, and use the improved PPO algorithm to train the DRL local model; Periodically aggregate DRL local model parameters of each base station using a central server, add Laplace noise to the aggregated DP-DRL global model parameters using a DP differential privacy technology, balance the privacy protection strength and model performance by adjusting the value of the privacy budget when adding noise, and generate a DP-DRL global model; Input the fused wireless communication data into the DP-DRL global model to output an initial joint allocation strategy; Allocate wireless communication resources based on the initial joint allocation strategy, statistically analyze system performance indicators through a sliding window, feed back to a fuzzy logic controller, adjust the weight of the reward function, and obtain a target joint allocation strategy.

2. The wireless communication resource allocation method based on artificial intelligence according to claim 1, wherein, The terminal link state data, user behavior data and environmental interference signal data in each base station are obtained to obtain multi-modal wireless communication data, which comprises: Real-time collect signal strength, signal-to-noise ratio, bit error rate and transmission delay parameters between the terminal and the base station to obtain terminal link state data; Collect user's business type, business use period, data transmission rate demand and service quality requirement data by integrating user behavior monitoring software on the terminal device to obtain user behavior data; Real-time detect the surrounding electromagnetic interference signals, including at least co-frequency interference, adjacent frequency interference, industrial noise and natural noise, using a spectrum analyzer and an electromagnetic interference detector deployed around the base station to obtain environmental interference signal data; Integrate the terminal link state data, user behavior data and environmental interference signal data to obtain multi-modal wireless communication data.

3. The method of claim 1, wherein the AI-based wireless communication resource allocation method is characterized by, The different data sources in the multi-modal wireless communication data are dynamically weighted using an attention mechanism to obtain fused wireless communication data, which comprises: Based on an LSTM network, a multi-modal attention network is constructed, which contains multiple attention sub-modules corresponding to terminal link state data, user behavior data and environmental interference signal data respectively; Calculate the attention weights of different data sources in different time and space dimensions using the attention sub-modules to dynamically adjust the importance of multi-modal wireless communication data to obtain initial wireless communication data; Fuse the initial wireless communication data weighted by the attention mechanism to generate fused wireless communication data using weighted summation, and dynamically adjust the weight coefficient according to the real-time calculated attention weights to obtain fused wireless communication data.

4. The method of claim 1, wherein the AI-based wireless communication resource allocation method is characterized by, The DRL local model is constructed in each base station, and the improved PPO algorithm is used to train the DRL local model, which comprises: The feature data in the fusion wireless communication data is taken as the state input of the DRL local model, and the dimension of the state space is dynamically adjusted according to the actually collected data features, and the state space of the DRL local model is defined; The action space includes a wireless communication resource allocation strategy, and each action corresponds to a specific resource allocation method, and the model action space is defined; The transmission delay and the packet loss rate are increased in the reward function for the voice call service with high real-time requirement; and the user throughput weight is increased for the video stream service with large bandwidth demand.

5. The wireless communication resource allocation method based on artificial intelligence as described in claim 1, characterized in that, The DRL local model parameters of each base station are periodically aggregated by the central server, DP differential privacy technology is used to add noise, and a DP-DRL global model is generated, including: The DRL local model parameters of each base station are collected and periodically aggregated, and the aggregation period is set according to the running requirements and network conditions of the system; The federated average algorithm is used as the model parameter aggregation algorithm, and the DRL local model parameters of each base station are weighted and averaged, and the weight is set according to the size of the base station; After the DRL local model parameters are aggregated, the DP differential privacy technology is used to add Laplace noise to the aggregated DP-DRL global model parameters; The value of the privacy budget is adjusted to balance the privacy protection strength and the model performance when adding noise, and the DP-DRL global model is generated.

6. The method of claim 1, wherein the wireless communication resource is allocated based on artificial intelligence. The fusion wireless communication data is input into the DP-DRL global model, and an initial joint allocation strategy is output, including: The fusion wireless communication data is input into the DP-DRL global model, and the DP-DRL global model outputs the probability distribution of the action space; Based on the probability distribution, the action with the maximum probability is selected as the allocation strategy based on the greedy strategy; The allocation strategy is converted into a specific wireless communication resource allocation strategy format to obtain the initial joint allocation strategy.

7. The method of claim 1, wherein the wireless communication resource is allocated based on artificial intelligence. Based on the initial joint allocation strategy, the wireless communication resources are allocated, and the system performance indicators are counted through a sliding window, including: Based on the initial joint allocation strategy, the wireless communication resources are allocated, including setting the size and sliding step of the sliding window, and collecting and counting the spectrum utilization rate, average user throughput, QoS satisfaction, energy efficiency and system fairness indicators.

8. An artificial intelligence-based wireless communication resource allocation system, characterized by, The wireless communication resource allocation system based on artificial intelligence includes the following modules: A data acquisition module is used to obtain terminal link state data, user behavior data and environmental interference signal data in each base station, obtain multi-modal wireless communication data, dynamically weight different data sources in the multi-modal wireless communication data by using an attention mechanism, and obtain fusion wireless communication data; A local training module is used to construct a DRL local model in each base station, introduce an experience replay mechanism and a gradient clipping technology based on a traditional PPO algorithm, obtain an improved PPO algorithm, and train the DRL local model by using the improved PPO algorithm; The federal aggregation module is configured to periodically aggregate DRL local model parameters of each base station by using a central server, add Laplace noise to DP-DRL global model parameters after aggregation by using a DP differential privacy technology, and balance privacy protection strength and model performance by adjusting a value of a privacy budget when adding the noise to generate a DP-DRL global model. The resource allocation module is configured to input the fused wireless communication data into the DP-DRL global model and output an initial joint allocation strategy. The online adaptive module is configured to allocate wireless communication resources based on the initial joint allocation strategy, count system performance indicators by using a sliding window, feed back to a fuzzy logic controller, adjust a reward function weight, and obtain a target joint allocation strategy.

9. The wireless communication resource allocation system based on artificial intelligence of claim 8, wherein, The data collection module further includes the following sub-modules: The construction sub-module is configured to construct a multi-modal attention network based on an LSTM network, and the multi-modal attention network includes multiple attention sub-modules corresponding to terminal link state data, user behavior data, and environmental interference signal data. The calculation sub-module is configured to calculate attention weights of different data sources in different time and space dimensions by using the attention sub-modules, dynamically adjust importance of multi-modal wireless communication data, and obtain initial wireless communication data. The fusion sub-module is configured to fuse the initial wireless communication data weighted by the attention mechanism, generate fused wireless communication data by using a weighted summation method, dynamically adjust a weight coefficient according to the real-time calculated attention weights, and obtain the fused wireless communication data.

10. The wireless communication resource allocation system based on artificial intelligence of claim 8, wherein, The resource allocation module further includes the following sub-modules: The input sub-module is configured to input the fused wireless communication data into the DP-DRL global model, and the DP-DRL global model outputs a probability distribution of an action space. The selection sub-module is configured to select an action with the maximum probability as an allocation strategy by using a greedy strategy based on the probability distribution. The obtaining sub-module is configured to convert the allocation strategy into a specific wireless communication resource allocation strategy format to obtain the initial joint allocation strategy. ​

Images