PON network AI dynamic regulation and control method and system fused with large model

Abstract

The embodiment of the invention relates to the technical field of PON networks, and particularly discloses a PON network AI dynamic regulation and control method and system fusing a large model. According to the embodiment of the invention, multi-source data acquisition is carried out, and data fusion and feature extraction are carried out on multi-source acquired data; based on the fusion large model, processing the structured feature data and the historical traffic data; performing strategy planning and constraint embedding to generate a plurality of candidate regulation and control strategies; selecting a target regulation and control strategy, and issuing and executing the strategy; and calculating a target strategy income, and performing strategy feedback optimization. Multi-source data acquisition can be carried out, large models are fused, future bandwidth requirements are predicted, strategies are generated and selected, an optimal target regulation and control strategy is selected to be issued and executed according to the current network state, the target strategy income is calculated, strategy feedback optimization is carried out, bandwidth allocation is reasonable, service quality is ensured, bandwidth resource waste is avoided, and the network quality is improved. And the regulation and control efficiency of the PON network is effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of PON network, and particularly relates to a PON network AI dynamic regulation method and system fusing a large model. BACKGROUND

[0002] A PON network is an optical fiber access technology based on a point-to-multipoint (P2MP) topology structure, and its core feature is to realize signal distribution of a single optical fiber to multiple user terminals through a passive optical splitter (Passive Optical Splitter), without the participation of active electronic devices (such as amplifiers and switches), and it is a mainstream solution of the current fiber access network (FTTx) and is widely used in home broadband, enterprise private line, mobile base station backhaul and other scenarios.

[0003] Currently, the traditional PON network regulation mainly adopts a static bandwidth allocation mode to allocate a fixed bandwidth for each optical network unit, and cannot predict and dynamically adjust future bandwidth demand according to actual business demand, which may affect the service quality due to insufficient bandwidth allocation, and may also cause some bandwidth to be in an idle state, resulting in waste of bandwidth resources. SUMMARY

[0004] The purpose of the embodiments of the application is to provide a PON network AI dynamic regulation method and system fusing a large model, which aims to solve the problems proposed in the background art.

[0005] To achieve the above-mentioned purpose, the embodiments of the application provide the following technical solutions: A PON network AI dynamic regulation method fusing a large model, the method specifically comprises the following steps: Multi-source data acquisition is performed to obtain multi-source collected data, and data fusion and feature extraction are performed on the multi-source collected data to obtain structured feature data; Historical traffic data is obtained, and the structured feature data and the historical traffic data are processed based on a preset fusion large model to obtain future bandwidth demand and abnormal detection data; Strategy planning and constraint embedding are performed according to the future bandwidth demand and the abnormal detection data to generate a plurality of candidate regulation strategies; The current network state is obtained, a target regulation strategy is selected from the plurality of candidate regulation strategies according to the current network state, and strategy issuing and execution are performed; The strategy expected effect and the strategy actual effect are obtained, the target strategy benefit is calculated, and strategy feedback optimization is performed according to the target strategy benefit.

[0006] As a further limitation of the technical scheme of the embodiment of the application, the multi-source data acquisition, the acquisition of multi-source acquisition data, and the data fusion and feature extraction of the multi-source acquisition data to obtain structured feature data specifically include the following steps: Multi-source data acquisition is performed to acquire multi-source acquisition data. The multi-source acquisition data is effectively extracted and standardized to generate multi-source standard data. The multi-source standard data is temporally and spatially aligned and associatedly fused to generate fused standard data. The fused standard data is analyzed and extracted for timing features, spatial features, and context features to obtain structured feature data.

[0007] As a further limitation of the technical scheme of the embodiment of the application, the multi-source acquisition data is composed of network layer data, service layer data, and environment layer data, wherein the network layer data includes bandwidth utilization, delay, and packet loss rate; the service layer data includes service type and QoS requirement; and the environment layer data includes fiber link loss, temperature, and device power consumption.

[0008] As a further limitation of the technical scheme of the embodiment of the application, the acquisition of historical traffic data, the processing of the structured feature data and the historical traffic data based on a preset fusion large model, and the acquisition of future bandwidth demand and anomaly detection data specifically include the following steps: Historical traffic data is acquired. Future bandwidth demand is predicted by a preset fusion large model in combination with the historical traffic data and the structured feature data. Real-time traffic is acquired. Anomaly deviation analysis is performed on the future bandwidth demand based on the real-time traffic to record anomaly detection data.

[0009] As a further limitation of the technical scheme of the embodiment of the application, the strategy planning and constraint embedding based on the future bandwidth demand and the anomaly detection data to generate multiple candidate control strategies specifically include the following steps: Strategy planning is performed based on the future bandwidth demand and the anomaly detection data to generate multiple planning control strategies. Constraint analysis is performed on the multiple planning control strategies to determine multiple constraint conditions. The multiple constraint conditions are embedded in the multiple corresponding planning control strategies to generate multiple candidate control strategies.

[0010] As a further limitation of the technical scheme of the embodiment of the application, the acquisition of the current network state, the selection of a target control strategy from the multiple candidate control strategies based on the current network state, and the strategy issuance and execution specifically include the following steps: acquire a current network state; dynamically select an optimal target regulation strategy from a plurality of the candidate regulation strategies according to the current network state; generate a target regulation instruction according to the target regulation strategy; issue and execute the target regulation instruction.

[0011] As a further limitation of the technical scheme of the embodiment of the present application, the acquisition of the strategy expected effect and the strategy actual effect, the calculation of the target strategy benefit, and the strategy feedback optimization according to the target strategy benefit specifically include the following steps: determine a corresponding strategy expected effect according to the target regulation strategy; perform actual monitoring to acquire a strategy actual effect; compare the strategy actual effect with the strategy expected effect to calculate a target strategy benefit; plan strategy optimization parameters according to the target strategy benefit, and perform strategy feedback optimization according to the strategy optimization parameters.

[0012] As a further limitation of the technical scheme of the embodiment of the present application, the calculation formula of the target strategy benefit is as follows: ; wherein, is the target strategy benefit, is an actual service quality, is an expected service quality, is an expected energy consumption, is an actual energy consumption, and is a preset influence coefficient.

[0013] A PON network AI dynamic regulation system fusing a large model, the system comprising a multi-source data acquisition unit, a demand prediction and anomaly detection unit, a regulation strategy generation unit, a strategy selection and execution unit, and a strategy feedback optimization unit, wherein: The multi-source data acquisition unit is configured to acquire multi-source data, acquire multi-source acquisition data, and perform data fusion and feature extraction on the multi-source acquisition data to obtain structured feature data. The demand prediction and anomaly detection unit is configured to acquire historical traffic data, process the structured feature data and the historical traffic data based on a preset fusion large model to acquire future bandwidth demand and anomaly detection data. The regulation strategy generation unit is configured to perform strategy planning and constraint embedding according to the future bandwidth demand and the anomaly detection data to generate a plurality of candidate regulation strategies. The policy selection execution unit is configured to acquire a current network state, select a target regulation policy from the plurality of candidate regulation policies according to the current network state, and perform policy issuing and execution. The policy feedback optimization unit is configured to acquire a policy expected effect and a policy actual effect, calculate a target policy benefit, and perform policy feedback optimization according to the target policy benefit.

[0014] As a further limitation of the technical scheme of the embodiment of the application, the multi-source data acquisition unit specifically comprises: The multi-source data acquisition module is configured to acquire multi-source data by multi-source data acquisition. The data preprocessing module is configured to effectively extract and standardize the multi-source data to generate multi-source standard data. The data fusion module is configured to perform time-space alignment and correlation fusion on the multi-source standard data to generate fused standard data. The feature extraction module is configured to analyze and extract time-series features, spatial features and context features of the fused standard data to obtain structured feature data.

[0015] Compared with the prior art, the application has the following beneficial effects: The embodiment of the application performs multi-source data acquisition, data fusion and feature extraction on multi-source data, processes structured feature data and historical traffic data based on a fused large model, performs policy planning and constraint embedding to generate a plurality of candidate regulation policies, selects a target regulation policy, performs policy issuing and execution, calculates a target policy benefit, and performs policy feedback optimization. The application can perform multi-source data acquisition, predict future bandwidth demand through a fused large model, generate and select a policy, select an optimal target regulation policy for issuing and execution according to a current network state, calculate a target policy benefit, and perform policy feedback optimization, so that bandwidth allocation is reasonable, service quality is ensured, bandwidth resource waste is avoided, and the regulation efficiency of a PON network is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only some embodiments of the application.

[0017] Figure 1 A flowchart of the method provided by the embodiment of the application is shown.

[0018] Figure 2 A flowchart of multi-source data acquisition processing in the method provided by the embodiment of the application is shown.

[0019] Figure 3A flow chart of acquiring future bandwidth demand and anomaly detection data in the method provided by the embodiment of the application is shown.

[0020] Figure 4 A flow chart of strategy planning and constraint embedding in the method provided by the embodiment of the application is shown.

[0021] Figure 5 A flow chart of selecting a target regulation strategy in the method provided by the embodiment of the application is shown.

[0022] Figure 6 A flow chart of performing strategy feedback optimization in the method provided by the embodiment of the application is shown.

[0023] Figure 7 An application architecture diagram of the system provided by the embodiment of the application is shown.

[0024] Figure 8 A structure block diagram of a multi-source data acquisition unit in the system provided by the embodiment of the application is shown. DETAILED DESCRIPTION

[0025] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0026] It can be understood that at present, the traditional PON network regulation mainly adopts a static bandwidth allocation manner to allocate a fixed bandwidth for each optical network unit, and cannot predict and dynamically adjust the future bandwidth demand according to the actual service demand, which is easy to affect the service quality due to insufficient bandwidth allocation, and can also cause some bandwidth to be in an idle state, thereby causing waste of bandwidth resources.

[0027] To solve the above problems, the embodiment of the application acquires multi-source acquisition data by performing multi-source data acquisition, acquires structured feature data by performing data fusion and feature extraction on the multi-source acquisition data, acquires historical traffic data, processes the structured feature data and the historical traffic data based on a preset fusion large model, acquires future bandwidth demand and anomaly detection data, performs strategy planning and constraint embedding according to the future bandwidth demand and the anomaly detection data, generates a plurality of candidate regulation strategies, acquires a current network state, selects a target regulation strategy from the plurality of candidate regulation strategies according to the current network state, and performs strategy issuing and execution, acquires a strategy expected effect and a strategy actual effect, calculates a target strategy benefit, and performs strategy feedback optimization according to the target strategy benefit. The multi-source data acquisition can be performed, the future bandwidth demand can be predicted through the fusion large model, the strategy generation and selection can be performed, the optimal target regulation strategy can be selected according to the current network state and then issued and executed, the target strategy benefit is calculated, and the strategy feedback optimization is performed, so that the bandwidth allocation is reasonable, the service quality is ensured, the bandwidth resource waste is avoided, and the regulation efficiency of the PON network is effectively improved.

[0028] Figure 1 A flowchart of the method provided by the embodiment of the application is shown.

[0029] Specifically, a PON network AI dynamic regulation method based on a fusion large model specifically includes the following steps: Step S101, multi-source data acquisition is performed to acquire multi-source acquisition data, and data fusion and feature extraction are performed on the multi-source acquisition data to obtain structured feature data.

[0030] In the embodiment of the application, the network layer, the service layer and the environment layer are all collected to acquire multi-source acquisition data including bandwidth utilization, delay, packet loss rate, service type, QoS demand, fiber link loss, temperature and device power consumption, effective data is extracted from the multi-source acquisition data, and the multi-source acquisition data is standardized according to a preset standardization requirement to generate multi-source standard data, then the multi-source standard data is processed by time-space alignment and associated fusion to generate fusion standard data, and the fusion standard data is analyzed and extracted for time sequence feature, spatial feature and context feature to obtain structured feature data.

[0031] It can be understood that the multi-source acquisition data is composed of network layer data, service layer data and environment layer data, wherein the network layer data includes bandwidth utilization, delay and packet loss rate, the service layer data includes service type and QoS demand, and the environment layer data includes fiber link loss, temperature and device power consumption.

[0032] It can be understood that the time sequence feature is a feature of short-term fluctuation mode and periodic trend of traffic, the space feature is a feature of correlation between regional traffic, and the context feature is a feature of service priority.

[0033] Specifically, Figure 2 A flowchart of multi-source data acquisition and processing in the method provided by the embodiment of the application is shown.

[0034] In the preferred embodiment provided by the application, the multi-source data acquisition, the acquisition of multi-source acquisition data, and the data fusion and feature extraction of the multi-source acquisition data to obtain structured feature data specifically include the following steps: Step S1011, multi-source data acquisition is performed to acquire multi-source acquisition data; Step S1012, effective extraction and standardization processing are performed on the multi-source acquisition data to generate multi-source standard data; Step S1013, time-space alignment and correlation fusion are performed on the multi-source standard data to generate fused standard data; Step S1014, time sequence feature, space feature, and context feature analysis and extraction are performed on the fused standard data to obtain structured feature data.

[0035] Further, the PON network AI dynamic regulation method of the fusion large model further includes the following steps: Step S102, historical traffic data is acquired, the structured feature data and the historical traffic data are processed based on the preset fusion large model, and future bandwidth demand and abnormal detection data are acquired.

[0036] In the embodiment of the application, by acquiring historical traffic data, combining the historical traffic data and the structured feature data, and then performing data processing on the historical traffic data and the structured feature data by the preset fusion large model, demand prediction is performed on a preset future time period, future bandwidth demand is obtained, real-time traffic is acquired, abnormal deviation analysis is performed on the future bandwidth demand based on the real-time traffic, threshold comparison is performed on the deviation between the real-time traffic and the predicted value corresponding to the future bandwidth demand, and abnormal detection data is recorded.

[0037] It can be understood that the fusion large model in the embodiment of the application is a large model that fuses LSTM and a graph neural network.

[0038] Specifically, Figure 3 A flowchart of acquiring future bandwidth demand and abnormal detection data in the method provided by the embodiment of the application is shown.

[0039] In the preferred embodiments provided by the present application, the obtaining historical traffic data, processing the structured feature data and the historical traffic data based on a preset fusion large model, and obtaining future bandwidth demand and anomaly detection data specifically include the following steps: Step S1021, obtaining historical traffic data; Step S1022, combining the historical traffic data and the structured feature data, and predicting future bandwidth demand through a preset fusion large model; Step S1023, obtaining real-time traffic; Step S1024, performing anomaly deviation analysis on the future bandwidth demand based on the real-time traffic, and recording anomaly detection data.

[0040] Further, the PON network AI dynamic regulation method of the fusion large model further includes the following steps: Step S103, performing strategy planning and constraint embedding according to the future bandwidth demand and the anomaly detection data, and generating a plurality of candidate regulation strategies.

[0041] In the embodiments of the present application, strategy planning is performed according to the future bandwidth demand and the anomaly detection data, a plurality of planning regulation strategies including dynamic wavelength allocation, bandwidth allocation, and route switching are generated, constraint analysis is performed on the plurality of planning regulation strategies, a plurality of constraint conditions including QoS guarantee and energy consumption limitation are determined, and the plurality of constraint conditions are embedded in the plurality of corresponding planning regulation strategies to generate a plurality of candidate regulation strategies.

[0042] Specifically, Figure 4 A flowchart of strategy planning and constraint embedding in the method provided by the embodiments of the present application is shown.

[0043] In the preferred embodiments provided by the present application, the performing strategy planning and constraint embedding according to the future bandwidth demand and the anomaly detection data, and generating a plurality of candidate regulation strategies specifically include the following steps: Step S1031, performing strategy planning according to the future bandwidth demand and the anomaly detection data, and generating a plurality of planning regulation strategies; Step S1032, performing constraint analysis on the plurality of planning regulation strategies, and determining a plurality of constraint conditions; Step S1033, embedding the plurality of constraint conditions in the plurality of corresponding planning regulation strategies, and generating a plurality of candidate regulation strategies.

[0044] Further, the PON network AI dynamic regulation method of the fusion large model further includes the following steps: Step S104, acquiring a current network state, selecting a target regulation strategy from the plurality of candidate regulation strategies according to the current network state, and performing policy issuing and execution.

[0045] In the embodiment of the application, the current network state is acquired, the current network state including real-time traffic, device load, etc. is acquired, the current network state and the plurality of candidate regulation strategies are analyzed based on the preset real-time decision model, the optimal target regulation strategy is dynamically selected from the plurality of candidate regulation strategies, and the corresponding target regulation instruction is generated according to the target regulation strategy. The target regulation instruction is issued to the OLT / ONU device for policy execution.

[0046] Specifically, Figure 5 A flowchart of selecting a target regulation strategy in the method provided by the embodiment of the application is shown.

[0047] In the preferred embodiment provided by the application, the acquiring of the current network state, the selecting of the target regulation strategy from the plurality of candidate regulation strategies according to the current network state, and the performing of policy issuing and execution specifically include the following steps: Step S1041, acquiring a current network state; Step S1042, dynamically selecting an optimal target regulation strategy from the plurality of candidate regulation strategies according to the current network state; Step S1043, generating a target regulation instruction according to the target regulation strategy; Step S1044, issuing and executing the target regulation instruction.

[0048] Further, the PON network AI dynamic regulation method based on the fusion large model further includes the following steps: Step S105, acquiring a policy expected effect and a policy actual effect, calculating a target policy benefit, and performing policy feedback optimization according to the target policy benefit.

[0049] In the embodiment of the application, the corresponding policy expected effect is determined according to the target regulation strategy, the actual monitoring of the time delay reduction rate, the packet loss rate change, the energy consumption change, etc. is performed in the process of executing the target regulation strategy, the policy actual effect is acquired, the target policy benefit is calculated by comparing the policy actual effect with the policy expected effect, the policy optimization parameter is planned according to the target policy benefit, and then the policy feedback optimization is performed according to the policy optimization parameter. Specifically, the calculation formula of the target policy benefit is: Among them, is the target policy benefit, is an actual service quality, ​for expected service quality, for expected energy consumption, for actual energy consumption, and for preset influence coefficient.

[0050] Specifically, Figure 6 A flow chart of performing policy feedback optimization in the method provided by the embodiment of the application is shown.

[0051] In the preferred embodiment provided by the application, the obtaining of the expected effect of the policy and the actual effect of the policy and the calculation of the target policy benefit, and the performing of policy feedback optimization according to the target policy benefit specifically include the following steps: Step S1051, determining the corresponding policy expected effect according to the target control policy; Step S1052, performing actual monitoring to obtain the policy actual effect; Step S1053, comparing the policy actual effect with the policy expected effect to calculate the target policy benefit; Step S1054, planning policy optimization parameters according to the target policy benefit, and performing policy feedback optimization according to the policy optimization parameters.

[0052] Further, Figure 7 An application architecture diagram of the system provided by the embodiment of the application is shown.

[0053] In another preferred embodiment provided by the application, a PON network AI dynamic control system integrating a large model includes: A multi-source data acquisition unit 101 is configured to acquire multi-source data, obtain multi-source acquisition data, and perform data fusion and feature extraction on the multi-source acquisition data to obtain structured feature data.

[0054] In the embodiment of the application, the multi-source data acquisition unit 101 acquires multi-source acquisition data including bandwidth utilization, time delay, packet loss rate, service type, QoS demand, optical fiber link loss, temperature, and device power consumption by collecting the network layer, the service layer, and the environment layer, effectively identifies the multi-source acquisition data, extracts effective data from the multi-source acquisition data, and performs standardized processing on the data according to preset standardized requirements to generate multi-source standard data. Then, the multi-source standard data is subjected to spatio-temporal alignment and associated fusion processing to generate fusion standard data, and the fusion standard data is subjected to analysis and extraction of time sequence features, spatial features, and context features to obtain structured feature data.

[0055] Specifically, Figure 8 A structural block diagram of the multi-source data acquisition unit 101 in the system provided by the embodiment of the application is shown.

[0056] In the preferred embodiments provided by the present application, the multi-source data acquisition unit 101 specifically comprises: A multi-source data acquisition module 1011 is configured to acquire multi-source data. A data preprocessing module 1012 is configured to effectively extract and standardize the multi-source data to generate multi-source standard data. A data fusion module 1013 is configured to perform spatio-temporal alignment and correlation fusion on the multi-source standard data to generate fused standard data. A feature extraction module 1014 is configured to analyze and extract the fused standard data to obtain structured feature data.

[0057] Further, the PON network AI dynamic regulation system of the fusion large model further comprises: A demand prediction and anomaly detection unit 102 is configured to acquire historical traffic data, process the structured feature data and the historical traffic data based on a preset fusion large model, and obtain future bandwidth demand and anomaly detection data.

[0058] In the embodiments of the present application, the demand prediction and anomaly detection unit 102 combines the historical traffic data and the structured feature data by acquiring the historical traffic data, and then processes the data by the preset fusion large model to predict the demand in a preset future time period, thereby obtaining the future bandwidth demand. Real-time traffic is acquired, and the future bandwidth demand is analyzed for abnormal deviation based on the real-time traffic. The deviation between the real-time traffic and the predicted value corresponding to the future bandwidth demand is compared with a threshold, and anomaly detection data is recorded.

[0059] A regulation strategy generation unit 103 is configured to perform strategy planning and constraint embedding based on the future bandwidth demand and the anomaly detection data to generate a plurality of candidate regulation strategies.

[0060] In the embodiments of the present application, the regulation strategy generation unit 103 performs strategy planning based on the future bandwidth demand and the anomaly detection data to generate a plurality of planning regulation strategies including dynamic wavelength allocation, bandwidth allocation, and routing switching. Constraint analysis is performed on the plurality of planning regulation strategies to determine a plurality of constraint conditions including QoS guarantee and energy consumption limitation. The plurality of constraint conditions are embedded into the plurality of corresponding planning regulation strategies to generate a plurality of candidate regulation strategies.

[0061] A strategy selection and execution unit 104 is configured to acquire a current network state, select a target regulation strategy from the plurality of candidate regulation strategies based on the current network state, and perform strategy delivery and execution.

[0062] In the embodiment of the present application, the policy selection execution unit 104 acquires the current network state, and acquires the current network state including real-time traffic, device load, etc., and then analyzes the current network state and the plurality of candidate regulation policies based on the preset real-time decision model, dynamically selects the optimal target regulation policy from the plurality of candidate regulation policies, and then generates the corresponding target regulation instruction according to the target regulation policy, and issues the target regulation instruction to the OLT / ONU device for policy execution.

[0063] The policy feedback optimization unit 105 is configured to acquire the policy expected effect and the policy actual effect, calculate the target policy benefit, and perform policy feedback optimization according to the target policy benefit.

[0064] In the embodiment of the present application, the policy feedback optimization unit 105 determines the corresponding policy expected effect according to the target regulation policy, and performs actual monitoring of the delay reduction rate, the packet loss rate change, the energy consumption change, etc. during the execution of the target regulation policy, acquires the policy actual effect, compares the policy actual effect with the policy expected effect, calculates the target policy benefit, plans the policy optimization parameter according to the target policy benefit, and then performs policy feedback optimization according to the policy optimization parameter. Specifically, the calculation formula of the target policy benefit is as follows: ; Wherein, is the target policy benefit, is the actual service quality, is the expected service quality, is the expected energy consumption, is the actual energy consumption, and is the preset influence coefficient.

[0065] It should be understood that although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps has no strict sequence limitation, and these steps can be executed in other orders. Moreover, at least a part of the steps in each embodiment can include a plurality of sub-steps or a plurality of stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least a part of other steps or sub-steps or stages of other steps.

[0066] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0067] Any combination of the technical features of the above-mentioned embodiments can be combined. In order to make the description simple, all possible combinations of the technical features in the above-mentioned embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0068] The above-mentioned embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the present application. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

[0069] The above-mentioned is only the preferred embodiment of the present application, and does not limit the present application. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for AI dynamic regulation of a PON network with a large model, characterized in that, The method specifically comprises the following steps: Multi-source data acquisition is performed to obtain multi-source acquisition data, and data fusion and feature extraction are performed on the multi-source acquisition data to obtain structured feature data; Historical traffic data is obtained, the structured feature data and the historical traffic data are processed based on a preset fusion large model, future bandwidth demand and anomaly detection data are obtained; Strategy planning and constraint embedding are performed according to the future bandwidth demand and the anomaly detection data, and a plurality of candidate regulation strategies are generated; A current network state is obtained, a target regulation strategy is selected from the plurality of candidate regulation strategies according to the current network state, and strategy issuing and execution are performed; Strategy expected effect and strategy actual effect are obtained, target strategy benefit is calculated, and strategy feedback optimization is performed according to the target strategy benefit.

2. The PON network AI dynamic regulation method of the fusion large model according to claim 1, characterized in that, The multi-source data acquisition, the obtaining of multi-source acquisition data, and the data fusion and feature extraction on the multi-source acquisition data to obtain structured feature data specifically comprise the following steps: Multi-source data acquisition is performed to obtain multi-source acquisition data; Effective extraction and standardization processing are performed on the multi-source acquisition data to generate multi-source standard data; Temporal and spatial alignment and associated fusion are performed on the multi-source standard data to generate fusion standard data; Temporal feature, spatial feature and context feature analysis and extraction are performed on the fusion standard data to obtain structured feature data.

3. The PON network AI dynamic regulation method of the fusion large model according to claim 2, characterized in that, The multi-source acquisition data is composed of network layer data, service layer data and environment layer data, wherein the network layer data includes bandwidth utilization, delay and packet loss rate; the service layer data includes service type and QoS demand; and the environment layer data includes fiber link loss, temperature and device power consumption.

4. The PON network AI dynamic regulation method of the fusion large model according to claim 1, characterized in that, The obtaining of historical traffic data, the processing of the structured feature data and the historical traffic data based on a preset fusion large model, and the obtaining of future bandwidth demand and anomaly detection data specifically comprise the following steps: Historical traffic data is obtained; Future bandwidth demand is predicted by a preset fusion large model in combination with the historical traffic data and the structured feature data; Real-time traffic is obtained; Abnormal deviation analysis is performed on the future bandwidth demand based on the real-time traffic, and anomaly detection data is recorded.

5. The PON network AI dynamic regulation method of the fusion large model according to claim 1, characterized in that, The strategy planning and constraint embedding according to the future bandwidth demand and the anomaly detection data to generate a plurality of candidate regulation strategies specifically comprise the following steps: Strategy planning is performed according to the future bandwidth demand and the anomaly detection data to generate a plurality of planning regulation strategies; Constraint analysis is performed on the plurality of planning regulation strategies to determine a plurality of constraint conditions; The plurality of constraint conditions are embedded into a plurality of corresponding planning regulation strategies to generate a plurality of candidate regulation strategies.

6. The PON network AI dynamic regulation method of the fusion large model according to claim 1, characterized in that, The obtaining of a current network state, the selection of a target regulation strategy from the plurality of candidate regulation strategies according to the current network state, and the strategy issuing and execution specifically comprise the following steps: A current network state is obtained; The optimal target regulation strategy is dynamically selected from the plurality of candidate regulation strategies according to the current network state; A target regulation instruction is generated according to the target regulation strategy; The target regulation instruction is issued and executed.

7. The PON network AI dynamic regulation method of the fusion large model according to claim 1, characterized in that, The acquisition strategy expected effect and strategy actual effect, calculate the target strategy benefit, according to the target strategy benefit, strategy feedback optimization specifically includes the following steps: According to the target control strategy, determine the corresponding strategy expected effect; Actual monitoring, obtain the strategy actual effect; The strategy actual effect and the strategy expected effect are compared, and the target strategy benefit is calculated; According to the target strategy benefit, planning strategy optimization parameters, and according to the strategy optimization parameters, strategy feedback optimization is carried out.

8. The PON network AI dynamic regulation method of the fusion large model according to claim 7, characterized in that, The calculation formula of the target strategy benefit is: ； wherein, is a target policy benefit, is an actual quality of service, is an expected quality of service, is an expected energy consumption, is an actual energy consumption, and is a preset influence coefficient.

9. A PON network AI dynamic regulation system of a fusion large model, characterized in that, The system includes multi-source data acquisition unit, demand prediction and anomaly detection unit, control strategy generation unit, strategy selection execution unit and strategy feedback optimization unit, wherein: Multi-source data acquisition unit, for multi-source data acquisition, obtains multi-source acquisition data, and carries out data fusion and feature extraction on the multi-source acquisition data to obtain structured feature data; Demand prediction and anomaly detection unit, for obtaining historical traffic data, based on the preset fusion big model, processing the structured feature data and the historical traffic data, obtaining future bandwidth demand and anomaly detection data; Control strategy generation unit, for strategy planning and constraint embedding according to the future bandwidth demand and the anomaly detection data, generating a plurality of candidate control strategies; Strategy selection execution unit, for obtaining the current network state, selecting the target control strategy from the plurality of candidate control strategies according to the current network state, and carrying out strategy issuing and executing; Strategy feedback optimization unit, for obtaining strategy expected effect and strategy actual effect, calculating target strategy benefit, according to the target strategy benefit, strategy feedback optimization.

10. The PON network AI dynamic regulation system of the fusion large model according to claim 9, wherein, The multi-source data acquisition unit specifically includes: Multi-source data acquisition module, for multi-source data acquisition, obtaining multi-source acquisition data; Data preprocessing module, for effective extraction and standardization processing of the multi-source acquisition data, generating multi-source standard data; Data fusion module, for spatio-temporal alignment and associated fusion of the multi-source standard data, generating fusion standard data; Feature extraction module, for analyzing and extracting time series features, spatial features and context features of the fusion standard data, obtaining structured feature data.

Images