A slice wireless resource configuration method and device and storage medium

Abstract

The application discloses a slice wireless resource configuration method and device and a storage medium, comprising: predicting and / or analyzing the state and / or performance of a network and / or a slice; and performing slice resource configuration according to the prediction result and / or analysis result. The application can solve the problem of slow convergence speed of reinforcement learning, realize dynamic slice resource configuration, match the performance requirement of the slice, improve the guarantee rate of the slice SLA, and realize dynamic self-adaptation of the slice resource based on the network environment.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method, apparatus and storage medium for slicing wireless resources. Background Technology

[0002] Network slicing, based on soft-defined networking and network function virtualization technologies, abstracts the same physical network into several customized transmission channels to empower diverse digital services with differentiated transmission needs across various industries in 5G. Among these, dynamically configuring radio access network slice resources is crucial to ensuring service SLA (Service Level Agreement) requirements.

[0003] RAN (Radio Access Network) slice resource allocation methods are mainly divided into model-based allocation methods, non-optimization algorithm-based methods, and reinforcement learning-based intelligent allocation methods. Currently, model-based optimization methods are widely used in RAN slicing problems, effectively managing small-scale networks under simplified network statistical models. Existing work primarily models the RAN slicing problem as an optimization problem, aiming to maximize network performance / operator revenue while satisfying the QoS (Quality of Service) constraints of specific slices. In addition, non-optimization model-based slice resource allocation methods mainly allocate slice resources according to the proportion of traffic throughput or distribute slice resources evenly. These methods are easy to implement but have weak adaptability. To improve the adaptability of slice resource allocation methods to complex traffic patterns and differentiated transmission requirements, reinforcement learning-based intelligent slice resource allocation methods have received widespread attention in recent years. This method mainly uses maximizing network resource utilization as a reward and the number of network resources allocated as an action, gradually increasing the cumulative reward value through closed-loop interaction with the network environment. Reinforcement learning-based slice resource allocation methods can automatically adapt to diverse service requirements and network environments to meet dynamic service transmission needs.

[0004] The shortcomings of existing technologies are that dynamic resource allocation methods based on reinforcement learning suffer from problems such as slow algorithm convergence speed due to the inability to predict network performance, and the inability to quickly provide slice allocation decisions. Summary of the Invention

[0005] This invention provides a method, apparatus, and storage medium for slicing wireless resources, which addresses the problem that dynamic resource allocation methods based on reinforcement learning suffer from slow algorithm convergence due to the inability to predict network performance status, thus failing to provide rapid slice allocation decisions.

[0006] This invention provides the following technical solutions:

[0007] A method for configuring sliced ​​radio resources includes:

[0008] Predict and / or analyze the state and / or performance of the network and / or slices;

[0009] Slice resources are configured based on prediction and / or analysis results.

[0010] During implementation, resource allocation for tiled slices is performed based on prediction and / or analysis results, including:

[0011] The first slice resource configuration is determined according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm.

[0012] A reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice.

[0013] In implementation, before determining the pre-configuration of the first slice resources based on the collected slice network data and the first slice resource configuration model, prediction of the network and / or slices includes:

[0014] Predict and / or analyze the status and / or performance of the network and / or slices based on the collected network status and / or performance data;

[0015] Predict and / or analyze slice status and / or performance based on the collected slice status and / or performance data;

[0016] Based on the predicted and / or analyzed results of slice performance and the slice performance requirement parameters, determine whether the current slice performance meets business requirements;

[0017] When the business requirements are not met, the first slice resource configuration shall be determined according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data.

[0018] During implementation, the collected slice and / or network status and / or performance data include one or a combination of the following:

[0019] Collect slice resource configuration data from base stations, collect network status and / or performance data from base stations, collect slice status and / or performance data from base stations, and obtain slice performance requirement parameters from service management and orchestration units and / or network management units and / or network optimization units.

[0020] During implementation, it further includes:

[0021] Use one or a combination of the following methods to process network status and / or performance data into sliced ​​status and / or performance data:

[0022] Mean calculation, state parameter normalization, and state parameter standardization.

[0023] In practice, based on the collected slices and / or network state and / or performance data, one or a combination of the following machine learning prediction algorithms are used to predict the network state and / or performance: LSTM algorithm, decision tree algorithm, support vector machine, random forest, CNN algorithm, and DNN algorithm.

[0024] In practice, statistical network and / or slice performance is achieved by statistically analyzing one or a combination of the following network and / or slice performance parameters:

[0025] Throughput, latency, bit error rate, and packet loss rate.

[0026] In practice, determining whether the current slice performance meets business requirements is done by judging the slice resource configuration quality based on the network service performance parameter thresholds required by the industry, thereby determining whether the current slice performance meets business requirements.

[0027] In practice, the first slice resource allocation model is obtained by training the first slice resource allocation model based on machine learning algorithms. The model is obtained by training one of the following algorithms or a combination thereof: DNN algorithm, LSTM algorithm, RNN algorithm, CNN algorithm.

[0028] In practice, the first slice resource configuration is the allocation of PRB quantity at the UE level and / or service level and / or slice level.

[0029] In practice, the reinforcement learning algorithm is one of the following algorithms or a combination thereof:

[0030] Multi-armed gambling machine algorithm, Q-learning.

[0031] In practice, the reinforcement learning algorithm is used to configure the resources of the first slice for the second slice, and the configuration action space is generated by calculating the uncertainty of the model.

[0032] In practice, the configuration action space is generated by calculating the jitter of the predicted values ​​of the deep learning model through model uncertainty calculation.

[0033] In practice, the selection of configuration actions is based on the principle of maximizing the slice SLA guarantee rate, and the slice resource configuration result is selected from the configuration action space.

[0034] During implementation, the base station configures radio resources based on the slice resource configuration results, and the configuration results are one of the following:

[0035] The number of PRB resources at the UE level and / or service level and / or slice level.

[0036] A sliced ​​wireless resource configuration device, comprising:

[0037] The processor is used to read programs from memory and execute the following procedures:

[0038] Predict and / or analyze the state and / or performance of the network and / or slices;

[0039] Slice resources are configured based on prediction and / or analysis results;

[0040] A transceiver is used to receive and send data under the control of a processor.

[0041] During implementation, resource allocation for tiled slices is performed based on prediction and / or analysis results, including:

[0042] The first slice resource configuration is determined according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm.

[0043] A reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice.

[0044] In implementation, before determining the first slice resource configuration according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data, the prediction and / or analysis of the network and / or slice status and / or performance includes:

[0045] Predict and / or analyze the status and / or performance of the network and / or slices based on the collected network status and / or performance data;

[0046] Predict and / or analyze slice status and / or performance based on the collected slice status and / or performance data;

[0047] Based on the predicted and / or analyzed results of slice performance and the slice performance requirement parameters, determine whether the current slice performance meets business requirements;

[0048] When the business requirements are not met, the first slice resource configuration shall be determined according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data.

[0049] During implementation, the collected slice and / or network status and / or performance data include one or a combination of the following:

[0050] Collect slice resource configuration data from base stations, collect network status and / or performance data from base stations, collect slice status and / or performance data from base stations, and obtain slice performance requirement parameters from service management and orchestration units and / or network management units and / or network optimization units.

[0051] During implementation, it further includes:

[0052] Use one or a combination of the following methods to process network status and / or performance data into sliced ​​status and / or performance data:

[0053] Mean calculation, state parameter normalization, and state parameter standardization.

[0054] In practice, based on the collected slices and / or network state and / or performance data, one or a combination of the following machine learning prediction algorithms are used to predict the network state and / or performance: LSTM algorithm, decision tree algorithm, support vector machine, random forest, CNN algorithm, and DNN algorithm.

[0055] In practice, statistical network and / or slice performance is achieved by statistically analyzing one or a combination of the following network and / or slice performance parameters:

[0056] Throughput, latency, bit error rate, and packet loss rate.

[0057] In practice, determining whether the current slice performance meets business requirements is done by judging the slice resource configuration quality based on the network service performance parameter thresholds required by the industry, thereby determining whether the current slice performance meets business requirements.

[0058] In practice, the first slice resource allocation model is obtained by training the first slice resource allocation model based on machine learning algorithms. The model is obtained by training one of the following algorithms or a combination thereof: DNN machine learning algorithm, LSTM algorithm, RNN algorithm, CNN algorithm.

[0059] In practice, the first slice resource pre-configuration is the allocation of PRB quantity at the UE level and / or service level and / or slice level.

[0060] In practice, the reinforcement learning algorithm is one of the following algorithms or a combination thereof:

[0061] Multi-armed gambling machine algorithm, Q-learning.

[0062] In practice, the reinforcement learning algorithm is used to configure the resources of the first slice for the second slice, and the configuration action space is generated by calculating the uncertainty of the model.

[0063] In practice, the configuration action space is generated by calculating the jitter of the predicted values ​​of the deep learning model through model uncertainty calculation.

[0064] In practice, the selection of configuration actions is based on the principle of maximizing the slice SLA guarantee rate, and the slice resource configuration result is selected from the configuration action space.

[0065] A sliced ​​wireless resource configuration device, comprising:

[0066] Prediction and analysis unit for predicting and / or analyzing the state and / or performance of the network and / or slices;

[0067] The configuration unit is used to configure slice resources based on prediction results and / or analysis results.

[0068] In implementation, the configuration unit includes:

[0069] The first configuration module is used to determine the first slice resource configuration according to the collected slice and / or network status and / or performance data according to the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm.

[0070] The second configuration module is used to perform second slice resource configuration based on the first slice resource configuration using a reinforcement learning algorithm.

[0071] In implementation, the prediction and analysis units include:

[0072] The pre-judgment module is used to predict the status and / or performance of the slices and / or networks based on the collected status and / or performance data of the slices and / or networks, and to statistically analyze the slice performance before determining the first slice resource configuration according to the first slice resource configuration model.

[0073] Based on the slice performance statistics and slice performance requirement parameters, determine whether the current slice performance meets business requirements;

[0074] When the business requirements are not met, the first configuration module is triggered to determine the first slice resource configuration according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data.

[0075] In implementation, the pre-judgment module further uses the collected slice and / or network status and / or performance data, including one or a combination of the following data:

[0076] Collect slice resource configuration data from base stations, collect network status and / or performance data from base stations, collect slice status and / or performance data from base stations, and obtain slice performance requirement parameters from service management and orchestration units and / or network management units and / or network optimization units.

[0077] In implementation, the pre-judgment module is further used to process network status and / or performance data into slice status and / or performance data using one or a combination of the following methods:

[0078] Mean calculation, state parameter normalization, and state parameter standardization.

[0079] In implementation, the pre-judgment module is further used to predict the state and / or performance of the network based on the collected slices and / or network state and / or performance data, using one or a combination of the following machine learning prediction algorithms: LSTM algorithm, decision tree algorithm, support vector machine, random forest, CNN algorithm, and DNN algorithm.

[0080] In implementation, the pre-judgment module is further used to statistically analyze the network and / or slice performance using one or a combination of the following slice and / or network performance parameters:

[0081] Throughput, latency, bit error rate, and packet loss rate.

[0082] In practice, the pre-judgment module is further used to determine the quality of slice resource configuration by judging the network service performance parameter thresholds required by the industry, thereby determining whether the current slice performance meets the business requirements.

[0083] In implementation, the first configuration module is further used to train the first slice resource configuration model based on machine learning algorithms, and the first slice resource configuration model is obtained by training based on one of the following algorithms or a combination thereof: DNN algorithm, LSTM algorithm, RNN algorithm, CNN algorithm.

[0084] In practice, the first configuration module is further used for the pre-configuration of the first slice resources, which is the allocation of PRB quantity at the UE level and / or service level and / or slice level.

[0085] In implementation, the second configuration module is further configured to use one or a combination of the following algorithms as the reinforcement learning algorithm:

[0086] Multi-armed gambling machine algorithm, Q-learning.

[0087] In practice, the second configuration module is further used to generate a configuration action space by calculating model uncertainty when configuring the resources of the first slice using a reinforcement learning algorithm to configure the resources of the second slice.

[0088] In practice, the second configuration module is further used to calculate the configuration action space by calculating the jitter of the predicted values ​​of the deep learning model when calculating the configuration action space through model uncertainty calculation.

[0089] In practice, the second configuration module is further used to select the slice resource configuration result from the configuration action space based on the principle of maximizing the slice SLA guarantee rate when selecting configuration actions.

[0090] During implementation, the base station configures radio resources based on the slice resource configuration results, and the configuration results are one of the following:

[0091] The number of PRB resources at the UE level and / or service level and / or slice level.

[0092] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described sliced ​​wireless resource configuration method.

[0093] The beneficial effects of this invention are as follows:

[0094] In the technical solution provided in the embodiments of the present invention, the state and / or performance of the network and / or slices are actively predicted and / or analyzed before slice resource configuration is performed based on the prediction and / or analysis results. That is, the network and slices are actively perceived before configuration is performed. Slice resources are dynamically adaptive based on the network environment. Therefore, slice configuration decisions can be quickly provided based on the network performance status, while matching slice performance requirements and improving slice SLA guarantee rate.

[0095] Furthermore, the slice resource allocation is divided into two stages: pre-configuration and reconfiguration. This scheme aims to reduce the computational complexity of online learning algorithms (such as reinforcement learning). In the slice resource pre-configuration stage, AI algorithms are used to mine the correspondence between long-term slice configuration experience and wireless environment and slice requirements for pre-configuration. In the slice resource reconfiguration stage, online learning methods such as reinforcement learning are used to achieve dynamic slice resource allocation. Reinforcement learning is a preferred method for achieving dynamic slice resource allocation, but its convergence speed is limited. Therefore, the pre-configuration of slice resources aims to solve the problem of slow convergence speed of reinforcement learning while ensuring that the reinforcement learning model can effectively search for the optimal solution (meeting the model's exploration needs), thus achieving dynamic slice resource allocation.

[0096] Furthermore, the slice resource configuration scheme can also be based on network status and slice performance awareness as inputs. These inputs include network status predictions, slice performance statistics, and slice performance decision results. The input of these three types of data helps the slice resource configuration module respond to short-term changes in the network environment and slice performance, while simultaneously matching slice performance requirements to improve slice SLA guarantee rates.

[0097] Furthermore, to address the issue of unpredictable network performance in scenarios where users are constantly moving, network status prediction and judgment are added, thereby enabling slice resources to dynamically adapt based on the network environment. Attached Figure Description

[0098] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0099] Figure 1 This is a schematic diagram illustrating the implementation process of the sliced ​​wireless resource configuration method in an embodiment of the present invention;

[0100] Figure 2 This is a schematic diagram of the basic process of the RAN slice resource dynamic configuration method in an embodiment of the present invention;

[0101] Figure 3 This is a schematic diagram of the RAN slice resource dynamic configuration scheme framework in an embodiment of the present invention;

[0102] Figure 4 This is a schematic diagram of the eMBB service slice traffic and SINR prediction model in an embodiment of the present invention;

[0103] Figure 5 This is a schematic diagram of the reconfiguration action space generation process in an embodiment of the present invention;

[0104] Figure 6 This is a schematic diagram of the structure of the sliced ​​wireless resource configuration device in an embodiment of the present invention. Detailed Implementation

[0105] The inventor noticed the following during the invention process:

[0106] The following problems exist in the method of configuring slicing resources in wireless access networks:

[0107] 1) Model-based configuration methods rely on prior business traffic models to establish optimization models, but the optimization models are difficult to converge and are not suitable for situations with complex and unknown traffic models and large network scale.

[0108] 2) Non-optimization model-based slice resource allocation methods allocate resources according to the proportion of service throughput or allocate slice resources equally. These methods are easy to implement, but it is difficult to dynamically perceive the differentiated transmission requirements of different services and the changing network channel environment.

[0109] 3) Reinforcement learning-based resource allocation methods exhibit strong adaptability to traffic models and diverse business requirements. However, current dynamic resource allocation methods based on reinforcement learning struggle to perceive network performance and suffer from slow model convergence. For scenarios involving continuous user movement, the network state changes with the user's movement, necessitating advance awareness of these performance changes. Furthermore, current reinforcement learning methods are insufficient to meet the requirement of rapid model convergence for scenarios with continuous user movement.

[0110] Based on this, the technical solution provided in this embodiment of the invention, for scenarios where users are constantly moving, determines the optimal slice configuration by perceiving network and slice performance, and by adopting a hierarchical intelligent configuration method of slice resource pre-configuration and slice resource reconfiguration. This solves the problems of the reinforcement learning-based RAN slice configuration method, which cannot predict the network performance status, has a slow algorithm convergence speed, and cannot quickly provide slice configuration decisions.

[0111] To achieve dynamic RAN slice resource configuration with network awareness, the specific implementation of the RAN slice resource dynamic configuration scheme based on machine learning algorithms provided in this embodiment of the invention will be described below with reference to the accompanying drawings.

[0112] Figure 1 The implementation flow diagram of the method for configuring sliced ​​wireless resources is shown in the figure, and may include:

[0113] Step 101: Predict and / or analyze the state and / or performance of the network and / or slices;

[0114] Step 102: Configure slice resources based on prediction results and / or analysis results.

[0115] During implementation, resource allocation for tiled slices is performed based on prediction and / or analysis results, including:

[0116] The first slice resource configuration is determined according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm.

[0117] A reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice.

[0118] Specifically, this solution is a hierarchical intelligent RAN slice resource configuration scheme. Aiming to reduce the computational complexity of online learning algorithms (such as reinforcement learning), this scheme comprises two stages: slice resource pre-configuration and slice resource reconfiguration. In the slice resource pre-configuration stage, AI (Artificial Intelligence) algorithms are used to mine the correspondence between long-term slice configuration experience and the wireless environment and slice requirements. In the slice resource reconfiguration stage, online learning methods such as reinforcement learning are used to achieve dynamic slice resource configuration.

[0119] In implementation, before determining the first slice resource configuration according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data, the prediction and / or analysis of the network and / or slice status and / or performance includes:

[0120] Predict and / or analyze the status and / or performance of the network and / or slices based on the collected network status and / or performance data;

[0121] Predict and / or analyze slice status and / or performance based on the collected slice status and / or performance data;

[0122] Based on the predicted and / or analyzed results of slice performance and the slice performance requirement parameters, determine whether the current slice performance meets business requirements;

[0123] When the business requirements are not met, the first slice resource configuration shall be determined according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data.

[0124] The following examples will illustrate this point.

[0125] Figure 2 A schematic diagram illustrating the basic process of dynamic configuration of RAN slice resources. Figure 3 The diagram illustrates the framework for a dynamic RAN slice resource configuration scheme. This scheme utilizes machine learning algorithms to achieve network performance awareness and dynamic slice resource configuration. The scheme is generally divided into three parts, as follows: Figure 2 As shown. The overall framework of the scheme is as follows. Figure 3 As shown.

[0126] in, Figure 2 The first part corresponds to Figure 3 The "Data Acquisition and Preprocessing Module" in the document; Figure 2 The second part corresponds to Figure 3 "Network and Slice Status and Performance Prediction and Analysis" in the document; Figure 2 The third part corresponds to Figure 3 The diagram shows the "slice resource pre-configuration module" and the "slice resource reconfiguration module". The names "data acquisition and preprocessing module", "network and slice status and performance prediction and analysis", "slice resource pre-configuration module", and "slice resource reconfiguration module" are for illustrative purposes only and do not imply that implementation can only be performed by modules with these names.

[0127] The specific implementation details are explained below:

[0128] Part 1: Data Acquisition and Preprocessing.

[0129] During implementation, the collected slice and / or network status and / or performance data include one or a combination of the following:

[0130] Collect slice resource configuration data from base stations, collect network status and / or performance data from base stations, collect slice status and / or performance data from base stations, and obtain slice performance requirement parameters from service management and orchestration units and / or network management units and / or network optimization units.

[0131] like Figure 3 Data acquisition and preprocessing module:

[0132] 1) Collect slice resource configuration data from the base station, such as the number of allocated PRBs (Physical Resource Blocks) at the slice level and / or service level and / or UE level;

[0133] 2) Collect network status and / or performance data from base stations, such as coverage data, uplink and / or downlink SINR (Signal to Interference plus Noise Ratio); reliability data, uplink and / or downlink CQI (Channel Quality Indicator) and uplink and / or downlink MCS (Modulation and Coding Scheme); network quality data, uplink and / or downlink uplink traffic, uplink and / or downlink rate, uplink and / or downlink latency, and uplink and / or downlink packet loss rate; and resource allocation data, uplink and / or downlink PRB resource utilization, etc., parameters characterizing network status and / or performance.

[0134] 3) Collect slice status and / or performance data from base stations;

[0135] 4) Obtain radio network slice SLA parameters from the Service Management and Orchestration Unit and / or Network Management Unit and / or Network Optimization Unit, such as from SMO (Service Management and Orchestration), NSSMF (Network Slice Subnet Management Function), and Non-RT RIC (Non-real-time RAN Intelligent Controller), such as the slice uplink and / or downlink dedicated PRB allocation percentage, the slice uplink and / or downlink maximum PRB allocation percentage, the per UE per slice downlink and / or uplink maximum PRB allocation percentage, the per slice indication resource sharing margin, the per slice priority value, the slice authorization configuration, the slice LCID (Logical Channel ID) configuration, the slice scheduling request, the slice semi-persistent scheduling configuration, the slice uplink and / or downlink throughput, the slice uplink and / or downlink latency, the slice uplink and / or downlink bit error rate, and the slice uplink and / or downlink packet loss rate, etc.

[0136] In practice, it may further include:

[0137] Use one or a combination of the following methods to process network state and / or performance data into sliced ​​state and / or performance data:

[0138] Mean calculation, state parameter normalization, and state parameter standardization.

[0139] Specifically, network status and / or performance data can be further preprocessed into slice status and / or performance data. Preprocessing methods include mean calculation, status parameter normalization and standardization, or implementation of UE (User Equipment) level and / or service level and / or slice level statistics.

[0140] Example:

[0141] Data acquisition and preprocessing functions can be deployed on a Non-RT RIC. A Non-RT RIC can perform the following operations:

[0142] a) Obtain RAN slice SLA targets from SMO or NSSMF;

[0143] b) Obtain network and slice status and performance data from the base station through the O1 interface, and obtain slice resource configuration information from the base station.

[0144] Part 2: Prediction and analysis of network and slice status and performance.

[0145] like Figure 3 The network and slice status and performance prediction and analysis module has the functions of network status and / or performance prediction and / or analysis, slice status and / or performance prediction and / or analysis, and slice performance determination. Specifically, it can be as follows:

[0146] 1) Network (slice) status (performance) prediction.

[0147] In practice, based on the collected network (or slice) state (or performance) data, one or a combination of the following machine learning prediction algorithms are used to predict the network (or slice) state (or performance): LSTM algorithm, decision tree algorithm, support vector machine, random forest, CNN algorithm, and DNN algorithm.

[0148] like Figure 3 The network (or slice) state (or performance) prediction function receives network (or slice) state (or performance) parameters sent by the data acquisition and preprocessing module, and predicts the network (or slice) state (or performance). Prediction methods can employ machine learning algorithms, such as LSTM (Long Short-Term Memory), decision tree algorithms, support vector machines, random forests, CNN (Convolutional Neural Network) algorithms, and DNN (Deep Neural Networks) algorithms.

[0149] Example:

[0150] The network status prediction module can use the LSTM prediction algorithm to predict traffic, channel quality, data buffering, etc. at the cell, UE, and slice levels for different types of slices and / or service types (eMBB (Enhanced Mobile Broadband) slices, mMTC (Massive Machine Type Communication) slices, and URLLC (Ultra Reliable & Low Latency Communication) slices).

[0151] When building an LSTM-based time series forecasting model, the corresponding time window can be selected for training the forecasting model based on different business types. This involves training the model using historical traffic and / or SINR data over a specific time period. During the model inference phase, the traffic and / or SINR value for the next time window is predicted from the current time window's traffic and / or SINR value. Figure 4This is a schematic diagram of the eMBB service slice traffic and SINR prediction model. If network state prediction for eMBB service slices is considered, then... Figure 4 As shown.

[0152] 2) Slice (network) status (performance) analysis.

[0153] In practice, analyzing the state or performance of a network or slice involves statistically analyzing one or a combination of the following network or slice state or performance parameters:

[0154] Coverage data includes uplink and / or downlink SINR (Signal to Interference plus Noise Ratio); reliability data includes uplink and / or downlink CQI (Channel Quality Indicator) and uplink and / or downlink MCS (Modulation and Coding Scheme); network quality data includes uplink and / or downlink uplink traffic, uplink and / or downlink rate, uplink and / or downlink latency, and uplink and / or downlink packet loss rate; resource allocation data includes uplink and / or downlink PRB resource utilization, etc., parameters characterizing the network status and / or performance.

[0155] The slice's dedicated uplink and / or downlink PRB allocation percentage, maximum uplink and / or downlink PRB allocation percentage, maximum downlink and / or uplink PRB allocation percentage per UE per slice, resource sharing margin per slice, priority value per slice, slice authorization configuration, slice LCID configuration, slice scheduling request, slice semi-persistent scheduling configuration, slice uplink and / or downlink throughput, slice uplink and / or downlink latency, slice uplink and / or downlink bit error rate, slice uplink and / or downlink packet loss rate, etc.

[0156] like Figure 3 The status and performance prediction and analysis function receives network or slice status or performance parameters sent by the data acquisition and preprocessing module, and performs network or slice status or performance statistics. Statistical methods include time window-based methods, which use time windows as time boundaries to statistically analyze network performance parameters. Slice performance statistics include throughput, latency, bit error rate, and packet loss rate.

[0157] 3) Slicing performance judgment.

[0158] In practice, determining whether the current slice performance meets business requirements is done by judging the slice resource configuration quality based on the network service performance parameter thresholds required by the industry, thereby determining whether the current slice performance meets business requirements.

[0159] like Figure 3 The slice performance judgment function determines whether the performance of the current slice meets business requirements based on slice performance statistics and slice performance requirement parameters sent by the data acquisition and preprocessing module. Judgment methods include threshold judgment, which assesses the quality of slice resource configuration based on industry-required network service performance parameter thresholds, such as maximum latency, minimum throughput, and maximum bit error rate.

[0160] Example:

[0161] Predictive and analytical capabilities for status and performance can be deployed in Near-RT RICs (Near-real-time RAN Intelligent Controllers). Near-RT RICs can perform the following operations:

[0162] a) Obtain network and slice status and performance data from the base station via the E2 interface;

[0163] b) Predict or analyze the state or performance of networks and slices using xApp;

[0164] c) Monitor network or RAN slice performance;

[0165] d) Obtain RAN slice SLA targets from SMO, NSSMF, or Non-RT RIC;

[0166] e) Receive xApp from SMO or Non-RT RIC.

[0167] To enable state and performance prediction and analysis, Non-RT RIC can perform the following operations:

[0168] a) Train AI / ML models that will be deployed in Non-RT RICs and / or near-RT RICs;

[0169] b) Supports the deployment and updating of AI / ML models to Near-RT RIC;

[0170] c) Receive xApp from SMO or Non-RT RIC;

[0171] d) Create an A1 strategy based on the A1 feedback;

[0172] e) Send A1 policy and extended information to Near-RT RIC.

[0173] 3. Dynamic configuration of slice resources.

[0174] like Figure 3The implementation of dynamic slice resource configuration is based on a hierarchical intelligent RAN slice resource configuration scheme oriented towards SLA guarantees. This scheme includes a slice resource pre-configuration module and a slice resource reconfiguration module. Reinforcement learning is the preferred method for achieving dynamic slice resource configuration, but its convergence speed is limited. Therefore, the pre-configuration of slice resources aims to address the slow convergence speed of reinforcement learning while ensuring that the reinforcement learning model can effectively search for the optimal solution (meeting the model's exploration requirements).

[0175] (I) Slice resource pre-configuration module.

[0176] 1: Training of the slice resource pre-configuration model.

[0177] In practice, the first slice resource allocation model is obtained by training the first slice resource allocation model based on machine learning algorithms. It is obtained by training one of the following algorithms or a combination thereof: DNN algorithm, LSTM algorithm, RNN (Recurrent Neural Networks) algorithm, and CNN algorithm.

[0178] like Figure 3 The slice resource pre-configuration model training function receives historical slice resource configuration information sent by the data acquisition and preprocessing module, as well as network state prediction values, slice performance statistics, and slice performance judgment results sent by the network and slice state and performance prediction and analysis module. Based on machine learning algorithms such as DNN (Deep Neural Networks), it realizes slice resource pre-configuration model training.

[0179] 2: Slice resource pre-configuration model inference.

[0180] In practice, the first slice resource pre-configuration is the allocation of PRB quantity at the UE level and / or service level and / or slice level.

[0181] like Figure 3 The slice resource pre-configuration model inference function receives network status prediction values, slice performance statistics, and slice performance decision results sent by the network and slice status and performance prediction and analysis. Based on the slice resource pre-configuration model, it performs slice resource pre-configuration model inference to obtain slice resource pre-configuration results. The pre-configuration results can be UE-level and / or service-level and / or slice-level PRB quantity allocation.

[0182] Example:

[0183] In the DNN-based slice pre-configuration model, a large amount of slice configuration experience is first collected to form an experience dataset. In this dataset, each set of configuration experience data consists of network state values, slice resource configuration results, and slice performance judgment results. Network state can be the traffic value within a sliding window or the average channel quality index (CQI). The duration of each slice window can be set according to service requirements, such as 1 second, with the sliding window time being the first 10 seconds of the current slice window. Slice resource configuration results refer to the configuration combination of the number of physical resource blocks (PRBs) used for different slices within the current slice window. One calculation scheme for slice performance judgment results is the product of the performance (throughput, latency, bit error rate, packet loss rate) SLA guarantee rate of each slice within the current slice window under the current network state and slice resource configuration. The slice SLA guarantee rate is defined as the ratio of the number of data packets generated by the service within the current slice window that simultaneously meet latency and user rate requirements to the total number of data packets generated within the current slice window, with a value range of [0,1].

[0184] The structure of a DNN mainly includes: an input layer, n fully connected layers (F1, F2, ..., Fn), and an output layer. The input layer takes into account the network state values ​​and slice performance decisions corresponding to configuration experience that meets slice performance requirements. The label data required for DNN model training consists of slice resource configuration results that meet slice performance requirements. The input layer is connected to the fully connected layer F1; all neurons in the fully connected layer F1 are fully connected to the neurons in the next fully connected layer F2; the output is the information on the combination of PRB resource quantities for each slice, i.e., the proportion of each slice's PRB allocation to the total upstream PRB.

[0185] (ii) Slice resource reconfiguration module.

[0186] In practice, the reinforcement learning algorithm is one of the following algorithms or a combination thereof:

[0187] Multi-armed gambling machine algorithm, Q-learning.

[0188] Reinforcement learning algorithms, such as the multi-armed gambling machine algorithm and Q-learning, can be used to reconfigure sliced ​​resources, as follows:

[0189] 1: Reconfigure motion space generation.

[0190] In practice, the reinforcement learning algorithm is used to configure the resources of the first slice for the second slice, and the configuration action space is generated by calculating the uncertainty of the model.

[0191] In practice, the configuration action space is generated by calculating the jitter of the predicted values ​​of the deep learning model through model uncertainty calculation.

[0192] like Figure 3 The slice resource reconfiguration module receives the slice resource preconfiguration results from the slice resource preconfiguration module and generates the reconfiguration action space through model uncertainty calculation. Specifically, this can involve calculating the jitter of the deep learning model's predicted values. In the action space generation module design, based on the slice resource preconfiguration results, the mean and variance of the preconfigured values ​​are obtained by calculating the uncertainty of the preconfigured model, and finally, the reconfiguration action space is generated. This reconfiguration action space significantly reduces the size of the available space, which is beneficial for improving the convergence speed of reinforcement learning, and also meets the need for reinforcement learning to explore the optimal solution.

[0193] Example:

[0194] Figure 5 The diagram illustrates the process of generating the reconfiguration action space. Dropout is used to calculate the uncertainty of the deep learning model output. By testing the outputs of multiple DNNs, the variance and mean of the number of PRB resource configuration actions output by the trained DNN on each slice are calculated. Furthermore, it is assumed that the variables in each dimension of the PRB slice resource configuration actions follow a normal distribution. Finally, based on the distribution of slice resource configuration actions, the slice resource pre-configuration action space is output. This DNN-based slice resource pre-configuration module aims to provide a slice resource reconfiguration action space based on effective historical slice resource configuration experience, addressing QoS guarantees such as latency and user rate. This slice resource reconfiguration action space significantly reduces the selection space for slice resource configuration, facilitating rapid convergence and decision-making in reinforcement learning.

[0195] 2: Action selection.

[0196] In practice, the reconfiguration action selection is based on the principle of maximizing the slice SLA guarantee rate, and the final slice resource configuration result is selected from the reconfiguration action space.

[0197] The action selection function selects the final slice resource configuration result from the action selection space based on the principle of maximizing the slice SLA guarantee rate. Specifically, the selection method can be to construct a joint SLA guarantee rate of indicators such as service SLA latency, SLA throughput, and SLA bit error rate as the model reward.

[0198] Example:

[0199] In the slice resource reconfiguration algorithm based on multi-armed gambling machines, the slice resource reconfiguration action space serves as the input to the dynamic slice resource configuration module based on MAB (Multi-Armed Block). The MAB algorithm employs a greedy strategy to select slice PRB (Programmable Node Boundary) configuration actions. Within the slice resource reconfiguration action space, it selects the configuration action with the highest slice SLA (Segment Level Agreement) guarantee rate, and finally outputs a set of slice PRB configuration combinations. Subsequently, the network environment provides feedback on information such as the slice SLA guarantee rate, spectral efficiency, and packet loss rate within this slice window, which the MAB uses to calculate the reward.

[0200] Example:

[0201] Dynamic configuration of slice resources can be deployed in Near-RT RICs. Near-RT RICs can perform the following operations:

[0202] a) Based on the O1 configuration, A1 policy, and E2 report, perform optimized RAN (E2) operations to meet RAN slicing requirements;

[0203] b) Supports the interpretation and execution of Non-RT RIC strategies;

[0204] c) Implement dynamic configuration of slice resources through xApp;

[0205] d) Send slice resource configuration actions to the base station through the E2 interface.

[0206] To enable dynamic configuration of slice resources, the RAN can perform the following operations:

[0207] a) Supports performance measurement of specific slices via O1;

[0208] b) Support for performance reporting of specific slices via E2;

[0209] c) Supports slice protection actions, such as slice-aware resource allocation and priority allocation.

[0210] Based on the same inventive concept, this invention also provides a sliced ​​wireless resource configuration device and a computer-readable storage medium. Since the principle of these devices in solving the problem is similar to that of the sliced ​​wireless resource configuration method, the implementation of these devices can refer to the implementation of the method, and the repeated parts will not be described again.

[0211] When implementing the technical solutions provided in the embodiments of the present invention, they can be implemented in the following manner.

[0212] Figure 6 The schematic diagram of the sliced ​​wireless resource configuration device is shown in the figure. The device includes:

[0213] Processor 600 is used to read the program from memory 620 and execute the following procedures:

[0214] Predict and / or analyze the state and / or performance of the network and / or slices;

[0215] Slice resources are configured based on prediction and / or analysis results;

[0216] Transceiver 610 is used to receive and send data under the control of processor 600.

[0217] During implementation, resource allocation for tiled slices is performed based on prediction and / or analysis results, including:

[0218] The first slice resource configuration is determined according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm.

[0219] A reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice.

[0220] In implementation, before determining the first slice resource configuration according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data, the prediction and / or analysis of the network and / or slice status and / or performance includes:

[0221] Predict and / or analyze the status and / or performance of the network and / or slices based on the collected network status and / or performance data;

[0222] Predict and / or analyze slice status and / or performance based on the collected slice status and / or performance data;

[0223] Based on the predicted and / or analyzed results of slice performance and the slice performance requirement parameters, determine whether the current slice performance meets business requirements;

[0224] When the business requirements are not met, the first slice resource configuration shall be determined according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data.

[0225] During implementation, the collected slice and / or network status and / or performance data include one or a combination of the following:

[0226] Collect slice resource configuration data from base stations, collect network status and / or performance data from base stations, collect slice status and / or performance data from base stations, and obtain slice performance requirement parameters from service management and orchestration units and / or network management units and / or network optimization units.

[0227] During implementation, it further includes:

[0228] Use one or a combination of the following methods to process network status and / or performance data into sliced ​​status and / or performance data:

[0229] Mean calculation, state parameter normalization, and state parameter standardization.

[0230] In practice, based on the collected slices and / or network state and / or performance data, one or a combination of the following machine learning prediction algorithms are used to predict the network state and / or performance: LSTM algorithm, decision tree algorithm, support vector machine, random forest, CNN algorithm, and DNN algorithm.

[0231] In practice, statistical network and / or slice performance is achieved by statistically analyzing one or a combination of the following network and / or slice performance parameters:

[0232] Throughput, latency, bit error rate, and packet loss rate.

[0233] In practice, determining whether the current slice performance meets business requirements is done by judging the slice resource configuration quality based on the network service performance parameter thresholds required by the industry, thereby determining whether the current slice performance meets business requirements.

[0234] In practice, the first slice resource allocation model is obtained by training the first slice resource allocation model based on machine learning algorithms. The model is obtained by training one of the following algorithms or a combination thereof: DNN algorithm, LSTM algorithm, RNN algorithm, CNN algorithm.

[0235] In practice, the first slice resource pre-configuration is the allocation of PRB quantity at the UE level and / or service level and / or slice level.

[0236] In practice, the reinforcement learning algorithm is one of the following algorithms or a combination thereof:

[0237] Multi-armed gambling machine algorithm, Q-learning.

[0238] In practice, the reinforcement learning algorithm is used to configure the resources of the first slice for the second slice, and the configuration action space is generated by calculating the uncertainty of the model.

[0239] In practice, the configuration action space is generated by calculating the jitter of the predicted values ​​of the deep learning model through model uncertainty calculation.

[0240] In practice, the selection of configuration actions is based on the principle of maximizing the slice SLA guarantee rate, and the slice resource configuration result is selected from the configuration action space.

[0241] During implementation, the base station configures radio resources based on the slice resource configuration results, and the configuration results are one of the following:

[0242] The number of PRB resources at the UE level and / or service level and / or slice level.

[0243] Among them, Figure 6 In this context, the bus architecture may include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 600) and memory (memory 620). The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface. Transceiver 610 may be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. Processor 600 is responsible for managing the bus architecture and general processing, and memory 620 may store data used by processor 600 during operation.

[0244] This invention also provides a sliced ​​wireless resource configuration device, comprising:

[0245] Prediction and analysis unit for predicting and / or analyzing the state and / or performance of the network and / or slices;

[0246] The configuration unit is used to configure slice resources based on prediction results and / or analysis results.

[0247] In implementation, the configuration unit includes:

[0248] The first configuration module is used to determine the pre-configuration of the first slice resources according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm.

[0249] The second configuration module is used to perform second slice resource configuration based on the first slice resource configuration using a reinforcement learning algorithm.

[0250] In implementation, the prediction and analysis units include:

[0251] The pre-judgment module is used to predict the status of the slices and / or networks and to statistically analyze the slice performance before determining the first slice resource configuration according to the first slice resource configuration model based on the collected slice and / or network status and / or network status and / or performance data.

[0252] Based on the slice performance statistics and slice performance requirement parameters, determine whether the current slice performance meets business requirements;

[0253] When the business requirements are not met, the first configuration module is triggered to determine the first slice resource configuration according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data.

[0254] In implementation, the pre-judgment module further uses the collected slice and / or network status and / or performance data, including one or a combination of the following data:

[0255] Collect slice resource configuration data from base stations, collect network status and / or performance data from base stations, collect slice status and / or performance data from base stations, and obtain slice performance requirement parameters from service management and orchestration units and / or network management units and / or network optimization units.

[0256] In implementation, the pre-judgment module is further used to process network status and / or performance data into slice status and / or performance data using one or a combination of the following methods:

[0257] Mean calculation, state parameter normalization, and state parameter standardization.

[0258] In implementation, the pre-judgment module is further used to predict the state and / or performance of the network based on the collected slices and / or network state and / or performance data, using one or a combination of the following machine learning prediction algorithms: LSTM algorithm, decision tree algorithm, support vector machine, random forest, CNN algorithm, and DNN algorithm.

[0259] In implementation, the pre-judgment module is further used to statistically analyze the network and / or slice performance using one or a combination of the following network and / or slice performance parameters:

[0260] Throughput, latency, bit error rate, and packet loss rate.

[0261] In practice, the pre-judgment module is further used to determine the quality of slice resource configuration by judging the network service performance parameter thresholds required by the industry, thereby determining whether the current slice performance meets the business requirements.

[0262] In implementation, the first configuration module is further used to train the first slice resource configuration model based on machine learning algorithms, and the first slice resource configuration model is obtained by training based on one of the following algorithms or a combination thereof: DNN algorithm, LSTM algorithm, RNN algorithm, CNN algorithm.

[0263] In practice, the first configuration module is further used for the pre-configuration of the first slice resources, which is the allocation of PRB quantity at the UE level and / or service level and / or slice level.

[0264] In implementation, the second configuration module is further configured to use one or a combination of the following algorithms as the reinforcement learning algorithm:

[0265] Multi-armed gambling machine algorithm, Q-learning.

[0266] In practice, the second configuration module is further used to generate a configuration action space by calculating model uncertainty when configuring the resources of the first slice using a reinforcement learning algorithm to configure the resources of the second slice.

[0267] In practice, the second configuration module is further used to calculate the configuration action space by calculating the jitter of the predicted values ​​of the deep learning model when calculating the configuration action space through model uncertainty calculation.

[0268] In practice, the second configuration module is further used to select the slice resource configuration result from the configuration action space based on the principle of maximizing the slice SLA guarantee rate when selecting configuration actions.

[0269] During implementation, the base station configures radio resources based on the slice resource configuration results, and the configuration results are one of the following:

[0270] The number of PRB resources at the UE level and / or service level and / or slice level.

[0271] For ease of description, the various parts of the device described above are divided into modules or units according to their functions. Of course, in implementing this invention, the functions of each module or unit can be implemented in one or more software or hardware components.

[0272] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described sliced ​​wireless resource configuration method.

[0273] For specific implementation details, please refer to the implementation of the sliced ​​wireless resource configuration method.

[0274] In summary, the technical solution provided by this invention is a hierarchical intelligent RAN slice resource configuration scheme. This scheme aims to reduce the computational complexity of online learning (such as reinforcement learning) algorithms and includes two stages: slice resource pre-configuration and slice resource reconfiguration. In the slice resource pre-configuration stage, AI algorithms are used to mine the correspondence between long-term slice configuration experience and the wireless environment and slice requirements. In the slice resource reconfiguration stage, online learning methods such as reinforcement learning are used to achieve dynamic slice resource configuration.

[0275] Furthermore, the input to the slice resource configuration scheme based on network status and slice performance awareness includes network status predictions, slice performance statistics, and slice performance decision results. The input of these three types of data helps the slice resource configuration module respond to short-term changes in the network environment and slice performance, while simultaneously matching slice performance requirements to improve slice SLA guarantee rates.

[0276] Furthermore, the design of the action space reconfiguration module generates the action space reconfiguration of slice resources based on the preconfiguration results of slice resources. This solves the problem of slow convergence speed of reinforcement learning while ensuring that the reinforcement learning model can effectively search for the optimal solution (meeting the exploration needs of the model).

[0277] This solution can:

[0278] For scenarios where users are constantly moving, a network status prediction module is added to enable slice resources to dynamically adapt based on the network environment;

[0279] A hierarchical intelligent configuration scheme that includes slice resource pre-configuration and slice resource reconfiguration is adopted to reduce the problems of slow algorithm convergence speed and inability to quickly provide slice configuration decisions in reinforcement learning-based slice resource configuration methods.

[0280] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0281] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0282] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0283] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0284] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for configuring sliced ​​wireless resources, characterized in that, include: Predict and / or analyze the state and / or performance of the network and / or slices; Slice resources are configured based on prediction and / or analysis results; Slice resource allocation is performed based on prediction and / or analysis results, including: The first slice resource configuration is determined according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm. A reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice.

2. The method as described in claim 1, characterized in that, Before determining the first slice resource configuration according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data, the prediction and / or analysis of the network and / or slice status and / or performance includes: Predict and / or analyze the status and / or performance of the network and / or slices based on the collected network status and / or performance data; Predict and / or analyze slice status and / or performance based on the collected slice status and / or performance data; Based on the predicted and / or analyzed results of slice performance and the slice performance requirement parameters, determine whether the current slice performance meets business requirements; When the business requirements are not met, the first slice resource configuration shall be determined according to the first slice resource configuration model based on the collected slice and / or network status and / or performance data.

3. The method as described in claim 1, characterized in that, The collected slice and / or network status and / or performance data includes one or a combination of the following: Collect slice resource configuration data from base stations, collect network status and / or performance data from base stations, collect slice status and / or performance data from base stations, and obtain slice performance requirement parameters from service management and orchestration units and / or network management units and / or network optimization units.

4. The method as described in claim 3, characterized in that, Further includes: Use one or a combination of the following methods to process network status and / or performance data into sliced ​​status and / or performance data: Mean calculation, state parameter normalization, and state parameter standardization.

5. The method as described in claim 2, characterized in that, Based on the collected slices and / or network state and / or performance data, one or a combination of the following machine learning algorithms is used to predict the state and / or performance of the network: Long Short-Term Memory (LSTM) algorithm, Decision Tree algorithm, Support Vector Machine, Random Forest, Convolutional Neural Network (CNN) algorithm, and Deep Neural Network (DNN) algorithm.

6. The method as described in claim 2, characterized in that, Statistical network and / or slice performance is defined by statistically analyzing one or a combination of the following network and / or slice performance parameters: Throughput, latency, bit error rate, and packet loss rate.

7. The method as described in claim 6, characterized in that, Determining whether the current slice performance meets business requirements is done by judging the slice resource configuration quality based on the network service performance parameter thresholds required by the industry, thereby determining whether the current slice performance meets business requirements.

8. The method according to any one of claims 1 to 7, characterized in that, The first slice resource allocation model is obtained by training the first slice resource allocation model based on the machine learning algorithm. It is obtained by training based on one or a combination of the following algorithms: DNN machine learning algorithm, LSTM algorithm, recurrent neural network (RNN) algorithm, and CNN algorithm.

9. The method according to any one of claims 1 to 7, characterized in that, The first slice resource configuration is the allocation of the number of physical resource blocks (PRBs) at the user equipment (UE) level and / or service level and / or slice level.

10. The method according to any one of claims 1 to 7, characterized in that, The reinforcement learning algorithm is one of the following algorithms or a combination thereof: Multi-armed gambling machine algorithm, Q-learning.

11. The method according to any one of claims 1 to 7, characterized in that, The reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice. The configuration action space is generated by calculating the uncertainty of the model.

12. The method as described in claim 11, characterized in that, The configuration action space is generated by calculating the jitter of the predicted values ​​of the deep learning model.

13. The method as described in claim 12, characterized in that, The selection of configuration actions is based on the principle of maximizing the Segment Service Level Agreement (SLA) guarantee rate, and the result of the slice resource configuration is selected from the configuration action space.

14. The method as described in claim 1, characterized in that, The base station configures radio resources based on the slice resource configuration results, and the configuration results are one of the following: The number of PRB resources at the UE level and / or service level and / or slice level.

15. A slicing wireless resource configuration device, characterized in that, include: The processor is used to read programs from memory and execute the following procedures: Predict and / or analyze the state and / or performance of the network and / or slices; Slice resources are configured based on prediction and / or analysis results; A transceiver is used to receive and send data under the control of a processor; Slice resource allocation is performed based on prediction and / or analysis results, including: The first slice resource configuration is determined according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm. A reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice.

16. A slicing wireless resource configuration device, characterized in that, include: Prediction and / or analysis unit for predicting and / or analyzing the state and / or performance of the network and / or slices; A configuration unit is used to configure slice resources based on prediction and / or analysis results; Slice resource allocation is performed based on prediction and / or analysis results, including: The first slice resource configuration is determined according to the collected slice and / or network status and / or performance data and the first slice resource configuration model. The first slice resource configuration model is obtained by training the first slice resource configuration model based on the historical slice and / or network status and / or performance data and a machine learning algorithm. A reinforcement learning algorithm is used to configure the resources of the second slice after configuring the resources of the first slice.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the sliced ​​wireless resource configuration method according to any one of claims 1 to 14.

Images