CN116887327A - A QoS prediction method and device - Google Patents

Abstract

The invention discloses a QoS prediction method and a QoS prediction device; the QoS prediction model is deployed into a network simulator based on network equipment, and the network simulator is used for constructing a digital twin network mapped with a physical network service flow; the network device loads a QoS prediction model in a network simulator for a network environment of the digital twin network. The method simulates the physical network environment based on the digital twin network in the network simulator, is favorable for realizing various changes based on the digital twin network simulating the physical network environment, obtains diversified QoS prediction results corresponding to the service flows under different physical network environments, is favorable for carrying out strategy adjustment aiming at the physical network environment of each situation in time, and improves the service quality of the end-to-end service flow.

Description

A QoS prediction method and device

Technical field

The present invention relates to the field of communications, and in particular to a quality of service (QoS) prediction method and device.

Background technique

Currently, the service quality of the network is predicted based on the current physical network environment, that is, QoS prediction is performed based on the current specific physical network environment. In order to ensure that network services maintain stable operation, the physical network environment cannot be changed at will for QoS prediction, so the QoS prediction results are relatively simple. However, there are many potential changes in the physical network environment in the future. For example, the network bandwidth will continue to decrease or the signal-to-noise ratio will deteriorate in the future. How to predict QoS for diverse network environments is a technical problem that needs to be solved.

Contents of the invention

The present invention provides a QoS prediction method and device, which is beneficial to realizing QoS prediction for diversified network environments, and thereby is beneficial to improving the service quality of end-to-end business flows.

In a first aspect, the present invention provides a QoS prediction method. The following takes a network device as the execution subject of the method as an example. The method includes: the network device deploys the QoS prediction model into a network simulator, and the network simulator is used to construct The digital twin network is mapped with the physical network business flow; the network device loads the QoS prediction model in the network simulator according to the network environment of the digital twin network.

In an optional implementation, the network device loads the QoS prediction model in the network simulator for the network environment of the digital twin network, including: the network device sets the network environment of the digital twin network in the network simulator; the network device targets the digital twin network. For each of the multiple network environments of the twin network, a QoS prediction model is loaded.

In an optional implementation, the QoS prediction model is obtained based on federated learning between network equipment and network data analytics function (NWDAF) network elements.

In an optional implementation, the method further includes: the network device determines a first prediction model corresponding to the QoS parameter based on the air interface wireless parameters of the network device; the network device sends the model parameters corresponding to the first prediction model to the NWDAF network element. ; The network device receives model parameters corresponding to the second prediction model from the NWDAF network element. The second prediction model is determined by the NWDAF network element based on the model parameters corresponding to the first prediction model from each network device in the plurality of network devices.

If the first indication information is received, the network device determines the QoS prediction model based on the model parameters corresponding to the second prediction model; otherwise, the network device updates the model corresponding to the first prediction model based on the model parameters corresponding to the second prediction model and the air interface wireless parameters. parameters, and sends updated model parameters corresponding to the first prediction model to the NWDAF network element. The first instruction information is used to instruct the network device to stop updating the model.

In an optional implementation, the method further includes: the network device sends network parameters corresponding to the network environment to a policy control function (PCF) network element, and QoS prediction results obtained by loading the QoS prediction model for the network environment; PCF Network elements are used to adjust physical network policies based on network parameters corresponding to the network environment and QoS prediction results.

In a second aspect, the present invention also provides a QoS prediction device, which includes:

The deployment unit is used to deploy the QoS prediction model into the network simulator, and the network simulator is used to build a digital twin network that maps the physical network service flow.

The prediction unit is used to load the QoS prediction model in the network simulator for the network environment of the digital twin network.

In an optional implementation, the prediction unit loads the QoS prediction model in the network simulator for the network environment of the digital twin network, specifically for: setting the network environment of the digital twin network in the network simulator; targeting the digital twin network For each of the multiple network environments, load the QoS prediction model.

In an optional implementation, the QoS prediction model is obtained by the QoS prediction device and the NWDAF network element based on federated learning.

In an optional implementation, the device further includes:

A determining unit configured to determine a first prediction model corresponding to the QoS parameter based on the air interface wireless parameters of the network device.

A sending unit, configured to send model parameters corresponding to the first prediction model to the NWDAF network element.

The receiving unit is configured to receive model parameters corresponding to the second prediction model from the NWDAF network element. The second prediction model is determined by the NWDAF network element based on the model parameters corresponding to the first prediction model from each device in the plurality of devices.

The determining unit is also configured to: when the receiving unit receives the first indication information, determine the QoS prediction model based on the model parameters corresponding to the second prediction model; otherwise, update the QoS prediction model based on the model parameters corresponding to the second prediction model and the air interface wireless parameters. A model parameter corresponding to a prediction model. The first instruction information is used to instruct the network device to stop updating the model.

The sending unit is also configured to send the updated model parameters corresponding to the first prediction model to the NWDAF network element after the determination unit updates the model parameters corresponding to the first prediction model.

In an optional implementation, the sending unit is also used to send network parameters corresponding to the network environment and QoS prediction results obtained by loading the QoS prediction model for the network environment to the PCF network element; the PCF network element is used to send network parameters corresponding to the network environment based on the network environment. Parameters and QoS prediction results adjust the physical network policy.

In a third aspect, the present invention also provides a communication device, including a memory and a processor. The memory is used to store a computer program. The computer program includes program instructions. The processor is configured to call the program instructions to cause the communication device to execute the first aspect. method described.

In a fourth aspect, the present invention also provides a computer-readable storage medium. Computer-readable instructions are stored in the computer-readable storage medium. When the computer-readable instructions are run on the communication device, the communication device causes the communication device to execute the first aspect. method described.

Compared with the prior art, the beneficial effects of the present invention are:

In the present invention, the network device can simulate the physical network environment based on the digital twin network in the network simulator, and perform QoS prediction on the network environment of the digital twin network in the network simulator. This method is conducive to simulating various changes in the physical network environment based on the digital twin network, which is conducive to QoS prediction for various situations of the network environment, and is conducive to timely adjustment of strategies for the physical network environment in each situation, improving end-to-end The service quality of end business flows.

Description of the drawings

Figure 1 is a schematic diagram of a network architecture provided by an embodiment of the present invention.

Figure 2 is a schematic diagram of federated learning provided by an embodiment of the present invention.

Figure 3 is a schematic diagram of a digital twin network architecture provided by an embodiment of the present invention.

Figure 4 is a schematic flowchart of a QoS prediction method provided by an embodiment of the present invention.

Figure 5 is a schematic diagram of another QoS prediction method provided by an embodiment of the present invention.

Figure 6 is a schematic diagram of a QoS prediction method framework provided by an embodiment of the present invention.

Figure 7 is a schematic structural diagram of a QoS prediction device provided by an embodiment of the present invention.

Figure 8 is a schematic structural diagram of a communication device provided by an embodiment of the present invention.

Detailed ways

The present invention will be described below with reference to the drawings in the present invention.

First, in order to better understand the present invention, a communication system to which the present invention is applicable is described.

The technical solution of the present invention can be applied to various communication systems, such as: long term evolution (LTE) system, fourth generation mobile communication technology (4th generation, 4G) system, new radio technology (newradio, NR) system, The fifth generation mobile communication technology (5th generation mobile networks, 5G) system, and with the continuous development of communication technology, the technical solution of the present invention can also be used in subsequent evolved communication systems, such as the sixth generation mobile communication technology (6th generation mobile networks) , 6G) system, seventh generation mobile communication technology (7th generation mobile networks, 7G) system, etc.

Please refer to Figure 1. Figure 1 is a schematic diagram of a network architecture provided by the present invention. The network architecture is an independent networking architecture of the 5G system. The technical solution of the present invention can be applied to this network architecture. The network architecture includes: terminal equipment, radio access network (RAN), 5G core network (5G core, 5GC) and data network (datanetwork, DN). Among them, the 5G core network can be divided into control plane function (CPF) and user plane function (UPF). CPF includes NWDAF, network exposure function (NEF), network repository function (NRF), PCF, unified data management function (UDM), application function (AF), and authentication server Function (authentication server function, AUSF), access and mobility management function (AMF), session management function (SMF). In addition, in the network architecture shown in Figure 1, N1, N2, N3, N4 and N6 are 5G network connection channels; terminal equipment and RAN equipment are connected through the air interface (over the air, OTA), and other equipment/network The connections between elements are all wired connections. Therefore, changes in the wireless parameters of the air interface have a great impact on the end-to-end service flow QoS. The local data of the RAN equipment is closely related to the 5G service quality that the current network can provide.

In the embodiment, NWDAF is a network function of the core network in the 5G network. NWDAF network elements (that is, network elements with NWDAF) can be used to collect and analyze data on various network functions and elements, such as collecting and analyzing data on user plane functions and session management functions. NWDAF network elements can also be responsible for providing real-time network intelligence for various network function network elements, such as traffic optimization, network slicing, QoS management, etc.

5GC is the core network of 5G network and one of the infrastructure supporting 5G wireless access technology. 5GC includes a series of network functions, such as user data management, session management, security functions, etc. 5GC can achieve high-speed data transmission, low latency, large-scale connections and more secure communications, thereby supporting various new application scenarios and services. .

The next generation base station (the next generation node B, gNB) is a RAN device that is connected to the core network and provides wireless access. gNB can support higher bandwidth and lower latency, while also supporting more connections and device types, including IoT devices, smart home devices and vehicles. gNB also adopts a more flexible architecture that can be deployed and configured according to network requirements to support different application scenarios.

PCF network elements (that is, network elements with PCF) can be used to use a unified policy framework to manage network behavior and coordinate user information in unified data repository (unified data repository, UDR) network elements (that is, network elements with UDR) to implement relevant strategies.

Air interface OTA is used to receive and send data based on radio waves in communication systems. Air interface OTA can usually be used to describe the process between sending and receiving radio signals, that is, the process of wireless signals being transmitted from terminal equipment to base stations or from base stations to terminal equipment.

Terminal equipment may also be called user equipment (UE). The terminal device can be a mobile phone (mobile phone), tablet computer (Pad), computer with wireless transceiver function, virtual reality (VR) terminal, augmented reality (AR) terminal, industrial control (industrial control) wireless terminals, wireless terminals in self-driving, wireless terminals in remote medical, wireless terminals in smart grid, wireless terminals in transportation safety, smart Wireless terminals in smart cities, etc.

The network device may be a device with a wireless transceiver function or a chip that can be disposed on the device. The network device may be a RAN device. For example, network equipment includes but is not limited to: evolved node B (evolved node B, eNB), gNB, radio network controller (radio network controller, RNC), node B (Node B, NB), network equipment controller (base station controller (BSC), network equipment transceiver station (BTS), home network equipment (for example, home evolved Node B, or home Node B, HNB), baseband unit (BBU), wireless relay node, Wireless backhaul nodes, transmission points (transmission and reception point, TRP or transmission point, TP), etc., can also be equipment used in 4G, 5G, 6G and other systems. There are no restrictions here.

In order to facilitate understanding of the embodiments disclosed in the present invention, the scenarios in the embodiments disclosed in the present invention are described by taking the scenario of 5G network as an example. It should be noted that the solutions in the embodiments disclosed in the present invention can also be applied to other wireless communication networks. , the corresponding name can also be replaced with the name of the corresponding function in other wireless communication networks.

The relevant concepts involved in the present invention are briefly introduced below.

1. Service quality

Quality of Service (i.e. QoS) can be used to provide end-to-end service quality guarantees for different application scenarios, different service types and different applications, thereby providing users with a better service experience. For example, for real-time communication applications such as real-time video streaming and voice calls, low-latency and high-bandwidth quality of service are required to ensure communication quality and user experience. For another example, for applications such as large-scale Internet of Things applications and cloud computing, high reliability and high security quality of service are required to ensure the reliability and security of data transmission. Therefore, predicting service quality in a dynamically changing network and making strategic adjustments in advance based on possible changes in service quality can more effectively ensure service quality and user experience.

In 5G networks, the main characteristics of QoS include: bandwidth, delay, reliability, security, and flexibility. Among them, 5G network provides high bandwidth and can meet the needs of high-speed data transmission. 5G networks provide low-latency services and enable high-speed real-time communications. 5G networks provide high-reliability services and enable high-reliability data transmission. 5G networks provide high-security services that can protect user data privacy and network security. 5G networks provide flexible QoS service quality that can be adjusted and optimized according to application scenarios and user needs.

In the 5G system, QoS parameters (ie, parameters related to QoS) include: bandwidth (Bandwidth), signal to interference plus noise ratio (SNR), bit error rate (bit error rate, BER), received signal reference Power (reference signal receiving power, RSRP), modulation and coding scheme (MCS), delay (Latency), packet loss rate (Packet Loss), capacity (Capacity), scheduling algorithm (Scheduling algorithm), retransmission Number of retransmissions. Taking the base station as an example, these parameters are explained below.

Among them, bandwidth is the available spectrum bandwidth allocated by the base station to each user. SNR is the ratio between the signal strength received by a base station and the noise level. SNR can be used to evaluate channel quality. BER is the ratio of bit errors that occur during transmission. BER is often used to measure the reliability of data. RSRP can be used to characterize wireless signal strength in wireless networks. MCS is the encoding method of physical layer data stream. There are sixteen kinds of MCS in the standard specification. The modulation methods from low to high according to the transmission efficiency can include: binary phase shift keying (BPSK), quadrature phase shift Keying (quadraturephase shift keying, QPSK), 16 quadrature amplitude modulation (quadrature amplitude modulation, QAM), 64QAM.

Latency is the time it takes for data to travel from source to destination and can be used to measure the quality of real-time applications such as voice and video. Packet loss rate is the ratio of packets lost during data transmission, and packet loss rate often affects the response time and quality of an application. Capacity is the number of users that a base station can handle simultaneously. Capacity is usually determined by the processing power and storage capacity of the processor in the base station. The scheduling algorithm is the algorithm used by the base station to allocate bandwidth and resources, which can be used to ensure that the service quality requirements of different users are met. The number of retransmissions is the number of times a data packet needs to be resent during data transmission. The number of retransmissions usually affects delay and packet loss rate.

2. Federated Learning

Federated learning (FL) is an emerging machine learning framework. FL allows multiple participants to jointly train machine learning models to achieve prediction or classification tasks while protecting data privacy. FL is an effective Distributed learning methods. In federated learning, each participant (such as a mobile device, sensor, or cloud server) saves its own local data and updates the global model by exchanging encrypted model parameters instead of sharing data directly. This method can not only protect data privacy, but also reduce communication overhead and reduce model generalization error.

Among them, federated learning does not require participants to upload local data to the central server, which can avoid data leakage and abuse. Therefore, federated learning has the advantage of data privacy protection. Moreover, in federated learning, only model parameters, rather than raw data, need to be exchanged between participants and the central server, so the communication overhead is relatively small. In addition, federated learning can increase training samples by combining data from different participants, thereby improving model generalization performance. Therefore, federated learning also has better model generalization performance.

For example, combined with Figure 2, each client among the three participants (client) #1, client #2, and client #3 can train the model locally and obtain the local FL corresponding to each client. Model (local FL model), each client sends the model parameters corresponding to the local FL model corresponding to the client to the federation center (such as the central server) (in Figure 2, client #1 sends ω ₁ , client # 2 sends ω ₂ and client #3 sends ω ₃ ); the federation center determines the global FL model based on the model parameters from client #1, client #2, and client #3, and reports it to the client Client #1, client #2, and client #3 respectively send the model parameters corresponding to the global FL model (the federation center sends g in Figure 2).

3. Digital twin network

Digital twin network (DTN) is a virtual twin corresponding to a physical network entity that is digitally created, and the virtual twin can be interactively mapped with the physical network entity in real time; DTN is a virtual twin based on real network data interaction. simulation environment. DTN can be built based on big data and AI. DTN is conducive to effectively ensuring network operation and maintenance in scenarios with heavy network load and large network scale, optimizing the network in real time, and assisting new business innovations such as network slicing and edge computing.

The core elements of DTN are: data, model, interaction, and mapping. Physical network layer data can be collected through real-time or non-real-time data collection methods and stored in the data warehouse to provide data support for building and empowering network twins, and form a feature-rich data model based on these data. . Among them, the data of the physical network layer may include: physical entity data, spatial data, resource data, protocols, interfaces, routing, signaling, processes, performance, logs, status, etc. In addition, multiple model instances can be created through flexible combination to serve various network applications. At the same time, physical network entities can be mapped through high-fidelity visualization pages through the network twin, and finally the visualization page, twin network layer, and physical network layer can be achieved. real-time interaction. At the same time, based on technologies such as artificial intelligence and big data analysis, the entire life cycle of physical networks can be analyzed, diagnosed, simulated and controlled.

Please refer to Figure 3. Figure 3 is a schematic diagram of a digital twin network architecture provided by an embodiment of the present invention. The digital twin network architecture includes: network application layer, twin network layer, and physical network layer. Among them, the network application layer can be used to implement network technology verification and optimization, network visualization, and network management and maintenance. The twin network layer can be used to implement service mapping models, network twin management, and network simulation verification. The network application layer and the twin network layer can be used to implement Data interaction.

4. Network emulator

The network simulator can be used to build a digital twin network. The network simulator can be a virtual unit implemented based on code in the network device, or it can be hardware deployed in the network device, without limitation.

In this embodiment of the present invention, the network emulator may be ns-3. ns-3 is a widely used open source network simulator that can be used to simulate and evaluate various communication network protocols and applications. ns-3 provides a modular set of libraries including network protocols, routing, wireless networks, sensor networks, mobile networks, applications and devices, etc. Using ns-3, users can create a virtual communications network environment to test and evaluate various network protocols and applications. In addition, the network simulator in the embodiment of the present invention can also be used for other simulators that can be used to build a digital twin network, without limitation.

Currently, 5G network service quality is predicted based on the current physical network environment, and the prediction results are based on the QoS prediction value obtained under the current specific physical network state. However, in order to ensure that network services maintain stable operation, the physical network environment cannot be changed at will to conduct relevant tests, so the QoS prediction results are relatively single; and there are many potential changes in the physical network in the future period (for example, the network bandwidth continues to decrease in the future period Or the signal-to-noise ratio deteriorates, etc.), the QoS change trend under specific changing conditions may not be effectively predicted.

Embodiments of the present invention provide a QoS prediction method, which is conducive to simulating various changes in the physical network environment based on the digital twin network, thereby being conducive to realizing QoS prediction for various situations of the network environment, and thereby being conducive to timely prediction of the physical network for each situation. Make strategic adjustments to the environment to improve the service quality of end-to-end business flows.

The QoS prediction method and device provided by the present invention will be described below with reference to the accompanying drawings.

Please refer to Figure 4, which is a schematic flow chart of a QoS prediction method provided by the present invention. The execution subject of the QoS prediction method may be a network device, or may also be a chip in the network device, etc., which is not limited here. Figure 5 takes the network device as the execution subject as an example for illustration. The QoS prediction method includes the following steps:

S101. The network device deploys the QoS prediction model into a network simulator. The network simulator is used to build a digital twin network that maps to the physical network service flow.

In an optional implementation, the QoS prediction model is obtained by network equipment and NWDAF network elements based on federated learning. Using federated learning to determine the QoS prediction model is beneficial for each network device to determine the QoS prediction model through interactive model parameters without sharing local data with other devices, thereby improving the data privacy security of network devices. It can be seen that the federated learning method can ensure the data privacy and security of network devices while taking into account the integrity of the training data set.

Optionally, the method may also include: each network device among the plurality of network devices determines a first prediction model corresponding to the QoS parameter based on the air interface wireless parameter of the network device; each network device among the plurality of network devices reports to the NWDAF The network element sends model parameters corresponding to the first prediction model. The NWDAF network element determines the second prediction model based on model parameters corresponding to the first prediction model from each network device in the plurality of network devices. The NWDAF network element sends model parameters corresponding to the second prediction model to each network device in the plurality of network devices.

If the model parameters corresponding to the second prediction model converge, the NWDAF network element may send first instruction information to multiple network devices, and the first instruction information is used to instruct the network devices to stop updating the model. If the model parameters corresponding to the second prediction model do not converge, the NWDAF network element does not send the first indication information to the multiple network devices.

For each network device in the plurality of network devices, if the network device receives the first indication information, the network device determines the QoS prediction model based on the model parameters corresponding to the second prediction model; otherwise, the network device determines the QoS prediction model based on the second prediction model. Corresponding model parameters and air interface wireless parameters, update model parameters corresponding to the first prediction model, and send updated model parameters corresponding to the first prediction model to the NWDAF network element. The NWDAF network element updates the model parameters corresponding to the second prediction model based on the updated model parameters corresponding to the first prediction model of each network device in the plurality of network devices, and sends the second prediction to each network device in the plurality of network devices. The updated model parameters corresponding to the model.

After the NWDAF network element updates the model parameters corresponding to the second prediction model, if the model parameters corresponding to the second prediction model converge, the NWDAF network element sends the first indication information to multiple network devices; if the model parameters corresponding to the second prediction model do not converge, Convergence, the NWDAF network element does not send the first indication information to multiple network devices. For each network device in the plurality of network devices, it may be determined to perform the operation of determining the QoS prediction model based on the model parameters corresponding to the most recently received second prediction model based on whether the first indication information is received, or to perform the operation of determining the QoS prediction model based on The operation of updating the model parameters corresponding to the first prediction model with the most recently received model parameters and air interface wireless parameters corresponding to the second prediction model.

It can be understood that, each time the network device updates the model parameters corresponding to the first prediction model, it performs the operation of sending the updated model parameters corresponding to the first prediction model to the NWDAF network element. Each time the NWDAF network element receives updated model parameters corresponding to the first prediction models from multiple network devices, it performs an update of model parameters based on the updated model parameters corresponding to the first prediction models of the multiple network devices. The operation of sending model parameters corresponding to the second prediction model to multiple network devices respectively, and determining whether to send the updated model parameters corresponding to the second prediction model to the multiple network devices based on whether the updated model parameters corresponding to the second prediction model converge. The operation of sending the first instruction information to each network device respectively. Until the NWDAF network element determines that the model parameters corresponding to the second prediction model converge, the NWDAF network element may stop executing the update of the model parameters corresponding to the second prediction model based on the model parameters from multiple network devices, and update the model parameters corresponding to the second prediction model to each of the multiple network devices. The network device sends the model parameters corresponding to the second prediction model. When receiving the first indication information, the network device stops performing the operation of updating the model parameters corresponding to the first prediction model, and performs the operation of determining the QoS prediction model based on the most recently received model parameters corresponding to the second prediction model.

In addition, if the number of QoS parameters to be predicted is M (M is a positive integer), the number of first prediction models determined by each network device is M, and the M first prediction models determined by each network device are related to the M QoS Parameters correspond one to one; the number of second prediction models determined by the NWDAF network element is M, and the M second prediction models correspond to M QoS parameters one-to-one; correspondingly, the number of QoS prediction models determined by each network device The number is M, and the M QoS prediction models correspond to the M QoS parameters one-to-one.

For example, a linear model is used as an example, the air interface wireless parameters are MCS, RSRP, and SINR, and the QoS parameters are bandwidth, delay, and packet loss rate. Each of the n network devices locally trains a first prediction model corresponding to bandwidth, a first prediction model corresponding to delay, and a first prediction model corresponding to packet loss rate. Among them, the i-th network device among the n network devices locally trains the first prediction model corresponding to the bandwidth as shown in formula (1), and the first prediction model corresponding to the delay can be shown as formula (2), The first prediction model corresponding to the packet loss rate can be as shown in formula (3).

Q _{Bandwidth_t} ＝b _1i ×MCS _t +b _2i ×RSRP _t +b _3i ×SINR _t +w _1i (1)

Q _{Latency_t} ＝l _1i ×MCS _t +l _2i ×RSRP _t +l _3i ×SINR _t +w _2i (2)

Q _{PacketLoss_t} ＝p _1i ×MCS _t +p _2i ×RSRP _t +p _3i ×SINR _t +w _3i (3)

Among them, Q _{Bandwidth_t} is the output of the first prediction model corresponding to the bandwidth determined by the i-th network device, indicating the bandwidth; b _1i , b _2i , b _3i are the first prediction model corresponding to the bandwidth determined by the i-th network device the corresponding model parameters. Q _{Latency_t} is the output of the first prediction model corresponding to the latency determined by the i-th network device, indicating the latency; l _1i , l _2i , l _3i are the first predictions corresponding to the latency determined by the i-th network device The model parameters corresponding to the model. Q _{PacketLoss_t} is the output of the first prediction model corresponding to the packet loss rate determined by the i-th network device, indicating the packet loss rate; p _1i , p _2i , p _3i are the output of the first prediction model corresponding to the packet loss rate determined by the i-th network device The model parameters corresponding to the first prediction model. w _1i is the bias parameter corresponding to the first prediction model corresponding to bandwidth determined by the i-th network device, w _2i is the bias parameter corresponding to the first prediction model corresponding to delay determined by the i-th network device , w _3i is the bias parameter corresponding to the first prediction model corresponding to the packet loss rate determined by the i-th network device. MCS _t , RSRP _t , and SINR _t are inputs to the first prediction model determined by the i-th network device, and respectively represent data related to MCS, data related to RSRP, and data related to SINR.

Each network device among the n network devices sends b _1i , b _2i , b _3i , l _1i , l _2i , l _3i , p _1i , p _2i , p _3i to the NWDAF network element. The NWDAF network element is based on data from multiple networks. Model parameters of the device are aggregated to obtain a second prediction model corresponding to the bandwidth (i.e., a global model corresponding to the bandwidth), a second prediction model corresponding to the delay (i.e., a global model corresponding to the delay), and a second prediction model corresponding to the packet loss. The second prediction model corresponding to the packet loss rate (that is, the global model corresponding to the packet loss rate). The model parameters of the second prediction model may be as shown in formula (4).

Among them, θ _i is the model parameter corresponding to the first prediction model sent by the i-th network device among the n network devices. For the second prediction model corresponding to the bandwidth, H _pre (x) represents the bandwidth, θ _i = [b _1i b _2i b _3i ]; for the second prediction model corresponding to the delay, θ _i = [l _1i l _2i l _3i ]; for the second prediction model corresponding to the packet loss rate, θ _i =[p _1i p _2i p _3i ].

The NWDAF network element sends the model parameters corresponding to the second prediction model to each of the n network devices. Each network device updates the model parameters corresponding to the first prediction model based on the received model parameters and sends them to the NWDAF network. The element sends updated model parameters corresponding to the first prediction model. The NWDAF network element repeatedly performs the foregoing operations until the second prediction model obtained by aggregation of the NWDAF network elements converges, that is, the model parameters corresponding to the second prediction model tend to be stable. When the model parameters corresponding to the second prediction model converge, the NWDAF network element may send first instruction information to each of the n network devices to instruct the network device to stop updating the model. Then, after receiving the first indication information, each network device among the n network devices can determine the QoS prediction model based on the model parameters corresponding to the most recently received second prediction model from the NWDAF network element. Wherein, for any one of the n network devices, the QoS prediction model determined by the network device can be as shown in formula (5).

H _pre (x)＝θ×x _t +w (5)

Among them, if formula (5) represents the QoS prediction model corresponding to the bandwidth, H _pre (x) represents the bandwidth, and θ is the second prediction model corresponding to the bandwidth received by the network device most recently since receiving the first indication information. The corresponding model parameter, w, is the bias parameter corresponding to the QoS prediction model corresponding to the bandwidth determined by the network device. If formula (5) represents the QoS prediction model corresponding to the delay, H _pre (x) represents the delay, and θ is the second prediction model corresponding to the delay received by the network device most recently since the first indication information was received. The corresponding model parameter, w, is the bias parameter corresponding to the QoS prediction model corresponding to the delay determined by the network device. If formula (5) represents the QoS prediction model corresponding to the packet loss rate, H _pre (x) represents the packet loss rate, and θ is the most recently received packet loss rate corresponding to the packet loss rate since the network device received the first indication information. 2. Model parameters corresponding to the prediction model, w is the bias parameter corresponding to the QoS prediction model corresponding to the packet loss rate determined by the network device.

Optionally, the method may also include: the NWDAF network element sending first information to multiple network devices, where the first information is used to indicate the model structure; correspondingly, each network device in the multiple network devices receives the first information. Each network device in the plurality of network devices may use the model structure indicated by the first information as the model structure of the first prediction model. The NWDAF network element may use the model structure indicated by the first information as the model structure of the second prediction model. In addition, optionally, the first information may include model input parameter types, model structure types, etc. Wherein, the model input parameter type is the air interface wireless parameter type of the network device (for example, MCS, RSRP, SINR, etc.), then each network device in the multiple network devices can determine the air interface wireless parameter type required by the first prediction model, Thereby collecting the specific values of these types of air interface wireless parameters. The model structure type may be, for example, a linear structure, etc.

In an optional implementation, the method may also include: the network device uses air interface wireless parameters under different service flows under various physical network conditions in different periods to increase the diversity of the data set, thereby making a determined QoS prediction model Has certain generalization ability.

S102. The network device loads the QoS prediction model in the network simulator according to the network environment of the digital twin network.

In an optional implementation, the network device loads the QoS prediction model in the network simulator for the network environment of the digital twin network, which may include: the network device sets the network environment of the digital twin network in the network simulator; QoS prediction models are loaded for each of the multiple network environments of the digital twin network. In other words, in the network simulator, the network environment of the digital twin network can be diversified based on different network change conditions, thereby achieving QoS prediction under various network conditions and obtaining diversified QoS prediction results.

For example, for any one of the n network devices, taking the linear model and the air interface wireless parameters as MCS, RSRP, and SINR as an example, load the QoS prediction model corresponding to the bandwidth for k situations of the network environment of the digital twin network. , the diversified QoS prediction results shown in formula (7) can be obtained.

Among them, s _k represents k situations of network environment, Represents diversified QoS prediction results, and b ₁ , b ₂ , and b ₃ are model parameters corresponding to the QoS prediction model corresponding to the bandwidth.

In an optional implementation, the method may further include: the network device sending network parameters corresponding to the network environment of the digital twin network to the PCF network element, and QoS predictions obtained by loading the QoS prediction model for the network environment of the digital twin network. As a result, the PCF network element can adjust the physical network strategy based on the network parameters corresponding to the network environment of the digital twin network and the QoS prediction results obtained from the network environment of the digital twin network.

In the scenario where the network device loads the QoS prediction model in the network simulator for each of the multiple network environments of the digital twin network, the network device can send each of the multiple network environments of the digital twin network to the PCF network element. The network parameters corresponding to each network environment, and the QoS prediction results obtained for each network environment of the digital twin network. In this way, the PCF network element can determine the network environment that matches the physical network environment from various network environments of the digital twin network, determine the QoS prediction results obtained by the network equipment for this network environment of the digital twin network, and then based on the QoS prediction results Adjust physical network policies. It can be seen that in the scenario where the network equipment has obtained diversified QoS prediction results, the network equipment can send diversified QoS prediction results to the PCF network element. The PCF network element can match the appropriate QoS based on the future change trend of the real physical network. Predict the results and make network policy adjustments in advance to face the deterioration of communication status caused by changes in network conditions (network environment), thereby ensuring network stability and business service quality.

To sum up, in this QoS prediction method, the network device deploys the QoS prediction model into the network simulator. The network simulator is used to build a digital twin network that maps the physical network business flow; the network device targets the digital twin in the network simulator. Network environment of the network, loading QoS prediction model. It can be seen that this method simulates the physical network environment based on the digital twin network in the network simulator, and achieves QoS prediction of the physical network environment by performing QoS prediction on the network environment of the digital twin network. This method is conducive to simulating various changes in the physical network environment in the future based on the digital twin network, and obtaining diversified QoS prediction results corresponding to business flows in different physical network environments; and, this method can also combine various changes in the digital twin network The network parameters corresponding to the network environment and the QoS prediction results under each network environment in the multiple network environments are fed back to the physical network, so that the PCF network element can make timely decisions based on the physical network environment and/or the future change trend of the physical network environment. Outbound policy adjustments are made to ensure the service quality of end-to-end business flows.

The following takes n network devices (network device #1 to network device #n) and NWDAF network elements to determine the QoS prediction model based on federated learning as an example to exemplify the QoS prediction method provided by the present invention. Please refer to Figure 5. Figure 5 is a schematic diagram of another QoS prediction method provided by an embodiment of the present invention. The QoS prediction method includes the following steps:

S201. The NWDAF network element sends first information to each of the n network devices, where the first information is used to indicate the model structure.

S202. Each network device among the n network devices collects the air interface wireless parameters of the network device according to the first information.

S203. Each network device among the n network devices performs model training based on the air interface wireless parameters of the network device to obtain a first prediction model. Specifically, the network device #1 to the network device #n obtain the first prediction model #1 to the first prediction model #n respectively.

S204. Each network device among the n network devices sends the model parameters corresponding to the first prediction model to the NWDAF network element. Correspondingly, the NWDAF network element receives model parameters corresponding to the first prediction model from each of the n network devices. Specifically, network device #1 sends model parameters corresponding to the first prediction model #1 to the NWDAF network element, network device #2 sends model parameters corresponding to the first prediction model #2 to the NWDAF network element,..., network device #n sends The NWDAF network element sends the model parameters corresponding to the first prediction model #n.

S205. The NWDAF network element determines the second prediction model based on the model parameters corresponding to the first prediction model of each of the n network devices. Specifically, the NWDAF network element determines the second prediction model based on the model parameters respectively corresponding to the first prediction model #1 to the first prediction model #n.

S206. The NWDAF network element sends the model parameters corresponding to the second prediction model to each of the n network devices. Correspondingly, each of the n network devices receives model parameters corresponding to the second prediction model from the NWDAF network element.

S207. The NWDAF network element determines whether the second prediction model has converged. If the NWDAF network element determines that the second prediction model converges, step S208 is executed. If the NWDAF network element determines that the second prediction model has not converged, step S208 is not performed.

S208. The NWDAF network element sends first instruction information to each of the n network devices. The first instruction information is used to instruct the network device to stop updating the model.

S209. For each network device among the n network devices, the network device determines whether the first indication information is received. If the network device does not receive the first indication information, step S210 is executed. If the network device receives the first indication information, steps S211 to S215 are executed.

S210. For each network device among the n network devices, the network device updates the model parameters corresponding to the first prediction model based on the model parameters corresponding to the second prediction model and the air interface wireless parameters of the network device, and performs steps S204 to S209.

Among them, in the re-executed step S204, each network device among the n network devices sends the updated model parameters corresponding to the first prediction model to the NWDAF network element; in the re-executed step S205, the NWDAF network element Based on the updated model parameters corresponding to the first prediction model of each network device among the n network devices, the model parameters corresponding to the second prediction model are updated; in step S206 performed again, the NWDAF network element Each network device in the device sends updated model parameters corresponding to the second prediction model.

Specifically, network device #1 updates the model parameters corresponding to the first prediction model #1, network device #2 updates the model parameters corresponding to the first prediction model #2,..., network device #n updates the model parameters corresponding to the first prediction model #n model parameters.

S211. Each network device among the n network devices determines a QoS prediction model based on the model parameters corresponding to the second prediction model.

S212. Each network device among the n network devices deploys the QoS prediction model to a network simulator. The network simulator is used to build a digital twin network that maps to the physical network service flow.

S213. Each network device among the n network devices loads the QoS prediction model in the network simulator for each of the multiple network environments of the digital twin network.

S214. Each network device among the n network devices sends to the PCF network element the network parameters corresponding to each of the multiple network environments of the digital twin network, and the network parameters corresponding to each of the multiple network environments of the digital twin network. QoS prediction results obtained by loading the QoS prediction model.

S215. For each of the n network devices, the PCF network element provides the network parameters corresponding to each network environment in the multiple network environments of the digital twin network based on the network device, and the network device for the multiple network environments of the digital twin network. The QoS prediction results obtained in each network environment adjust the physical network policy.

Understandably, in conjunction with Figure 6, the NWDAF network element and n network devices can determine the QoS prediction model based on federated learning, and each of the n network devices can be used to build a digital twin network that maps to the physical network service flow. Deploy the QoS prediction model in the network simulator and load the QoS prediction model in the network simulator to make QoS predictions for the network environment of the digital twin network, and then feed back the obtained QoS prediction results to the PCF network elements in the physical network. It is conducive to ensuring the data privacy and security of network equipment while enabling QoS prediction for various network change situations, and feeding back the network parameters corresponding to various network environments of the digital twin network and the QoS prediction results corresponding to each network environment. physical network, so that PCF network elements can make timely policy adjustments to ensure end-to-end service quality.

Please refer to FIG. 7 . FIG. 7 is a schematic structural diagram of a QoS prediction device provided by an embodiment of the present invention. The QoS prediction device may be a network device or a device (such as a chip) with network device functions. Specifically, as shown in Figure 7, the QoS prediction device 700 may include a deployment unit 701 and a prediction unit 702. in:

The deployment unit 701 is used to deploy the QoS prediction model into a network simulator, and the network simulator is used to build a digital twin network that maps to the physical network service flow.

The prediction unit 702 is used to load the QoS prediction model in the network simulator for the network environment of the digital twin network.

In an optional implementation, the prediction unit 702 loads the QoS prediction model in the network simulator for the network environment of the digital twin network, specifically for: setting the network environment of the digital twin network in the network simulator; For each of the multiple network environments of the twin network, a QoS prediction model is loaded.

In an optional implementation, the QoS prediction model is obtained by the QoS prediction device 700 and the NWDAF network element based on federated learning.

In an optional implementation, the QoS prediction device 700 further includes:

The determining unit 703 is configured to determine the first prediction model corresponding to the QoS parameter based on the air interface wireless parameters of the network device.

The sending unit 704 is configured to send the model parameters corresponding to the first prediction model to the NWDAF network element.

The receiving unit 705 is configured to receive model parameters corresponding to the second prediction model from the NWDAF network element. The second prediction model is the NWDAF network element based on the first prediction model from each QoS prediction device 700 in the plurality of QoS prediction devices 700. The model parameters are determined.

The determining unit 703 is further configured to: when the receiving unit 705 receives the first indication information, determine the QoS prediction model based on the model parameters corresponding to the second prediction model; the first indication information is used to instruct the QoS prediction device 700 to stop updating the model; When the receiving unit 705 does not receive the first indication information, it updates the model parameters corresponding to the first prediction model based on the model parameters corresponding to the second prediction model and the air interface wireless parameters.

The sending unit 704 is also configured to send the updated model parameters corresponding to the first prediction model to the NWDAF network element after the determining unit 703 updates the model parameters corresponding to the first prediction model.

In an optional implementation, the sending unit 704 is also configured to send the network parameters corresponding to the network environment of the digital twin network and the QoS prediction results obtained by loading the QoS prediction model for the network environment of the digital twin network to the PCF network element; The PCF network element is used to adjust the physical network strategy based on the network parameters and QoS prediction results corresponding to the network environment of the digital twin network.

For a more detailed description of the above-mentioned QoS prediction device 700 and the technical effects it brings, please refer to the relevant descriptions in the above-mentioned method embodiments, and will not be described again here.

Please refer to FIG. 8 , which is a schematic structural diagram of a communication device provided by an embodiment of the present invention. The communication device may be a network device or a device (such as a chip) with network device functions. Specifically, as shown in FIG. 8 , the communication device 800 may include a processor 801 and a memory 802 . The memory 802 is used to store a computer program. The computer program includes program instructions. The processor 801 is configured to call the program instructions to cause the communication device 800 to execute the aforementioned method.

The processor 801 can be a central processing unit (CPU). The processor 801 can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (application specific integrated circuits). , ASIC), off-the-shelf programmable gate array (field-programmable gate array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

In one way, the communication device 800 is used to perform the functions of the network device in the aforementioned method embodiment:

The processor 801 is used to deploy the QoS prediction model into a network simulator, and the network simulator is used to build a digital twin network that maps the physical network service flow.

The processor 801 is also used to load the QoS prediction model in the network simulator for the network environment of the digital twin network.

In an optional implementation, the processor 801 loads the QoS prediction model in the network simulator for the network environment of the digital twin network, specifically for: setting the network environment of the digital twin network in the network simulator; For each of the multiple network environments of the twin network, a QoS prediction model is loaded.

In an optional implementation, the QoS prediction model is obtained by the communication device 800 and the NWDAF network element based on federated learning.

In an optional implementation, the processor 801 is also configured to determine a first prediction model corresponding to the QoS parameter based on the air interface wireless parameters of the communication device 800 . The device also includes: a transceiver 803, configured to send model parameters corresponding to the first prediction model to the NWDAF network element. The transceiver 803 is also configured to receive model parameters corresponding to the second prediction model from the NWDAF network element. The second prediction model is a model corresponding to the NWDAF network element based on the first prediction model from each communication device 800 in the plurality of communication devices 800 . The parameters are determined.

The processor 801 is also configured to: when the transceiver 803 receives the first indication information, determine the QoS prediction model based on the model parameters corresponding to the second prediction model; the first indication information is used to instruct the communication device 800 to stop updating the model; in the transceiver 803 When the first indication information is not received, update the model parameters corresponding to the first prediction model based on the model parameters corresponding to the second prediction model and the air interface wireless parameters.

The transceiver 803 is also configured to send the updated model parameters corresponding to the first prediction model to the NWDAF network element after the processor 801 updates the model parameters corresponding to the first prediction model.

In an optional implementation, the transceiver 803 is also used to send network parameters corresponding to the network environment of the digital twin network and QoS prediction results obtained by loading the QoS prediction model for the network environment of the digital twin network to the PCF network element; PCF The network element is used to adjust the physical network strategy based on the network parameters and QoS prediction results corresponding to the network environment of the digital twin network.

For a more detailed description of the above-mentioned communication device 800 and the technical effects it brings, please refer to the relevant descriptions in the above-mentioned method embodiments, and will not be described again here.

Embodiments of the present invention also provide a computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instruction is run on a processor, the method flow of the above method embodiment is implemented.

An embodiment of the present invention also provides a computer program product. When the computer program product is run on a processor, the method flow of the above method embodiment is implemented.

It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it still The technical solutions described in the foregoing embodiments can be modified, or some or all of the technical features can be equivalently replaced; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the scope of the technical solutions of the embodiments of the present invention. .

Claims (12)

1. A quality of service QoS prediction method, characterized in that it is applied to network equipment, and the method includes: Deploy the QoS prediction model into a network simulator, which is used to build a digital twin network that maps to the physical network service flow; Load the QoS prediction model in the network simulator for the network environment of the digital twin network. 2. The quality of service QoS prediction method according to claim 1, characterized in that, in the network simulator, for the network environment of the digital twin network, the step of loading the QoS prediction model includes: Set the network environment of the digital twin network in the network simulator; QoS prediction models are loaded for each of the multiple network environments of the digital twin network. 3. The quality of service QoS prediction method according to claim 1 or 2, characterized in that the QoS prediction model is obtained by network equipment and network data analysis function NWDAF network elements based on federated learning. 4. The quality of service QoS prediction method according to claim 3, characterized in that the method further includes: Based on the air interface wireless parameters of the network device, determine a first prediction model corresponding to the QoS parameters; Send the model parameters corresponding to the first prediction model to the NWDAF network element; Receive model parameters corresponding to the second prediction model from the NWDAF network element. The second prediction model is obtained by the NWDAF network element based on model aggregation corresponding to the first prediction model from each network device in the plurality of network devices. global model; If the first indication information is received, determine the QoS prediction model based on the model parameters corresponding to the second prediction model, where the first indication information is used to instruct the network device to stop updating the model; Otherwise, based on the model parameters corresponding to the second prediction model and the air interface wireless parameters, update the model parameters corresponding to the first prediction model, send the updated model parameters corresponding to the first prediction model to the NWDAF network element, and repeat the aforementioned operations until NWDAF The network element determines that the model parameters corresponding to the second prediction model have converged, and the NWDAF network element sends the first indication information to multiple network devices. 5. The quality of service QoS prediction method according to claim 1 or 2, characterized in that the method further includes: Send the network parameters corresponding to the network environment and the QoS prediction results obtained by loading the QoS prediction model for the network environment to the policy control function PCF network element; The PCF network element is used to adjust the physical network policy based on network parameters corresponding to the network environment and QoS prediction results. 6. A quality of service QoS prediction device, characterized in that it includes: A deployment unit configured to deploy the QoS prediction model into a network simulator, where the network simulator is used to build a digital twin network that maps to the physical network service flow; The prediction unit is used to load the QoS prediction model in the network simulator for the network environment of the digital twin network. 7. The service quality QoS prediction device according to claim 6, characterized in that the prediction unit loads the QoS prediction model in the network simulator according to the network environment of the digital twin network, and is specifically used for: Set the network environment of the digital twin network in the network simulator; QoS prediction models are loaded for each of the multiple network environments of the digital twin network. 8. The quality of service QoS prediction device according to claim 6 or 7, characterized in that the QoS prediction model is obtained by network equipment and network data analysis function NWDAF network elements based on federated learning. 9. The quality of service QoS prediction device according to claim 8, characterized in that the device further includes: A determining unit configured to determine the first prediction model corresponding to the QoS parameters based on the air interface wireless parameters of the network device; A sending unit, configured to send model parameters corresponding to the first prediction model to the NWDAF network element; A receiving unit configured to receive model parameters corresponding to the second prediction model from the NWDAF network element, where the second prediction model is the model parameter corresponding to the first prediction model of the NWDAF network element based on each network device in the plurality of network devices. definite; The determining unit is further configured to: when the receiving unit receives the first indication information, determine the QoS prediction model based on the model parameters corresponding to the second prediction model; when the receiving unit does not receive the first indication information, determine the QoS prediction model based on the first indication information. The model parameters and air interface wireless parameters corresponding to the second prediction model are updated, and the model parameters corresponding to the first prediction model are updated; the first instruction information is used to instruct the device to stop updating the model; The sending unit is also configured to send the updated model parameters corresponding to the first prediction model to the NWDAF network element after the determination unit updates the model parameters corresponding to the first prediction model. 10. The quality of service QoS prediction device according to claim 6 or 7, characterized in that the device further includes: The sending unit is also used to send the network parameters corresponding to the network environment and the QoS prediction results obtained by loading the QoS prediction model for the network environment to the policy control function PCF network element; The PCF network element is used to adjust the physical network policy based on network parameters corresponding to the network environment and QoS prediction results. 11. A communication device, characterized in that it includes a memory and a processor, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to cause the The communication device performs the method according to any one of claims 1 to 5. 12. A computer-readable storage medium, characterized in that computer-readable instructions are stored in the computer-readable storage medium, and when the computer-readable instructions are run on a communication device, the communication device causes the communication device to execute the right The method according to any one of claims 1 to 5.