Radio resource scheduling method, apparatus, device, medium and program product

Abstract

The invention provides a wireless resource scheduling method and device, equipment, a medium and a program product, and relates to the field of wireless transmission, and the method comprises the steps: collecting a current network resource use condition; receiving measurement data sent by the terminal during operation of the target service, and evaluating the service quality of the target service based on the measurement data to obtain service quality information; synchronizing the service quality information and the network resource use condition to a first control unit so as to generate service quality target information through the first control unit; calling a second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model in combination with the service quality target information, the service quality information and the configuration information of the terminal; the first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used for scheduling wireless resources for the terminal. And the response speed of the base station to the wireless resource scheduling is improved.

Description

Technical Field

[0001] The present application relates to the field of wireless transmission, and in particular to a wireless resource scheduling method, apparatus, device, medium and program product. Background Art

[0002] Quality of Service (QoS) is a service guarantee mechanism used to ensure user service performance in mobile communication networks. In existing technologies, to achieve QoS for users, base stations typically use radio resource scheduling algorithms to schedule or reallocate radio resources. However, traditional radio resource scheduling algorithms require manual, fixed rules, which prevents base stations from adjusting resources in a timely manner and results in long response times. Summary of the Invention

[0003] The embodiments of the present application provide a wireless resource scheduling method, apparatus, device and storage medium to solve the problem that the base station cannot adjust resources in a timely manner and has a long response time in the existing wireless resource scheduling method.

[0004] In order to solve the above technical problems, this application is implemented as follows:

[0005] In a first aspect, an embodiment of the present application provides a radio resource scheduling method, the method comprising:

[0006] Collect current network resource usage;

[0007] receiving measurement data of a target service during operation sent by a terminal, and evaluating the service quality of the target service based on the measurement data to obtain service quality information;

[0008] synchronizing the service quality information and the network resource usage to a first control unit, so as to generate service quality target information through the first control unit;

[0009] Invoking a second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and combining the service quality target information, the service quality information, and the configuration information of the terminal;

[0010] The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

[0011] In a second aspect, an embodiment of the present application further provides a wireless resource scheduling device. The wireless resource scheduling device includes:

[0012] Collection module, used to collect current network resource usage;

[0013] An evaluation module, configured to receive measurement data of a target service during operation sent by a terminal, and evaluate the service quality of the target service based on the measurement data to obtain service quality information;

[0014] a generating module, configured to synchronize the quality of service information and the network resource usage to a first control unit, so as to generate quality of service target information through the first control unit;

[0015] A calling module, configured to call the second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and in combination with the quality of service target information, the quality of service information, and the terminal and configuration information;

[0016] The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

[0017] In a third aspect, an embodiment of the present application provides an electronic device, including a transceiver and a processor, wherein the processor is configured to:

[0018] Collect current network resource usage;

[0019] receiving measurement data of a target service during operation sent by a terminal, and evaluating the service quality of the target service based on the measurement data to obtain service quality information;

[0020] synchronizing the service quality information and the network resource usage to a first control unit, so as to generate service quality target information through the first control unit;

[0021] Invoking a second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and combining the service quality target information, the service quality information, and the configuration information of the terminal;

[0022] The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

[0023] In a fourth aspect, an embodiment of the present application provides an electronic device comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the wireless resource scheduling method as described in the first aspect above are implemented.

[0024] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the wireless resource scheduling method described in the first aspect above are implemented.

[0025] In a sixth aspect, an embodiment of the present application provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the wireless resource scheduling method as described in the first aspect above.

[0026] The wireless resource scheduling method of an embodiment of the present application includes collecting current network resource usage; receiving measurement data of a target service sent by a terminal during operation, evaluating the service quality of the target service based on the measurement data to obtain service quality information; synchronizing the service quality information and the network resource usage to a first control unit, so that the first control unit generates service quality target information; calling a second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model, combining the service quality target information, the service quality information, and the terminal configuration information; wherein the first control unit is a non-real-time wireless access network intelligent control unit, and the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal. The method generates the first resource scheduling instruction by calling the near-real-time wireless access network intelligent control unit based on the artificial intelligence model, combining multi-dimensional data such as the service quality target information, the service quality information, and the terminal configuration information; while improving the accuracy of wireless resource scheduling, it also improves the response speed of the base station to wireless resource scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0028] Figure 1 This is one of the flow charts of the wireless resource scheduling method provided in an embodiment of the present application;

[0029] Figure 2 This is a schematic diagram of the O-RAN system architecture provided by an embodiment of the present application;

[0030] Figure 3 This is a schematic diagram of an RNN model provided in one embodiment of the present application;

[0031] Figure 4This is a structural diagram of a wireless resource scheduling device provided by an embodiment of the present application;

[0032] Figure 5 This is a structural diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0034] The embodiment of the present application provides a method for scheduling wireless resources. Figure 1 , Figure 1 This is a flow chart of the wireless resource scheduling method provided by an embodiment of the present application. Figure 1 As shown, the following steps are included:

[0035] Step 101: Collect current network resource usage;

[0036] In this step, the above-mentioned network resource usage can be understood as wireless resource occupancy, allocation and load-related information at the base station and cell levels, which may include the cell's single network slice selection assistance information (Single Network Slice Selection Assistance Information, S-NSSAI), the cell's PRBs utilization rate and the number of access users in the cell, etc., which is the basis for evaluating whether the current wireless resources are sufficient and whether priority scheduling is required.

[0037] Step 102: Receive measurement data of the target service during operation sent by the terminal, and evaluate the service quality of the target service based on the measurement data to obtain service quality information;

[0038] In this step, the measurement data of the target service sent by the above terminal during operation can be understood as the underlying measurement report and feedback information sent by the user terminal (UE) to the base station node when running the target service. The specific type of the above target service is not specifically limited in the embodiment of the present application, and may include low-priority ordinary traffic services, such as service 1: 5QI=8, Priority Level=80 Transmission Control Protocol (TCP) real-time video service; and high-priority large-traffic services, such as service 2: 5QI=2, Priority Level=40 traditional real-time live broadcast service, etc. The above measurement data includes UE's channel quality information, such as Channel Quality Indicator (CQI), Reference Signal Receiving Power (RSRP), service transmission status, such as downlink service throughput, Data Radio Bearer (DRB) packet loss rate, etc.

[0039] The above-mentioned service quality information can be understood as the target service QoS performance indicator obtained based on the terminal measurement data evaluation, which may include the UE identity information, the UE network slice information S-NSSAI, the service quality identifier (QoSIdentifier, 5QI) of the target service, the priority information of the target service, the uplink and downlink rate, bit error rate, delay, etc. of the target service. The above-mentioned service quality information can be used as a direct basis for judging whether the QoS of the target service meets the standard, such as whether the uplink and downlink rate is higher than the preset threshold.

[0040] Step 103: Synchronize the service quality information and the network resource usage to a first control unit, so as to generate service quality target information through the first control unit.

[0041] In this step, the above-mentioned first control unit can be understood as a non-real-time radio access network intelligent controller (Non-Real-Time RAN Intelligent Controller, Non-RT RIC), which usually works in conjunction with the application (rApp) running on the Non-RT RIC, and is mainly responsible for policy formulation and network optimization on a long time scale, including generating "service quality target information" based on service quality information and network resource usage, which may include the QoS targets of the target service: 5QI, priority, guaranteed bit rate (Guaranteed Flow Bit Rate, GFBR), maximum bit rate (Maximum Flow Bit Rate, MFBR), etc.

[0042] Step 104: Invoke the second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and in combination with the service quality target information, the service quality information, and the configuration information of the terminal;

[0043] The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

[0044] In this step, the second control unit can be understood as a near-real-time radio access network intelligent controller (Near-Real-Time RAN Intelligent Controller, Near-RT RIC), which usually works in conjunction with the application (xApp) running on the Near-RT RIC and can be responsible for resource management and optimization on a short time scale, including generating real-time resource scheduling instructions based on the AI ​​model and response latency much lower than the first control unit.

[0045] The above-mentioned preset artificial intelligence model can take the service quality target information, the service quality information and the configuration information of the terminal as input, thereby outputting the resource scheduling result. The above-mentioned artificial intelligence model may include a recurrent neural network model long short-term memory network (Long Short-Term Memory, LSTM), a convolutional neural network with multi-dimensional feature extraction capabilities, and a reinforcement learning model that can learn the optimal scheduling strategy by interacting with the network environment, etc. As long as these models can generate resource scheduling instructions that meet the requirements based on UE configuration information, base station measurement information and QoS targets, different types of artificial intelligence models do not affect the realization of the basic functions of the wireless resource scheduling method of the embodiment of the present application.

[0046] The configuration information of the above-mentioned terminal can be used to ensure that the resource scheduling instructions comply with the terminal hardware and protocol capabilities, which may specifically include UE slice-level information (such as S-NSSAI), UE identity information (UE ID), UE DRB information (such as DRB mapping relationship), UE QoS flow-level information (such as QoS flow priority), UE capability-level information (such as the upper limit of the transmission rate supported by the UE), etc.

[0047] The above-mentioned first resource scheduling instruction can be understood as the final scheduling instruction output by the second control unit, which can be used to guide the base station node to perform resource allocation, and may specifically include DRB mapping indication information of the UE or target service, wireless resource allocation indication information, such as slice-level PRBs configuration, CQI configuration, SPS configuration, etc., wireless access control indication information (such as SR reconfiguration), etc.

[0048] In some optional implementations, the steps in the embodiments of the present application may be performed using an open radio access network (O-RAN). The O-RAN architecture is an open, modular radio access network architecture that aims to promote network flexibility and scalability. The O-RAN architecture can introduce a variety of new intelligent functional modules, which can be referenced. Figure 2 Specifically, the O-RAN architecture includes modules such as the Non-RT RIC and Near-RT RIC on the platform side, and modules such as the Open Distributed Unit and Open Centralized Unit on the base station side. Through the collaborative operation of these modules, the O-RAN architecture can provide more flexible and efficient network services, including QoS assurance.

[0049] For example, assume that within a 5G system, there is a base station that complies with the O-RAN standard architecture and supports O-RAN standard-defined functions. This base station can control multiple 5G cells, which can provide services to multiple 5G user terminals within the system. These UEs also support O-RAN standard-defined functions. In addition, the base station is controlled by an intelligent platform that complies with the O-RAN standard architecture and supports O-RAN standard-defined functions.

[0050] The intelligent platform architecture may include: Service Management and Orchestration (SMO), which may be responsible for the management and orchestration of network services, including lifecycle management, performance management, fault management, etc.; Non-RT RIC / rApp, rApp, Near-RT RIC / xApp, and xApp. Some nodes can also be defined, including: the O-RAN Radio Unit (O-RU), which is responsible for transmitting and receiving wireless signals, including RF processing and front-end signal processing, and can be connected to the O-RAN Distributed Unit (O-DU) via the Open Fronthaul interface; the O-DU, which is responsible for processing mid-layer protocols, including partial processing of the physical layer, the MAC layer, and the RLC layer, and is connected to the O-RAN Central Unit-Control Plane (O-CU-CP) and the O-RAN Central Unit-User Plane (O-CU-UP) via the F1 interface; the O-CU-CP, which is responsible for processing high-layer protocols, including the RRC and PDCP control planes, and is connected to the O-CU-UP via the F1 interface; and the O-CU-UP, which is responsible for processing user plane data, including the SDAP layer and the PDCP user plane. The O-DU, O-CU-CP, and O-CU-UP can be collectively referred to as E2 nodes. The interfaces between the platform architecture and nodes may include: O1 interface, responsible for information exchange between SMO and E2 nodes and between SMO and Near-RT RIC; A1 interface, responsible for communication between SMO and Non-RT RIC, for the transmission of policies and instructions; E2 interface, responsible for communication between Near-RT RIC and nodes, for real-time control and optimization; F1 interface, responsible for communication between O-DU and O-CU, for the transmission of control and user data; Open Fronthaul interface, responsible for communication between O-RU and O-DU, for the transmission of wireless signals.

[0051] In the wireless resource scheduling method of the embodiment of the present application, by collecting the current network resource usage, the network side can grasp the network resource status in real time; the non-real-time wireless access network intelligent control unit is responsible for non-real-time target setting, generates service quality target information, and clarifies the reasonable direction for subsequent resource scheduling; then, by calling the near-real-time wireless access network intelligent control unit based on the artificial intelligence model, combined with multi-dimensional data such as service quality target information, the service quality information and the configuration information of the terminal, a near-real-time first resource scheduling instruction is generated; while improving the accuracy of wireless resource scheduling, the response speed of the network side to wireless resource scheduling is improved.

[0052] Optionally, the target service includes a first service and a second service, and the priority of the second service is higher than that of the first service; when the radio resources scheduled by the resource scheduling optimization instruction exceed the total amount of radio resources or the terminal capability configuration, the method further includes:

[0053] Invoking the second control unit to send a quality of service optimization request message to the first control unit to generate adjusted quality of service target information, where the adjusted quality of service target information is adapted to the current total amount of wireless resources and terminal capability configuration;

[0054] The second control unit is called to generate a second resource scheduling instruction based on the artificial intelligence model, combined with the adjusted service quality target information, the service quality information and the terminal configuration information, and the second resource scheduling instruction is used to schedule wireless resources for the terminal.

[0055] The first service can be understood as a lower-priority service in the system that requires basic QoS guarantees. The second service can be understood as a higher-priority service in the system that has a higher priority than the first service. Its QoS requirements are prioritized and must be met first during resource allocation. The total amount of wireless resources can be understood as all wireless resources that can be allocated in the current cell. This is the core basis for determining whether the scheduling instruction is feasible. For example, when the sum of the PRBs allocated for the first and second services exceeds the "total number of PRBs that can be allocated in the cell," it means that the "resource scheduling optimization instruction exceeds the total amount of wireless resources."

[0056] The terminal configuration capabilities described above can be understood as the upper limit of wireless transmission capabilities supported by the terminal hardware and protocol layers, such as the maximum transmission rate, modulation and demodulation methods supported by the UE, or the upper limit on the number of DRBs that the UE can carry. If a scheduling instruction requires the terminal to transmit beyond its capabilities, such as if the allocated rate exceeds the upper limit of the UE hardware, this is considered "not in compliance with the terminal capability configuration."

[0057] The aforementioned QoS optimization request information can be understood as information sent by the second control unit (Near-RT RIC / xApp) to the first control unit (Non-RT RIC / rApp) requesting adjustment of the QoS target. This information may include cell radio resource information, identity information of the first or second service, the current QoS target and priority of the first or second service, or the resource scheduling strategy derived from the aforementioned artificial intelligence model. It is primarily used to provide feedback to the first control unit that "the current scheduling instruction is not feasible." The adjusted QoS target information is the new QoS target information generated by the first control unit based on the QoS optimization request information and adapted to current resources and terminal capabilities.

[0058] For example, at the beginning of the process, there is a connected UE1 in the system, which is configured with only a low-priority, quality-of-service, normal traffic service (i.e., the first service). Another connected UE2 in the system is configured with only a high-priority, high-traffic service (i.e., the second service). These UEs have not yet started any data services, i.e., have not yet started uplink or downlink data transmission.

[0059] Assuming that the system has fewer available wireless resources, the basic service quality of the first service needs to be guaranteed in this scenario. If the network side detects that the actual QoS of the above-mentioned first service does not meet the basic service quality conditions, the network side can formulate a second resource scheduling instruction to improve the actual QoS of the first service.

[0060] In some other optional implementations, the Near-RT RIC / xApp is unable to use the derivation results to generate a reasonable second resource scheduling instruction (E2 policy / E2 command), which may be due to factors such as scheduling priority or the above-mentioned artificial intelligence model derivation result that the total amount of wireless resources allocated to the first service and the second service exceeds the total amount of system resources.

[0061] In the wireless resource scheduling method of the embodiment of the present application, by clarifying that the target service includes the first and second services of different priorities, differentiated identification of service QoS requirements can be achieved, providing a basis for allocating resources according to priority, and ensuring that the QoS of high-priority services is met first; when the resource scheduling instruction exceeds the total amount or does not match the terminal capability, the second control unit sends an optimization request to the first control unit, which can timely feedback scheduling anomalies, avoid invalid instructions and waste resources, and provide an accurate basis for policy adjustment;

[0062] By generating adjusted QoS targets that adapt the total amount of resources and terminal capabilities, the QoS targets can be made more in line with actual conditions, to a certain extent avoiding the failure of guarantees caused by the targets being out of touch with reality, and providing feasible guidance for subsequent scheduling; the second control unit combines the adjusted targets, service quality information and terminal configuration, and relies on the AI ​​model to generate the second scheduling instructions, which can dynamically optimize the instructions to ensure that the instructions are executable and meet business priorities, thereby achieving reasonable QoS guarantees for different services under limited resources.

[0063] Optionally, the preset artificial intelligence model is a recurrent neural network model;

[0064] The supervision information of the input layer of the recurrent neural network model includes terminal service configuration information, the measurement data, the network resource usage, the service quality information and the service quality target information;

[0065] The supervision information of the output layer of the recurrent neural network model includes data radio bearer DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information; the DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information are used to constitute the resource scheduling instruction.

[0066] In the wireless resource scheduling method of the embodiment of the present application, the above-mentioned recurrent neural network (RNN) model is the core algorithm model for generating resource scheduling optimization instructions. The supervision information of the above-mentioned input layer can be understood as the input data vector during RNN model training and inference; the supervision information of the above-mentioned output layer can be understood as the output data vector during RNN model training and inference. The above-mentioned DRB mapping indication information is used to clarify which DRB bearer the QoS flow corresponding to the UE's target service should be mapped to, for example: mapping the QoS flow of the high-priority second service to a dedicated DRB, and mapping the QoS flow of the low-priority first service to a shared DRB, etc., to ensure that the service data is transmitted through the adapted bearer and meets the QoS requirements. The above-mentioned wireless resource allocation indication information may include: slice-level PRBs configuration information, CQI configuration information, scheduling request reconfiguration information, semi-static scheduling configuration information, etc., which can clearly specify how many PRBs are allocated to the terminal's target service, what scheduling method is used, or directly determine the resource supply amount. The above-mentioned wireless access control indication information can be understood as being used to ensure that the terminal can initiate resource requests normally according to the scheduling policy and ensure that the scheduling instructions are implemented. Specifically, it may include UE access priority configuration, scheduling request trigger threshold adjustment, etc.

[0067] Since RNN can process complex time series input vectors, such as UE configuration information, base station measurement information and QoS optimization strategy, it can effectively capture time dependencies through the loop structure. Its output layer can generate the corresponding resource scheduling result vector, optimize network resource allocation, and achieve dynamic adjustment of QoS goals. The above RNN model can be shown as follows: Figure 3 As shown:

[0068] Let z t is the net input of the hidden layer at time t, h t is the hidden layer at time t, that is, the hidden layer input passed to time t+1, o t is the unnormalized output layer input at time t, is the output layer at time t, y t is the supervision information at time t, then the mathematical expression of the RNN model is:

[0069] z t =Wh t-1 +Ux t +b

[0070] h t =f(z t )

[0071] o t =Vh t +c

[0072]

[0073] Among them, U is the weight matrix from the input layer to the hidden layer, W is the weight matrix between adjacent hidden layers, V is the weight matrix from the hidden layer to the output layer, b and c are bias vectors, f(·) is the hidden layer activation function, g(·) is the output layer activation function, which is used to output the probability distribution, L t (·) is the loss function.

[0074] In some optional implementations, the model input layer can be the basic information that Near-RT RIC / xApp needs to obtain when it detects QoS anomalies of service 1 and optimizes the scheduling strategy. The supervisory input x of the input layer at time t is t is a vector composed of information such as UE service configuration, cell resource status and QoS target obtained from UE general configuration information, base station general measurement information and the first service quality optimization strategy. The composition of the supervision input vector can be x t=[UE service configuration, UE measurement information, cell identity information, cell load, current QoS indicator, QoS optimization target], where each element can be further refined into a specific numerical value or classification label. For example, the current QoS indicator can be a vector containing multiple sub-items, such as QoS optimization target = [5QI, priority, MFBR, GFBR].

[0075] In some optional implementations, the model output layer can be set to the optimization strategy finally obtained by Near-RT RIC / xApp when it detects the QoS anomaly of service 1 and optimizes the scheduling strategy. Then the supervisory output y of the output layer at time t is t is the supervisory input x of the input layer at time t t The corresponding resource scheduling result at time t. The resource scheduling result at time t can be y t =[DRB mapping indication information of UE / service, radio resource allocation indication information of UE / service, radio access control indication information of UE / service], where each element can be further refined into a specific numerical value or classification label. For example, the radio resource allocation indication information of UE / service can be a vector containing multiple sub-items, such as the radio resource allocation indication information of UE / service = [slice-level PRBs configuration information, CQI configuration information, SR reconfiguration information, SPS configuration information].

[0076] In the wireless resource scheduling method of the embodiment of the present application, by clearly presetting the artificial intelligence model as a recurrent neural network model, a highly adaptable algorithmic basis is provided for the generation of resource scheduling strategies; by limiting the input layer supervision information of the recurrent neural network model to include terminal service configuration, measurement data, network resource usage, service quality information and service quality target information, comprehensive input of multi-dimensional data is achieved, providing sufficient data support for the model to accurately derive scheduling strategies; by clearly stating that the output layer supervision information includes DRB mapping, wireless resource allocation and wireless access control indication information, and these information constitute resource scheduling optimization instructions, direct connection between the model output and actual scheduling requirements is achieved, and the generated instructions can directly guide wireless resource scheduling, reducing efficiency loss and errors in the instruction conversion link.

[0077] Optionally, the training process of the recurrent neural network model includes:

[0078] Input supervised dataset D = {x i ,y i}(i=1,2,…,N),x i is the input vector, y i is the output vector, N is the number of samples;

[0079] Initializing the number of training rounds, bias vectors, and weight parameters of the recurrent neural network model;

[0080] The training is performed cyclically, wherein the order of each supervisory data in the supervisory data set is disrupted in each round, and a feedforward calculation is performed on each supervisory data, wherein the feedforward calculation includes: processing a hidden layer net input based on a hidden layer activation function to obtain a hidden layer output, and processing an output layer input based on an output layer activation function to obtain a predicted output;

[0081] Based on the hidden layer net input, hidden layer output, output layer input and predicted output, the parameter partial derivatives are calculated by the preset training algorithm, and the predicted output and y are quantified in combination with the loss function. i Deviation;

[0082] Based on the parameter partial derivative and the bias, the bias vector and the weight parameter are updated to obtain the trained recurrent neural network model.

[0083] The above-mentioned input vector can be understood as the supervision information of the input layer at time t, which can be composed of UE service configuration, UE measurement information, cell identity information, cell load conditions, current QoS indicators, and QoS optimization goals, and is used to provide input features for model training; the above-mentioned output vector can be understood as the supervision information of the output layer at time t, which can be composed of DRB mapping indication information, wireless resource allocation indication information, and wireless access control indication information, and is used to provide labeling results for model training.

[0084] The number of training rounds mentioned above can be understood as the number of times the model fully traverses the supervised dataset. Through multi-round training, i.e., traversing the dataset T times, the model gradually adjusts its parameters to minimize the loss function, avoiding the underfitting of parameters caused by single-round training and ensuring that the model can stably learn the mapping relationship between input and output. Shuffling the sample order can prevent the model from learning the order of sample inputs rather than the true relationship between input features and outputs.

[0085] The above-mentioned bias vector and weight parameters can be understood as the core parameters used to calculate the input-output mapping relationship in the RNN model, which can include "weight matrix U (input layer to hidden layer), W (between adjacent hidden layers), V (hidden layer to output layer)" and "bias vectors b, c". Among them, the above-mentioned weight matrices U, W and V can be used to quantify the influence of different input features on the output results. For example: U determines the contribution of "UE measurement information" in the input vector to the hidden layer output; the above-mentioned bias vectors b and c can be used to adjust the baseline value of the input data to avoid model fitting deviations caused by differences in the value range of the input features. The above-mentioned feedforward calculation refers to the forward data calculation process from the input layer to the output layer in model training.

[0086] In some optional implementations, the hidden layer activation function f(·) may be a Sigmoid function:

[0087]

[0088] Here, e is the base of the natural logarithm, and the output ranges from 0 to 1. The Sigmoid function has an S-shaped curve that can map any real value to a value between 0 and 1. It can be used as the output layer for binary classification problems or as the gated activation function in RNNs, such as LSTM and GRU.

[0089] In some optional implementations, let the output layer activation function g(·) be the Softmax function:

[0090] The Softmax function is a commonly used normalization function that can be used in the output layer of a neural network for multi-classification problems. The Softmax function converts a vector containing arbitrary real numbers into a probability distribution where each element has a value between 0 and 1 and the sum of all elements is 1.

[0091] The formula of the Softmax function is as follows:

[0092] Given an input vector z=(z1,z2,...,z k ), the output z of the Softmax function (z) The calculation is as follows:

[0093]

[0094] Where, e is the base of natural logarithm, z i is the i-th element of the input vector, and k is the dimension of the vector.

[0095] In the wireless resource scheduling method of the embodiment of the present application, by inputting a supervision data set containing input vectors and output vectors, a labeled sample of the input-to-output mapping is provided to the recurrent neural network model, laying a data foundation for the subsequent accurate generation of scheduling instructions; by disrupting the order of the supervision data and performing feedforward calculations in each training round, and combining the hidden layer and output layer activation functions to process the data, it not only avoids the deviation of the model learning sample order, but also introduces nonlinear fitting capabilities through the activation function, allowing the model to adapt to complex resource scheduling mapping relationships, thereby improving the accuracy of the predicted output.

[0096] Optionally, the preset training algorithm is a Back Propagation Through Time (BPTT) algorithm, and the loss function is a cross entropy function.

[0097] In the wireless resource scheduling method of the embodiment of the present application, the above-mentioned back propagation algorithm over time can be understood as an adapted form of the back propagation algorithm (BP) in RNN, which expands the RNN into a multi-layer feedforward neural network (FNN) according to the time dimension, so that the hidden layer of the RNN at the tth moment corresponds to the tth hidden layer of the FNN, and then calculates the partial derivatives of the loss function with respect to the parameters of each layer through back propagation, and finally realizes the iterative update of the parameters. The above-mentioned back propagation algorithm over time can use the hidden layer and output layer information at all times to calculate the partial derivatives of the loss function with respect to the parameters. The above-mentioned cross entropy is a concept used in information theory to measure the difference between two probability distributions. In machine learning, cross entropy is often used to measure the difference between two probability distributions, and is usually used as a loss function in classification problems. For two probability distributions P and Q, the calculation formula of cross entropy H(P,Q) is as follows:

[0098] H(P,Q)=-∑ X P(x)logQ(x)

[0099] Here, P(x) represents the probability of event x observed in the actual probability distribution, and Q(x) represents the probability of event x predicted by the model.

[0100] The smaller the cross entropy, the smaller the difference between the two probability distributions. In machine learning, cross entropy is often used as a loss function for classification problems, especially in neural networks. For multi-class classification problems, assuming y i represents the probability of the actual category i, and Represents the probability of category i predicted by the model, then the calculation formula of the cross entropy loss function J is as follows:

[0101]

[0102] where y is the actual category distribution (usually a one-hot encoded vector), and is the model’s predicted probability distribution. The goal of the cross entropy loss function is to minimize the difference between the model’s predicted probability distribution and the actual category distribution.

[0103] In some optional implementations, the RNN can be approximated as a multi-layer feedforward neural network (FNN) unfolded in the time dimension. The hidden layer at time t in the RNN corresponds to the tth hidden layer in the FNN. Therefore, the basic BP algorithm process of the FNN can be applied to the RNN. In this case, the BP algorithm can be called the Back Propagation Through Time (BPTT) algorithm. Because each layer in the RNN shares parameters, the calculation of the partial derivatives of each parameter requires the use of the hidden layers and output layers at all times.

[0104] Assuming that the input length of each sample is T, the loss function L at time t is t The partial derivatives of W, U, V, b, and c are shown below:

[0105]

[0106] Among them, δ t,k is the loss function L at time t t Net input z to the hidden layer at time k k The partial derivative of

[0107] And when k = t,

[0108]

[0109] When 1<=k<=t-1,

[0110]

[0111] Finally, the hidden layer activation function f(·), the output layer activation function g(·) and the loss function L t (·) is brought into the above formula, then the partial derivative L of the loss function with respect to the parameters of each layer in the neural network is t It can be rewritten as follows:

[0112]

[0113] in,

[0114]

[0115] In the wireless resource scheduling method of the embodiment of the present application, by clearly setting the preset training algorithm as the time-based backpropagation algorithm and adapting the parameter sharing characteristics of the recurrent neural network, the partial derivatives of the parameters across the time dimension can be effectively calculated to ensure the effectiveness of the model training; by clearly setting the loss function as the cross-entropy function and adapting the probability distribution output characteristics of the output layer activation function, the deviation between the predicted scheduling results and the actual results can be accurately quantified, and the model parameters can be quickly adjusted to reduce the deviation, which can improve the model training convergence efficiency and the final scheduling inference accuracy.

[0116] Optionally, the network resource usage includes the network slice information of the cell, the utilization rate of the cell's physical resource blocks, the total number of allocated resource blocks in the cell, the total number of allocable resource blocks in the cell, and the number of users accessing the cell.

[0117] In the wireless resource scheduling method of the embodiment of the present application, the network slice information of the above-mentioned cell can be understood as single network slice selection assistance information (S-NSSAI), which is an identifier for distinguishing different network slices, and can help the control unit identify whether the slice-level resource allocation is balanced, so as to avoid excessive resource occupation of a certain slice, resulting in a decrease in the QoS of other slice services. The above-mentioned physical resource block (PRB) is the basic measurement unit of wireless resources in the 5G network. The utilization rate of the physical resource block of the above-mentioned cell can be used as an indicator to judge the "cell resource load level". A high utilization rate (such as more than 80%) indicates that resources are tight and high-priority services need to be prioritized; a low utilization rate (such as less than 30%) indicates that resources are sufficient and the QoS requirements of more services can be guaranteed. The total number of resource blocks allocated to the above-mentioned cell refers to the total number of PRBs allocated to all access terminals (UE) and services in the current cell. It is used to calculate the PRB utilization rate and evaluate the remaining resources. It can be used as a direct basis for the control unit to determine whether there are sufficient resources to meet the QoS requirements of the service. The total number of allocatable resource blocks in the cell refers to the maximum number of PRBs that can be allocated within the current cell within a radio frame. The number of PRBs allocated by any scheduling instruction cannot exceed this value to avoid situations where "resource scheduling exceeds the system's carrying capacity." The number of users connected to the cell reflects the user density of the current cell and is directly related to the resource load. When the number of users is high, the average number of PRBs that can be allocated to a single user decreases, making resource contention more likely. When the number of users is low, resources are relatively abundant, and more PRBs can be allocated to user services to improve QoS. This serves as a reference for the control unit to formulate resource allocation strategies.

[0118] In the wireless resource scheduling method of the embodiment of the present application, by clarifying the network resource usage including the cell network slice information, the resource occupancy status of different slices can be grasped more accurately. By including the cell physical resource block utilization rate, the total number of allocated and allocable resource blocks, the cell resource load and remaining supply capacity can be quantitatively evaluated to avoid over-allocation or idle waste of resources, which can provide direct data support for judging "whether high-priority services need to be prioritized"; by incorporating the number of users accessing the cell, the user density can be reflected to ensure the rationality of resource scheduling under different user scales; by refining the specific dimensions of network resource usage as a whole, the resource status perception can be more comprehensive and quantitative, laying a data foundation for the first control unit to formulate scientific QoS goals and the second control unit to generate precise scheduling instructions, thereby improving the accuracy and reliability of QoS protection.

[0119] See also Figure 4 , Figure 4This is a structural diagram of a wireless resource scheduling device provided in another embodiment of the present application.

[0120] like Figure 4 As shown, the wireless resource scheduling device 400 includes:

[0121] Collection module 401, used to collect current network resource usage;

[0122] An evaluation module 402 is configured to receive measurement data of a target service during operation sent by a terminal, and evaluate the service quality of the target service based on the measurement data to obtain service quality information;

[0123] A generating module 403 is configured to synchronize the service quality information and the network resource usage to a first control unit, so as to generate service quality target information through the first control unit;

[0124] The calling module 404 is configured to call the second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and in combination with the service quality target information, the service quality information, and the terminal configuration information;

[0125] The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

[0126] Optionally, the target service includes a first service and a second service, and the priority of the second service is higher than that of the first service; when the radio resources scheduled by the resource scheduling optimization instruction exceed the total amount of radio resources or the terminal capability configuration, the radio resource scheduling apparatus 400 may further be used to:

[0127] Invoking the second control unit to send a quality of service optimization request message to the first control unit to generate adjusted quality of service target information, where the adjusted quality of service target information is adapted to the current total amount of wireless resources and terminal capability configuration;

[0128] The second control unit is called to generate a second resource scheduling instruction based on the artificial intelligence model, combined with the adjusted service quality target information, the service quality information and the terminal configuration information, and the second resource scheduling instruction is used to schedule wireless resources for the terminal.

[0129] Optionally, the preset artificial intelligence model is a recurrent neural network model;

[0130] The supervision information of the input layer of the recurrent neural network model includes terminal service configuration information, the measurement data, the network resource usage, the service quality information and the service quality target information;

[0131] The supervision information of the output layer of the recurrent neural network model includes data radio bearer DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information; the DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information are used to constitute the resource scheduling instruction.

[0132] Optionally, the training process of the recurrent neural network model includes:

[0133] Input supervised dataset D = {x i ,y i}(i=1,2,…,N),x i is the input vector, y i is the output vector, N is the number of samples;

[0134] Initializing the number of training rounds, bias vectors, and weight parameters of the recurrent neural network model;

[0135] The training is performed cyclically, wherein the order of each supervisory data in the supervisory data set is disrupted in each round, and a feedforward calculation is performed on each supervisory data, wherein the feedforward calculation includes: processing a hidden layer net input based on a hidden layer activation function to obtain a hidden layer output, and processing an output layer input based on an output layer activation function to obtain a predicted output;

[0136] Based on the hidden layer net input, hidden layer output, output layer input and predicted output, the parameter partial derivatives are calculated by the preset training algorithm, and the predicted output and y are quantified in combination with the loss function. i Deviation;

[0137] Based on the parameter partial derivative and the bias, the bias vector and the weight parameter are updated to obtain the trained recurrent neural network model.

[0138] Optionally, the preset training algorithm is a back-propagation over time algorithm, and the loss function is a cross-entropy function.

[0139] Optionally, the network resource usage includes the network slice information of the cell, the utilization rate of the cell's physical resource blocks, the total number of allocated resource blocks in the cell, the total number of allocable resource blocks in the cell, and the number of users accessing the cell.

[0140] For details, see Figure 5 As shown, an embodiment of the present application further provides an electronic device, including a bus 501 , a transceiver 502 , an antenna 503 , a bus interface 504 , a processor 505 and a memory 506 .

[0141] Processor 505, configured to:

[0142] Collect current network resource usage;

[0143] receiving measurement data of a target service during operation sent by a terminal, and evaluating the service quality of the target service based on the measurement data to obtain service quality information;

[0144] synchronizing the service quality information and the network resource usage to a first control unit, so as to generate service quality target information through the first control unit;

[0145] Invoking a second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and combining the service quality target information, the service quality information, and the configuration information of the terminal;

[0146] The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

[0147] exist Figure 5 In the embodiment, the bus architecture (represented by bus 501) is shown. Bus 501 may include any number of interconnected buses and bridges. Bus 501 links together various circuits including one or more processors represented by processor 505 and memory represented by memory 506. Bus 501 may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all well known in the art and are therefore not described further herein. Bus interface 504 provides an interface between bus 501 and transceiver 502. Transceiver 502 may be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 505 is transmitted over a wireless medium via antenna 503. Furthermore, antenna 503 receives data and transmits the data to processor 505.

[0148] The processor 505 is responsible for managing the bus 501 and general processing, and may also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 506 may be used to store data used by the processor 505 when performing operations.

[0149] Optionally, the processor 505 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or a complex programmable logic device (CPLD).

[0150] Optionally, the target service includes a first service and a second service, and the priority of the second service is higher than that of the first service; when the radio resources scheduled by the resource scheduling optimization instruction exceed the total amount of radio resources or the terminal capability configuration, the processor 505 is specifically configured to:

[0151] Invoking the second control unit to send a quality of service optimization request message to the first control unit to generate adjusted quality of service target information, where the adjusted quality of service target information is adapted to the current total amount of wireless resources and terminal capability configuration;

[0152] The second control unit is called to generate a second resource scheduling instruction based on the artificial intelligence model, combined with the adjusted service quality target information, the service quality information and the terminal configuration information, and the second resource scheduling instruction is used to schedule wireless resources for the terminal.

[0153] Optionally, the processor 505 is specifically configured to:

[0154] Merging the first data into first page data, and constructing a fourth summary tree based on the first page data;

[0155] Sending the first digest tree and the fourth digest tree to an on-chain verification module for optimistic proof verification;

[0156] Obtain a first confirmation notification fed back by the on-chain verification module.

[0157] Optionally, the preset artificial intelligence model is a recurrent neural network model;

[0158] The supervision information of the input layer of the recurrent neural network model includes terminal service configuration information, the measurement data, the network resource usage, the service quality information and the service quality target information;

[0159] The supervision information of the output layer of the recurrent neural network model includes data radio bearer DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information; the DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information are used to constitute the resource scheduling instruction.

[0160] Optionally, the training process of the recurrent neural network model includes:

[0161] Input supervised dataset D = {x i ,y i}(i=1,2,…,N),x i is the input vector, y i is the output vector, N is the number of samples;

[0162] Initializing the number of training rounds, bias vectors, and weight parameters of the recurrent neural network model;

[0163] The training is performed cyclically, wherein the order of each supervisory data in the supervisory data set is disrupted in each round, and a feedforward calculation is performed on each supervisory data, wherein the feedforward calculation includes: processing a hidden layer net input based on a hidden layer activation function to obtain a hidden layer output, and processing an output layer input based on an output layer activation function to obtain a predicted output;

[0164] Based on the hidden layer net input, hidden layer output, output layer input and predicted output, the parameter partial derivatives are calculated by the preset training algorithm, and the predicted output and y are quantified in combination with the loss function. i Deviation;

[0165] Based on the parameter partial derivative and the bias, the bias vector and the weight parameter are updated to obtain the trained recurrent neural network model.

[0166] Optionally, the preset training algorithm is a back-propagation over time algorithm, and the loss function is a cross-entropy function.

[0167] Optionally, the network resource usage includes the network slice information of the cell, the utilization rate of the cell's physical resource blocks, the total number of allocated resource blocks in the cell, the total number of allocable resource blocks in the cell, and the number of users accessing the cell.

[0168] It should be noted that the electronic device provided in the embodiment of the present application is a device capable of executing the above-mentioned wireless resource scheduling method. Therefore, all implementation methods in the above-mentioned wireless resource scheduling method embodiment are applicable to the electronic device and can achieve the same or similar beneficial effects. To avoid repetition, this embodiment will not be described in detail.

[0169] An embodiment of the present application also provides an electronic device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, the various processes of the above-mentioned wireless resource scheduling method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0170] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the various processes of the above-mentioned wireless resource scheduling method embodiment are implemented, and the same technical effects are achieved. To avoid repetition, the details are not described here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0171] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the various processes of the above-mentioned wireless resource scheduling method embodiment and can achieve the same technical effect. To avoid repetition, they will not be repeated here.

[0172] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0173] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0174] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A wireless resource scheduling method, characterized in that: The method comprises: Collect current network resource usage; receiving measurement data of a target service during operation sent by a terminal, and evaluating the service quality of the target service based on the measurement data to obtain service quality information; synchronizing the service quality information and the network resource usage to a first control unit, so as to generate service quality target information through the first control unit; Invoking a second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and combining the service quality target information, the service quality information, and the configuration information of the terminal; The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

2. The method according to claim 1, characterized in that The target service includes a first service and a second service, and the priority of the second service is higher than that of the first service; In a case where the radio resources scheduled by the resource scheduling optimization instruction exceed the total amount of radio resources or the terminal capability configuration, the method further includes: Invoking the second control unit to send a quality of service optimization request message to the first control unit to generate adjusted quality of service target information, where the adjusted quality of service target information is adapted to the current total amount of wireless resources and terminal capability configuration; The second control unit is called to generate a second resource scheduling instruction based on the artificial intelligence model, combined with the adjusted service quality target information, the service quality information and the terminal configuration information, and the second resource scheduling instruction is used to schedule wireless resources for the terminal.

3. The method according to claim 1, characterized in that The preset artificial intelligence model is a recurrent neural network model; The supervision information of the input layer of the recurrent neural network model includes terminal service configuration information, the measurement data, the network resource usage, the service quality information and the service quality target information; The supervision information of the output layer of the recurrent neural network model includes data radio bearer DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information; the DRB mapping indication information, wireless resource allocation indication information and wireless access control indication information are used to constitute the resource scheduling instruction.

4. The method according to claim 3, characterized in that The training process of the recurrent neural network model includes: Input supervised dataset D = {x i ,y i }(i=1,2,…,N),x i is the input vector, y i is the output vector, N is the number of samples; Initializing the number of training rounds, bias vectors, and weight parameters of the recurrent neural network model; The training is performed cyclically, wherein the order of each supervisory data in the supervisory data set is disrupted in each round, and a feedforward calculation is performed on each supervisory data, wherein the feedforward calculation includes: processing a hidden layer net input based on a hidden layer activation function to obtain a hidden layer output, and processing an output layer input based on an output layer activation function to obtain a predicted output; Based on the hidden layer net input, hidden layer output, output layer input and predicted output, the parameter partial derivatives are calculated by the preset training algorithm, and the predicted output and y are quantified in combination with the loss function. i Deviation; Based on the parameter partial derivative and the bias, the bias vector and the weight parameter are updated to obtain the trained recurrent neural network model.

5. The method according to claim 4, characterized in that The preset training algorithm is a back propagation over time algorithm, and the loss function is a cross entropy function.

6. The method according to any one of claims 1 to 5, characterized in that The network resource usage includes the network slicing information of the cell, the utilization rate of the cell's physical resource blocks, the total number of allocated resource blocks in the cell, the total number of allocable resource blocks in the cell, and the number of users accessing the cell.

7. A wireless resource scheduling device, characterized in that: include: Collection module, used to collect current network resource usage; An evaluation module, configured to receive measurement data of a target service during operation sent by a terminal, and evaluate the service quality of the target service based on the measurement data to obtain service quality information; a generating module, configured to synchronize the quality of service information and the network resource usage to a first control unit, so as to generate quality of service target information through the first control unit; A calling module, configured to call the second control unit to generate a first resource scheduling instruction based on a preset artificial intelligence model and in combination with the quality of service target information, the quality of service information, and the terminal and configuration information; The first control unit is a non-real-time wireless access network intelligent control unit, the second control unit is a near-real-time wireless access network intelligent control unit, and the first resource scheduling instruction is used to schedule wireless resources for the terminal.

8. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the wireless resource scheduling method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the wireless resource scheduling method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the wireless resource scheduling method according to any one of claims 1 to 6.

Images