CN-122027029-A - Deep learning-based base station optical network life cycle management method and system

Abstract

The invention relates to the technical field of life cycle management and discloses a base station optical network life cycle management method and system based on deep learning, wherein the base station optical network life cycle management method based on deep learning comprises the steps of collecting manufacturer information, ageing parameter definition, historical temperature sequence and temperature sensitivity coefficient of each optical fiber section of a link; the method comprises the steps of generating a semantic alignment matrix by using a pre-training language model coding parameter text, eliminating manufacturer parameter definition deviation, converting actual running time into accumulated accelerated aging equivalent time at a reference temperature based on a heat-activated aging model, generating a standardized sequence by mapping attenuation data, fitting and extracting aging fingerprint characteristics, inputting a Weibull model to calculate residual life distribution and overall failure probability, and inversely transforming and identifying a link life bottleneck section.

Inventors

• YANG HUITANG
• ZHU WEIPING
• CHEN WEI

Assignees

• 中国铁塔股份有限公司绍兴市分公司

Dates

Publication Date

20260512

Application Date

20260127

Claims (10)

1. 1. The method for managing the life cycle of the base station optical network based on deep learning is characterized by comprising the following steps: Acquiring manufacturer identification and aging parameter definition text of each section of optical fiber of a target link, and acquiring historical temperature sequences and temperature sensitivity coefficients of each section of optical fiber; inputting the aging parameter definition text of each manufacturer into a pre-training language model for vectorization coding, and generating an aging characteristic semantic alignment matrix based on similarity calculation among semantic vectors; Based on the historical temperature sequence of each section of optical fiber and the temperature sensitivity coefficient of a corresponding manufacturer, calculating the aging acceleration factor of each section of optical fiber by using a thermal activation aging model, and converting the actual running time of each section into accumulated aging acceleration equivalent time at a uniform reference temperature; acquiring an attenuation coefficient history measurement sequence of each section of optical fiber, replacing an actual time axis with an accumulated accelerated aging equivalent time axis, mapping attenuation measured values of each manufacturer to a unified aging characteristic space by utilizing an aging characteristic semantic alignment matrix, and generating a standardized equivalent time-attenuation coefficient sequence; performing curve fitting on the standardized equivalent time-attenuation coefficient sequences of all the sections, extracting aging characteristic parameters, and generating aging fingerprint characteristic vectors of all the sections; Inputting the aging fingerprint feature vector of each segment into a Weibull survival analysis model, calculating the residual life distribution parameters of each segment in an equivalent time domain, and calculating the failure probability of the whole link based on a serial reliability principle; and inversely transforming the residual life predicted value of each equivalent time domain into an actual time domain based on the aging acceleration factor of each segment, and identifying and outputting the paragraph with the shortest actual residual life in the link as a life bottleneck segment.
2. 2. The method of claim 1, wherein the aging parameter definition text comprises a decay rate definition, an end-of-life standard, and a test condition, wherein the generating of the aging characteristic semantic alignment matrix based on similarity calculation between semantic vectors comprises generating an aging characteristic semantic alignment matrix characterizing the aging parameter semantic mapping weights between vendors by calculating semantic correspondence of the aging parameters between vendors using cosine similarity based on semantic vectors of the aging parameters of each vendor.
3. 3. The method of claim 1, wherein the thermally activated aging model is an arrhenius equation model, and the aging acceleration factor is calculated based on a difference between the inverse of the unified reference temperature and the inverse of the actual operating temperature, vendor-specific activation energy parameters, and a natural exponential function.
4. 4. The method of claim 1, wherein the cumulative accelerated aging equivalent time is obtained by time integrating or weighted accumulation of aging acceleration factors for each time period.
5. 5. The method of claim 1, wherein the aging characteristic parameters include an aging rate parameter, an aging inflection parameter, and a rate of change statistic.
6. 6. The method of claim 1, wherein the output of the weibull survival analysis model includes shape parameters and scale parameters, and wherein the reliability function for each segment of optical fiber is calculated based on the shape parameters and scale parameters.
7. 7. The method of claim 6, wherein calculating the link overall failure probability based on the tandem reliability principle comprises continuously multiplying the reliability functions of the optical fibers in the link to obtain a link overall reliability function, and calculating a link overall failure probability curve based on the link overall reliability function.
8. 8. The method of claim 1, wherein inversely transforming the predicted remaining lifetime for each segment of equivalent time domain to an actual time domain comprises dividing the predicted remaining lifetime for the current time domain based on an average aging acceleration factor for each segment of optical fiber to obtain the predicted remaining lifetime for the actual time domain.
9. 9. The method of claim 1, further comprising outputting a link overall life prediction value, an actual remaining life ordering of each segment, a vendor identification of a life bottleneck segment, and an environmental risk level.
10. 10. A deep learning-based base station optical network lifecycle management system for performing the method recited in any of claims 1-9, comprising: The data acquisition module is used for acquiring manufacturer identification, aging parameter definition text, historical temperature sequence, temperature sensitivity coefficient and attenuation coefficient historical measurement sequence of each section of optical fiber of the target link; The semantic alignment module is used for inputting the aging parameter definition text of each manufacturer into the pre-training language model for vectorization coding and generating an aging characteristic semantic alignment matrix; the time alignment module is used for calculating the aging acceleration factors of the optical fibers of each section based on the heat-activated aging model and converting the actual running time into accumulated aging acceleration equivalent time; the characteristic mapping module is used for replacing a time axis of the attenuation coefficient history measurement sequence with an equivalent time axis and mapping the equivalent time axis to a unified aging characteristic space by utilizing an aging characteristic semantic alignment matrix; the characteristic extraction module is used for performing curve fitting on the standardized equivalent time-attenuation coefficient sequence and extracting an aging fingerprint characteristic vector; the survival analysis module is used for calculating the distribution parameters of the residual life of each section and the overall failure probability of the link based on the Weibull survival analysis model; and the output module is used for inversely transforming the equivalent time domain predicted value into an actual time domain and identifying a life bottleneck section.

Description

Deep learning-based base station optical network life cycle management method and system Technical Field The invention relates to the technical field of life cycle management, in particular to a method and a system for managing the life cycle of a base station optical network based on deep learning. Background In a base station backhaul optical network of a large-scale operator, an optical cable link is generally composed of optical fiber segments of a plurality of manufacturers, and each segment is laid in a different temperature environment area, including a high temperature area such as a desert, a tropical climate zone, and the like. The definitions of the aging parameters of the optical fibers of each manufacturer are different from the measurement standard, for example, the change rate of attenuation coefficients, the aging threshold value, the life definition and other parameters are inconsistent. Meanwhile, the aging acceleration effects of the optical fibers of various manufacturers are different in different temperature environments, and the aging acceleration characteristics of the optical fibers of various manufacturers are different in high temperature environments. The prior art method cannot uniformly model ageing data of optical fiber sections of different manufacturers in different temperature environments. If the joint analysis is directly carried out on the multi-manufacturer mixed link, the superposition effect of semantic deviation and temperature correction deviation can be generated, wherein the semantic deviation is caused by the inconsistency of the definition of aging parameters by each manufacturer, and the temperature correction deviation is caused by the variability of the temperature response characteristics of the optical fibers of each manufacturer. The superposition of the two deviations causes serious misalignment of the overall life prediction of the hybrid link, and the weak section which is firstly deteriorated in the link cannot be accurately identified, so that the accuracy and timeliness of operation and maintenance decision are affected. Disclosure of Invention The invention provides a base station optical network life cycle management method and system based on deep learning, which solve the technical problem of joint analysis semantic deviation caused by parameter expression difference among manufacturers in the related technology. The invention provides a base station optical network life cycle management method based on deep learning, which comprises the following steps: Acquiring manufacturer identification and aging parameter definition text of each section of optical fiber of a target link, and acquiring historical temperature sequences and temperature sensitivity coefficients of each section of optical fiber; inputting the aging parameter definition text of each manufacturer into a pre-training language model for vectorization coding, and generating an aging characteristic semantic alignment matrix based on similarity calculation among semantic vectors; Based on the historical temperature sequence of each section of optical fiber and the temperature sensitivity coefficient of a corresponding manufacturer, calculating the aging acceleration factor of each section of optical fiber by using a thermal activation aging model, and converting the actual running time of each section into accumulated aging acceleration equivalent time at a uniform reference temperature; acquiring an attenuation coefficient history measurement sequence of each section of optical fiber, replacing an actual time axis with an accumulated accelerated aging equivalent time axis, mapping attenuation measured values of each manufacturer to a unified aging characteristic space by utilizing an aging characteristic semantic alignment matrix, and generating a standardized equivalent time-attenuation coefficient sequence; performing curve fitting on the standardized equivalent time-attenuation coefficient sequences of all the sections, extracting aging characteristic parameters, and generating aging fingerprint characteristic vectors of all the sections; Inputting the aging fingerprint feature vector of each segment into a Weibull survival analysis model, calculating the residual life distribution parameters of each segment in an equivalent time domain, and calculating the failure probability of the whole link based on a serial reliability principle; and inversely transforming the residual life predicted value of each equivalent time domain into an actual time domain based on the aging acceleration factor of each segment, and identifying and outputting the paragraph with the shortest actual residual life in the link as a life bottleneck segment. The aging parameter definition text comprises attenuation change rate definition, life end standard and test condition, and the generation of the aging characteristic semantic alignment matrix based on similarity calculation among semantic vectors comprise