CN115442216B - Network slicing fault self-healing method, device, equipment and computer storage medium - Google Patents

Abstract

The invention discloses a network slice fault self-healing method, a device, equipment and a computer program product, wherein the method comprises the following steps: acquiring multi-dimensional index data of a network slice, and detecting the multi-dimensional index data; when abnormal data are detected, generating a slice self-healing action and a slice fault action according to the multidimensional index data and judging the slice self-healing action and the slice fault action to determine a behavior score of the generated slice self-healing action and the slice fault action; and carrying out iterative optimization on the generated self-healing actions of the slice and the fault actions of the slice according to the behavior scores so as to determine a target self-healing action capable of self-healing the fault of the network slice and carry out fault repair on the network slice. The invention respectively generates the slice self-healing action and the slice fault action to form countermeasures by the double intelligent agents, forces the generated slice self-healing action to continuously improve the fault repairing capability of the slice self-healing action, improves the performance of the fault self-healing strategy, and further improves the fault self-healing recovering capability of the network slice.

Description

Network slicing fault self-healing method, device, equipment and computer storage medium

technical field

The present invention relates to the field of communication technology, in particular to a network slice fault self-healing method, device, equipment and computer program product.

Background technique

Existing self-healing methods for network slicing faults need to rely on the experience of technicians and manually set self-healing strategies. The self-healing strategy is mainly to simply set thresholds for various index data to trigger slice fault self-healing actions. The self-healing strategy relies on manual settings. When the rules are implemented, the self-healing efficiency of faults is low and error-prone, and the policy cannot be adjusted in time when the network slice changes. Moreover, in the existing slice fault self-healing process, it is first necessary to analyze the fault type and fault impact range, then determine the appropriate fault handling process, and implement fault repair or fault self-healing based on artificially defined strategies. The self-healing process is relatively long. More time is spent on fault analysis and location, resulting in a longer service impact time, which affects user experience. Therefore, the existing network slicing fault self-healing strategy that relies on manually setting rules has poor fault self-healing performance.

Contents of the invention

The main purpose of the present invention is to provide a network slicing fault self-healing method, device, equipment and computer program product, aiming to solve the existing technical problem of poor fault self-healing performance of the existing network slicing fault self-healing strategy relying on manual setting rules .

In addition, in order to achieve the above object, the present invention also provides a network slice fault self-healing method, the network slice fault self-healing method includes the following steps:

Obtain the multi-dimensional index data of the network slice, and detect the obtained multi-dimensional index data;

When abnormal index data is detected in the multi-dimensional index data of the network slice, a slice self-healing action and a slice fault action are generated according to the abnormal index data;

evaluating the generated slice self-healing actions and slice failure actions to determine the behavior scores of the slice self-healing actions and the slice failure actions;

Iterative optimization is performed on the generated slice self-healing action and slice fault action according to the behavior score to determine a target self-healing action, and the target self-healing action is used to repair the fault of the network slice.

Optionally, the step of generating a slice self-healing action and a slice fault action according to the multidimensional index data includes:

Input the multi-dimensional index data into the preset target self-healing model, wherein the target self-healing model is obtained by iteratively training the preset self-healing model to be trained by using the historical multi-dimensional index data of network slices, and the target Self-healing models include action generators;

According to the multi-dimensional index data, an action generator of the target self-healing model is used to generate a slice self-healing action and a slice fault action.

Optionally, the action generator includes a self-healing action generator and a fault action generator, and the action generator using the target self-healing model generates a slice self-healing action and a slice fault action according to the multidimensional index data steps, including:

Determine the current full information state of the network slice according to the multidimensional index data, wherein the full information state includes a first state observable by the self-healing action generator, and a first state observable by the fault action generator Two states;

inputting the first state in the full information state into the self-healing action generator, and using the self-healing action generator to generate slice self-healing actions;

The second state in the full information state is input into the fault action generator, and the slice fault action is generated by the fault action generator.

Optionally, the target self-healing model further includes a self-healing action evaluator and a fault action evaluator, which evaluates the generated slice self-healing action and slice fault action to determine the slice self-healing action and the The steps for slicing the behavioral score of fault actions include:

input the first state in the full information state, the slice self-healing action and the slice fault action into the self-healing action evaluator in the target self-healing model, so as to evaluate the slice self-healing action Judging, determining the behavior score of the self-healing action of the slice;

determining the fault damage radius according to the slice fault action;

inputting the second state in the full information state, the fault damage radius, the slice self-healing action and the slice fault action into the fault action evaluator in the target self-healing model, so as to evaluate the Slice fault actions are judged to determine the behavior score of the slice fault actions.

Optionally, performing iterative optimization on the generated slice self-healing action and slice fault action according to the behavior score to determine a target self-healing action, and using the target self-healing action to perform fault repair on the network slice steps, including:

Feedback the behavior score of the slice self-healing action to the self-healing action generator, and feed back the behavior score of the slice fault action to the fault action generator, return and execute the self-healing action using the target The self-healing action generator in the model generates a slice self-healing action, and uses the fault action generator in the target self-healing model to generate a slice fault action, so as to perform the slice self-healing action and the slice fault action Iterative optimization, until the behavior score of the slice self-healing action and the behavior score of the slice failure action meet the preset conditions, and obtain the target self-healing action;

Switch the state of the network slice according to the target self-healing action, so as to switch the network slice from the current full information state to the target full information state corresponding to the target self-healing action, and perform fault on the network slice repair.

Optionally, performing iterative optimization on the generated slice self-healing action and slice fault action according to the behavior score to determine a target self-healing action, and using the target self-healing action to perform fault repair on the network slice After the steps, include:

Inputting the target self-healing action into a preset reward function to obtain a reward value corresponding to the target self-healing action;

An experience replay data set is generated according to the reward value, and model parameters of the target self-healing model are updated according to the experience replay data set.

Optionally, before the step of generating a slice self-healing action and a slice fault action according to the target index data, it may further include:

Obtain the historical indicator data of the network slice and perform preprocessing to obtain sample data;

Obtain model architecture parameters, and establish a basic self-healing model according to the model architecture parameters;

Using the sample data to iteratively train the basic self-healing model to obtain a target self-healing model.

In addition, in order to achieve the above purpose, the present invention also provides a network slice fault self-healing device, the network slice fault self-healing device includes:

The data detection module is used to obtain the multi-dimensional index data of the network slice, and inspect the obtained multi-dimensional index data;

An action generation module, configured to generate slice self-healing actions and slice fault actions according to the multi-dimensional index data when abnormal data is detected in the multi-dimensional index data of the network slice, and to generate slice self-healing actions and slice fault actions Actions are judged to determine the behavior scores of the slice self-healing action and the slice failure action;

The dual-agent confrontation module is used to iteratively optimize the generated slice self-healing action and slice fault action according to the behavior score, so as to determine the target self-healing action, and use the target self-healing action to perform fault on the network slice repair.

In addition, in order to achieve the above object, the present invention also provides a network slice fault self-healing device, the network slice fault self-healing device includes: a memory, a processor, and an A network slice fault self-healing program, when the network slice fault self-healing program is executed by the processor, the steps of the above-mentioned network slice fault self-healing method are implemented.

In addition, in order to achieve the above object, the present invention also provides a computer program product, the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned network slice fault self-healing method are implemented.

A network slicing fault self-healing method, device, equipment and computer program product proposed by the embodiments of the present invention. Compared with the poor self-healing performance of the self-healing strategy in the existing network slicing fault self-healing method, in the embodiment of the present invention, by obtaining the multi-dimensional index data of the network slice, and detecting the acquired multi-dimensional index data; when When abnormal data is detected in the multi-dimensional index data of the network slice, generate slice self-healing actions and slice fault actions according to the multi-dimensional index data, and evaluate the generated slice self-healing actions and slice fault actions to determine the The behavior scores of the slice self-healing action and the slice fault action; iteratively optimize the generated slice self-healing action and slice fault action according to the behavior score to determine the target self-healing action, and use the target self-healing action Perform fault repair on the network slice. When it is detected that there is an anomaly in most of the index data of the network slice, it will actively create a fault by generating a slice fault action, and fight against the generated slice self-healing action, forcing the generated slice self-healing action to improve its own fault repair ability, thereby improving The fault self-healing performance of the network slicing self-healing strategy is improved, and the fault problem in the network slicing can be identified and repaired in time, and the fault self-healing can be recovered before the network slicing fault causes interruption, avoiding serious consequences, and improving the self-healing recovery of the network slicing ability.

Description of drawings

FIG. 1 is a schematic diagram of the hardware structure of an implementation of a network slicing fault self-healing device provided by an embodiment of the present invention;

FIG. 2 is a schematic flowchart of a first embodiment of a network slice fault self-healing method according to the present invention;

3 is a schematic diagram of the confrontation process of the action generator in the second embodiment of the network slice fault self-healing method of the present invention;

4 is a schematic diagram of the neural network hierarchical architecture of the action generator and the action evaluator in the third embodiment of the network slice fault self-healing method of the present invention;

FIG. 5 is a schematic diagram of functional modules of an embodiment of a network slice fault self-healing device according to the present invention.

The realization of the purpose of the present invention, functional characteristics and advantages will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

In the following description, use of suffixes such as 'module', 'part' or 'unit' for denoting elements is only for facilitating description of the present invention and has no specific meaning by itself. Therefore, 'module', 'part' or 'unit' may be used in combination.

The main solution of the embodiment of the present invention: by acquiring the multi-dimensional index data of the network slice, and detecting the obtained multi-dimensional index data; when abnormal data is detected in the multi-dimensional index data of the network slice, according to the multi-dimensional index data Generate slice self-healing actions and slice fault actions from the data, and evaluate the generated slice self-healing actions and slice fault actions to determine the behavior scores of the slice self-healing actions and the slice fault actions; The generated slice self-healing action and slice fault action are iteratively optimized to determine a target self-healing action, and use the target self-healing action to perform fault repair on the network slice. By actively creating faults, the dual agents that generate slice self-healing actions and slice fault actions can fight against each other, improve the fault self-healing performance of the network slice self-healing strategy, identify and repair faults in network slices in a timely manner, and prevent network slice failures Self-healing and recovery of faults can be made before interruptions are caused to avoid serious consequences, thereby improving the self-healing and recovery capabilities of network slicing.

The main technical terms involved in the embodiments of the present invention:

Network slicing (networkslice, NS): Network slicing refers to the customization of different logical networks according to different service requirements on physical or virtual network infrastructure. Network slicing can be a complete end-to-end network including terminal equipment, access network, transmission network, core network and application server, which can provide complete communication services and have certain network capabilities. Network slices can also be any combination of terminal devices, access networks, transmission networks, core networks, and application servers.

Chaos Engineering: Chaos Engineering is a technical means that can ensure the availability of the system and improve the resilience of the technical architecture. It aims to kill the failure in the infancy, that is, to identify the failure before the failure causes interruption, By actively creating faults, testing the behavior of the system under various pressures, proactively identifying and repairing faults to avoid serious consequences.

NSMF: Network Slice Management Function (Network Slice Management Function), responsible for receiving network slice requirements, managing the life cycle, performance, faults, etc. of network slices, arranging the composition of network slices, and decomposing network slice requirements into each network slice Network or network function requirements, and send a network slice subnet management request to each NSSMF.

NSSMF: Network Slice Subnet Management Function (Network Slice Subnet Management Function), receives the network slice subnet deployment requirements issued by NSMF, manages the network slice subnet, arranges the composition of the network slice subnet, and divides the network slice subnet The SLA (Service Level Agreement, service level agreement) requirements of the network service are mapped to the QoS (Quality of Service, service quality) requirements of the network service, and the deployment request of the network service is issued.

The embodiment of the present invention considers that in the existing related schemes, the network slice fault self-healing strategy relies on manually set rules, and the self-healing restoration process takes a long time and is prone to errors. When a slice fault changes, more time is spent on the fault Analysis and positioning, unable to adjust self-healing strategies in time. In order to solve the above problems, chaos engineering is introduced in the self-healing strategy of network slicing. By actively creating faults, testing the behavior of the system under various pressures, proactively identifying and repairing faults, and avoiding serious consequences. However, the existing network slicing fault self-healing methods based on chaos engineering mainly rely on artificial faults, and the rules of faults are easily learned by the self-healing system, resulting in a decline in the generalization ability and self-healing recovery ability of the self-healing strategy. Therefore, existing network slicing fault self-healing strategies generally have the problem of poor fault self-healing performance.

Therefore, the embodiment of the present invention proposes a solution. When an abnormality is detected in the multi-dimensional index data of the network slice, the fault action is actively created by generating a slice fault action, and it will fight against the self-healing action of the generated slice to force the self-healing action of the generated slice. Continuously improve its own fault repair capability, thereby improving the performance of the fault self-healing strategy, identifying and repairing faults before network slicing faults cause interruptions, improving the fault self-healing and recovery capabilities of network slicing, and avoiding serious consequences.

Specifically, referring to FIG. 1 , FIG. 1 is a schematic diagram of functional modules of a terminal device belonging to a network slicing fault self-healing device according to an embodiment of the present invention. The terminal device (also called a terminal or device) may be a PC, a smart phone, or a tablet computer. Mobile terminal equipment with data processing functions such as portable computers.

As shown in FIG. 1 , the terminal device may include: a processor 1001 , such as a CPU, a network interface 1004 , a user interface 1003 , a memory 1005 , and a communication bus 1002 . Wherein, the communication bus 1002 is used to realize connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. Optionally, the network interface 1004 may include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001 .

Optionally, the terminal may further include a camera, an RF (Radio Frequency, radio frequency) circuit, a sensor, an audio circuit, a WiFi module, and the like. Among them, sensors such as light sensors, motion sensors and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display screen according to the brightness of the ambient light, and the proximity sensor may turn off the display screen and/or backlight. As a kind of motion sensor, the gravitational acceleration sensor can detect the magnitude of acceleration in various directions (generally three axes), and can detect the magnitude and direction of gravity when it is stationary, and can be used for applications that recognize the posture of mobile terminals (such as horizontal and vertical screen switching, Related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tap), etc.; of course, the mobile terminal can also be equipped with other sensors such as gyroscope, barometer, hygrometer, thermometer, infrared sensor, etc. No longer.

Those skilled in the art can understand that the terminal structure shown in FIG. 1 does not constitute a limitation on the terminal, and may include more or less components than those shown in the figure, or combine some components, or arrange different components.

As shown in FIG. 1 , the memory 1005 as a computer program product may include an operating system, a network communication module, a user interface module, and a network slice fault self-healing program.

In the terminal shown in Figure 1, the network interface 1004 is mainly used to connect to the background server and perform data communication with the background server; the user interface 1003 is mainly used to connect to the client (client) and perform data communication with the client; and the processor 1001 may be used to call a network slice fault self-healing program stored in the memory 1005, and when the network slice fault self-healing program is executed by a processor, operations in the network slice fault self-healing method provided in the following embodiments are implemented.

Based on the above-mentioned device hardware structure, an embodiment of the network slicing fault self-healing method of the present invention is proposed.

Referring to FIG. 2, in the first embodiment of the network slice fault self-healing method of the present invention, the network slice fault self-healing method includes:

Step S10, acquiring multi-dimensional index data of the network slice, and detecting the acquired multi-dimensional index data;

With the development of communication technology, especially the popularization and application of the 5th-generation mobile communication (the 5th-generation, 5G) technology, the concept of network slicing is introduced to cope with the difference in network performance requirements of different communication services. In order to ensure the continuity of network services, network slicing needs to have certain self-healing recovery capabilities, so that faults can be self-healed before network slicing faults cause network interruption.

The present invention proposes a network slicing fault self-healing method that can be applied to 5G network slicing. Based on dual-agent reinforcement learning and chaos engineering, fault actions and self-healing actions are generated by actively creating faults, and fault actions and self-healing actions are used The confrontation between the two agents improves the generalization ability of slice self-healing, and forces the generated self-healing actions to continuously improve its own fault repair ability, thereby improving the self-healing recovery ability of network slices. Specifically, the multi-dimensional index data of the network slice is acquired first, and the acquired multi-dimensional index data is detected to determine whether there is anomaly in the network slice. It can be seen that NSMF and NSSMF are generally installed in the network environment where network slicing is located. NSMF is responsible for managing network slicing, mainly responsible for monitoring multi-dimensional index data of network slicing. Network slicing generally includes terminal equipment, access network, transmission network, core network and application servers, while the access network, transmission network, and core network in network slicing include multiple network sub-slices managed by NSSMF, such as wireless access network sub-slices, transmission network sub-slices, and core network sub-slices. In this embodiment, when obtaining the multi-dimensional indicator data of the network slice, the KPI (Key Performance Indicator, key performance indicator or target-based quantitative management index) data of the network sub-slice is mainly obtained from the NSMF, including the wireless access network sub-slice, KPIs of network sub-slices such as transmission network sub-slices and core network sub-slices. KPIs of radio access network sub-slices include radio access network transmission delay, average uplink/downlink user throughput rate, and uplink/downlink cell average throughput rate , CPU average occupancy rate, number of online users, QoS flow establishment success rate, call establishment success rate, etc.; KPIs of transmission network sub-slices include transmission network transmission delay, bandwidth utilization rate, packet loss rate, data transmission volume, and bit error rate etc.; KPIs of core network sub-slices include core network transmission delay, utilization rate of virtualized storage resources, utilization rate of virtualized network resources, utilization rate of virtualized computing resources, number of error codes, request success rate, etc. Detect the acquired KPI data. When abnormal KPI data is detected, for example, the value of a certain KPI obviously exceeds its normal range, it indicates that there is a fault in the network slice, or there is a potential fault. Identify the interruption before it is interrupted and automatically repair it, so that the failure of the network slice can be self-healed and avoid serious consequences such as downtime.

Step S20, when abnormal data is detected in the multi-dimensional index data of the network slice, generate slice self-healing actions and slice fault actions according to the multi-dimensional index data, and judge the generated slice self-healing actions and slice fault actions , to determine the behavior scores of the slice self-healing action and the slice failure action;

In this embodiment, the network slicing fault self-healing method is provided with a self-healing model. When abnormal data is detected in the acquired index data, the model can generate slice self-healing actions and slice faults according to the acquired multi-dimensional index data. actions, and then evaluate the generated fault actions and self-healing actions, and then determine the behavior scores of the generated slice self-healing actions and slice fault actions. When generating slice failure actions, the model is converged toward the direction of the high score of the slice failure action, and then generates fault actions that make network slices difficult to repair. The directions with high scores converge to generate self-healing actions that can repair network slice failures. Then judge the generated slice self-healing action and slice fault action, and make the two confront each other according to the behavior scores of the slice self-healing action and slice fault action, forcing the generated slice self-healing action to continuously improve its fault repair ability.

The refinement of step S20, including steps A1-A2:

Step A1, inputting the multi-dimensional index data into the preset target self-healing model, wherein the target self-healing model is obtained by iteratively training the preset self-healing model to be trained by using the historical multi-dimensional index data of the network slice, The target self-healing model includes an action generator;

Step A2, using the action generator of the target self-healing model to generate a slice self-healing action and a slice fault action according to the multi-dimensional index data.

Furthermore, when generating slice self-healing actions and slice fault actions, the acquired multi-dimensional index data can be preprocessed first. The preprocessing includes but is not limited to normalization processing. Taking normalization processing as an example, the acquired The indicator data is scaled proportionally, and uniformly mapped to a smaller interval, that is, the data is scaled to a given minimum and maximum value, usually between 0 and 1, to obtain the target indicator data, and then to obtain When the target index data is processed, the convergence speed of data and the accuracy of data processing can be improved. Input the preprocessed multidimensional index data into the preset target self-healing model. The target self-healing model is obtained by iterative training of the basic self-healing model to be trained by using the historical multidimensional index data of the network slice. The action generator including dual agents can generate slice self-healing actions and slice fault actions respectively according to the input multi-dimensional index data.

Further, the action generator includes a self-healing action generator and a fault action generator. In step A2, according to the multidimensional index data, the step of generating slice self-healing action and slice fault action by using the action generator of the target self-healing model is refined , including steps A21-A23:

Step A21, determine the current full information state of the network slice according to the multi-dimensional index data, wherein the full information state includes the first state observable by the self-healing action generator, and the fault action generator can the second state of observation;

Step A22, input the first state in the full information state into the self-healing action generator, and use the self-healing action generator to generate a slice self-healing action;

Step A23, inputting the second state in the full information state into the fault action generator, and using the fault action generator to generate slice fault actions.

In this embodiment, the action generator of the target self-healing model includes a self-healing action generator (actor1) and a fault action generator (actor2), and uses the preset action generator of the target self-healing model to generate slice self-healing actions and When a slice fails to act, first determine the current full information state of the network slice according to the input target index data. The full information state includes the first state (s ₁ ) that actor1 can observe and the second state (s ₂ ). Then use actor1 to generate slice self-healing actions according to the state s ₁ of the network slice it can observe, and use actor2 to generate slice failure actions according to the state s ₂ of the network slice it can observe.

Step S30, perform iterative optimization on the generated slice self-healing action and slice fault action according to the behavior score to determine a target self-healing action, and use the target self-healing action to perform fault repair on the network slice.

Using the target self-healing model, different action generators generate corresponding actions according to the partial states of the network slices that they can observe, evaluate the generated slice self-healing actions and slice failure actions, and determine the behavior scores of the generated actions , iteratively optimizes the generated slice self-healing actions and slice fault actions according to the behavior score, forcing the self-healing action generator to generate slice self-healing actions with higher fault repair capabilities, and then determine the network slice fault self-healing action Target self-healing action. The fault repair of the network slice is performed by using the target self-healing action to make the fault self-healing of the network slice. Wherein, the way of repairing the failure of the network slice includes but not limited to switching the state of the network slice according to the determined target self-healing action.

In this embodiment, by acquiring the multi-dimensional index data of the network slice, and detecting the acquired multi-dimensional index data; when abnormal data is detected in the multi-dimensional index data of the network slice, a slice is generated according to the multi-dimensional index data Self-healing actions and slice fault actions, and evaluate the generated slice self-healing actions and slice fault actions to determine the behavior scores of the slice self-healing actions and slice fault actions; The self-healing action and the slice failure action are iteratively optimized to determine a target self-healing action, and the network slice is repaired by using the target self-healing action. When it is detected that there is an anomaly in most of the index data of the network slice, it will actively create a fault by generating a slice fault action, and fight against the generated slice self-healing action, forcing the generated slice self-healing action to improve its own fault repair ability, thereby improving The fault self-healing performance of the network slicing self-healing strategy is improved, and the fault problem in the network slicing can be identified and repaired in time, and the fault self-healing can be recovered before the network slicing fault causes interruption, avoiding serious consequences, and improving the self-healing recovery of the network slicing ability.

Further, on the basis of the above-mentioned embodiments of the present invention, a second embodiment of the network slice fault self-healing method of the present invention is proposed.

This embodiment is a refinement of step S20 in the first embodiment. The difference between this embodiment and the above-mentioned embodiments of the present invention is that the target self-healing model in this embodiment also includes a self-healing action evaluator and a fault action evaluator. , in step S20 of the above-mentioned embodiment, evaluating the generated slice self-healing action and slice failure action to determine the refinement of the behavior scores of the slice self-healing action and the slice failure action: including steps B1-B3:

Step B1, input the first state in the full information state, the slice self-healing action and the slice failure action into the self-healing action evaluator in the target self-healing model, so as to evaluate the slice self-healing action The healing action is judged to determine the behavior score of the slice self-healing action;

Step B2, determining the fault damage radius according to the slice fault action;

Step B3, inputting the second state in the full information state, the fault damage radius, the slice self-healing action and the slice fault action into the fault action evaluator in the target self-healing model to Evaluate the slice fault action, and determine the behavior score of the slice fault action.

Based on the above-mentioned embodiments, in this embodiment, when judging the generated slice self-healing actions and slice fault actions, the full information state of the network slice and the generated slice self-healing The evaluation is performed in the evaluator of the healing model, and then the behavior scores of the generated slice self-healing actions and slice fault actions are determined. Specifically, the target self-healing model also includes a self-healing action evaluator and a failure action evaluator, the slice self-healing action (a ₁ ) to be generated, the state s ₁ of the network slice that the self-healing action generator actor1 can observe, and the generated The slice fault action (a ₂ ) of the slice is input into the self-healing action evaluator (critic1) to judge _the slice self-healing action, and obtain the behavior score Q ₁ (s ₁ ,a ₁ ,a ₂ ) of the slice self-healing action a 1 . Determine the fault damage radius (c ₂ ) according to the generated slice fault actions, and combine the network slice state s ₂ , fault damage radius c ₂ , slice fault action a ₂ and slice self-healing action a ₁ that can be observed by the fault action generator actor2 Input to the fault action evaluator (critic2) to judge the slice fault action, and obtain the behavior score _{Q 2 (} s ₂ , c ₂ , a ₂ , a ₁ ) of the slice fault action a 2 .

Further, the refinement of step S30 in the above embodiment also includes steps C1-C2:

Step C1, feed back the behavior score of the slice self-healing action to the self-healing action generator, and feed back the behavior score of the slice fault action to the fault action generator, return and execute the The self-healing action generator in the target self-healing model generates a slice self-healing action, and uses the fault action generator in the target self-healing model to generate a slice fault action, so that the slice self-healing action and the slice The fault action is optimized and updated until the behavior score of the slice self-healing action and the behavior score of the slice fault action meet a preset condition, and a target self-healing action is obtained;

Step C2, switch the network slice from the current full information state to the target full information state corresponding to the target self-healing action, so as to repair the fault of the network slice.

Furthermore, after determining the behavior scores of slice self-healing actions and slice failure actions, the target self-healing actions are determined according to the behavior scores, so that when performing fault repairs on network slices, the behavior scores of slice self-healing actions and slice failure actions need to be Respectively feed back to the corresponding action generators actor1 and actor2, return and execute using the self-healing action generator in the target self-healing model to generate slice self-healing actions, and use the fault action generator in the target self-healing model The step of generating a slice fault action is to generate a new slice self-healing action and a slice fault action. Then judge the newly generated slice self-healing action and slice fault action, and continuously optimize and update the generated slice self-healing action and slice fault action through the DNQ (Deep Q-Learning, deep Q-learning) loop structure to form actor1 and Actor2 dual-agent confrontation improves the stability and convergence of the self-healing model. The target self-healing action is determined until the generated slice self-healing action and slice fault action meet the preset condition. The preset condition may be that the slice self-healing action has the highest behavior score and can repair the fault generated by the slice fault action.

Referring to Figure 3, Figure 3 is a schematic diagram of the confrontation process of slice self-healing actions and slice failure actions generated by the target self-healing model in this embodiment. In Figure 3, after the network slice outputs a full information state s _all , actor1 and actor2 only It can obtain part of the state information that can be observed by itself, while critic1 and critic2 can obtain the full information state, and at the same time obtain the strategic actions taken by the two agents (intelligent bodies), namely actor1 and actor2. That is to say, although different actors cannot obtain all the information, nor can they obtain the strategies of other actors, but the evaluator corresponding to each actor can observe all the information and guide the corresponding actor to optimize its own strategy. Specifically, the behavior scores obtained by the critic1 and critic2, that is, the values _Q1 and _Q2 are respectively fed back to the action generators actor1 and actor2, and the action generators actor1 and actor2 determine the current behavior according to the determined behavior strategy corresponding to the current behavior score Whether the generated action is the optimal policy, if not, generate a new action to optimize and update its own policy. Among them, the deterministic behavior policy of the action generators actor1 and actor2 is DPG (Deterministic Policy Gradient, deterministic policy gradient), so that each step of its own behavior can directly obtain a definite value through the function μ. DPG is shown in the following formula 1 Show:

a _t = μ(s _t |θ ^μ ) (1)

In formula 1, a _t is the action selected at time t, s _t is the state of the environment at time t, θ ^μ is the weight value, and this function μ is the optimal behavior strategy, and training it can obtain a deterministic optimal behavior strategy. The optimal behavior policy function, the deterministic policy gradient formula is shown in the following formula 2:

in, Indicates the gradient. It should be noted that the deterministic policy in the present invention is based on DPG's DDPG (DeepDeterministic Policy Gradient, deep deterministic policy gradient), which combines the deep learning neural network with the policy learning method of DPG, that is, the value function and Policy functions are expressed through neural networks. Compared with DPG, DDPG uses the neural network to simulate the policy function μ and Q function, that is, the policy network and Q network, and then uses the method of deep learning to train the above neural network, and obtains the action generator actor1 and actor2 of the target self-healing model to select and generate Deterministic policy gradient for actions.

Further, the deterministic policy gradient shown in the foregoing formula 1-2 is only a preferred policy of this embodiment, and is used to illustrate the embodiment of the present invention, and is not used to limit the present invention. In this embodiment, the generated slice self-healing action and slice failure action are optimized and updated according to the behavior score obtained by the evaluator, and then after the target self-healing action is determined, according to the target full information state of the network slice corresponding to the target self-healing action , switch the network slice from the current full information state to the target full information state, so as to complete the fault repair of the network slice. Among them, the switching of the full information state of the network slice is, for example, the wireless network sub-slice: switch to the standby service channel; the transmission network sub-slice: switch to the standby logical port; the core network sub-slice: switch to the standby virtual machine, etc.

In this embodiment, the dual-agents are controlled to generate slice self-healing actions and slice fault actions, and the evaluator is used to judge the generated slice self-healing actions and slice fault actions. Based on the deterministic policy gradient, the evaluated behavior scores Optimize and update the generated slice self-healing actions and slice fault actions, so that the behavior of the dual agents of the action generator can generate confrontation actions, forcing the self-healing action generator to generate self-healing actions with higher fault repair capabilities, and then determine the optimal The behavior strategy is the target self-healing action, which improves the generalization ability of the self-healing strategy, switches the state of the network slice according to the determined target self-healing action, and improves the fault self-healing ability of the network slice.

Based on the above-mentioned embodiments, a third embodiment of the network slicing fault self-healing method of the present invention is proposed. This embodiment is the step after step S30 in the above-mentioned embodiment. After step S30, steps D1-D2 are also included:

Step D1, inputting the target self-healing action into a preset reward function to obtain the reward value corresponding to the target self-healing action;

Step D2, generating an experience replay data set according to the reward value, and updating the model parameters of the target self-healing model according to the experience replay data set.

Based on the above-mentioned embodiment, in this embodiment, after the network slice is repaired according to the determined target self-healing action, the target self-healing action is input into the preset reward function, so as to obtain the reward corresponding to the target self-healing action Value, according to the reward value to generate the experience replay data set, and use the generated experience replay data set to update the model parameters of the target self-healing model.

Specifically, both the action generator and the evaluator of the target self-healing model include an estimation network and a target network, and the generated experience playback data such as (s, a, r, s'), where s is the network before the target self-healing action switch The full information state of the slice, s' is the full information state of the network slice after the target self-healing action is switched, a=(a ₁ , a ₂ ) is the behavior of actor1 and actor2 based on the full information state of the network slice, r Indicates the reward value obtained by the actor from the environment after performing the behavior. The generated experience playback data set records the actions, rewards, and results of the next state in each state of the network slice (s, a, r, s'). When the experience playback data set has a certain storage capacity, when the data storage is full, the first stored data can be overwritten.

When using the experience replay data set to update the model parameters, it is mainly to update the estimated network. The parameters of the target network can be obtained by tracking the estimated network parameters. When updating the estimated network, first initialize the action generator actor network μ (sθ ^μ ) and the critic network Q(s,a|θ ^Q ), initialize the weight values as θ ^μ and θ ^Q respectively, and then initialize the target network Q'=Q(s,a|θ ^Q ) and μ'= μ(s|θ ^μ ), randomly select N pieces of experience replay data from the experience replay data set, and update the parameters of the evaluator according to the objective function shown in the following formula 3:

y _i ＝r _i +γQ'(s _i+1 ,μ'(s _i+1 |θ ^μ′ |θ ^Q′ )) (3)

Among them, y _i represents the target network, _ri represents the reward value, θ ^μ' and θ ^Q' represent the target weight value, γ represents the discount factor, and the parameters of the evaluator are updated by minimizing the loss function shown in the following formula 4:

In formula 4, L is the loss function. In this embodiment, the Q value is obtained based on the square loss of the actual and estimated Q values. The estimated Q value is based on the current state s and the action a outputted by the action estimation network. The input estimation network is obtained, and the actual Q value is obtained by adding the actual reward r and the discounted value of the Q value. The actual Q value inputs the state s' of the next moment and the action a' obtained by the action reality network into the reality The network gets. The parameters of the evaluator are updated by minimizing the loss function. For the update of the actor of the action generator, based on the deterministic policy gradient shown in the above formula 1, the deterministic policy gradient is used to update the actor network. The purpose of the actor is to obtain an action with a high Q value as much as possible. Therefore, the loss of the actor can be as simple as The understanding is that the larger the feedback Q value obtained, the smaller the loss, and the smaller the feedback Q value obtained, the greater the loss. Therefore, using the parameter gradient of the actor itself Action Gradients Combined with Evaluator /> According to the gradient update method shown in the following formula 5, the gradient of the actor network /> Update to make the actor update its own parameters in the direction that is more likely to obtain a larger Q value.

Finally, by tracking and learning the parameter update of the estimated network, the target network parameters are updated according to the following formula 6. Formula 6 is the weight update method of the critic and actor target network:

θ′ _i ←τθ _i +(1-τ)θ′ _i (6)

In this embodiment, τ is set as a number very close to 1, so that the parameter θ of the target network will not fluctuate greatly, thereby improving the stability of the model.

Before step S20, steps E1-E3 are also included:

Step E1, obtaining the historical indicator data of the network slice and performing preprocessing to obtain sample data;

Step E2, obtaining model architecture parameters, and establishing a basic self-healing model according to the model architecture parameters;

Step E3, using the sample data to iteratively train the basic self-healing model to obtain a target self-healing model.

Furthermore, based on the above embodiments, before generating the slice self-healing action and the slice fault action according to the processed target index data, the model needs to be pre-trained. When pre-training the model, first obtain the historical multi-dimensional index data of the network slice and perform preprocessing to obtain the sample data for model pre-training, then obtain the policy function, evaluation function and model architecture parameters, and build a basic self-healing model. Then use the sample data to iteratively train the basic model to obtain the target self-healing model. Among them, the architectural parameters of the model include the number of layers of the neural network of the model, and the number of neurons set in each layer of the neural network.

Based on the target self-healing model in the above embodiments, the basic self-healing model built in this embodiment is a dual-agent DDPG model, including two action generators (actors) and two critics (critics), each actor or The critic contains two neural networks with the same structure, the target network (target_net) and the estimated network (eval_net), only the parameter update frequency is different. Taking actor1 and critic1 above as examples, the basic self-healing model built is shown in Figure 4. Figure 4 is a schematic diagram of the neural network hierarchy of the self-healing action generator and self-healing action evaluator of the basic self-healing model to be trained. In Figure 4, the input layer of actor1 is used to input the state s ₁ of the current slice network, and the hidden layer includes 3 fully connected layers (Dense), with 256 and 128 neurons respectively. After each fully connected layer, a discarding layer is introduced to effectively avoid model overfitting. In the discarding layer, neurons are discarded with probability p and other neurons are retained with probability q=1-p. In this embodiment, discarding is set Probability p=0.2, that is, randomly ignore 20% of neurons, making them invalid. The output layer is a fully connected layer (Dense): set 3 neurons to output the self-healing action of the network sub-slice, including the wireless network sub-slice (switched to the standby service channel), the transmission network sub-slice (switched to the standby logical port) , Core network sub-slicing (switching to the standby virtual machine).

Criter critic1 sets up two input layers respectively, one input layer 1 is used to receive the full information state of the network slice in the latest T time period, and the other input layer 2 is used to receive the generated slice self-healing action and slice failure action. Set up two fully connected layers (Dense) under layer 1, respectively set 256 and 128 neurons, set up a fully connected layer (Dense) under input layer 2, and set 16 neurons, and then pass the merge layer (merge ) to combine actions and states, and finally set up a fully connected layer containing 128 neurons and an output layer containing only 1 neuron for the final output of Q ₁ (s ₁ ,a ₁ , a ₂ ) value.

After building the neural network hierarchical structure of the basic self-healing model, use the obtained sample data to train the built and self-healing models. During the training, two intelligent actors are trained separately. Specifically, wireless access will be included. The historical network slice multi-dimensional KPI (s ₁ ) of the network sub-slice, transmission network sub-slice, and core network sub-slice is input to actor1 composed of a fully connected neural network, and the generated self-healing action corresponding to the network slice is output. Then input the state KPI (s ₁ ) of the network slice, the generated slice self-healing action (a ₁ ), and the slice fault action (a ₂ ) into the critic1 composed of a multi-branch fully connected neural network, and output the evaluation of this The Q value Q ₁ (s ₁ , a ₁ , a ₂ ) of the action selection. Feedback the Q ₁ value to actor1, and actor1 selects the self-healing action that can minimize the business impact to the greatest extent according to the Q ₁ value, and the actor1 model weight after training convergence can be used as the slice self-healing action generator.

Then input the historical network slice multi-dimensional KPI (s ₂ ) and fault damage radius (c ₂ ) including the radio access network sub-slice, transmission network sub-slice, and core network sub-slice KPI into actor2 composed of a fully connected neural network , output the generated fault generation action (a ₂ ) for this slice. Input network slice status KPI (s ₂ ), fault damage radius (c ₂ ), generated slice self-healing action (a ₁ ), and slice fault action (a ₂ ) to the critic2 composed of multi-branch fully connected neural network In , output the Q value Q ₂ (s ₂ ,c ₂ a ₂ ,a ₁ ) for evaluating this action choice. The Q ₂ value is fed back to the action generator, and actor2 selects the slice fault action that can make the slice difficult to self-heal to the greatest extent under a certain fault damage radius according to the Q ₂ value. The actor2 model weights after training convergence can be used as slice fault action generators. In the process of model pre-training, the experience playback data (s, a, r, s') of all stages in the training process are saved to form an experience playback data set, which is used to update the basic data set of parameters during the model running process. The sample data completes the iterative training of the built basic self-healing model, that is, the target self-healing model is obtained.

In this embodiment, the built basic self-healing model is iteratively trained by using the historical multi-dimensional index data of the network slice to obtain the target self-healing model, and after completing the fault repair of the network slice according to the target self-healing action, use The target self-healing action generates an experience playback data set, and then updates the model parameters, improving the self-healing performance of the target self-healing model.

In addition, referring to FIG. 5 , an embodiment of the present invention also proposes a network slice fault self-healing device, and the network slice fault self-healing device includes:

A data detection module 10, configured to obtain multidimensional index data of network slices, and inspect the obtained multidimensional index data;

The action generation module 20 is configured to generate a slice self-healing action and a slice fault action according to the multi-dimensional index data when abnormal data is detected in the multi-dimensional index data of the network slice, and to generate slice self-healing actions and slice fault actions. The fault action is judged to determine the behavior score of the slice self-healing action and the slice fault action;

The dual-agent confrontation module 30 is configured to iteratively optimize the generated slice self-healing action and slice failure action according to the behavior score, so as to determine the target self-healing action, and use the target self-healing action to perform Bug fixes.

Optionally, the action generating module 20 is further configured to:

Input the multi-dimensional index data into the preset target self-healing model, wherein the target self-healing model is obtained by iteratively training the preset self-healing model to be trained by using the historical multi-dimensional index data of network slices, and the target Self-healing models include action generators;

According to the target index data, an action generator of the target self-healing model is used to generate a slice self-healing action and a slice fault action.

Optionally, the action generating module 20 is further configured to:

Determine the current full information state of the network slice according to the multidimensional index data, wherein the full information state includes a first state observable by the self-healing action generator, and a first state observable by the fault action generator Two states;

inputting the first state in the full information state into the self-healing action generator, and using the self-healing action generator to generate slice self-healing actions;

The second state in the full information state is input into the fault action generator, and the slice fault action is generated by the fault action generator.

Optionally, the action generating module 20 is further configured to:

input the first state in the full information state, the slice self-healing action and the slice fault action into the self-healing action evaluator in the target self-healing model, so as to evaluate the slice self-healing action Judging, determining the behavior score of the self-healing action of the slice;

determining the fault damage radius according to the slice fault action;

inputting the second state in the full information state, the fault damage radius, the slice self-healing action and the slice fault action into the fault action evaluator in the target self-healing model, so as to evaluate the Slice fault actions are judged to determine the behavior score of the slice fault actions.

Optionally, the dual-agent confrontation module 30 is also used for:

Feedback the behavior score of the slice self-healing action to the self-healing action generator, and feed back the behavior score of the slice fault action to the fault action generator, return and execute the self-healing action using the target The self-healing action generator in the model generates a slice self-healing action, and uses the fault action generator in the target self-healing model to generate a slice fault action, so as to perform the slice self-healing action and the slice fault action Optimizing and updating until the behavior score of the slice self-healing action and the behavior score of the slice failure action meet a preset condition, and obtain a target self-healing action;

Switching the network slice from the current full information state to the target full information state corresponding to the target self-healing action, so as to perform fault repair on the network slice.

Optionally, the network slice fault self-healing device further includes a self-healing policy update module, configured to:

Inputting the target self-healing action into a preset reward function to obtain a reward value corresponding to the target self-healing action;

An experience replay data set is generated according to the reward value, and model parameters of the target self-healing model are updated according to the experience replay data set.

Optionally, the network slicing fault self-healing device also includes a model training module for:

Obtain the historical indicator data of the network slice and perform preprocessing to obtain sample data;

Obtain model architecture parameters, and establish a basic self-healing model according to the model architecture parameters;

Using the sample data to iteratively train the basic self-healing model to obtain a target self-healing model.

In addition, an embodiment of the present invention also proposes a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, operations in the network slice fault self-healing method provided in the foregoing embodiments are implemented.

The various embodiments of the device and the computer program product of the present invention can refer to the various embodiments of the network slicing fault self-healing method of the present invention, which will not be repeated here.

It should be noted that in this article, relational terms such as first and second etc. are only used to distinguish one entity/operation/object from another entity/operation/object and do not necessarily require or imply these the existence of any such actual relationship or order between entities/operations/objects; the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, An article or system includes not only those elements, but also other elements not expressly listed, or elements inherent in such a process, method, article, or system. Without further limitations, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article or system comprising that element.

As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment. The apparatus embodiments described above are merely illustrative, where units illustrated as separate components may or may not be physically separate. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. It can be understood and implemented by those skilled in the art without creative effort.

The serial numbers of the above embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product in essence or in other words, the part that contributes to the prior art, and the computer software product is stored in a storage medium (such as ROM/RAM) as described above. , magnetic disk, optical disk), including several instructions to make a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the network slicing fault self-healing method described in various embodiments of the present invention .

The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process conversion made by using the description of the present invention and the contents of the accompanying drawings, or directly or indirectly used in other related technical fields , are all included in the scope of patent protection of the present invention in the same way.

Claims (9)

1. The network slice fault self-healing method is characterized by comprising the following steps of:

acquiring multi-dimensional index data of the network slice, and detecting the acquired multi-dimensional index data _；

When abnormal data are detected to exist in the multi-dimensional index data of the network slice, generating a slice self-healing action and a slice fault action according to the multi-dimensional index data, and judging the generated slice self-healing action and slice fault action to determine the action scores of the slice self-healing action and the slice fault action;

the step of generating a slice self-healing action and a slice fault action according to the multidimensional index data comprises the following steps:

inputting the multi-dimensional index data into a preset target self-healing model, wherein the target self-healing model is obtained by performing iterative training on the preset self-healing model to be trained by utilizing the historical multi-dimensional index data of a network slice, and the target self-healing model comprises an action generator;

generating a slicing self-healing action and a slicing fault action by using an action generator of the target self-healing model according to the multidimensional index data;

and performing iterative optimization on the generated slice self-healing action and slice fault action according to the behavior score to determine a target self-healing action, and performing fault repair on the network slice by using the target self-healing action.

2. The network slice fault self-healing method of claim 1, wherein the action generator comprises a self-healing action generator and a fault action generator, the step of generating slice self-healing actions and slice fault actions using the action generator of the target self-healing model according to the multi-dimensional index data comprising:

determining a current full information state of the network slice according to the multi-dimensional index data, wherein the full information state comprises a first state observable by the self-healing action generator and a second state observable by the fault action generator;

inputting a first state in the full information states into the self-healing action generator, and generating a slicing self-healing action by using the self-healing action generator;

and inputting a second state in the all-information states into the fault action generator, and generating a slicing fault action by using the fault action generator.

3. The network slice fault self-healing method of claim 2, wherein the target self-healing model further comprises a self-healing action evaluator and a fault action evaluator, the step of evaluating the generated slice self-healing action and slice fault action to determine a behavioral score of the slice self-healing action and the slice fault action comprising:

Inputting a first state in the full information state, the slice self-healing action and the slice fault action into a self-healing action judging device in the target self-healing model so as to judge the slice self-healing action and determine a behavior score of the slice self-healing action;

determining a fault destruction radius according to the slice fault action;

and inputting a second state in the full information state, the fault destruction radius, the slice self-healing action and the slice fault action into a fault action judging device in the target self-healing model so as to judge the slice fault action and determine a behavior score of the slice fault action.

4. The network slice fault self-healing method of claim 3, wherein the step of iteratively optimizing the generated slice self-healing actions and slice fault actions according to the behavioral scores to determine a target self-healing action and utilizing the target self-healing action to perform fault remediation on the network slice comprises:

feeding back the behavior score of the slice self-healing action to the self-healing action generator, feeding back the behavior score of the slice fault action to the fault action generator, returning to and executing the step of generating the slice self-healing action by using the self-healing action generator in the target self-healing model and generating the slice fault action by using the fault action generator in the target self-healing model, so as to perform iterative optimization on the slice self-healing action and the slice fault action until the behavior score of the slice self-healing action and the behavior score of the slice fault action meet preset conditions, and obtaining the target self-healing action;

And switching the state of the network slice according to the target self-healing action so as to switch the network slice from the current full-information state to the target full-information state corresponding to the target self-healing action, and repairing the fault of the network slice.

5. The network slice fault self-healing method of claim 4, wherein the step of iteratively optimizing the generated slice self-healing actions and slice fault actions according to the behavioral scores to determine a target self-healing action and utilizing the target self-healing action to perform fault remediation on the network slice comprises:

inputting the target self-healing action into a preset rewarding function to obtain a rewarding value corresponding to the target self-healing action;

and generating an experience playback data set according to the reward value, and updating model parameters of the target self-healing model according to the experience playback data set.

6. The network slice fault self-healing method of claim 1, wherein prior to the step of generating slice self-healing actions and slice fault actions from the multi-dimensional index data, further comprising:

acquiring historical index data of a network slice and preprocessing the historical index data to obtain sample data;

Obtaining model architecture parameters, and establishing a basic self-healing model according to the model architecture parameters;

and carrying out iterative training on the basic self-healing model by using the sample data to obtain a target self-healing model.

7. A network slice fault self-healing device, characterized in that the network slice fault self-healing device comprises:

the data detection module is used for acquiring the multi-dimensional index data of the network slice and detecting the acquired multi-dimensional index data;

the action generating module is used for generating a slice self-healing action and a slice fault action according to the multi-dimensional index data when abnormal data exist in the multi-dimensional index data of the network slice, and judging the generated slice self-healing action and slice fault action to determine the action scores of the slice self-healing action and the slice fault action; inputting the multi-dimensional index data into a preset target self-healing model, wherein the target self-healing model is obtained by performing iterative training on the preset self-healing model to be trained by utilizing the historical multi-dimensional index data of a network slice, and the target self-healing model comprises an action generator; generating a slicing self-healing action and a slicing fault action by using an action generator of the target self-healing model according to the multidimensional index data;

And the dual-agent countermeasure module is used for carrying out iterative optimization on the generated slice self-healing action and slice fault action according to the behavior score so as to determine a target self-healing action and carrying out fault repair on the network slice by utilizing the target self-healing action.

8. A network slice fault self-healing device, the network slice fault self-healing device comprising: memory, a processor and a network slice fault self-healing program stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the network slice fault self-healing method according to any one of claims 1 to 6.

9. A storage medium having stored thereon a computer program which, when executed by a processor, implements the network slice fault self-healing method according to any one of claims 1-6.