Wireless resource allocation methods, apparatus, computer equipment and readable storage media

Abstract

This application provides a wireless resource allocation method, apparatus, computer device, and readable storage medium. The method includes: acquiring channel state data between multiple target base stations and multiple target terminals; inputting the channel state data into a model to obtain a corresponding wireless resource allocation result; a first model is obtained by adjusting a pre-trained second model according to target model parameters; the target model parameters are the values ​​of model parameter variables that minimize the total loss value obtained by the second model after allocating wireless resources for multiple sample channel state data, determined in the sample null space by combining an objective function; the objective function includes a parameter optimization function that includes the functional relationship between the total loss variable and the model parameter variables while minimizing the total loss value as the optimization objective, and a constraint function used to limit the equality relationship between the first loss and the second loss; thereby improving the real-time performance and accuracy of wireless resource allocation.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a wireless resource allocation method, apparatus, computer equipment, and readable storage medium. Background Technology

[0002] Wireless resource allocation refers to the process of rationally scheduling and optimizing the allocation of limited resources such as spectrum, power, and time in a wireless communication system to maximize network performance and user experience. Facing dynamically changing wireless environments and user demands, efficient resource allocation can improve data transmission rates, expand network capacity, and ensure service quality and fairness, making it one of the key technologies for modern mobile communication network optimization. With the widespread adoption of mobile devices and the surge in data traffic, efficient wireless resource allocation is crucial for ensuring network stability and improving user experience.

[0003] However, resource allocation faces numerous challenges in complex wireless communication environments. First, spectrum resources are limited and need to be balanced among numerous users and applications. Second, user needs and network conditions are constantly changing, requiring resource allocation to adapt quickly to these changes. In addition, factors such as network energy consumption, coverage, and quality of service must also be taken into account.

[0004] In related technologies, to optimize resource allocation and improve overall system performance, power control and channel allocation are generally optimized using medium- to long-term network measurement information. Specifically, the network environment can be continuously monitored, and data on user behavior, traffic patterns, and channel conditions can be collected. This data is then used to analyze network performance, and finally, based on this rich measurement data, a more accurate model is built to predict future network conditions, thereby enabling more intelligent and dynamic resource management decisions. However, this traditional method has high computational complexity and struggles to adapt to rapidly changing network environments, resulting in low real-time performance and accuracy in wireless resource allocation. Summary of the Invention

[0005] The main objective of this application is to provide a wireless resource allocation method, apparatus, computer device, and readable storage medium that can improve the real-time performance and accuracy of wireless resource allocation.

[0006] To achieve the above objectives, a first aspect of this application proposes a wireless resource allocation method, the method comprising:

[0007] Acquire channel state data between multiple target base stations and multiple target terminals;

[0008] The channel state data is input into a pre-trained first model to obtain the wireless resource allocation results of the multiple target base stations for the multiple target terminals;

[0009] Wherein, the first model is obtained by adjusting the pre-trained second model according to the target model parameters; the target model parameters are the values ​​of the model parameter variables that are determined in the null space of the samples when the second model minimizes the total loss value after allocating wireless resources for multiple sample channel state data, in combination with the objective function.

[0010] The second model, obtained by adjusting parameters according to historical stages, is determined after processing multiple historical sample channel state data to obtain an input data matrix. The objective function includes a parameter optimization function and a constraint function.

[0011] The parameter optimization function, while aiming to minimize the total loss value, includes the functional relationship between the total loss variable and the model parameter variable.

[0012] The constraint function is used to define the equality relationship between the first loss and the second loss; the first loss is calculated based on the intermediate model corresponding to the current value of the model parameter variable, and after allocating radio resources for multiple historical sample channel state data at different historical stages; the second loss is calculated based on the second model corresponding to the historical model parameters, and after allocating radio resources for the multiple historical sample channel state data.

[0013] Accordingly, a second aspect of the embodiments of this application provides a wireless resource allocation apparatus, the apparatus comprising:

[0014] The acquisition module is used to acquire channel state data between multiple target base stations and multiple target terminals;

[0015] An input module is used to input the channel state data into a pre-trained first model to obtain the wireless resource allocation results of the multiple target base stations for the multiple target terminals. The first model is obtained by adjusting a pre-trained second model according to target model parameters. The target model parameters are the values ​​of model parameter variables that minimize the total loss value obtained by the second model after allocating wireless resources for multiple sample channel state data, determined in the sample null space by combining an objective function. The sample null space is determined by processing multiple historical sample channel state data to obtain an input data matrix after the second model, obtained by parameter adjustment based on historical stages, has been processed. The objective function includes a parameter optimization function and a constraint function. The parameter optimization function, while minimizing the total loss value as the optimization objective, includes the functional relationship between the total loss variable and the model parameter variables. The constraint function is used to limit the equality relationship between the first loss and the second loss. The first loss is calculated based on the intermediate model corresponding to the current value of the model parameter variables, after allocating wireless resources for multiple historical sample channel state data at different historical stages. The second loss is calculated based on the second model corresponding to the historical model parameters, after allocating wireless resources for the multiple historical sample channel state data.

[0016] In some embodiments, the wireless resource allocation device further includes a training module for:

[0017] Obtain the sample training set corresponding to the current stage, wherein the sample training set includes multiple sample channel state data;

[0018] The optimization objective is to minimize the total loss value, and a parameter optimization function is constructed based on the functional relationship between the total loss variable and the model parameter variable. The total loss variable is obtained by summing the losses of multiple current samples after allocating radio resources for multiple sample channel state data based on the intermediate model corresponding to the current value of the model parameter variable.

[0019] Based on the intermediate model corresponding to the current values ​​of the model parameter variables, the first loss is calculated after allocating radio resources to multiple historical sample channel state data from different historical stages.

[0020] Based on the second model corresponding to the historical model parameters, after allocating radio resources to the multiple historical sample channel state data, the second loss is calculated.

[0021] Based on the equality relationship between the first loss and the second loss, a constraint function is constructed;

[0022] Based on the constraint function and the parameter optimization function, construct the objective function;

[0023] Based on the objective function, the target model parameters are determined in the sample null space to minimize the total loss value obtained by the second model after allocating radio resources for the multiple sample channel state data.

[0024] The second model is adjusted based on the target model parameters to obtain the first model.

[0025] In some implementations, the training module is further configured to:

[0026] Obtain at least one resource allocation target for wireless resource allocation in the current stage; wherein, the resource allocation target includes data rate, communication latency rate, and allocation balance rate;

[0027] Using the intermediate model corresponding to the current model parameter variable values, radio resource allocation is performed sequentially on each sample channel state data to obtain the first radio resource allocation result.

[0028] Based on the at least one resource allocation objective and in conjunction with the first wireless resource allocation result, at least one sub-loss is calculated after wireless resource allocation for each sample channel state data.

[0029] Based on the sum of the at least one sub-loss, the current sample loss corresponding to the value of the current model parameter variable is obtained when the intermediate model processes the channel state data of each sample.

[0030] In some implementations, the training module is further configured to:

[0031] The second model processes multiple historical channel state data in each network layer to obtain the input data matrix corresponding to each network layer.

[0032] In each network layer, the input data matrix is ​​approximated by a low rank to obtain an updated input data matrix;

[0033] Based on the updated input data matrix, the null space of the samples corresponding to each network layer is determined.

[0034] In some implementations, the training module is further configured to:

[0035] Using the second model, the multiple historical channel state data are processed according to the historical model parameters corresponding to each network layer to obtain the input data matrix corresponding to each network layer;

[0036] Repeat the process in the next network layer of the second model, processing the multiple historical channel state data according to the next historical model parameters corresponding to the next network layer to obtain the next input data matrix corresponding to the next network layer, until the input data matrix of the last network layer of the intermediate model is obtained.

[0037] In some implementations, the training module is further configured to:

[0038] In each network layer, singular value decomposition is performed on the corresponding input data matrix to obtain the left singular vector matrix, the singular value diagonal matrix, and the transpose of the right singular vector matrix of the input data matrix.

[0039] Obtain a preset extraction value, and extract the same number of matrix columns as the preset extraction value from the left singular vector matrix to obtain an updated left singular vector matrix;

[0040] Extract the same number of singular values ​​as the preset extraction value from the singular value diagonal matrix to obtain an updated singular value diagonal matrix;

[0041] Extract the same number of matrix rows as the preset extraction value from the transpose of the right singular vector matrix to obtain the transpose of the updated right singular vector matrix;

[0042] The updated input data matrix is ​​obtained based on the updated left singular vector matrix, the updated singular value diagonal matrix, and the transpose of the updated right singular vector matrix.

[0043] In some embodiments, the wireless resource allocation device further includes a disabling module for:

[0044] Obtain multiple scaling parameters corresponding to multiple network layers of the second model;

[0045] Obtain the absolute values ​​of the multiple scaling parameters corresponding to the multiple scaling parameters, and determine multiple target scaling parameters based on the order of the magnitude of the absolute values ​​of the multiple scaling parameters;

[0046] Identify multiple target neurons corresponding to the multiple target scaling parameters, and temporarily disable the corresponding multiple target neurons in the multiple network layers.

[0047] Accordingly, a third aspect of the present application provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the wireless resource allocation method described in any one of the embodiments of the first aspect of the present application.

[0048] Accordingly, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the wireless resource allocation method described in any one of the embodiments of the second aspect of the present application.

[0049] This application embodiment acquires channel state data between multiple target base stations and multiple target terminals; inputs the channel state data into a pre-trained first model to obtain the radio resource allocation results of multiple target base stations to multiple target terminals; wherein, the first model is obtained by adjusting a pre-trained second model according to target model parameters; the target model parameters are the values ​​of model parameter variables that minimize the total loss value obtained by the second model in the sample null space after performing radio resource allocation on multiple sample channel state data, combined with the objective function; the sample null space is determined by the second model obtained by adjusting parameters according to historical stages and processing multiple historical sample channel state data to obtain an input data matrix; the objective function includes a parameter optimization function and a constraint function; the parameter optimization function, while minimizing the total loss value as the optimization objective, includes the functional relationship between the total loss variable and the model parameter variable; the constraint function is used to limit the equality relationship between the first loss and the second loss; the first loss is calculated based on the intermediate model corresponding to the current value of the model parameter variable, after performing radio resource allocation on multiple historical sample channel state data at different historical stages; the second loss is calculated based on the second model corresponding to the historical model parameters, after performing radio resource allocation on the multiple historical sample channel state data. This approach fully leverages the periodicity and repetitiveness of channel distribution. Based on multiple sample channel state data from the current stage, a pre-trained and well-performing second model is rapidly trained, eliminating the need for frequent learning of already adapted distributions and improving training efficiency and adaptability. Furthermore, the second model, with parameters adjusted based on historical stages, processes multiple historical channel state data to obtain an input data matrix, determining the sample null space. Optimization within this null space at the current stage allows for a smooth transition between stages, avoiding catastrophic forgetting of historically learned knowledge. This enables the model to efficiently adapt to dynamically changing wireless environments while retaining historical knowledge, improving resource allocation accuracy. Simultaneously, optimizing model parameters by constructing an objective function allows the model to consider both long-term and short-term resource allocation goals, resulting in more precise allocation that better meets network performance and user experience requirements. Compared to traditional methods, this approach does not rely on complex numerical optimization techniques but utilizes the powerful learning capabilities and parallel computing characteristics of deep learning. This enables the model to allocate channel resources quickly and accurately, improving the real-time performance and accuracy of resource allocation. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the architecture of the wireless resource allocation system provided in the embodiments of this application;

[0051] Figure 2 This is a flowchart of the wireless resource allocation method provided in the embodiments of this application;

[0052] Figure 3 This is a flowchart of the model training process provided in the embodiments of this application;

[0053] Figure 4 This is a diagram showing the channel distribution changes at different stages, as provided in the embodiments of this application.

[0054] Figure 5 This is a graph showing the performance comparison between the method of this application and other learning methods provided in the embodiments of this application;

[0055] Figure 6 This is a graph showing the comparison of training time between the method of this application and other learning methods provided in the embodiments of this application;

[0056] Figure 7 This is a schematic diagram of the functional modules of the wireless resource allocation device provided in the embodiments of this application;

[0057] Figure 8 This is a schematic diagram of the hardware structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0061] Wireless resource allocation refers to the process of rationally scheduling and optimizing the allocation of limited resources such as spectrum, power, and time in a wireless communication system to maximize network performance and user experience. Facing dynamically changing wireless environments and user demands, efficient resource allocation can improve data transmission rates, expand network capacity, and ensure service quality and fairness, making it one of the key technologies for modern mobile communication network optimization. With the widespread adoption of mobile devices and the surge in data traffic, efficient wireless resource allocation is crucial for ensuring network stability and improving user experience.

[0062] However, resource allocation faces numerous challenges in complex wireless communication environments. First, spectrum resources are limited and need to be balanced among numerous users and applications. Second, user needs and network conditions are constantly changing, requiring resource allocation to adapt quickly to these changes. In addition, factors such as network energy consumption, coverage, and quality of service must also be taken into account.

[0063] In related technologies, to optimize resource allocation and improve overall system performance, power control and channel allocation are generally optimized using medium- to long-term network measurement information. Specifically, the network environment can be continuously monitored, and data on user behavior, traffic patterns, and channel conditions can be collected. This data is then used to analyze network performance, and finally, based on this rich measurement data, a more accurate model is built to predict future network conditions, thereby enabling more intelligent and dynamic resource management decisions. However, this traditional method has high computational complexity and struggles to adapt to rapidly changing network environments, resulting in low real-time performance and accuracy in wireless resource allocation.

[0064] Based on this, embodiments of this application provide a wireless resource allocation method, apparatus, computer device, and readable storage medium, which can improve the real-time performance and accuracy of resource allocation.

[0065] The wireless resource allocation method, apparatus, computer equipment, and readable storage medium provided in the embodiments of this application are specifically described through the following embodiments. First, the wireless resource allocation system in the embodiments of this application is described.

[0066] Please refer to Figure 1 In some embodiments, this application provides a wireless resource allocation system, including a terminal 11 and a server 12.

[0067] For example, terminal 11 can be a smartphone, tablet, IoT device, etc., and server 12 can be a base station, edge computing server, cloud computing center server, etc.

[0068] In some implementations, terminal 11 may include a communication module, a data processing module, and a resource allocation request module. Specifically, the communication module can be responsible for wireless communication with the base station, receiving downlink signals from the base station, such as Channel State Information (CSI) and resource allocation instructions; and simultaneously sending uplink signals, such as its own service demand information and feedback information. For example, in a 5G communication system, the communication module of terminal 11 supports New Radio (NR) technology, enabling high-speed, low-latency data transmission.

[0069] Furthermore, the data processing module can preprocess the received channel state data, such as through channel estimation and noise cancellation, to improve the accuracy and reliability of the data. Simultaneously, the data processing module can generate corresponding resource requirement information, such as required bandwidth and latency parameters, based on its own service needs, such as video playback and file download.

[0070] Furthermore, the resource allocation request module can integrate the processed channel status data with its own resource requirement information to form a resource allocation request, and send it to the server 12 through the communication module.

[0071] In some implementations, server 12 may be a computer device and may include a data receiving and parsing module, a resource allocation model (such as a first model, a second model, etc.), a resource allocation decision module, an instruction generation and sending module, etc.

[0072] Specifically, the data receiving and parsing module can receive resource allocation requests from multiple terminals 11, and parse the channel state data and resource requirement information in the requests to extract key parameters. For example, it can parse out information such as the Channel Quality Indicator (CQI) and the required service type (such as voice, data, etc.) for each terminal 11, providing a basis for subsequent resource allocation.

[0073] For example, the resource allocation decision module can be used to allocate wireless resources based on the parsed channel state data and resource demand information. This model can be a deep learning model based on continuous learning, which avoids catastrophic forgetting by optimizing in the sample null space, enabling it to quickly adapt to dynamically changing wireless environments while retaining the memory of previously learned knowledge, thus improving the accuracy and efficiency of resource allocation.

[0074] Furthermore, the resource allocation decision module can generate specific resource allocation decisions based on the model's output, such as the spectrum resources, power resources, and time resources allocated to each terminal 11. Then, the instruction generation and transmission module can convert the resource allocation decisions into specific instructions, such as the frequency band, power level, and time slice information for resource allocation, and send them to the corresponding terminal 11 via the communication module. Simultaneously, control instructions can also be sent as needed, such as adjusting the transmit power of terminal 11 or switching frequency bands, to optimize network performance.

[0075] The wireless resource allocation system enables efficient and accurate allocation of wireless resources, thereby improving the overall performance of wireless communication networks and the user experience.

[0076] The wireless resource allocation method in this application can be illustrated through the following embodiments.

[0077] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent will be obtained first. Furthermore, the collection, use, and processing of this data will comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user will be obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent will the necessary user-related data for the normal operation of the embodiments of this application be obtained.

[0078] In this embodiment, the description will focus on the wireless resource allocation device, which can be integrated into a computer device. See also... Figure 2 , Figure 2 This is a flowchart illustrating the steps of a wireless resource allocation method provided in this application embodiment. Taking the wireless resource allocation device specifically integrated into a terminal or server as an example, the specific process when the processor on the terminal or server executes the program instructions corresponding to the wireless resource allocation method is as follows:

[0079] Step 101: Obtain channel state data between multiple target base stations and multiple target terminals.

[0080] In some implementations, to ensure that the model can learn and predict based on the latest and most accurate channel state data, channel state data between the target base station and multiple target terminals can be obtained to achieve efficient, intelligent and adaptive wireless resource allocation.

[0081] In this context, the target base station can be a fixed device in a wireless communication network, responsible for transmitting and receiving wireless signals with user terminals within a certain coverage area. The target base station is the infrastructure of the wireless network and is typically installed at high locations (such as rooftops or towers) to ensure wide signal coverage.

[0082] The target terminal can be a device that the user uses directly to access the wireless communication network and perform operations such as voice calls, data transmission, and Internet access. The target terminal can communicate with the base station through its built-in wireless communication module (such as Wi-Fi, 4G / 5G, etc.) to send and receive wireless signals to the target base station.

[0083] The channel state data can be the optimization variables for resource allocation and the channel data of the current stage, so that the first model can adjust the resource allocation scheme according to the channel characteristics of the current stage.

[0084] In some implementations, the channel state data includes not only channel data but also optimization variables for resource allocation, such as parameters like spectrum, power, and time allocation, which are the targets of the first model's optimization. The specific data categories included in the channel state data can be adjusted according to actual circumstances, and this application does not impose specific limitations on this.

[0085] For example, the optimization variables for resource allocation can be determined by the target base station based on the network design and operation strategies, such as spectrum resource allocation (frequency selection), power control (transmit power adjustment), and time allocation (time slot allocation). Furthermore, the target terminal (such as a mobile phone or tablet) can measure the channel quality with the base station through its built-in wireless communication module. The measurement results include signal-to-noise ratio, channel gain, and signal strength, and the measured channel state data is fed back to the target base station via the uplink. The system can centrally collect and process channel state data from different target base stations and target terminals to provide decision support for resource allocation.

[0086] By acquiring channel state data, we can gain a more accurate understanding of the channel distribution across the entire network, which facilitates subsequent allocation of wireless resources using the first model, thereby optimizing network performance and user experience.

[0087] Step 102: Input the channel state data into the pre-trained first model to obtain the wireless resource allocation results of multiple target base stations to multiple target terminals;

[0088] The first model is obtained by adjusting the pre-trained second model according to the target model parameters; the target model parameters are the values ​​of the model parameter variables that are determined in the null space of the samples when the second model allocates wireless resources for multiple sample channel state data to minimize the total loss value, in combination with the objective function.

[0089] The second model, obtained by adjusting parameters based on historical stages, is determined after processing multiple historical sample channel state data to obtain the input data matrix. The objective function includes a parameter optimization function and a constraint function.

[0090] The parameter optimization function, while aiming to minimize the total loss value, also includes the functional relationship between the total loss variable and the model parameter variables.

[0091] The constraint function is used to limit the equality relationship between the first loss and the second loss; the first loss is calculated based on the intermediate model corresponding to the current model parameter variable value, and is obtained after allocating radio resources for multiple historical sample channel state data from different historical stages; the second loss is calculated based on the second model corresponding to the historical model parameter, and is obtained after allocating radio resources for multiple historical sample channel state data.

[0092] In some implementations, in order to enable the first model to make intelligent decisions based on real-time channel conditions, channel state data can be input into the pre-trained first model for the allocation of wireless resources, thereby improving the overall network performance and user experience.

[0093] The first model can be a model that is trained on the channel distribution of the current stage by the second model, and has the ability to allocate channel state data of different historical stages and the current stage well.

[0094] The wireless resource allocation result can be the output of the first model, which is used to indicate the specific allocation amount of wireless resources (such as frequency, power, and time) among multiple target base stations and multiple target terminals.

[0095] The second model can be the base model of the first model. The second model has been optimized in terms of parameters in the historical stage and has the ability to efficiently and accurately allocate historical channel state data of all historical stages. By training the second model with sample channel state data of the current stage, the first model adapted to the channel state of the current stage can be obtained.

[0096] The target model parameters can be the model parameters obtained after training on the sample channel state data of the current stage during the continuous learning process. The target model parameters can enable the first model to achieve the optimal resource allocation effect when processing the data of the current stage, while retaining the memory of the data of the historical stage.

[0097] The objective function can be used to guide the optimization process of the model, including the parameter optimization function and the constraint function.

[0098] The sample null space can be the null space corresponding to the weight matrix of each network layer of the second model after training. By restricting the update operation of the model parameters in the current stage to the sample null space, the second model can be adjusted only in the direction that will not change the prediction results of historical data, thus effectively avoiding the catastrophic forgetting problem. That is, the model will not lose its memory of the data in the previous stage when learning the data of the new stage.

[0099] Among them, the sample channel state data can be used to train the second model. Both the sample channel state data and the channel state data correspond to the current stage, that is, the channel distribution of the two is the same. This allows the second model to adjust its model parameters based on the sample channel state data and then have the ability to allocate wireless resources efficiently and accurately according to the channel state data.

[0100] The total loss value can be a metric used by users to measure the overall performance of the model throughout the training process, and is used to quantify the model's ability to allocate wireless resources.

[0101] The values ​​of the model parameter variables can be the specific values ​​of the adjustable parameters in the intermediate model. During the model training process, the values ​​of the model parameter variables can be adjusted in real time to enable the intermediate model to achieve better training results.

[0102] Among them, historical channel state data can be the data from the previous historical stage before the current stage, which was used to train the second model and then input into the second model for radio resource allocation.

[0103] The input data matrix can be a matrix composed of all the input data during the forward propagation process in each network layer of the second model, with each network layer corresponding to one input data matrix.

[0104] The parameter optimization function can be an optimization problem, with the goal of continuously optimizing the values ​​of the model parameter variables under the constraints of the constraint function, and minimizing the total loss value corresponding to all sample channel state data as much as possible.

[0105] The constraint function can be a constraint condition defined during the optimization of the parameter optimization function. It is used to limit the first loss obtained by the intermediate model from processing the sample channel state data of each historical stage when training the second model in the current stage to be equal to the second loss obtained by the second model from processing the corresponding historical sample channel state data of each historical stage. This ensures that while the intermediate model is learning new data, it can still make accurate and efficient wireless resource allocation based on the historical sample channel state data, and will not forget old knowledge.

[0106] The total loss variable can be a measure representing the loss of the entire model. The total loss variable can be obtained by adding up the losses of multiple current samples obtained by the intermediate model predicting multiple sample channel state data.

[0107] The model parameter variables can be the weights and biases in the intermediate model. The model parameter variables are continuously adjusted during training to minimize the loss function while satisfying the constraint function.

[0108] The first loss can be the intermediate model corresponding to the values ​​of the model parameter variables in the current training state, and the loss after allocating radio resources to the historical sample channel state data corresponding to each historical stage.

[0109] The second loss can be the loss after allocating radio resources to the channel state data of the historical samples corresponding to each historical stage, based on the second model trained in the previous historical stage of the current stage.

[0110] The intermediate model can refer to a version of the model using the current configuration of model parameter variables, which is located in the process of transforming from the second model to the first model. The intermediate model does not refer to a fixed model. Any model involved in the process of adjusting the model parameter variables of the second model to obtain the first model can be called an intermediate model.

[0111] The historical sample channel state data can be the sample channel state data used to train the model in the historical stage corresponding to the second model and in the historical stage before the second model. Each historical stage corresponds to a different model and different historical sample channel state data.

[0112] The historical model parameters can be the parameter configurations of the second model at the corresponding historical stage.

[0113] For example, suppose there is a wireless communication system that includes 3 target base stations (base station 1, base station 2, base station 3) and 6 target terminals (terminal A, terminal B, terminal C, terminal D, terminal E, terminal F). These base stations and terminals are distributed in a certain geographical area, and each base station is responsible for providing wireless communication services to terminals within a certain range.

[0114] In some implementations, channel state data includes not only channel gain, noise level, and interference, but also optimization variables for resource allocation, such as parameters like spectrum, power, and time allocation. This data can be obtained through channel estimation and network management systems.

[0115] For example, if the channel state data includes the following information:

[0116] The channel gain from base station 1 to terminal A is 0.8, the noise level is 0.1, and the interference level is 0.2; the channel gain from base station 1 to terminal B is 0.7, the noise level is 0.2, and the interference level is 0.1; the channel gain from base station 2 to terminal C is 0.9, the noise level is 0.1, and the interference level is 0.3; the channel gain from base station 2 to terminal D is 0.6, the noise level is 0.3, and the interference level is 0.2; the channel gain from base station 3 to terminal E is 0.7, the noise level is 0.2, and the interference level is 0.1; and the channel gain from base station 3 to terminal F is 0.8, the noise level is 0.1, and the interference level is 0.2.

[0117] The channel state data input to the first model also includes resource allocation optimization variables such as spectrum resources, power resources, and time resources. Therefore, the aforementioned channel state data and resource allocation optimization variables can be integrated into an input matrix. Each row represents the channel state data and resource allocation optimization variables from a base station to a terminal, and each column represents a feature. Inputting the integrated channel state data into the pre-trained first model yields the wireless resource allocation result. Assuming the wireless resource allocation result is a matrix, where each row represents the resource allocation from a target base station to a target terminal, and each column represents a resource allocation variable (such as spectrum allocation, power allocation, or time allocation), the following specific allocation results can be obtained based on the output matrix corresponding to the wireless resource allocation result:

[0118] Base station 1 allocates 3MHz of spectrum resources, 0.4W of power resources, and 3ms of time resources to terminal A; base station 1 allocates 2MHz of spectrum resources, 0.3W of power resources, and 2ms of time resources to terminal B; base station 2 allocates 4MHz of spectrum resources, 0.5W of power resources, and 4ms of time resources to terminal C; base station 2 allocates 3MHz of spectrum resources, 0.2W of power resources, and 3ms of time resources to terminal D; base station 3 allocates 2MHz of spectrum resources, 0.4W of power resources, and 2ms of time resources to terminal E; and base station 3 allocates 3MHz of spectrum resources, 0.3W of power resources, and 3ms of time resources to terminal F.

[0119] In this way, the pre-trained first model can quickly and accurately allocate wireless resources to multiple target terminals for multiple target base stations based on the channel characteristics and resource allocation optimization variables of the current stage, thereby optimizing network performance and user experience.

[0120] Furthermore, the first model can be trained using a pre-trained second model, which has already been trained on some historical channel state data and is capable of efficiently and accurately allocating radio resources for all historical sample channel state data in all historical stages. To ensure that the first model still possesses the ability to efficiently and accurately allocate radio resources for all historical sample channel state data in all historical stages, the second model can be used to process multiple historical channel state data in the previous historical stage to obtain an input data matrix. Based on this input data matrix, the null space of the samples in the previous historical stage can be determined. By restricting parameter updates during the current stage's training process within the null space of the samples in the previous stage, catastrophic forgetting of the trained first model can be effectively avoided.

[0121] For example, the objective function can be used to find a set of parameters W at the current stage t, such that the training set D of samples at the current stage... t Minimize the total loss value on the training set D of historical samples for all historical stages q. q Historical sample channel state data x q The current intermediate model processes historical sample channel state data x for each historical stage q. q First loss l(x) q ,W), and the second model processes x q The second loss l(x) q W t-1 Equal constraints. For example, the objective function has the following form:

[0122]

[0123] Among them, W t The first model's parameters are obtained after the second model is trained; l(x,W) represents the loss value obtained by the intermediate model after processing the channel state data of each sample in the training set, corresponding to the values ​​of the model parameter variables at the current stage. q W) is the first loss obtained by the intermediate model after processing the channel state data of each historical sample in the historical sample training set, l(x) q W t-1 D is the second loss obtained by processing the channel state data of each historical sample in the historical sample training set by the second model; t Let x represent the training set of samples in the current stage t. q W represents the historical sample channel state data in historical stage q. t-1 D represents the model parameters of the second model. qThe historical sample training set represents the historical stage. The current stage is stage t, the second model is stage t-1, the model used to train the second model is stage t-2, and so on.

[0124] In some implementations, the parameter optimization function is used to find a set of model parameters W at the current stage t that minimizes the total loss on the training set of samples at the current stage. For example, the parameter optimization function can take the following form:

[0125]

[0126] in, Let D represent the training set of samples for the current stage t. t The summation of multiple losses corresponding to all sample channel state data, i.e., the total loss value variable, is argmin. W Indicates the search for achieving The model parameter W that reaches its minimum value.

[0127] In some implementations, the constraint function is used to ensure that, during the training of the second model, the intermediate model corresponding to the model parameter variables adjusted in real time obtains a first loss equal to the second loss of the second model when processing historical sample channel state data of all historical stages q, so as to avoid the intermediate model causing a decrease in performance on historical data when optimizing the current stage data. For example, the constraint function can take the following form:

[0128]

[0129] Where, l(x) q W) is the first loss obtained by the intermediate model after processing the channel state data of each historical sample in the historical sample training set, l(x) q W t-1 The second loss is obtained by processing the channel state data of each historical sample in the historical sample training set by the second model.

[0130] In some implementations, the first loss can be less than or equal to the second loss to ensure that the intermediate model and the first model perform at least as well as the second model on historical data. That is:

[0131]

[0132] By training a second model, obtained from the previous historical stage, on the current stage's sample training set, the periodicity and repetitiveness of the channel distribution can be fully utilized. This avoids the model frequently relearning from already adapted distributions and allows it to quickly adapt to the current stage's channel distribution. Specifically, the second model, as the base model, has undergone parameter optimization in historical stages and can efficiently and accurately process historical channel state data. To adapt to the current stage's channel distribution, the second model is trained using the current stage's sample channel state data to obtain the first model. This ensures optimal resource allocation when processing current stage data while retaining the memory of historical stage data.

[0133] In this process, the objective function plays a crucial role, comprising a parameter optimization function and a constraint function. The parameter optimization function minimizes the total loss at the current stage when the intermediate model processes sample channel state data, while the constraint function ensures that the intermediate model's performance when processing historical sample channel state data is comparable to or better than the second model; that is, the first loss is equal to or less than the second loss. By restricting parameter updates at the current stage to the sample null space, the intermediate model is only allowed to adjust in directions that do not alter the prediction results of historical data, thus effectively avoiding catastrophic forgetting.

[0134] Furthermore, in specific implementation, the input data integrating channel state data and resource allocation optimization variables is input into the pre-trained first model. The first model outputs the wireless resource allocation result, indicating how to allocate wireless resources among multiple target base stations and multiple target terminals. In this way, the first model can quickly and accurately allocate wireless resources based on the channel characteristics and resource allocation optimization variables of the current stage, thereby optimizing network performance and user experience.

[0135] Please refer to Figure 3 In some implementations, to enable the intermediate model to retain its memory of old data while continuously learning new data, the second model can be trained based on the sample training set corresponding to the current stage, thereby achieving effective resource allocation in a dynamically changing wireless environment. For example, the first model can be trained in the following way:

[0136] Step 201: Obtain the sample training set corresponding to the current stage, wherein the sample training set includes multiple sample channel state data;

[0137] Step 202: Minimize the total loss value as the optimization objective, and construct the parameter optimization function based on the functional relationship between the total loss variable and the model parameter variable; wherein, the total loss variable is obtained by summing the losses of multiple current samples calculated after allocating radio resources for multiple sample channel state data based on the intermediate model corresponding to the current model parameter variable value;

[0138] Step 203: Based on the intermediate model corresponding to the current model parameter variable values, after allocating radio resources for multiple historical sample channel state data from different historical stages, the first loss is calculated.

[0139] Step 204: Based on the second model corresponding to the historical model parameters, after allocating radio resources for multiple historical sample channel state data, the second loss is calculated.

[0140] Step 205: Based on the equality relationship between the first loss and the second loss, construct the constraint function;

[0141] Step 206: Construct the objective function based on the constraint function and the parameter optimization function;

[0142] Step 207: Combine the objective function to determine the target model parameters in the null space of the sample, which minimize the total loss value obtained by the second model after allocating radio resources for multiple sample channel state data.

[0143] Step 208: Adjust the parameters of the second model based on the target model parameters to obtain the first model.

[0144] The current stage can be a specific period in a multi-stage training process, during which the second model is being trained to adapt to new channel states. The current stage collects a new sample training set to adjust model parameters for optimizing radio resource allocation. The sample channel state data used to train the second model and the channel state data later input into the first model are both within the current stage.

[0145] The training set is a collection of channel data used to guide the second model in learning to effectively allocate wireless resources in the current phase. In the current phase, the training set includes multiple sample channel state data to reflect the characteristics of the wireless communication environment, such as channel gain, noise level, and interference level.

[0146] The current sample loss can be the loss value calculated for a single sample channel state data during model training.

[0147] Please refer to Figure 4For example, when the historical stage is stage 1, channel distribution A can correspond to dataset D1, and the model can be trained using D1. After the channel distribution changes, the model needs to be retrained based on the dataset of the corresponding stage. For instance, when the channel distribution A in stage 1 changes to channel distribution B in the current stage (stage t), the dataset D corresponding to stage t needs to be used. t The second model is trained to obtain the first model. When the channel distribution B changes to the channel distribution C, the dataset D corresponding to channel distribution C needs to be obtained. T The first model is trained, and so on. This application does not list them all in the embodiments.

[0148] For example, each sample channel state data can be input into the second model (the model parameters are adjusted during training, here it is an intermediate model), and the allocated sum rate is calculated based on the first radio resource allocation result output by the second model as the current sample loss. Specifically, the current sample loss l(x,W) has the following form:

[0149] l(x,W)=-SumRate(x,W);

[0150] Where x represents the sample channel state data, and W represents the current value of the model parameter variable.

[0151] In some implementations, the total loss value (total loss variable) can be obtained by summing the losses of all current samples obtained after the model processes all sample channel state data in the training set with the same model parameter values. For example, if the training set contains 5 sample channel state data, the total loss value can be obtained by summing the losses of the 5 current samples after the model processes these 5 sample channel state data.

[0152] Furthermore, the parameter optimization function can take the following form:

[0153]

[0154] in, Let D represent the training set of samples for the current stage t. t The summation of multiple losses corresponding to all sample channel state data, i.e., the total loss value variable, is argmin. W Indicates the search for achieving The model parameter W that reaches its minimum value.

[0155] Furthermore, an intermediate model can process the historical sample channel state data contained in the historical sample training set corresponding to each historical stage to obtain the corresponding first loss. Simultaneously, a second model can process the historical sample channel state data contained in the historical sample training set corresponding to each historical stage to obtain the corresponding second loss. The constraint function can constrain the first loss to be less than or equal to the second loss. It should be noted that when the first loss of the intermediate model and the second loss of the second model are established as equal, both the first and second losses are calculated based on the same historical sample channel state data. For example, the first loss is the loss of the intermediate model processing historical sample channel state data 1, and the second loss is the loss of the second model processing historical sample channel state data 1. This ensures that during the optimization process, the intermediate model's performance on historical data is at least equivalent to that of the second model, effectively preventing the forgetting of old knowledge.

[0156] In some implementations, the first loss, the second loss, and the current sample loss all correspond to the same loss function formula. For example, the first loss, the second loss, and the current sample loss are all calculated using the sum rate after wireless resource allocation, etc., to maintain consistency in model parameter evaluation and ensure accurate model performance comparisons. Since the form of the loss function has already been explained in the previous section on the current sample loss, please refer to the above text for details; it will not be listed here again.

[0157] In some implementations, the constraint function is used to ensure that, during the training of the second model, the intermediate model corresponding to the model parameter variables adjusted in real time obtains a first loss equal to the second loss of the second model when processing historical sample channel state data of all historical stages q, so as to avoid the intermediate model causing a decrease in performance on historical data when optimizing the current stage data. For example, the constraint function can take the following form:

[0158]

[0159] Where, l(x) q W) is the first loss obtained by the intermediate model after processing the channel state data of each historical sample in the historical sample training set, l(x) q W t-1 The second loss is obtained by processing the channel state data of each historical sample in the historical sample training set by the second model.

[0160] In some implementations, the objective function can be constructed based on the constraint function and the parameter optimization function:

[0161] For example, the objective function can take the following form:

[0162]

[0163] Among them, W t The first model's parameters are obtained after the second model is trained; l(x,W) represents the loss value obtained by the intermediate model after processing the channel state data of each sample in the training set, corresponding to the values ​​of the model parameter variables at the current stage. q W) is the first loss obtained by the intermediate model after processing the channel state data of each historical sample in the historical sample training set, l(x) q W t-1 D is the second loss obtained by processing the channel state data of each historical sample in the historical sample training set by the second model; t Let x represent the training set of samples in the current stage t. q W represents the historical sample channel state data in historical stage q. t-1 D represents the model parameters of the second model. q The historical sample training set represents the historical stage. The current stage is stage t, the second model is stage t-1, the model used to train the second model is stage t-2, and so on.

[0164] Specifically, the objective function can be used to determine a set of model parameters W in the null space of samples corresponding to each network layer at the current stage t, such that the training set D of samples at the current stage... t Minimize the total loss value on the training set D of historical samples for all historical stages q. q Historical sample channel state data x q The current intermediate model processes historical sample channel state data x for each historical stage q. q First loss l(x) q ,W), and the second model processes x q The second loss l(x) q W t-1 Equal constraints are applied to ensure that the performance of historical data remains unchanged while minimizing the total loss value in the current stage.

[0165] In some implementations, the following calculation expression can be used in the linear layer of the intermediate model:

[0166] WX = Y;

[0167] Where W represents the values ​​of the model parameter variables in the current stage, X is the input data matrix, and Y is the output of the intermediate model. To ensure that the intermediate model's learning of the training set corresponding to the channel state in the current stage does not affect the performance of previous stages, the model parameter update ΔW can be restricted to the null space of the data X from the previous historical stage, i.e., satisfying:

[0168] ΔWX=0, (W+ΔW)X=WX;

[0169] By updating the model parameters in the null space of the samples at the current stage, the intermediate model can avoid changing its prediction results for historical data when training on new data, thus effectively avoiding catastrophic forgetting.

[0170] Furthermore, when calculating the null space of samples corresponding to each network layer, the optimized neural network model from each training phase can be used to process all data samples (number of samples being n) from the current phase after each training phase. t Perform a complete forward propagation, recording the input data matrix of each layer during the forward propagation. Taking the l-th layer as an example, the input data matrix is ​​a matrix with dimension O. l-1 ×n t The matrix, where O l-1 n is the output dimension of the neurons in layer l-1, that is, the number of features output by layer l-1. t It represents the number of sample channel state data at the current stage.

[0171] In some implementations, after each training phase (e.g., after training on all data for the current phase), the input data matrix X of each network layer in the model can be calculated and saved as data from the previous phase. This input data matrix will be used to constrain the parameter updates of each network layer during the next training phase, keeping it in the null space of X. In this way, effective memorization of historical information can be achieved between training phases, while providing a more reliable foundation for learning the data of the current phase.

[0172] Furthermore, after optimizing and obtaining the target model parameters (i.e., the values ​​of the model parameter variables that are updated in the sample null space, satisfying the first loss equal to the second loss, and minimizing the total loss value), the second model can be adjusted based on the target parameters to obtain a first model that can better adapt to the current stage of data while maintaining the memory of historical data. In this way, continuous learning and optimization of wireless resource allocation can be achieved in the ever-changing channel state.

[0173] By employing the above methods, not only is the model's ability and speed of learning channel data at the current stage improved, but its ability to remember historical data is also enhanced, enabling the system to maintain robustness and efficiency in complex and ever-changing wireless communication environments, and significantly improving the overall performance and service quality of the system.

[0174] In some implementations, the method may further include:

[0175] (A.3.1) Obtain at least one resource allocation target for the current stage of wireless resource allocation; wherein the resource allocation target includes rate, communication latency rate, and allocation balance rate;

[0176] (A.3.2) Using the intermediate model corresponding to the current model parameter variable values, radio resource allocation is performed on each sample channel state data in sequence to obtain the first radio resource allocation result;

[0177] (A.3.3) Based on at least one resource allocation objective and in conjunction with the first wireless resource allocation result, calculate at least one sub-loss after wireless resource allocation for each sample channel state data;

[0178] (A.3.4) Based on the sum of at least one sub-loss, obtain the current sample loss when the intermediate model processes each sample channel state data, corresponding to the value of the current model parameter variable.

[0179] Resource allocation objectives can refer to specific performance metrics or standards that are desired to be optimized during the wireless resource allocation process. These objectives may include, but are not limited to, system capacity (i.e., total throughput), communication latency, and allocation balance.

[0180] Here, the sum of rates can be the sum of the total transmission rates that all target terminals can obtain in a given wireless communication network.

[0181] The communication latency rate can be the time required for a data packet to be sent and received.

[0182] Among them, the allocation balance rate can be the fair distribution of resources among different target terminals. A high allocation balance rate means that resources are distributed more evenly, and no target terminal is over-prioritized or ignored.

[0183] The first wireless resource allocation result can be a resource allocation scheme calculated by the current intermediate model for each target terminal based on the input sample channel state data. For example, the intermediate model can allocate specific frequency bandwidth, power level, and transmission time to each target terminal.

[0184] The current sample loss can be a loss value calculated based on the resource allocation result of a single sample channel state data and at least one resource allocation objective, using an intermediate model. For example, if the objective is to maximize the sum and rate, then the current sample loss can be a negative of the sum and rate values. The total loss value is obtained by summing the current sample losses corresponding to all sample channel state data in the sample training set.

[0185] In some implementations, taking the resource allocation target as the sum rate as an example, each sample channel state data is input into the intermediate model, and the allocated sum rate is calculated as the current sample loss based on the first wireless resource allocation result output by the intermediate model. Specifically, the current sample loss l(x,W) corresponding to any sample channel state data has the following form:

[0186] l(x,W)=-SumRate(x,W);

[0187] Where x represents the sample channel state data, and W represents the current value of the model parameter variable.

[0188] In some implementations, there can be multiple resource allocation objectives at the current stage. For example, in addition to the sum of the sum of the rates, the current sample loss can also be calculated using the communication latency rate or the allocation balancing rate. Furthermore, the current sample loss can be obtained by adding the sub-losses corresponding to the sum of the sum of the rates, the communication latency rate, and the allocation balancing rate. The specific form of calculating the current sample loss can be flexibly set according to the resource allocation objectives, and this application does not impose specific limitations on this.

[0189] By employing the above methods, we can not only ensure that the model does not forget historical data while optimizing the data of the current stage, but also flexibly adapt to multiple resource allocation goals of the current stage, achieving efficient and fair allocation of resources, thereby continuously improving network performance and user experience in a dynamically changing wireless communication environment.

[0190] In some implementations, to ensure that the output of the second model remains unchanged across historical data when learning in a new stage (the current stage), updates to the model parameters can be restricted to the null space of samples from the previous historical stage. This effectively avoids catastrophic forgetting and allows the model to continuously learn new knowledge without affecting its predictions of historical data. For example, before step 207, the following may also be included:

[0191] (B.1) The second model is used to process the corresponding historical channel state data in each network layer to obtain the input data matrix corresponding to each network layer;

[0192] (B.2) In each network layer, the input data matrix is ​​approximated by a low rank to obtain the updated input data matrix;

[0193] (B.3) Based on the updated input data matrix, determine the null space of the samples corresponding to each network layer.

[0194] The input data matrix can refer to a matrix composed of all sample channel state data received by each network layer in a neural network.

[0195] Furthermore, after each training phase, the optimized neural network model from that phase can be used to process all data samples (number of samples being n) from the current phase. t A complete forward propagation is performed, during which the input data matrix X of each layer is recorded. These input data matrices will be used in the next training stage to constrain the parameter updates of each network layer, keeping them within the null space of X. In this way, effective memorization of historical information can be achieved between training stages, while providing a more reliable foundation for learning from the current stage's data. Taking the network layer as l as an example, the input data matrix is ​​a matrix with dimension O... l-1 ×n t The matrix, where O l-1 n is the output dimension of the neurons in layer l-1, that is, the number of features output by layer l-1. t It represents the number of sample channel state data at the current stage.

[0196] Furthermore, a low-rank approximation is performed on the input data matrix of each network layer to obtain an updated input data matrix, which helps to remove unimportant features and retain the most important information, thereby simplifying the intermediate model and reducing the number of parameters.

[0197] Furthermore, when calculating the null space of samples corresponding to each network layer, after each training phase (e.g., after training all data for the current phase), the input data matrix X of each network layer in the model can be calculated and saved as the data of the previous phase. This input data matrix will be used to constrain the parameter updates of each network layer during the next training phase, keeping it within the null space of X. In this way, effective memorization of historical information can be achieved between training phases, while providing a more reliable foundation for learning the data of the current phase.

[0198] In some implementations, the following calculation expression can be used in the linear layer of the intermediate model:

[0199] WX = Y;

[0200] Where W represents the values ​​of the model parameter variables in the current stage, X is the input data matrix, and Y is the output of the intermediate model. To ensure that the intermediate model's learning of the training set corresponding to the channel state in the current stage does not affect the performance of previous stages, the model parameter update ΔW can be restricted to the null space of the data X from the previous historical stage, i.e., satisfying:

[0201] ΔWX=0, (W+ΔW)X=WX;

[0202] By updating the model parameters in the null space of the samples at the current stage, the model can retain its memory of historical data during the learning process of the new stage, enabling continuous learning and optimization of wireless resource allocation, and improving the robustness of the model.

[0203] In some implementations, to provide a basis for subsequent low-rank approximation processing and sample null space determination, the trained model can process the sample channel state data of the corresponding stage after each training stage, obtaining the input data matrix of each layer layer by layer. This allows the model to retain its memory of historical data and avoid catastrophic forgetting during subsequent training stages. For example, (B.1) may include:

[0204] (B.1.1) Using the second model, based on the historical model parameters corresponding to each network layer, multiple historical channel state data are processed to obtain the input data matrix corresponding to each network layer;

[0205] (B.1.2) Repeat the process in the next network layer of the second model, and process multiple historical channel state data according to the next historical model parameters corresponding to the next network layer to obtain the next input data matrix corresponding to the next network layer, until the input data matrix of the last network layer of the intermediate model is obtained.

[0206] In some implementations, each layer of the model receives the output of the previous layer as its input and generates a new feature representation for the next layer to use. The input data matrix of each network layer contains the information that the network needs to process and is a key component of the model learning process. The parameter update of each network layer is based on the input data matrix of that layer and the gradient of the loss function. Therefore, each layer needs an input data matrix to guide its parameter update.

[0207] For example, if the second model includes an input layer, two hidden layers (hidden layer 1 and hidden layer 2) and an output layer (this is just a simple example; the model structure may be more complex in reality), when using the second model to process historical channel state data (or sample historical channel state data, the input data is determined according to the actual situation) to obtain the input data matrix of each layer, historical sample channel state data can be received through the input layer. Assuming that these data have 3 features (e.g., channel gain, noise level, and interference level) and a total of 100 data points, then the input data matrix X1 of the input layer is a 3×100 matrix.

[0208] Hidden layer 1 receives the output of the input layer as its input. Assume hidden layer 1 has 4 neurons, each connected to all 3 features of the input layer. The second model processes this input data using its historical parameters from hidden layer 1, generating the output of hidden layer 1. These outputs are then used as the input to hidden layer 2. Therefore, the input data matrix X2 of hidden layer 1 will be a 4×100 matrix. Hidden layer 2 repeats the above process, receiving the output of hidden layer 1 as its input. Assume hidden layer 2 also has 4 neurons, processing the output from hidden layer 1. The second model processes this input data using its historical parameters from hidden layer 2, generating the output of hidden layer 2. These outputs are ultimately used as the input to the output layer. Therefore, the input data matrix X3 of hidden layer 2 will also be a 4×100 matrix.

[0209] By using the above method, the input data matrix of each layer can be obtained layer by layer. These matrices will be used to determine the null space of subsequent samples to ensure that the model can maintain its memory of historical data and avoid catastrophic forgetting during training in the new stage. The input data matrix of each layer not only guides the updating of the parameters of that layer, but also provides the necessary information for the next layer, thereby achieving effective learning and memory retention throughout the network.

[0210] In some implementations, to reduce the complexity and computational cost of parameter updates during subsequent model training while retaining the most important information, low-rank approximation can be used to reduce the dimensionality of the input data matrix. This effectively avoids the problems of high computational cost and long training time caused by high-dimensional data, while ensuring that the model retains its memory of historical data, thus improving the model's adaptability and robustness in dynamic wireless environments. For example, (B.2) may include:

[0211] (B.2.1) In each network layer, singular value decomposition is performed on the corresponding input data matrix to obtain the left singular vector matrix, the singular value diagonal matrix, and the transpose of the right singular vector matrix of the input data matrix.

[0212] (B.2.2) Obtain the preset extraction values, and extract the same number of matrix columns as the preset extraction values ​​from the left singular vector matrix to obtain the updated left singular vector matrix;

[0213] (B.2.3) Extract the same number of singular values ​​as the preset extraction values ​​from the singular value diagonal matrix to obtain the updated singular value diagonal matrix;

[0214] (B.2.4) Extract the same number of matrix rows as the preset extraction value from the transpose of the right singular vector matrix to obtain the updated transpose of the right singular vector matrix;

[0215] (B.2.5) Based on the updated left singular vector matrix, the updated singular value diagonal matrix, and the transpose of the updated right singular vector matrix, the updated input data matrix is ​​obtained.

[0216] The left singular vector matrix can be obtained by performing singular value decomposition (SVD) on the input data matrix (denoted by A), resulting in A = U∑V. T Here, U is the left singular vector matrix. The column vectors of the left singular vector matrix are orthogonal, and these column vectors are the left multiplication of matrix A by its transpose (i.e., AA). T The eigenvectors of the left singular vector matrix are the same as the number of rows in the input data matrix A, and the number of columns is equal to the rank of A.

[0217] The singular value diagonal matrix can be the diagonal matrix Σ obtained by performing singular value decomposition on the input data A. The elements on the diagonal of the singular value diagonal matrix are the singular values ​​of the input data matrix, sorted from largest to smallest. These singular values ​​reflect the importance or energy of each principal component.

[0218] The transpose of the right singular vector matrix can be the transpose of another orthogonal matrix V obtained after performing singular value decomposition on the input data A. T The column vectors of the right singular vector matrix are orthogonal, and these column vectors are the right product of the transpose of the input data matrix A (i.e., AA). T The eigenvectors of the input data matrix A are eigenvectors. The transpose of the right singular vector matrix has the same dimension as the number of columns of the input data matrix A, and the number of rows is equal to the rank of the input data matrix A.

[0219] The preset extraction value can be a pre-defined number used to extract a specific number of columns, singular values, or rows from the matrix after singular value decomposition. The preset extraction value can be determined based on the needs of the actual application. For example, in data dimensionality reduction, this value can be set according to how many principal components to retain. For instance, if you want to retain the three most important components of the input data matrix, then the preset extraction value would be set to 3.

[0220] The updated left singular vector matrix can be obtained by extracting a certain number of columns from the left singular vector matrix U based on preset extracted values. Assuming the preset extracted values ​​are k, the updated left singular vector matrix is ​​a matrix composed of the first k columns of U, and its dimension is the number of rows in the input data matrix A multiplied by k.

[0221] The updated singular value diagonal matrix can be obtained by extracting a certain number of singular values ​​from the original singular value diagonal matrix Σ based on a preset extraction value. If the preset extraction value is k, then the updated singular value diagonal matrix is ​​a k×k diagonal matrix, and the elements on the diagonal are the k largest singular values ​​in the original singular value diagonal matrix.

[0222] The transpose of the updated right singular vector matrix can be obtained by extracting preset values ​​and then converting the transpose V of the right singular vector matrix. T Extract the corresponding number of rows. Assuming the preset extraction value is k, then the transpose of the updated right singular vector matrix... It is a matrix consisting of the first k rows, and its dimension is k times the number of columns of the input data matrix A.

[0223] In some implementations, the preset extraction values ​​for each network layer can be the same or different, and the magnitude of the preset extraction values ​​can be flexibly adjusted according to the actual situation. For example, if the first hidden layer may need to capture the basic features of the input data, and more singular values ​​need to be retained to obtain richer information, then the first 80% of the singular values ​​can be retained. Or, if the third hidden layer is close to the output layer, it may need more abstract features, and only the first 30% of the singular values ​​can be retained, and so on. In practice, the preset extraction values ​​can be adjusted based on experimental results to find the optimal balance between information retention and computational efficiency. The preset extraction values ​​for each layer can be set according to the layer's contribution to the final output and the complexity of the data in that layer, to ensure the overall performance and computational efficiency of the network.

[0224] For example, taking the second model as an example, it has been trained to process channel state data in a wireless communication system and perform resource allocation. When a low-rank approximation is needed for the second model, for network layer A in the second model, if network layer A corresponds to a d×n input data matrix X, where s is the feature dimension and n is the number of samples, then the singular value decomposition of X can be expressed as: X=UΣV T Where U is a d×d left singular vector matrix, Σ is a d×n singular value diagonal matrix, and V T It is the transpose of an n×n right singular vector matrix.

[0225] Furthermore, if the preset extraction value k is set to 3, that is, the first k largest singular values ​​and their corresponding singular vectors are retained, the first k columns are extracted from U to form a new d×k left singular vector matrix U. k Extract the k largest singular values ​​from Σ to form a new k×k singular value diagonal matrix Σ. k From V T Extract the first k rows to form the transpose of a new k×n right singular vector matrix. The updated left singular vector matrix, the updated singular value diagonal matrix, and the transpose of the updated right singular vector matrix are combined to obtain the updated input data matrix.

[0226] In this way, computational complexity can be reduced without losing important information, while retaining important information and expanding the dimension of the null space. This not only increases the freedom of parameter updates in subsequent stages, but also enhances the model's adaptability and stability to new environments.

[0227] In some implementations, as training progresses, the null dimension of the input data decreases rapidly, limiting the feasible space for parameter updates and thus degrading model performance. To ensure the model remains robust and efficient in complex and ever-changing wireless communication environments, structural pruning can increase the degrees of freedom for parameter updates, reducing model complexity and computational cost, improving training efficiency, and enabling the model to quickly adapt to new channel environments. For example, a wireless resource allocation method may further include:

[0228] (C.1) Obtain multiple scaling parameters corresponding to multiple network layers of the second model;

[0229] (C.2) Obtain the absolute values ​​of multiple scaling parameters corresponding to multiple scaling parameters, and determine multiple target scaling parameters based on the order of the magnitude of the absolute values ​​of multiple scaling parameters;

[0230] (C.3) Determine multiple target neurons corresponding to multiple target scaling parameters, and temporarily disable the corresponding multiple target neurons in multiple network layers.

[0231] The scaling parameter can be a learnable parameter used to adjust the scale of the normalized features. The scaling parameter can be represented by γ, and it typically works in conjunction with the bias parameter β to restore the normalized data distribution.

[0232] The absolute value of the scaling parameter can be its absolute value, used to measure its influence on the neuron's output. A larger absolute value means that the neuron contributes more to the model's output; conversely, a smaller absolute value indicates a lower contribution.

[0233] The target scaling parameter can be a scaling parameter whose absolute value is less than the scaling threshold during selective disabling. The neuron corresponding to the target scaling parameter will be temporarily disabled to optimize model performance.

[0234] The target neuron can be a neuron with a target scaling parameter, i.e., a neuron with a small absolute value of the scaling parameter that is considered to contribute little to the overall performance of the model. During pruning, the target neuron and all its associated parameters are temporarily set to zero, thus rendering the target neuron completely ineffective in computation.

[0235] For example, all network layers of the second model can be traversed, and for each network layer containing a Batch Normalization (BN) layer, the scaling parameters of that layer can be extracted. Within each network layer, the absolute value of each scaling parameter can be calculated, and all absolute values ​​of scaling parameters within that network layer can be sorted in ascending order.

[0236] Furthermore, a scaling threshold or scaling ratio can be determined, and the scaling parameter with the smallest absolute value can be selected as the target scaling parameter. For example, the scaling parameters in the top 10% with the lowest absolute values ​​can be selected as the target scaling parameters, and the target neuron corresponding to the target scaling parameter can be determined. All relevant parameters (including γ and β) of the target neuron and its corresponding BN layer are set to zero to ensure that these parameters no longer generate any output during model training in the current stage. Furthermore, the scaling threshold or scaling ratio can be set according to actual conditions; for example, the scaling ratio can be 20%, 30%, etc., and this embodiment does not impose specific limitations on this.

[0237] In some implementations, the disabled neurons can be disabled at the current stage or continuously disabled, depending on the specific circumstances.

[0238] By using the above methods, the lower bound of the null space is increased, providing greater optimization space for subsequent stages. At the same time, it also improves the second model's learning ability on new data, maintains the model's memory of historical data, avoids the problem of catastrophic forgetting, and helps the model maintain robustness and efficiency in complex and ever-changing environments.

[0239] Please refer to Figure 5 and Figure 6 In some implementations, the wireless resource allocation method of this application can be evaluated through testing. For example, considering a downlink scenario involving 10 target base stations and 10 target terminals, with downlink transmission rate optimization as the objective, it is assumed that the channel undergoes three different distribution phases, each with independent and significantly varying channel characteristics. The maximum transmit power of the target base stations is set to 1W, and each target terminal has the same weight to ensure fairness in resource allocation.

[0240] Figure 5 The performance comparison results of the wireless resource allocation method in this application with other learning methods are presented. From Figure 5As can be seen, compared with memory-based learning methods, the performance improvement of the method in this application is close to one-third. Compared with transfer learning methods, the performance improvement of the method in this application is even more significant. This indicates that the method proposed in this application has stronger adaptability and optimization effect in dynamic environments.

[0241] Figure 6 The training time comparison results of the wireless resource allocation method of this application with other learning methods are presented. Figure 6 It is known that the training time of memory-based learning methods increases exponentially when the number of stages is large. In contrast, the wireless resource allocation method proposed in this application only shows a linear increase in training time, which fully demonstrates the significant advantage of the proposed method in terms of training efficiency.

[0242] This application embodiment acquires channel state data between multiple target base stations and multiple target terminals; inputs the channel state data into a pre-trained first model to obtain the radio resource allocation results of multiple target base stations to multiple target terminals; wherein, the first model is obtained by adjusting a pre-trained second model according to target model parameters; the target model parameters are the values ​​of model parameter variables that minimize the total loss value obtained by the second model in the sample null space after performing radio resource allocation on multiple sample channel state data, combined with the objective function; the sample null space is determined by the second model obtained by adjusting parameters according to historical stages and processing multiple historical sample channel state data to obtain an input data matrix; the objective function includes a parameter optimization function and a constraint function; the parameter optimization function, while minimizing the total loss value as the optimization objective, includes the functional relationship between the total loss variable and the model parameter variable; the constraint function is used to limit the equality relationship between the first loss and the second loss; the first loss is calculated based on the intermediate model corresponding to the current value of the model parameter variable, after performing radio resource allocation on multiple historical sample channel state data at different historical stages; the second loss is calculated based on the second model corresponding to the historical model parameters, after performing radio resource allocation on the multiple historical sample channel state data. This approach fully leverages the periodicity and repetitiveness of channel distribution. Based on multiple sample channel state data from the current stage, a pre-trained and well-performing second model is rapidly trained, eliminating the need for frequent learning of already adapted distributions and improving training efficiency and adaptability. Furthermore, the second model, with parameters adjusted based on historical stages, processes multiple historical channel state data to obtain an input data matrix, determining the sample null space. Optimization within this null space at the current stage allows for a smooth transition between stages, avoiding catastrophic forgetting of historically learned knowledge. This enables the model to efficiently adapt to dynamically changing wireless environments while retaining historical knowledge, improving resource allocation accuracy. Simultaneously, optimizing model parameters by constructing an objective function allows the model to consider both long-term and short-term resource allocation goals, resulting in more precise allocation that better meets network performance and user experience requirements. Compared to traditional methods, this approach does not rely on complex numerical optimization techniques but utilizes the powerful learning capabilities and parallel computing characteristics of deep learning. This enables the model to allocate channel resources quickly and accurately, improving the real-time performance and accuracy of resource allocation.

[0243] Please see Figure 7 This application also provides a wireless resource allocation device that can implement the above-described wireless resource allocation method. The wireless resource allocation device includes:

[0244] Acquisition module 71 is used to acquire channel state data between multiple target base stations and multiple target terminals;

[0245] Input module 72 is used to input channel state data into a pre-trained first model to obtain the radio resource allocation results of multiple target base stations for multiple target terminals. The first model is obtained by adjusting a pre-trained second model according to target model parameters. The target model parameters are the values ​​of the model parameter variables that minimize the total loss value obtained by the second model after allocating radio resources for multiple sample channel state data, determined in the sample null space by combining the objective function. The second model, obtained by adjusting parameters according to historical stages, is determined after processing multiple historical sample channel state data to obtain the input data matrix. The objective function includes a parameter optimization function and a constraint function. The parameter optimization function, while minimizing the total loss value as the optimization objective, includes the functional relationship between the total loss variable and the model parameter variables. The constraint function is used to limit the equality relationship between the first loss and the second loss. The first loss is calculated based on the intermediate model corresponding to the current model parameter variable values ​​after allocating radio resources for multiple historical sample channel state data at different historical stages. The second loss is calculated based on the second model corresponding to the historical model parameters after allocating radio resources for multiple historical sample channel state data.

[0246] The specific implementation of this wireless resource allocation device is basically the same as the specific embodiment of the wireless resource allocation method described above, and will not be repeated here. Subject to meeting the requirements of the embodiments of this application, the wireless resource allocation device may also be equipped with other functional modules to implement the wireless resource allocation method in the above embodiments.

[0247] This application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned wireless resource allocation method. This computer device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0248] Please see Figure 8 , Figure 8 The hardware structure of a computer device according to another embodiment is illustrated. The computer device includes:

[0249] The processor 81 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0250] The memory 82 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 82 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 82 and is called and executed by the processor 81 using the wireless resource allocation method of the embodiments of this application.

[0251] Input / output interface 83 is used to implement information input and output;

[0252] The communication interface 84 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0253] Bus 85 transmits information between various components of the device (e.g., processor 81, memory 82, input / output interface 83, and communication interface 84);

[0254] The processor 81, memory 82, input / output interface 83, and communication interface 84 are connected to each other within the device via bus 85.

[0255] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described wireless resource allocation method.

[0256] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0257] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0258] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0259] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0260] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0261] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0262] It should be understood that in this application, "at least one" and "several" refer to one or more, and "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0263] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0264] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0265] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0266] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0267] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for allocating wireless resources, characterized in that, The method includes: Acquire channel state data between multiple target base stations and multiple target terminals; The channel state data is input into a pre-trained first model to obtain the wireless resource allocation results of the multiple target base stations for the multiple target terminals; Wherein, the first model is obtained by adjusting the pre-trained second model according to the target model parameters; the target model parameters are the values ​​of the model parameter variables that are determined in the null space of the samples when the second model minimizes the total loss value after allocating wireless resources for multiple sample channel state data, in combination with the objective function. The second model, obtained by adjusting parameters according to historical stages, is determined after processing multiple historical sample channel state data to obtain an input data matrix. The objective function includes a parameter optimization function and a constraint function. The parameter optimization function, while aiming to minimize the total loss value, includes the functional relationship between the total loss variable and the model parameter variable. The constraint function is used to define the equality relationship between the first loss and the second loss; the first loss is calculated based on the intermediate model corresponding to the current value of the model parameter variable, and after allocating radio resources for multiple historical sample channel state data at different historical stages; the second loss is calculated based on the second model corresponding to the historical model parameters, and after allocating radio resources for the multiple historical sample channel state data.

2. The wireless resource allocation method according to claim 1, characterized in that, The first model was trained in the following way: Obtain the sample training set corresponding to the current stage, wherein the sample training set includes multiple sample channel state data; The optimization objective is to minimize the total loss value, and a parameter optimization function is constructed based on the functional relationship between the total loss variable and the model parameter variable. The total loss variable is obtained by summing the losses of multiple current samples after allocating radio resources for multiple sample channel state data based on the intermediate model corresponding to the current value of the model parameter variable. Based on the intermediate model corresponding to the current values ​​of the model parameter variables, the first loss is calculated after allocating radio resources to multiple historical sample channel state data from different historical stages. Based on the second model corresponding to the historical model parameters, after allocating radio resources to the multiple historical sample channel state data, the second loss is calculated. Based on the equality relationship between the first loss and the second loss, a constraint function is constructed; Based on the constraint function and the parameter optimization function, construct the objective function; Based on the objective function, the target model parameters are determined in the sample null space to minimize the total loss value obtained by the second model after allocating radio resources for the multiple sample channel state data. The second model is adjusted based on the target model parameters to obtain the first model.

3. The wireless resource allocation method according to claim 2, characterized in that, The method further includes: Obtain at least one resource allocation target for wireless resource allocation in the current stage; wherein, the resource allocation target includes data rate, communication latency rate, and allocation balance rate; Using the intermediate model corresponding to the current model parameter variable values, radio resource allocation is performed sequentially on each sample channel state data to obtain the first radio resource allocation result. Based on the at least one resource allocation objective and in conjunction with the first wireless resource allocation result, at least one sub-loss is calculated after wireless resource allocation for each sample channel state data. Based on the sum of the at least one sub-loss, the current sample loss of the intermediate model for processing each sample channel state data is obtained, corresponding to the value of the current model parameter variable.

4. The wireless resource allocation method according to claim 2, characterized in that, Before determining the target model parameters that minimize the total loss value obtained by the second model after allocating radio resources for the multiple sample channel state data in the sample null space, in conjunction with the objective function, the method further includes: The second model processes multiple historical channel state data in each network layer to obtain the input data matrix corresponding to each network layer. In each network layer, the input data matrix is ​​approximated by a low rank to obtain an updated input data matrix; Based on the updated input data matrix, the null space of the samples corresponding to each network layer is determined.

5. The wireless resource allocation method according to claim 4, characterized in that, The process of processing multiple historical channel state data in each network layer using the second model to obtain the input data matrix corresponding to each network layer includes: Using the second model, the multiple historical channel state data are processed according to the historical model parameters corresponding to each network layer to obtain the input data matrix corresponding to each network layer; Repeat the process in the next network layer of the second model, processing the multiple historical channel state data according to the next historical model parameters corresponding to the next network layer to obtain the next input data matrix corresponding to the next network layer, until the input data matrix of the last network layer of the intermediate model is obtained.

6. The wireless resource allocation method according to claim 4, characterized in that, In each of the network layers, the input data matrix is ​​approximated in low rank to obtain an updated input data matrix, including: In each network layer, singular value decomposition is performed on the corresponding input data matrix to obtain the left singular vector matrix, the singular value diagonal matrix, and the transpose of the right singular vector matrix of the input data matrix. Obtain a preset extraction value, and extract the same number of matrix columns as the preset extraction value from the left singular vector matrix to obtain an updated left singular vector matrix; Extract the same number of singular values ​​as the preset extraction value from the singular value diagonal matrix to obtain an updated singular value diagonal matrix; Extract the same number of matrix rows as the preset extraction value from the transpose of the right singular vector matrix to obtain the transpose of the updated right singular vector matrix; The updated input data matrix is ​​obtained based on the updated left singular vector matrix, the updated singular value diagonal matrix, and the transpose of the updated right singular vector matrix.

7. The wireless resource allocation method according to claim 4, characterized in that, The method further includes: Obtain multiple scaling parameters corresponding to multiple network layers of the second model; Obtain the absolute values ​​of the multiple scaling parameters corresponding to the multiple scaling parameters, and determine multiple target scaling parameters based on the order of the magnitude of the absolute values ​​of the multiple scaling parameters; Identify multiple target neurons corresponding to the multiple target scaling parameters, and temporarily disable the corresponding multiple target neurons in the multiple network layers.

8. A wireless resource allocation device, characterized in that, The device includes: The acquisition module is used to acquire channel state data between multiple target base stations and multiple target terminals; An input module is used to input the channel state data into a pre-trained first model to obtain the wireless resource allocation results of the multiple target base stations for the multiple target terminals. The first model is obtained by adjusting a pre-trained second model according to target model parameters. The target model parameters are the values ​​of model parameter variables that minimize the total loss value obtained by the second model after allocating wireless resources for multiple sample channel state data, determined in the sample null space by combining an objective function. The sample null space is determined by processing multiple historical sample channel state data to obtain an input data matrix after the second model, obtained by parameter adjustment based on historical stages, has been processed. The objective function includes a parameter optimization function and a constraint function. The parameter optimization function, while minimizing the total loss value as the optimization objective, includes the functional relationship between the total loss variable and the model parameter variables. The constraint function is used to limit the equality relationship between the first loss and the second loss. The first loss is calculated based on the intermediate model corresponding to the current value of the model parameter variables, after allocating wireless resources for multiple historical sample channel state data at different historical stages. The second loss is calculated based on the second model corresponding to the historical model parameters, after allocating wireless resources for the multiple historical sample channel state data.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the wireless resource allocation method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the wireless resource allocation method according to any one of claims 1 to 7.

Images