TW202026914A

TW202026914A - System and method for analyzing potential degradation probability of broadband service equipment

Info

Publication number: TW202026914A
Application number: TW107147757A
Authority: TW
Inventors: 李昀潔; 蘇立鼎
Original assignee: 中華電信股份有限公司
Priority date: 2018-12-28
Filing date: 2018-12-28
Publication date: 2020-07-16
Also published as: TWI695282B

Abstract

The present invention relates to a system and a method for analyzing the potential degradation probability of broadband service equipment. The method comprises collecting and checking historical service measurement data or configuration data of network devices and their device interfaces in the broadband service network; establishing a modeling data set based on the association set of historical service measurement data or configuration data; establishing an analysis data set based on the correlation sets of the historical service measurement data or configuration data; an equipment obstacle classification model is trained by using the modeling data set, and then the historical service measurement data or configuration data is imported into the combination of the equipment obstacle classification model and the analysis data sets to generate equipment obstacle classification data; and using the modeling data set to train the device degradation probability model, and then importing the device obstacle classification data into the device obstacle degradation probability model to generate a device degradation probability list.

Description

分析寛頻服務設備潛在劣化機率之系統及其方法 System and method for analyzing potential degradation probability of broadband service equipment

本發明係關於一種分析網路設備劣化之技術，尤指一種分析網路寛頻服務設備潛在劣化機率之系統及其方法。 The present invention relates to a technique for analyzing the degradation of network equipment, and in particular to a system and method for analyzing the potential degradation probability of network broadband service equipment.

目前，電信運營商皆透過佈建廣大之寛頻服務網路系統於提供網路服務給客戶，其透過串連寛頻服務網路中各端點間之網路設備並將各用戶之終端設備串接，然而對於電信運營商而言，如何管理此寛頻服路系統並持續維持品質實為一困難之議題。 At present, telecommunications operators provide network services to customers by deploying a wide range of broadband service network systems. They connect the network equipment between the endpoints in the broadband service network and connect the terminal equipment of each user. However, for telecom operators, how to manage this broadband service system and continue to maintain quality is a difficult issue.

再者，在確保一般用戶體驗品質(Quality of Experience)之前提下，在數量龐大的網路設備中如果使用資料來定位具品質問題之設備實屬不易，例如預先判斷具潛在劣化可能之網路設備後讓網路設備管理人員能更提前維修/更換該台設備，避免影響用戶體驗，然而因為寛頻服務網路設備資料特性與樣貌皆不同，不論是其設備廠牌型號、設備設定值、設備所屬之網路管理群組、設備所提供服務之服務組態等皆非透過單一或少數規則即可達到上述之品保目的。 Furthermore, before ensuring the quality of the general user experience (Quality of Experience), it is not easy to use data to locate quality problem equipment in a large number of network equipment, such as pre-determining the network with potential degradation. After the equipment, the network equipment management personnel can repair/replace the equipment in advance to avoid affecting the user experience. However, because the broadband service network equipment data characteristics and appearance are different, regardless of the equipment brand model and equipment settings , The network management group to which the equipment belongs, and the service configuration of the services provided by the equipment can achieve the above-mentioned quality assurance purposes without a single or a few rules.

在維運網路工作中，電信運營商雖可透過監測服務品質數據掌握服務網路品質概況，並且可直接鎖定特定幾個品質參數確保服務質量(Quality of Service,QOS)，例如直接監測網路設備連通狀況及其服務網管路徑等參數確保設備連線無問題、直接監測網路設備介面連通狀況及其服務網管路徑等參數確保設備介面連線無問題。然而，直接監測各樣之關鍵品保參數來提前警示管理單位，在錯綜複雜的服務網路架構中，因各樣品保參數及其資料皆與其他品保參數及其資料有複雜互相關聯或遞移關係，直接監控不易，因而導致透過上述監測方法無法完整確保網路架構之品質。 In the maintenance and operation of the network, although the telecom operator can grasp the quality of the service network by monitoring the service quality data, and can directly lock certain quality parameters to ensure the quality of service (Quality of Service, QOS), such as directly monitoring the network Parameters such as the connection status of the equipment and its service network management path ensure that there is no problem with the equipment connection, and directly monitor the connection status of the network equipment interface and the service network management path and other parameters to ensure that the equipment interface connection is free. However, various key quality assurance parameters are directly monitored to alert management units in advance. In the intricate service network structure, each sample assurance parameter and its data are complicatedly interrelated or transitive with other quality assurance parameters and its data. However, direct monitoring is not easy, and the above-mentioned monitoring methods cannot completely ensure the quality of the network architecture.

有鑑於此，如何提供一種快速及/或有效地分析寛頻服務設備潛在劣化機率之系統與方法，成為許多系統開發商的重要課題。 In view of this, how to provide a system and method for quickly and/or effectively analyzing the potential degradation probability of broadband service equipment has become an important issue for many system developers.

為解決至少上述問題，本案提供一種分析寛頻服務設備潛在劣化機率之系統，係包括：資料處理模組，係用以收集與檢查一寬頻服務網路中複數個網路設備及其複數個設備介面之歷史服務量測資料或組態資料；資料建模模組，係根據該些歷史服務量測資料或組態資料之關聯集，建立一建模資料集；資料分析模組，係根據該些歷史服務量測資料或組態資料之關聯集，建立一分析資料集；設備障礙分類模型模組，係以機器學習演算法利用該建模資料集訓練出一設備障礙分類模型，再將該些歷史服務量測資料或組態資料匯入該設備障礙分類模型與該分析資料集之組合中，以產生一設備障礙分類資料；以及設備劣化機率模型模組，係以機器學習演算法利用該建模資料集訓練出一設備劣化機率模型，再將該設備障礙分類資料匯入該設備劣化機率模型中，以產生一設備劣化機率清單。 In order to solve at least the above-mentioned problems, this case provides a system for analyzing the potential degradation probability of broadband service equipment, which includes: a data processing module, which is used to collect and inspect a plurality of network devices and their plural devices in a broadband service network The historical service measurement data or configuration data of the interface; the data modeling module establishes a modeling data set based on the association set of the historical service measurement data or configuration data; the data analysis module is based on the The correlation set of some historical service measurement data or configuration data is used to establish an analysis data set; the equipment obstacle classification model module uses the machine learning algorithm to use the modeling data set to train an equipment obstacle classification model. Some historical service measurement data or The configuration data is imported into the combination of the equipment obstacle classification model and the analysis data set to generate an equipment obstacle classification data; and the equipment degradation probability model module is used to train a machine learning algorithm using the modeling data set Equipment degradation probability model, and then import the equipment obstacle classification data into the equipment degradation probability model to generate a list of equipment degradation probability.

本發明復提出一種分析寛頻服務設備潛在劣化機率之方法，係包括：收集與檢查一寬頻服務網路中複數個網路設備及其複數個設備介面之歷史服務量測資料或組態資料；根據該些歷史服務量測資料或組態資料之關聯集，建立一建模資料集；根據該些歷史服務量測資料或組態資料之關聯集，建立一分析資料集；以機器學習演算法利用該建模資料集訓練出一設備障礙分類模型，再將該些歷史服務量測資料或組態資料匯入該設備障礙分類模型與該分析資料集之組合中，以產生一設備障礙分類資料；以及以機器學習演算法利用該建模資料集訓練出一設備劣化機率模型，再將該設備障礙分類資料匯入該設備劣化機率模型中，以產生一設備劣化機率清單。 The present invention further proposes a method for analyzing the potential degradation probability of broadband service equipment, which includes: collecting and checking historical service measurement data or configuration data of multiple network devices and multiple device interfaces in a broadband service network; Create a modeling data set based on the correlation set of the historical service measurement data or configuration data; create an analysis data set based on the correlation set of the historical service measurement data or configuration data; use machine learning algorithms Use the modeling data set to train an equipment obstacle classification model, and then import the historical service measurement data or configuration data into the combination of the equipment obstacle classification model and the analysis data set to generate an equipment obstacle classification data ; And using the modeling data set to train a device degradation probability model with a machine learning algorithm, and then import the equipment obstacle classification data into the device degradation probability model to generate a list of equipment degradation probability.

在前述之分析寛頻服務設備潛在劣化機率之系統及其方法中，復包括模型更新模組，係收集該設備劣化機率清單，藉以驗證是否需要更新模型組合。 In the aforementioned system and method for analyzing the potential degradation probability of broadband service equipment, a model update module is further included to collect the degradation probability list of the equipment to verify whether the model combination needs to be updated.

在前述之分析寛頻服務設備潛在劣化機率之系統及其方法中，復包括設備維運模組，係用以驗證且標註該些歷史服務量測資料或組態資料是否為潛在劣化之標籤。 In the aforementioned system and method for analyzing the potential degradation probability of broadband service equipment, the equipment maintenance module is also included to verify and mark whether the historical service measurement data or configuration data are labels for potential degradation.

在前述之分析寛頻服務設備潛在劣化機率之系統及其方法中，該設備障礙分類模型為一ID3決策樹(ID3 Decision Tree)、一C4.5決策樹(C4.5 Decision Tree)、一C5.0決策樹(C5.0 Decision Tree)、一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。 In the aforementioned system for analyzing the potential degradation probability of broadband service equipment and In the method, the equipment obstacle classification model is an ID3 decision tree (ID3 Decision Tree), a C4.5 decision tree (C4.5 Decision Tree), a C5.0 decision tree (C5.0 Decision Tree), and a CART Tree (Classification And Regression Tree), a random forest (Random Forest) or a gradient boosting decision tree (Gradient Boosting Decision Tree).

在前述之分析寛頻服務設備潛在劣化機率之系統及其方法中，該設備劣化機率模型為一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。 In the aforementioned system and method for analyzing the potential degradation probability of broadband service equipment, the device degradation probability model is a CART tree (Classification And Regression Tree), a random forest (Random Forest), or a gradient boosting decision tree (Gradient Boosting). Decision Tree).

在前述之分析寛頻服務設備潛在劣化機率之系統及其方法中，該使用資料建模模組針對該設備介面找出其對應之上鏈或下鏈對接設備，其中，以靠近網路核心之對接設備定義為該上鏈對接設備，以遠離網路核心之對接設備定義為該下鏈對接設備。 In the aforementioned system and method for analyzing the potential degradation probability of broadband service equipment, the data modeling module uses the data modeling module to find the corresponding uplink or downlink docking device for the device interface. Among them, the one close to the network core The docking device is defined as the uplink docking device, and the docking device far away from the network core is defined as the downlink docking device.

由上可知，本發明之分析寛頻服務設備潛在劣化機率之系統及其方法，透過綜整設備歷史量測資料，並解析服務網路中設備間之相依性，透過結合網路設備間之鏈結關係將各個設備服務品質數據與其組態資料關聯，同時因網路設備間連結皆透過其介面進行串接，因此也組合了各設備之介面組態與其服務品質數據，並且使用了機器學習演算法幫助克服各特徵間之遞移相依性，判讀網路設備劣化之潛在可能性，其藉以達到快速及/或精確地預測可能潛在劣化之設備端點之技術功效。 It can be seen from the above that the system and method for analyzing the potential degradation probability of broadband service equipment of the present invention integrates historical measurement data of equipment and analyzes the dependencies between equipment in the service network by combining the links between network equipment The connection relationship associates the service quality data of each device with its configuration data. At the same time, because the links between network devices are connected through its interface, it also combines the interface configuration of each device and its service quality data, and uses machine learning algorithms The method helps to overcome the gradual dependence of each feature, to determine the potential for network equipment degradation, and to achieve the technical effect of quickly and/or accurately predicting potentially degraded equipment endpoints.

11‧‧‧資料處理模組 11‧‧‧Data Processing Module

12‧‧‧資料建模模組 12‧‧‧Data Modeling Module

13‧‧‧資料分析模組 13‧‧‧Data Analysis Module

14‧‧‧設備劣化機率模型模組 14‧‧‧Device Deterioration Probability Model Module

15‧‧‧設備障礙分類模型模組 15‧‧‧Equipment obstacle classification model module

16‧‧‧設備維運模組 16‧‧‧Equipment maintenance module

17‧‧‧模型更新模組 17‧‧‧Model Update Module

S1~S7‧‧‧步驟 S1~S7‧‧‧Step

本案揭露之具體實施例將搭配下列圖式詳述，這些說明顯示在下列圖式：第1圖為本發明分析寛頻服務設備潛在劣化機率的系統架構示意圖。 The specific embodiments disclosed in this case will be described in detail with the following drawings, and these descriptions are shown in the following drawings: Figure 1 is a schematic diagram of the system architecture of the present invention for analyzing the potential degradation probability of broadband service equipment.

第2圖係為本發明分析寛頻服務設備潛在劣化機率方法之流程圖。 Figure 2 is a flowchart of the method for analyzing the potential degradation probability of broadband service equipment according to the present invention.

第3A-3C圖係為本發明決策樹結構之範例示意圖。 Figures 3A-3C are schematic diagrams of examples of the decision tree structure of the present invention.

以下藉由特定的具體實施例說明本發明之實施方式，熟悉此技藝之人士可由本說明書所揭示之內容輕易地瞭解本發明之其他優點及功效。 The following specific examples illustrate the implementation of the present invention. Those familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification.

須知，本說明書所附圖式所繪示之結構、比例、大小等，均僅用以配合說明書所揭示之內容，以供熟悉此技藝之人士之瞭解與閱讀，並非用以限定本發明可實施之限定條件，故不具技術上之實質意義，任何結構之修飾、比例關係之改變或大小之調整，在不影響本發明所能產生之功效及所能達成之目的下，均應仍落在本發明所揭示之技術內容得能涵蓋之範圍內。同時，本說明書中所引用之如「前」、「後」及「一」等之用語，亦僅為便於敘述之明瞭，而非用以限定本發明可實施之範圍，其相對關係之改變或調整，在無實質變更技術內容下，當視為本發明可實施之範疇。 It should be noted that the structures, proportions, sizes, etc. shown in the drawings in this manual are only used to match the contents disclosed in the manual for the understanding and reading of those familiar with the art, and are not intended to limit the implementation of the present invention. Therefore, it does not have any technical significance. Any structural modification, proportional relationship change, or size adjustment should still fall within the scope of the present invention without affecting the effects and objectives that can be achieved. The technical content disclosed by the invention can be covered. At the same time, the terms such as "front", "rear" and "one" quoted in this specification are only for ease of description and are not used to limit the scope of implementation of the present invention, changes in their relative relationships or Adjustments shall be regarded as the scope of the implementation of the present invention without substantial changes to the technical content.

請參照第1圖，係為本發明分析寛頻服務設備潛在劣化機率的系統架構示意圖。 Please refer to Figure 1, which is a schematic diagram of the system architecture of the present invention for analyzing the potential degradation probability of broadband service equipment.

如第1圖所示，本發明分析寛頻服務設備潛在劣化機率的系統包括資料處理模組11、資料建模模組12、資料分析模組13、設備劣化機率模型模組14、設備障礙分類模型模組15、設備維運模組16與模型更新模組17。 As shown in Figure 1, the system of the present invention for analyzing the potential degradation probability of broadband service equipment includes a data processing module 11, a data modeling module 12, a data analysis module 13, a device degradation probability model module 14, and equipment failure classification Model module 15, equipment maintenance and operation module 16, and model update module 17.

資料處理模組11係用以收集與檢查一寬頻服務網路中複數個網路設備及其複數個設備介面之歷史服務量測資料或組態資料。在一實施例中，將一歷史服務量測資料及組態資料對應至各個網路設備；將一歷史服務量測資料及組態資料對應至各個網路設備介面；且檢查收集之時間區間是否足夠。 The data processing module 11 is used to collect and check historical service measurement data or configuration data of a plurality of network devices and their plurality of device interfaces in a broadband service network. In one embodiment, a piece of historical service measurement data and configuration data is mapped to each network device; a piece of historical service measurement data and configuration data is mapped to each network device interface; and it is checked whether the collected time interval enough.

舉例而言，令設備或設備介面為{d|

d

D}，D代表寛頻服務網路之所有設備及其介面之集合，針對每台設備及設備介面皆有設備維運模組16依據現場實務經驗及採取若干適切之各式查測方式驗證而標註為是否潛在劣化之標籤C^d={0,1}，意即每台設備及設備介面標籤為

C ^d。為方便說明，以下所稱之類別或目標參數皆等同於此段敘述所說之設備及設備介面標籤，此問題為典型二元分類問題(Binary Classification Problem)，此二元類別表示設備及設備介面是否潛在劣化，實務上為了能夠精確測量問題，因此在結果呈現中會採用機率模型計算作為產出，換句話說，計算出之結果為針對每台設備及設備介面計算出潛在劣化機率值而非單一貼註。 For example, let the device or device interface be {d|

d

D}, D stands for the collection of all equipment and interfaces of the broadband service network. There is an equipment maintenance module for each equipment and equipment interface. 16 It is verified based on on-site practical experience and adopting various appropriate inspection methods. The label C ^d ={0,1} marked as potential degradation, which means that each device and device interface label is

C ^d . For the convenience of explanation, the categories or target parameters mentioned below are equivalent to the equipment and equipment interface labels mentioned in this paragraph. This problem is a typical Binary Classification Problem. This binary classification represents equipment and equipment interface. Whether it is potential degradation, in practice, in order to be able to accurately measure the problem, the probability model calculation is used as the output in the result presentation. In other words, the calculated result is the potential degradation probability value calculated for each device and device interface instead of Single note.

歷史服務量測資料包含設備及設備介面之訊務服務資料與服務狀態資料。設備訊務服務資料舉凡設備風扇轉速、設備溫度、設備CPU使用率、設備記憶體使用量等；設備介面訊務服務資料舉凡設備介面封包遺失數、設備介面循環冗餘碼值(CRC)、設備介面封包流量、設備介面封包錯誤數等；設備服務狀態資料舉凡設備失聯告警、設備網管路徑錯誤、設備浸水告警、維運改接通告、維運查修通告等；設備介面服務狀態資料舉凡設備介面斷連告警、設備介面瞬斷告警、設備介面速率未達告警等，此等資料由資料處理模組11所收集，令該服務量測資料為s且資料向量為d^s，前述資料可表示為

，i代表第i台設備或設備介面。 Historical service measurement data includes communication service data and service status data of equipment and equipment interface. Equipment communication service information includes equipment fan speed, equipment temperature, equipment CPU usage, equipment memory usage, etc.; equipment interface communication service information includes equipment interface packet loss, equipment interface cyclic redundancy code value (CRC), equipment Interface packet flow, number of device interface packet errors, etc.; equipment service status data, such as equipment loss alarm, equipment network management path error, equipment flooding alarm, maintenance and operation change notice, maintenance and inspection notice, etc.; equipment interface service status data, such as equipment Interface disconnection alarm, device interface instantaneous interruption alarm, device interface rate under-speed alarm, etc. These data are collected by the data processing module 11, so that the service measurement data is s and the data vector is d ^s . The aforementioned data can be expressed for

, I represents the i-th device or device interface.

組態資料主要為標示設備及設備介面之實體狀態與服務狀態。設備之組態資料包含設備廠牌、設備服務狀態、設備建置日期、設備網路位階等；設備介面包含介面服務狀態、介面接線廠牌、介面VLAN等；此等資料由資料處理模組11所收集，令該服務組態資料為c且資料向量d^c，前述資料可表示為

，i代表第i台設備或設備介面。 The configuration data mainly indicates the physical status and service status of the device and the device interface. The configuration data of the equipment includes equipment brand, equipment service status, equipment build date, equipment network level, etc.; equipment interface includes interface service status, interface wiring brand, interface VLAN, etc.; these data are provided by data processing module 11 Collected, let the service configuration data be c and the data vector d ^c , the aforementioned data can be expressed as

, I represents the i-th device or device interface.

由於分析上需要確保有足夠資料量使得模型訓練或分類效能達到預期，因此需使用資料處理模組11確認收集資料之時間區間是否足夠，確保數量足夠用於訓練模型，該時間區間可依據分析劣化目的之範疇來訂定所需區間，此實施方式假定實際量測時間點為量測基準點算起往前推算3天，若該模組所收集之資料未達規定之時間區間則待時間到達時重新收集，令時間區間為t，令

為組合服務量測資料及服務組態資料之第i個設備或設備介面，亦即

Since the analysis needs to ensure that there is enough data to make the model training or classification performance reach the expected, it is necessary to use the data processing module 11 to confirm whether the time interval for collecting data is sufficient and to ensure that the amount is sufficient for training the model. This time interval can be degraded according to the analysis The scope of the purpose is to define the required interval. This implementation assumes that the actual measurement time point is 3 days forward from the measurement reference point. If the data collected by the module does not reach the specified time interval, it will wait until the time arrives. Recollect at time, let the time interval be t, let

The i-th device or device interface that combines service measurement data and service configuration data, that is

資料建模模組12係根據該些歷史服務量測資料或組態資料之關聯集，建立一建模資料集。資料處理模組11收集與檢查一寬頻服務網路中複數個網路設備及其複數個設備介面之歷史服務量測資料或組態資料這兩種資料之後，使用資料建模模組12組合組態資料及其服務量測資料，將上述所收集之服務量測資料依據設備及設備介面分別對應其組態資料，換言之，所收集之服務量測資料並無揭示該設備或該設備介面所屬組態資料，例如設備斷連告警資料不會包含其設備服務狀態，且部份組態不會有相應之服務量測資料，反之亦然，此為遺漏值(Missing Value)資料。遺漏值資料大部份為資料本身問題或特殊因素導致無法收集因此需將其剔除，因此需要比對資料並結合完整資料以確保後續步驟順利。 The data modeling module 12 creates a modeling data set based on the association set of the historical service measurement data or configuration data. The data processing module 11 collects and checks the historical service measurement data or configuration data of a plurality of network devices and their plurality of device interfaces in a broadband service network, and then uses the data modeling module 12 combination group State data and its service measurement data. The service measurement data collected above corresponds to the configuration data of the device and device interface respectively. In other words, the collected service measurement data does not reveal the device or the group that the device interface belongs to. Status data, such as equipment disconnection alarm data will not include its equipment service status, and some configurations will not have corresponding service measurement data, and vice versa, this is Missing Value data. Most of the missing value data is due to the problem of the data itself or special factors that make it impossible to collect, so it needs to be eliminated. Therefore, it is necessary to compare the data and combine the complete data to ensure smooth subsequent steps.

使用資料建模模組12針對各個設備介面找出其對應之上鏈或下鏈對接設備。在一實施例中，上下鏈之定義為以靠近網路核心為上鏈，遠離者為下鏈。實體網路設備介面之對接方式共分為三種，一對一對接、多對一對接及多對多對接，例如某甲設備下接介面對接另外兩介面乙介面及丙介面，則需要將乙介面及丙介面各自關聯至甲介面之下鏈對接資料及各自將乙介面及丙介面之上鏈介面關聯至甲介面，另一方面網路設備佈建方式主要匯彙集收容於局端或電信運營商之機房，可能會有發生無下鏈設備之情形發生，亦即該設備介面為接取用戶用，而網路設備則因無直接對接因此不需關聯，此兩種狀況則在對接特徵中補零即可。 The data modeling module 12 is used to find the corresponding uplink or downlink docking device for each device interface. In one embodiment, the upper and lower links are defined as being close to the core of the network as the uplink, and those far away are the downlink. There are three docking methods for physical network device interfaces, one-to-one, many-to-one, and many-to-many. For example, if a device's downstream interface is connected to the other two interfaces, interface B and interface C, you need to connect interface B The interface and the C interface are respectively associated with the link data of the interface A and the upper link interface of the interface B and the interface C are respectively associated with the interface A. On the other hand, the network equipment deployment methods are mainly collected and housed in the bureau. In the computer room of the terminal or the telecom operator, there may be situations where there is no offline device, that is, the device interface is used to access the user, and the network device does not need to be associated because there is no direct connection. These two situations are Add zeros in the docking feature.

承上述內容，針對各個寛頻服務網路設備介面關聯其對應之設備介面資料，使用資料建模模組12依序執行「結合上鏈對接設備介面組態資料與歷史服務量測資料」、「結合下鏈對接設備介面組態資料與歷史服務量測資料」、「結合自身設備組態資料與歷史服務量測資料」等步驟，於「結合上鏈對接設備介面組態資料與歷史服務量測資料」、「結合下鏈對接設備介面組態資料與歷史服務量測資料」、「結合自身設備組態資料與歷史服務量測資料」步驟中針對各個設備介面找出其對應之上鏈及下鏈對接設備介面及其自身設備。 In accordance with the above content, for each broadband service network device interface to associate its corresponding device interface data, use the data modeling module 12 to sequentially execute "Combine on-chain docking device interface configuration data and historical service measurement data", " Combine the device interface configuration data and historical service measurement data of the offline docking device, "Combine its own device configuration data and the historical service measurement data" and other steps, and click the "Combine the device interface configuration data and historical service measurement data of the uplink docking device" In the steps of “Data”, “Combine the device interface configuration data and historical service measurement data of the offline docking”, “Combine the configuration data of your own device and the historical service measurement data”, find out the corresponding uplink and downlink for each device interface. Link the device interface and its own device.

待完成上述關聯後，建立適合可用之資料模型組合，將上述說明所關聯之集合使用資料建模模組12建立建模資料集建立一建模資料集。在一實施例中，建模資料集可依據目的切分訓練資料(Training Data)、驗證資料(Validation Data)及測試資料(Testing Data)，訓練資料為建立初步資料模型，驗證資料為調整資料模型超參數(Hyper Parameters)進而取得高效能之模型，測試資料則為測試資料集之泛化能力。 After the above-mentioned association is completed, a suitable and usable data model combination is established, and the data modeling module 12 is used to create a modeling data set to create a modeling data set using the set related to the above description. In one embodiment, the modeling data set can be divided into Training Data, Validation Data, and Testing Data according to the purpose. The training data is the establishment of a preliminary data model, and the validation data is the adjustment data model. Hyperparameters (Hyper Parameters) then obtain high-performance models, and the test data is the generalization ability of the test data set.

資料分析模組13係根據該些歷史服務量測資料或組態資料之關聯集，建立一分析資料集。待資料建模模組12 結合組合組態資料及其服務量測資料後，需針對各個寛頻服務網路設備及設備介面關聯其對應之設備資料，將設備介面之資料向量與其對接及自身設備之資料向量作結合。 The data analysis module 13 creates an analysis data set based on the correlation set of the historical service measurement data or configuration data. Data Modeling Module 12 After combining the combined configuration data and service measurement data, it is necessary to associate the corresponding device data for each broadband service network device and device interface, and combine the data vector of the device interface with its docking and the data vector of its own device.

在一實施例中，將其分析資料集切分成訓練及測試資料，舉例而言，可切7成為訓練資料，3成為測試資料。而訓練資料令其稱作設備及設備介面特徵矩陣M，為方便說明，以下所稱矩陣皆等同於本發明說明揭示中所提及之資料集。 In one embodiment, the analysis data set is divided into training and test data. For example, 7 can be cut into training data and 3 can be cut into test data. The training data is called the device and device interface feature matrix M. For the convenience of description, the matrix referred to below is equivalent to the data set mentioned in the disclosure of the present invention.

設備障礙分類模型模組15，係以機器學習演算法利用該建模資料集訓練出一設備障礙分類模型，再將該些歷史服務量測資料或組態資料匯入該設備障礙分類模型與該分析資料集之組合中，以產生一設備障礙分類資料。在一實施例中，該設備障礙分類模型為一ID3決策樹(ID3 Decision Tree)、一C4.5決策樹(C4.5 Decision Tree)、一C5.0決策樹(C5.0 Decision Tree)、一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。 The equipment obstacle classification model module 15 uses a machine learning algorithm to train an equipment obstacle classification model using the modeling data set, and then imports the historical service measurement data or configuration data into the equipment obstacle classification model and the Analyze the combination of data sets to generate a classification data of equipment obstacles. In one embodiment, the equipment obstacle classification model is an ID3 decision tree (ID3 Decision Tree), a C4.5 decision tree (C4.5 Decision Tree), a C5.0 decision tree (C5.0 Decision Tree), A CART tree (Classification And Regression Tree), a random forest (Random Forest) or a gradient boosting decision tree (Gradient Boosting Decision Tree).

在一實施例中，利用設備障礙分類模型模組15訓練設備障礙分類模型，此模型可應用任一種分類器(Classification Model)作為此模型，此實施方式假定應用C4.5決策樹演算法進行訓練與建模，該演算法精神主要利用資料熵(Information Entropy)之資訊量化特性，透過計算資訊量增長值(Information Gain)、資訊量增長率(Information Gain Ratio)及分割資訊量(Split Information) 指標，建構可應用於分類問題(Classification Problem)之分類器。 In an embodiment, the equipment obstacle classification model module 15 is used to train the equipment obstacle classification model. This model can be used with any classifier (Classification Model) as the model. This implementation assumes that the C4.5 decision tree algorithm is used for training. And modeling, the spirit of the algorithm mainly uses the information quantification characteristics of Information Entropy, by calculating the information gain (Information Gain), the information gain ratio (Information Gain Ratio) and the split information (Split Information) Indicators, to construct a classifier that can be applied to the Classification Problem.

舉例來說，先設定一根節點(root)，之後透過下列算式選取適合之節點(Node)與葉節點(Leaf Node)作為設備障礙分類模型模組15，以下簡介分類器訓練原理，令設備及設備介面特徵矩陣為M，計算各類別資訊量(Information)，其表示為：

，C ^d為是否潛在劣化之標籤集合，freq(

,M)表第k個標籤類別的數量，基於資訊理論(Information Theory)之資料通訊編碼通常以位元(bit)為單位，因此對數基數設為2，求得資訊量(Information)後，針對各個特徵分別計算切割特徵後之資訊量，其表示為：

，其中j代表為特徵A的值，而M_j代表特徵A之j值所構成之資料子集合，將原先計算之資訊量info(M)與各個特徵切割後之Info_A(M)計算後，求取各特徵之資訊增長值(Information Gain)，其表示為：gain(A)=Info(M)-Info _A(M)，求取該值時，若單以資訊增長值(Information Gain)作為切割樹子節點之依據，恐容易受到樣本多樣之特徵所影響(Bias)，例如部份特定SNMP trap會因機型不同而導致特徵容易被模型切分，因此需要加上懲罰因子限制選取規則降低受其偏差所影響，利用各特徵之樣本組成集合，計算分割資訊量(Split Information)，其表示為：

，此指標主要目的為求取各個特徵之資訊熵(Entropy)，此目的為計算出特徵之資訊分佈集合之資訊量(impurity)，求取該指標後，可求取資訊增長率(Information Gain Ratio)，其表示為：

，依照上述若干算式分別計算各個特徵之資訊增長率GainRatio(A)並擇其高者作一新節點(Node)，直到資訊增長率GainRatio(A)等於0或小於門檻值，則設為一新葉節點(Leaf)，重覆上述步驟直到無法切割新葉結點時為止，當步驟完成後所求得之決策樹即為設備障礙分類模型模組15，該設備障礙分類模型模組15需使用前述所提及之測試資料確認該設備障礙分類模型模組15具可用性後方可確定此步驟完成，決策樹結構係舉例於如第3A-3C圖所示。 For example, first set a node (root), and then use the following formula to select the appropriate node (Node) and leaf node (Leaf Node) as the equipment obstacle classification model module 15. The following introduces the principle of classifier training, so that the equipment and The feature matrix of the device interface is M, and the information of each category is calculated, which is expressed as:

, C ^d is the label set of potential degradation, freq (

,M ) represents the number of the k-th label category. The data communication code based on Information Theory is usually in bits. Therefore, the logarithmic base is set to 2. After obtaining the amount of information (Information), Each feature calculates the amount of information after cutting the feature, which is expressed as:

, Where j represents the value of feature A, and M _j represents the data subset formed by the j value of feature A. After calculating the originally calculated information info (M) and the Info _A (M) after each feature is cut, Calculate the information gain value (Information Gain) of each feature, which is expressed as: gain ( A ) = Info ( M ) -Info _A ( M ), when calculating the value, if the information gain value (Information Gain) is taken as The basis for cutting the sub-nodes of the tree is likely to be affected by the diverse characteristics of the sample (Bias). For example, some specific SNMP traps may be easily segmented by the model due to different models. Therefore, a penalty factor must be added to limit the selection rules to reduce the Affected by the deviation, the sample of each feature is used to form a set to calculate the amount of split information, which is expressed as:

The main purpose of this indicator is to obtain the information entropy (Entropy) of each feature. The purpose is to calculate the information volume (impurity) of the information distribution set of the feature. After obtaining this indicator, the information growth rate (Information Gain Ratio) ), which is expressed as:

Calculate the information growth rate GainRatio(A) of each feature according to the above formulas and choose the higher one as a new node (Node), until the information growth rate GainRatio(A) is equal to 0 or less than the threshold, then set to a new Leaf node (Leaf), repeat the above steps until the new leaf node cannot be cut, the decision tree obtained after the step is completed is the equipment obstacle classification model module 15, which needs to be used The aforementioned test data confirms that the equipment obstacle classification model module 15 is usable before confirming that this step is completed. The decision tree structure is shown in Figures 3A-3C.

待完成設備障礙分類模型模組15後，加入設備障礙特徵，此特徵目的為標示設備，藉以釐清潛在劣化設備與障礙設備之區別。換句話說，許多障礙設備所呈現之資料樣態(Pattern)與潛在劣化設備所呈現之樣態(Pattern)相似，單從原有特徵上來分析上略顯不足，因此需要加上此特徵區別此兩者差異。 After the equipment obstacle classification model module 15 is completed, the equipment obstacle feature is added. The purpose of this feature is to mark the equipment, so as to clarify the difference between potentially degraded equipment and obstacle equipment. In other words, the data pattern presented by many obstructive devices is similar to the pattern presented by potentially degraded devices. The analysis of the original features is slightly insufficient, so this feature needs to be added to distinguish this The difference between the two.

設備劣化機率模型模組14，係以機器學習演算法利用該建模資料集訓練出一設備劣化機率模型，再將該設備障礙分類資料匯入該設備劣化機率模型中，以產生一設備劣化機率清單。在一實施例中，該設備劣化機率模型為一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。將上述新加特徵之資料集，使用設備劣化機率模型模組14訓練設備劣化機率模型，此模型可應用任一種機率模型(Probability Model)，此說明假定使用梯度提昇決策樹(Gradient Boosting Decision Tree)作為模型，此機率模型透過建立多顆分類回歸決策樹(Classification and Regression Tree)並透過各決策樹之加權總合而得到一潛在劣化機率值，假設現已建立t顆分類回歸決策樹，令f_t(d_i)表示第t顆決策樹對第i設備或設備介面所計算之加權潛在劣化機率值，因此欲建立第t+1顆樹時可表示為：

，其中

表示前t顆決策樹之聯合輸出結果，

(d_i)為欲加入之第t+1顆樹之輸出函數，此時我們可將機率模型的損失函數L(Loss function)表示為：

The equipment degradation probability model module 14 uses a machine learning algorithm to use the modeling data set to train a equipment degradation probability model, and then incorporates the equipment obstacle classification data into the equipment degradation probability model to generate an equipment degradation probability List. In one embodiment, the device degradation probability model is a CART tree (Classification And Regression Tree), a Random Forest (Random Forest) or a Gradient Boosting Decision Tree (Gradient Boosting Decision Tree). Use the equipment degradation probability model module 14 to train the equipment degradation probability model on the data set with the above-mentioned newly added features. This model can be applied to any probability model (Probability Model). This description assumes the use of a gradient boosting decision tree (Gradient Boosting Decision Tree) As a model, this probability model obtains a potential degradation probability value through the establishment of multiple classification and regression trees (Classification and Regression Tree) and the weighted sum of the decision trees. Assuming that t classification and regression decision trees have been established, let f _t (d _i ) represents the weighted potential degradation probability value calculated by the t-th decision tree for the ith device or device interface, so when you want to build the t+1-th tree, it can be expressed as:

,among them

Represents the joint output result of the first t decision trees,

(d _i ) is the output function of the t+1th tree to be added. At this time, we can express the loss function L (Loss function) of the probability model as:

由於前t顆決策樹固定，據此可依據泰勒展開式近似損失函數如下：

，其中g_i、h_i分別為l(f_t(d_i),C_i ^d)之一階與二階導數，在優化第t+1顆樹時，l(f_t(d_i),C_i ^d)為一常數可忽略不計。假設第t+1顆樹之結構已知，其總共有K葉，每片葉上之預測分數為s_k，我們可將損失函數L改寫為：

，其中q為將d對應至葉的函數，I_k為第k葉上所有樣本的集合。由上式可以發現，若新決策樹的結構固定，可將損失函數視為K個獨立的二次方程式，可分別透過微分求極值優化各二次方程式達到優化整個損失函數的效果，當各葉之預測分數

時，可得到最佳的損失函數值，依照此損失函數值所求得之樹模型即可得設備劣化機率模型模組14，該模組需使用前述所提及之測試資料確認該模組具可用性後方可確定此步驟完成。 Since the first t decision trees are fixed, the loss function can be approximated according to the Taylor expansion as follows:

, Where g _{_i,} h _i respectively, _{_{l (f t (d i)}} , C i d) one order and the second derivative, when optimization t + 1 Ke _{_{tree, l (f t (d i}} ), C i ^d ) is a constant and can be ignored. Assuming that the structure of the t+1 th tree is known, there are a total of K leaves, and the prediction score on each leaf is s _k , we can rewrite the loss function L as:

, Where q is the function that corresponds d to the leaf, and I _k is the set of all samples on the kth leaf. It can be found from the above formula that if the structure of the new decision tree is fixed, the loss function can be regarded as K independent quadratic equations, and each quadratic equation can be optimized through differentiation to the extreme value to achieve the effect of optimizing the entire loss function. Predicted Score of Leaf

The optimal loss function value can be obtained, and the tree model obtained according to the loss function value can obtain the equipment degradation probability model module 14. The module needs to use the aforementioned test data to confirm that the module has Only after availability can this step be confirmed.

上述所說明為訓練階段之步驟及方法，此階段結束後方可得到兩種資料模型模組-設備障礙分類模型模組15及設備劣化機率模型模組14。當有新資料進來，可配合資料分析模組13計算潛在劣化機率。 The above descriptions are the steps and methods of the training stage. After this stage, two data model modules-equipment obstacle classification model module 15 and equipment degradation probability model module 14 can be obtained. When new information comes in, you can cooperate with the capital The material analysis module 13 calculates the potential degradation probability.

在加入設備障礙特徵後，係將所得之設備障礙資料加入至資料分析模組13，以供設備劣化機率模型模組14使用設備劣化機率模型計算劣化機率；該設備劣化機率模型為一機率模型(Probability Model)，透過該設備劣化機率模型，將輸入之資料透過其演算方程式後而得一機率值，再藉由該機率值產生一設備劣化機率清單，藉以表示設備潛在劣化之可能性。 After adding the equipment obstacle characteristics, the obtained equipment obstacle data is added to the data analysis module 13, so that the equipment degradation probability model module 14 uses the equipment degradation probability model to calculate the degradation probability; the equipment degradation probability model is a probability model ( Probability Model), through the equipment degradation probability model, the input data is passed through its calculation equation to obtain a probability value, and then a equipment degradation probability list is generated from the probability value to indicate the possibility of potential equipment degradation.

在一實施例中，可建立劣化清單並收集回饋樣本，將產生之該些預測機率資料，建立可供設備維運模組16可識別或參照之資料清單，並使設備維運模組16依據現場實際狀況回報狀況，舉凡更換設備、更換線材、更換Port位等並轉換成是否潛在劣化標籤。換言之，設備維運模組16係用以驗證且標註該些歷史服務量測資料或組態資料是否為潛在劣化之標籤。 In one embodiment, a deterioration list can be created and feedback samples can be collected. The predicted probability data generated can be used to create a list of data that can be identified or referred to by the equipment maintenance and operation module 16, and the equipment maintenance and operation module 16 is based on Report the actual conditions on site, including replacement of equipment, replacement of wires, replacement of port positions, etc., and convert them into potential deterioration labels. In other words, the equipment maintenance and operation module 16 is used to verify and mark whether the historical service measurement data or configuration data is a label of potential degradation.

模型更新模組17，係收集該設備劣化機率清單，藉以驗證是否需要更新模型組合。舉例而言，使用模型更新模組17收集上述步驟中藉由設備維運模組16所填資料，此資料稱為回饋樣本，並驗證是否達到樣本更新門檻，藉以驗證是否需要更新模型組合，此更新門檻可依據實際目的而設定，藉以得出一新模型組合，此說明假定樣本更新門檻值為正確率(Accuracy)<0.95或預測率(Precision)<0.6，意即當新產出之劣化清單經過設備維運模組16驗證後所得之預測率(Precision)小於60%或資料集之正確率 (Accuracy)小於95%，則需使用資料建模模組12重新啟動訓練階段以得到新的模型組合。 The model update module 17 collects a list of the equipment degradation probability to verify whether the model combination needs to be updated. For example, the model update module 17 is used to collect the data filled in by the equipment maintenance module 16 in the above steps. This data is called a feedback sample, and it is verified whether the sample update threshold is reached to verify whether the model combination needs to be updated. The update threshold can be set according to the actual purpose to obtain a new model combination. This description assumes that the sample update threshold is the accuracy rate (Accuracy) <0.95 or the prediction rate (Precision) <0.6, which means that the deterioration list of the new output The prediction rate (Precision) obtained after verification by the equipment maintenance module 16 is less than 60% or the accuracy rate of the data set (Accuracy) is less than 95%, the data modeling module 12 needs to be used to restart the training phase to obtain a new model combination.

綜上所述，本發明主要分為兩種架構，分別為建模階段及分析階段，前者主要用於建立資料模型，後者主要用於分析計算網路設備潛在劣化機率。藉由透過綜整設備歷史量測資料，並解析服務網路中設備間之相依性，透過結合網路設備間之鏈結關係將各個設備服務品質數據與其組態資料關聯，同時因網路設備間連結皆透過其介面進行串接，故亦組合了各設備之介面組態與其服務品質數據，並且使用了機器學習演算法幫助克服各特徵間之遞移相依性，判讀網路設備劣化之潛在可能性，故能達到精確預測可能潛在劣化之設備端點之技術功效。 In summary, the present invention is mainly divided into two architectures, namely the modeling phase and the analysis phase. The former is mainly used to establish a data model, and the latter is mainly used to analyze and calculate the potential degradation probability of network equipment. By integrating the historical measurement data of the equipment and analyzing the dependencies between the equipment in the service network, the service quality data of each equipment is associated with its configuration data by combining the link relationship between the network equipment. The inter-connections are all connected through its interface, so the interface configuration of each device and its service quality data are also combined, and machine learning algorithms are used to help overcome the transitional dependence between features and determine the potential for network equipment degradation Possibility, so it can achieve the technical effect of accurately predicting the equipment endpoints that may be potentially degraded.

本發明復提供一種分析寛頻服務設備潛在劣化機率方法，其方法流程圖如第2圖所示。 The present invention further provides a method for analyzing the potential degradation probability of broadband service equipment. The flowchart of the method is shown in Figure 2.

在步驟S1中，收集與檢查一寬頻服務網路中複數個網路設備及其複數個設備介面之歷史服務量測資料或組態資料。 In step S1, collect and check historical service measurement data or configuration data of multiple network devices and multiple device interfaces in a broadband service network.

在步驟S2中，根據該些歷史服務量測資料或組態資料之關聯集，建立一建模資料集。 In step S2, a modeling data set is established based on the association set of the historical service measurement data or configuration data.

在步驟S3中，根據該些歷史服務量測資料或組態資料之關聯集，建立一分析資料集。 In step S3, an analysis data set is established based on the association set of the historical service measurement data or configuration data.

在步驟S4中，以機器學習演算法利用該建模資料集訓練出一設備障礙分類模型，再將該些歷史服務量測資料或組態資料匯入該設備障礙分類模型與該分析資料集之組合中，以產生一設備障礙分類資料。 In step S4, a machine learning algorithm uses the modeling data set to train an equipment obstacle classification model, and then the historical service measurement data or configuration data is imported into the equipment obstacle classification model and the analysis data set. combination In order to generate a classification data of equipment obstacles.

在步驟S5中，以機器學習演算法利用該建模資料集訓練出一設備劣化機率模型，再將該設備障礙分類資料匯入該設備劣化機率模型中，以產生一設備劣化機率清單。 In step S5, a machine learning algorithm is used to train an equipment degradation probability model using the modeling data set, and then the equipment obstacle classification data is imported into the equipment degradation probability model to generate a equipment degradation probability list.

在步驟S6中，驗證且標註該些歷史服務量測資料或組態資料是否為潛在劣化之標籤。 In step S6, verify and mark whether the historical service measurement data or configuration data are labels of potential degradation.

在步驟S7中，收集該設備劣化機率清單，藉以驗證是否需要更新模型組合。 In step S7, a list of the equipment degradation probability is collected to verify whether the model combination needs to be updated.

綜上所述，本發明之分析寛頻服務設備潛在劣化機率之系統及其方法，主係判讀網路設備劣化之潛在可能性。其透過綜整設備歷史量測資料，並解析服務網路中設備間之相依性，透過結合網路設備間之鏈結關係將各個設備服務品質數據與其組態資料關聯，同時因網路設備間連結皆透過其介面進行串接，故亦組合了各設備之介面組態與其服務品質數據，並且使用了機器學習演算法幫助克服各特徵間之遞移相依性，而能達到快速及/或精確地預測可能潛在劣化之設備端點之技術功效。 To sum up, the system and method for analyzing the potential degradation probability of broadband service equipment of the present invention mainly determines the potential degradation of network equipment. It integrates the historical measurement data of the equipment and analyzes the dependencies between the equipment in the service network, and associates the service quality data of each equipment with its configuration data by combining the link relationship between the network equipment. The links are all connected through its interface, so the interface configuration of each device and its service quality data are also combined, and machine learning algorithms are used to help overcome the shift dependency between features, and achieve fast and/or accurate Predict the technical efficacy of the equipment endpoints that may potentially degrade.

上述實施例係用以例示性說明本發明之原理及其功效，而非用於限制本發明。任何熟習此項技藝之人士均可在不違背本發明之精神及範疇下，對上述實施例進行修改。因此本發明之權利保護範圍，應如後述之申請專利範圍所列。 The above-mentioned embodiments are used to exemplify the principles and effects of the present invention, but not to limit the present invention. Anyone familiar with this technique can modify the above-mentioned embodiments without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the rights of the present invention should be listed in the scope of patent application described later.

11‧‧‧資料處理模組 11‧‧‧Data Processing Module

12‧‧‧資料建模模組 12‧‧‧Data Modeling Module

13‧‧‧資料分析模組 13‧‧‧Data Analysis Module

16‧‧‧設備維運模組 16‧‧‧Equipment maintenance module

17‧‧‧模型更新模組 17‧‧‧Model Update Module

Claims

一種分析寛頻服務設備潛在劣化機率之系統，係包括：資料處理模組，係用以收集與檢查一寬頻服務網路中複數個網路設備及其複數個設備介面之歷史服務量測資料、組態資料或劣化標籤資料；資料建模模組，係根據該些歷史服務量測資料、組態資料或劣化標籤資料之關聯集，建立一建模資料集；資料分析模組，係根據該些歷史服務量測資料、組態資料或劣化標籤資料之關聯集，建立一分析資料集；設備障礙分類模型模組，係以機器學習演算法利用該建模資料集訓練出一設備障礙分類模型，再將該些歷史服務量測資料、組態資料或劣化標籤資料匯入該設備障礙分類模型與該分析資料集之組合中，以產生一設備障礙分類資料；以及設備劣化機率模型模組，係以機器學習演算法利用該建模資料集訓練出一設備劣化機率模型，再將該設備障礙分類資料匯入該設備劣化機率模型中，以產生一設備劣化機率清單。 A system for analyzing the potential degradation probability of broadband service equipment includes: a data processing module for collecting and checking historical service measurement data of multiple network devices and multiple device interfaces in a broadband service network, Configuration data or degraded label data; data modeling module is based on the correlation set of these historical service measurement data, configuration data or degraded label data to create a modeling data set; data analysis module is based on the The correlation set of some historical service measurement data, configuration data or degraded label data is used to establish an analysis data set; the equipment obstacle classification model module uses the machine learning algorithm to use the modeling data set to train an equipment obstacle classification model , And then import the historical service measurement data, configuration data or degradation label data into the combination of the equipment obstacle classification model and the analysis data set to generate a equipment obstacle classification data; and the equipment degradation probability model module, The machine learning algorithm uses the modeling data set to train an equipment degradation probability model, and then imports the equipment obstacle classification data into the equipment degradation probability model to generate a equipment degradation probability list.

如申請專利範圍第1項所述之分析寛頻服務設備潛在劣化機率之系統，復包括模型更新模組，係收集該設備劣化機率清單，藉以驗證是否需要更新模型組合。 For example, the system for analyzing the potential degradation probability of broadband service equipment described in the first item of the patent application includes a model update module, which collects a list of the equipment degradation probability to verify whether it is necessary to update the model combination.

如申請專利範圍第1項所述之分析寛頻服務設備潛在劣化機率之系統，復包括設備維運模組，係用以驗證且標註該些歷史服務量測資料、組態資料或劣化標籤資料是否為潛在劣化之標籤。 Analyze the potential of broadband service equipment as described in item 1 of the scope of patent application The degradation probability system also includes equipment maintenance and operation modules, which are used to verify and mark whether the historical service measurement data, configuration data, or degradation tag data are potentially degraded tags.

如申請專利範圍第1項所述之分析寛頻服務設備潛在劣化機率之系統，其中，該設備障礙分類模型為一ID3決策樹(ID3 Decision Tree)、一C4.5決策樹(C4.5 Decision Tree)、一C5.0決策樹(C5.0 Decision Tree)、一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。 For example, the system for analyzing the potential degradation probability of broadband service equipment described in item 1 of the scope of patent application, wherein the equipment obstacle classification model is an ID3 Decision Tree and a C4.5 Decision Tree (C4.5 Decision Tree). Tree), a C5.0 decision tree (C5.0 Decision Tree), a CART tree (Classification And Regression Tree), a random forest (Random Forest) or a gradient boosting decision tree (Gradient Boosting Decision Tree).

如申請專利範圍第1項所述之分析寛頻服務設備潛在劣化機率之系統，其中，該設備劣化機率模型為一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。 For example, the system for analyzing the potential degradation probability of broadband service equipment as described in item 1 of the scope of patent application, wherein the device degradation probability model is a CART tree (Classification And Regression Tree), a random forest (Random Forest) or a gradient boost Decision tree (Gradient Boosting Decision Tree).

如申請專利範圍第1項所述之分析寛頻服務設備潛在劣化機率之系統，其中，該使用資料建模模組針對該設備介面找出其對應之上鏈或下鏈對接設備，其中，以靠近網路核心之對接設備定義為該上鏈對接設備，以遠離網路核心之對接設備定義為該下鏈對接設備。 For example, the system for analyzing the potential degradation probability of broadband service equipment described in item 1 of the scope of patent application, wherein the usage data modeling module finds the corresponding uplink or downlink docking device for the device interface, where The docking device close to the network core is defined as the uplink docking device, and the docking device far away from the network core is defined as the downlink docking device.

一種分析寛頻服務設備潛在劣化機率之方法，係包括：收集與檢查一寬頻服務網路中複數個網路設備及其複數個設備介面之歷史服務量測資料、組態資料或劣化標籤資料；根據該些歷史服務量測資料、組態資料或劣化標籤資料之關聯集，建立一建模資料集；根據該些歷史服務量測資料、組態資料或劣化標籤資料之關聯集，建立一分析資料集；以機器學習演算法利用該建模資料集訓練出一設備障礙分類模型，再將該些歷史服務量測資料、組態資料或劣化標籤資料匯入該設備障礙分類模型與該分析資料集之組合中，以產生一設備障礙分類資料；以及以機器學習演算法利用該建模資料集訓練出一設備劣化機率模型，再將該設備障礙分類資料匯入該設備劣化機率模型中，以產生一設備劣化機率清單。 A method for analyzing the potential degradation probability of broadband service equipment includes: collecting and checking historical service measurement data, configuration data, or degradation tag data of multiple network devices and multiple device interfaces in a broadband service network; Create a modeling data set based on the association set of the historical service measurement data, configuration data or degraded label data; establish an analysis based on the association set of the historical service measurement data, configuration data or degraded label data Data set: Use the modeling data set to train a device obstacle classification model with a machine learning algorithm, and then import the historical service measurement data, configuration data or degradation label data into the device obstacle classification model and the analysis data In the combination of the set, a device failure classification data is generated; and a machine learning algorithm is used to use the modeling data set to train a device degradation probability model, and then the device failure classification data is imported into the device degradation probability model to Generate a list of equipment degradation probabilities.

如申請專利範圍第7項所述之分析寛頻服務設備潛在劣化機率之方法，復包括收集該設備劣化機率清單，藉以驗證是否需要更新模型組合。 For example, the method of analyzing the potential degradation probability of broadband service equipment described in item 7 of the scope of patent application includes collecting a list of the degradation probability of the equipment to verify whether it is necessary to update the model combination.

如申請專利範圍第7項所述之分析寛頻服務設備潛在劣化機率之方法，復包括驗證且標註該些歷史服務量測資料、組態資料或劣化標籤資料是否為潛在劣化之標籤。 For example, the method for analyzing the potential degradation probability of broadband service equipment as described in item 7 of the scope of patent application includes verifying and marking whether the historical service measurement data, configuration data or degraded label data are potentially degraded labels.

如申請專利範圍第7項所述之分析寛頻服務設備潛在劣化機率之方法，其中，該設備障礙分類模型為一ID3決策樹(ID3 Decision Tree)、一C4.5決策樹(C4.5 Decision Tree)、一C5.0決策樹(C5.0 Decision Tree)、一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。 For example, the method for analyzing the potential degradation probability of broadband service equipment described in item 7 of the scope of patent application, wherein the equipment obstacle classification model is an ID3 Decision Tree and a C4.5 Decision Tree (C4.5 Decision Tree). Tree), a C5.0 decision tree (C5.0 Decision Tree), a CART tree (Classification And Regression Tree), a random forest (Random Forest) or a gradient boosting decision tree (Gradient Boosting Decision Tree).

如申請專利範圍第7項所述之分析寛頻服務設備潛在劣化機率之方法，其中，該設備劣化機率模型為一CART樹(Classification And Regression Tree)、一隨機森林(Random Forest)或一梯度提昇決策樹(Gradient Boosting Decision Tree)。 For example, the method for analyzing the potential degradation probability of broadband service equipment as described in item 7 of the scope of patent application, wherein the device degradation probability model is a CART tree (Classification And Regression Tree), a random forest (Random Forest) or a gradient boost Decision tree (Gradient Boosting Decision Tree).

如申請專利範圍第7項所述之分析寛頻服務設備潛在劣化機率之方法，復包括針對該設備介面找出其對應之上鏈或下鏈對接設備，其中，以靠近網路核心之對接設備定義為該上鏈對接設備，以遠離網路核心之對接設備定義為該下鏈對接設備。 For example, the method of analyzing the potential degradation probability of broadband service equipment as described in item 7 of the scope of patent application includes finding the corresponding uplink or downlink docking device for the device interface, among which, the docking device close to the network core It is defined as the uplink docking device, and the docking device far away from the network core is defined as the downlink docking device.