Training method of box body identification model and hidden danger identification method of optical splitting box body
Through the training and feature screening of the box recognition model, the hidden dangers of optical sub-boxes in the Gibit passive optical network are identified, which solves the positioning problem of low-light problems of optical network units, improves maintenance efficiency and reduces the frequency of failure.
Patent Information
- Application Number
- CN202410966725.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-15
AI Technical Summary
In Gibit passive optical networks, the low-light problem of optical network units, especially the phenomenon of jumping and weak light, leads to deterioration of business performance and difficulty in accurately judgement and positioning, affecting installation and peacekeeping line maintenance.
The box recognition model is used to obtain and feature encode the attribute parameters of the box, use a random forest model to screen important features, train the box recognition model, and then judge the confidence of the optical box and identify potential hidden dangers box. .
It improves the positioning efficiency of low-light problems, reduces the proportion of jumping and weak light, reduces the average interruption time and failure duration of a single user, and avoids waste of resources.
Smart Images

Figure CN118964974A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the fields of deep learning, data processing, and computer technology, and in particular to a training method for a box recognition model and a method for identifying hidden dangers of an optical separation box. Background Art
[0002] In the maintenance scenario of wired broadband services based on Gigabit-capable PON (GPON) networks, the optical network unit (ONU) weak light problem is often one of the important factors that lead to the degradation of service performance. In the daily process of dealing with the ONU weak light problem of home and enterprise broadband users, in addition to the "stable weak light" problem of some users caused by equipment problems and line problems, there is also the phenomenon that the ONU receiving optical power of some existing users will only produce "jump weak light" in certain scenarios or specific time periods. "Stable weak light" means that the user is lower than -27dbm for at least one week in a row, and "jump weak light" means that the user has optical power degradation in different time periods of the same day and then returns to the initial stable value, or degradation occurs in specific weather factors (such as rainy days, nights). The above ONU optical power jump (OPJ) phenomenon often affects installation and maintenance or line maintenance. When the current optical power is tested to be up to standard, it is often impossible to accurately judge and locate the real cause of weak light and line problem points.
[0003] Therefore, how to accurately identify boxes with hidden dangers, improve the efficiency of locating weak light problems, provide data support for installation or line maintenance, and thus reduce the proportion of jumpy weak light has become one of the important research directions. Summary of the invention
[0004] The present disclosure aims to solve one of the technical problems in the related art at least to a certain extent. To this end, the first embodiment of the present disclosure proposes a training method for a box recognition model, comprising:
[0005] Acquire multiple attribute parameters of the first feature type of the preset optical distribution box sample, and perform feature encoding and feature derivation on the attribute parameters to obtain candidate feature variables of the second feature type, wherein the attribute parameters of the first feature type include box attribute parameters, environment attribute parameters, attribute parameters of the downband optical network unit included in the optical distribution box, and attribute parameters of the uplink optical line terminal;
[0006] Performing feature screening on candidate feature variables of the second feature type based on a preset random forest model to obtain target feature variables of the target feature type;
[0007] Input the target feature variable into a preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box sample;
[0008] The candidate box recognition model is trained based on the confidence of the downband optical network unit until the training is completed to obtain the target box recognition model.
[0009] In the disclosed embodiment, the attribute parameters are feature encoded and feature derived to obtain candidate feature variables of the second feature type. The candidate feature variables of the second feature type are feature screened based on a preset random forest model to obtain target feature variables of the target feature type. Feature types with higher importance can be screened out, and the influence of attribute parameters of different feature types on the prediction results can be comprehensively learned. The target feature variables are input into the preset candidate cabinet recognition model to obtain the confidence of multiple down-band optical network units included in the optical distribution cabinet sample. The candidate cabinet recognition model is trained based on the confidence of the down-band optical network unit to obtain the target cabinet recognition model. The target cabinet recognition model can be used to accurately judge the cabinet with hidden dangers, improve the efficiency of locating weak light problems, and provide data support for installation or line maintenance, thereby reducing the proportion of jump weak light.
[0010] The second aspect of the present disclosure provides a method for identifying hidden dangers of an optical distribution box, including:
[0011] Obtaining attribute parameters of the target feature type of the optical sub-box to be identified;
[0012] Performing feature encoding on the attribute parameters of the target feature type to obtain the target feature variable of the target feature type;
[0013] Input the target feature variable into the target box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box to be identified;
[0014] Based on the confidence of the downband optical network unit, it is judged whether the optical distribution box to be identified is a hidden danger box;
[0015] The target box recognition model is trained by the box recognition model training method proposed in the first aspect embodiment.
[0016] In the disclosed embodiment, the target characteristic variable is input into the target box identification model, the confidence of multiple down-band optical network units included in the optical sub-box to be identified is obtained, and whether the optical sub-box to be identified is a hidden danger box is judged based on the confidence of the down-band optical network unit. This can speed up the positioning of weak light problems and reduce the proportion of jumping weak light, reduce the average interruption time of a single user, the fault duration and the proportion of weak light ONU, and avoid wasting resources.
[0017] The third aspect of the present disclosure provides a training device for a box recognition model, including:
[0018] A first acquisition module is used to acquire multiple attribute parameters of a first feature type of a preset optical distribution box sample, and perform feature encoding and feature derivation on the attribute parameters to obtain candidate feature variables of a second feature type, wherein the attribute parameters of the first feature type include box attribute parameters, environment attribute parameters, attribute parameters of a downband optical network unit included in the optical distribution box, and attribute parameters of an uplink optical line terminal;
[0019] A feature screening module, used for screening the candidate feature variables of the second feature type based on a preset random forest model to obtain a target feature variable of the target feature type;
[0020] A second acquisition module is used to input the target feature variable into a preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box sample;
[0021] The training module is used to train the candidate box recognition model based on the confidence of the downband optical network unit until the training is completed to obtain the target box recognition model.
[0022] The fourth aspect of the present disclosure provides a device for identifying hidden dangers of an optical distribution box, including:
[0023] A first acquisition module is used to acquire attribute parameters of a target feature type of a light separation box to be identified;
[0024] A feature encoding module is used to perform feature encoding on the attribute parameters of the target feature type to obtain a target feature variable of the target feature type;
[0025] A second acquisition module is used to input the target feature variable into a preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box to be identified;
[0026] A judgment module, used for judging whether the optical sub-box to be identified is a hidden danger box based on the confidence of the down-band optical network unit;
[0027] The target box recognition model is trained by the box recognition model training device proposed in the third aspect embodiment.
[0028] A fifth aspect of the present disclosure provides an electronic device, including:
[0029] at least one processor; and
[0030] a memory communicatively connected to at least one processor; wherein,
[0031] The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor so that the at least one processor can execute the training method of the box recognition model provided in the embodiment of the first aspect of the present disclosure, or execute the hidden danger identification method of the optical distribution box provided in the embodiment of the second aspect of the present disclosure.
[0032] The sixth aspect embodiment of the present disclosure proposes a computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions are used to enable a computer to execute the training method of the box recognition model provided in the first aspect embodiment of the present disclosure, or to execute the hidden danger identification method of the optical distribution box provided in the second aspect embodiment of the present disclosure.
[0033] The seventh aspect embodiment of the present disclosure proposes a computer program product, including a computer program. When the computer program is executed by a processor, it implements the training method of the box recognition model provided in the first aspect embodiment of the present disclosure, or implements the hidden danger identification method of the optical distribution box provided in the second aspect embodiment of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 is a flow chart of a method for training a box recognition model according to an embodiment of the present disclosure;
[0035] Figure 2 is a schematic diagram of attribute parameters of an embodiment of the present disclosure;
[0036] Figure 3 is a schematic diagram of a feature derivative of an embodiment of the present disclosure;
[0037] Figure 4 is a schematic diagram of training set division according to an embodiment of the present disclosure;
[0038] Figure 5 is a flow chart of a method for training a box recognition model according to an embodiment of the present disclosure;
[0039] Figure 6 is a structural schematic diagram of a box recognition model according to an embodiment of the present disclosure;
[0040] Figure 7 is a flow chart of a method for identifying hidden dangers of an optical distribution box according to an embodiment of the present disclosure;
[0041] Figure 8 is a structural block diagram of a training device for a box recognition model according to an embodiment of the present disclosure;
[0042] Fig. 9 This is a structural block diagram of a hidden danger identification device for an optical sub-box according to an embodiment of the present disclosure;
[0043] Fig.10It is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION
[0044] Embodiments of the present disclosure are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present disclosure, and should not be construed as limiting the present disclosure.
[0045] For ease of understanding, the terms in the embodiments of the present disclosure are explained below:
[0046] Box: In a fiber access network, a box usually refers to a physical housing or box used to install fiber optic equipment, distribute fiber optic lines, and protect fiber optic connection components. These boxes usually contain equipment such as fiber distribution frames and fiber optic terminal boxes, which are used to manage and protect fiber optic connections.
[0047] Downstream Optical Network Unit (ONU): ONU is a device in the optical fiber access network that converts optical signals into electrical signals and then transmits them to the end user's device, such as home broadband access equipment. Downstream (lower link) means that ONU is located downstream of the optical access network and is responsible for communicating with the end user's device.
[0048] Uplink Optical Line Terminal OLT (Optical Line Terminal): OLT is another device in the optical access network, which is used to aggregate and manage the optical signals of multiple ONUs and connect to the operator's core network. Uplink means that OLT is located upstream of the optical access network and is responsible for communicating with the operator's network.
[0049] XGPON (10G-PON) is the abbreviation of 10G Next-Generation Passive Optical Network. It is a fiber-optic communication technology mainly used for the transmission and distribution of high-speed data, audio and video services.
[0050] Passive Optical Network (PON) is a single-fiber bidirectional optical access network that uses a point-to-multipoint (P2MP) structure.
[0051] The following describes the training method of the box recognition model and the hidden danger recognition method of the optical separation box according to the embodiments of the present disclosure in conjunction with the accompanying drawings.
[0052] Figure 1 is a flow chart of a method for training a box recognition model according to an embodiment of the present disclosure. Figure 1As shown, the method comprises the following steps:
[0053] S101, obtaining a plurality of attribute parameters of a first feature type of preset light classification box samples, and performing feature coding and feature derivation on the attribute parameters to obtain candidate feature variables of a second feature type.
[0054] Among them, Figure 2 As shown, the attribute parameters of the first characteristic type include cabinet attribute parameters, environmental attribute parameters, attribute parameters of the downband optical network unit included in the optical distribution cabinet, and attribute parameters of the uplink optical line terminal. For example, the environmental attribute parameters may include characteristic types such as weather, temperature, and relative humidity; the cabinet attribute parameters may include characteristic types such as geographical environment, cabinet capacity, and port utilization; the downband ONU attribute parameters may include characteristic types such as the manufacturer, model, installation method, ONU transmit optical power, ONU receive optical power, ONU upstream rate, and ONU downstream rate; the attribute parameters of the uplink OLT may include characteristic types such as whether it is XGPON, PON port transmit optical power, and PON receive optical power.
[0055] In some embodiments, based on the parameter type of the attribute parameter, the attribute parameter is feature encoded to obtain the initial feature variable of the first feature type. In the embodiment of the present disclosure, the parameter type can be divided into discrete and continuous types. Discrete variables include, for example, weather, geographical environment, manufacturer, model, cabinet capacity, installation method, whether it is XGPON, etc., which can be feature encoded according to the enumeration value of the feature to obtain the initial feature variable; continuous variables include, for example, temperature, relative humidity, port utilization, ONU transmit optical power, ONU receive optical power, ONU upstream rate, ONU downstream rate, PON port transmit optical power, PON receive optical power, etc., which can be divided into boxes and encoded according to the interval division to obtain the initial feature variable. The specific encoding value is as follows: Figure 2 shown.
[0056] In some embodiments, the first feature type includes a specified third feature type and a fourth feature type, and the initial feature variables of the third feature type and the initial feature variables of the fourth feature type are subjected to second-order cross-feature derivation to obtain candidate feature variables of the second feature type. For example, the weather feature variable is used to perform second-order cross-feature derivation with the cabinet capacity, port utilization, ONU received optical power, ONU upstream rate, ONU downstream rate, etc. Among them, there are 5 categories of weather features, 2 categories of cabinet capacity, and 5*2=10 categories after second-order derivation. There are 6 categories of port utilization, and 5*6=30 categories after second-order derivation. There are 7 categories of ONU received optical power, and 5*7=35 categories after second-order derivation. There are 5 categories of ONU upstream rate, and 5*5=25 categories after second-order derivation. There are 5 categories of ONU downstream rate, and 5*5=25 categories after second-order derivation.
[0057] like Figure 3 As shown in the figure, taking weather and ONU received optical power as examples, there are 5 categories of weather variables and 7 categories of ONU received optical power, and a total of 35 categories are derived from the two. Similarly, the cross-derivation operation of weather and cabinet capacity, port utilization, ONU upstream rate, and ONU downstream rate is consistent with the above. After feature derivation, all feature variables are uniquely encoded to obtain candidate feature variables of the second feature type.
[0058] S102, performing feature screening on candidate feature variables of the second feature type based on a preset random forest model to obtain target feature variables of the target feature type.
[0059] In some implementations, in order to improve the training efficiency of the box recognition model, a target feature type that has a greater impact on the predicted sample type may be screened from the second feature type.
[0060] In the disclosed embodiment, the candidate feature variables of the second feature type are calculated based on a preset random forest model to obtain importance parameters corresponding to each feature type. The importance parameters can indicate the degree of influence of the feature type on the predicted sample type. Feature screening is performed based on the importance parameters to obtain target feature variables of the target feature type.
[0061] S103, inputting the target feature variable into a preset candidate cabinet recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-cabinet sample.
[0062] In the embodiment of the present disclosure, the candidate box recognition model includes N tree model layers and an output layer. Each tree model layer includes multiple different types of tree models. The structure of each tree model layer can be the same or different, and the embodiment of the present disclosure is not limited to this.
[0063] In the disclosed embodiment, the candidate box recognition model is used to make inference predictions on each down-band optical network unit included in the optical sub-box sample based on the target feature variables, and obtain the confidence level of each down-band optical network unit (ie, obtain the inference value).
[0064] S104, training the candidate box recognition model based on the confidence of the downband optical network unit until the training is completed to obtain a target box recognition model.
[0065] In the embodiment of the present disclosure, the optical classification cabinet samples include two types of samples, namely, potential danger cabinet samples and non-potential danger cabinet samples.
[0066] The predicted sample type of the optical sub-box sample is obtained based on the confidence of the down-band optical network unit. In some embodiments, the predicted sample type of the optical sub-box sample indicates that the optical sub-box sample is a hidden danger box sample. In some embodiments, the predicted sample type of the optical sub-box sample indicates that the optical sub-box sample is not a hidden danger box sample.
[0067] For example, if the predicted sample type of the optical classification bin sample is inconsistent with the reference sample type, the parameters of the candidate bin recognition model are adjusted, and the candidate bin recognition model is trained using the next optical classification bin sample until the training is completed to obtain the target bin recognition model.
[0068] In some embodiments, the model is trained based on a preset loss function, and when the loss function converges to a preset loss threshold, the training is determined to be completed, and the target box recognition model is obtained. In some embodiments, after the number of iterations reaches a preset iteration number threshold, the training is determined to be completed, and the target box recognition model is obtained. This disclosure is not limited to this.
[0069] In the disclosed embodiment, the attribute parameters are feature encoded and feature derived to obtain candidate feature variables of the second feature type. The candidate feature variables of the second feature type are feature screened based on a preset random forest model to obtain target feature variables of the target feature type. Feature types with higher importance can be screened out, and the influence of attribute parameters of different feature types on the prediction results can be comprehensively learned. The target feature variables are input into the preset candidate cabinet recognition model to obtain the confidence of multiple down-band optical network units included in the optical distribution cabinet sample. The candidate cabinet recognition model is trained based on the confidence of the down-band optical network unit to obtain the target cabinet recognition model. The target cabinet recognition model can be used to accurately judge the cabinet with hidden dangers, improve the efficiency of locating weak light problems, and provide data support for installation or line maintenance, thereby reducing the proportion of jump weak light.
[0070] In some embodiments, a serious imbalance in the ratio of positive and negative samples may cause the model to be unable to learn the minority class samples insufficiently, and thus may not be able to fully learn and understand the characteristics of the minority class samples, which may lead to poor generalization ability of the model in real scenarios and affect the training effect. Therefore, in the embodiment of the present disclosure, in order to improve the training of the box recognition model, when the ratio of positive and negative samples is unbalanced, the training set is divided, including: obtaining a first number of hidden danger box samples and a second number of non-hidden danger box samples, in response to the absolute value of the difference between the first number and the second number being greater than a preset number threshold, determining a larger number of target box samples in the hidden danger box samples and the non-hidden danger box, obtaining a third number of downband optical network units included in the target box samples, clustering the target box samples based on the third number, obtaining multiple cluster categories, and dividing the hidden danger box samples into training sets based on the cluster categories. Specifically, as described in Figure 4 As shown in the figure, k-means (kmeans) clustering is performed on the minority class samples, and then according to the relationship between the resource center system box and onu, each box randomly selects a sample from the cluster category (class, cls) with the largest number of samples. If the number of samples in the cluster categories is the same, a cluster category is randomly selected, and then a sample is randomly selected from the category.
[0071] Figure 5 is a flow chart of a method for training a box recognition model according to an embodiment of the present disclosure. Figure 5 As shown, the method comprises the following steps:
[0072] S501, obtaining a plurality of attribute parameters of a first feature type of preset light classification box samples, and performing feature coding and feature derivation on the attribute parameters to obtain candidate feature variables of a second feature type.
[0073] For the introduction of step S501, please refer to the relevant content in the above embodiment, which will not be repeated here.
[0074] S502, inputting the candidate feature variables of the second feature type into the random forest model for pre-training to obtain the importance parameter of each second feature type.
[0075] In some embodiments, the candidate feature variables of the second feature type after the second-order derivation are substituted into the random forest model for pre-training, and after the pre-training is completed, the importance parameters of the influence of each feature type on the prediction result are obtained.
[0076] S503: Based on the importance parameter of the second feature type, perform feature screening on the second feature type and its candidate feature variables to obtain a target feature variable of the target feature type.
[0077] The target feature variables of the target feature type whose feature importance parameters are greater than the preset parameter threshold are selected as the training samples of the model, such as weather and ONU received optical power, cabinet capacity and ONU received optical power, and other second-order derivative features whose feature importance parameters are greater than 0.3, as training samples input into the model.
[0078] In some implementations, the parameter threshold may be 0.3. In other implementations, the parameter threshold may also be other values, which is not limited in the present disclosure.
[0079] S504, inputting the target feature variable into the first tree model layer of the candidate box recognition model, and having each tree model of the first tree model layer predict the target feature variable to obtain a plurality of first feature variables.
[0080] The candidate box recognition model includes N tree model layers and an output layer, and each tree model layer includes multiple tree models of different types.
[0081] like Figure 6 As shown, the candidate box recognition model includes two tree model layers, namely the first tree model layer and the second tree model layer. The candidate box recognition model is a two-layer stacked model architecture. The first layer is composed of 7 tree models (XGBoost, Catboost, LightGBM, RFGini, RFEntr, ExtraTreeGini, ExtraTreeEntr), among which XGBoost, Catboost and LightGBM are boosted tree models, RFGini and RFEntr are random tree models, ExtraTreeGini and ExtraTreeEntr are the same tree model but with different cost metrics, namely Gini coefficient and entropy metric (Entr entropy metric). The generalization and accuracy of the model are improved by combining different types of tree models and different generation values under the same tree model. The first tree model layer is mainly used to train and improve the performance and accuracy of a single model. The input of the first tree model layer is the feature parameters of the training sample selected in step 503, that is, the target feature variable of the target feature type. The 7 tree models of the first tree model layer predict the target feature variable respectively to obtain multiple first feature variables.
[0082] S505, concatenate the multiple first feature variables and input them into the next tree model layer for prediction, until multiple N-th feature variables output by the N-th layer are obtained.
[0083] Wherein, N is an integer greater than 1.
[0084] like Figure 6As shown, the second-layer tree model layer and the first-layer tree model layer have the same structure as an example. The input of the second-layer tree model layer is the concatenation of the output vectors of each tree model in the first layer. The second-layer tree model layer mainly improves the overall generalization ability of the model by introducing multiple model linkage training.
[0085] S506, weighting the Nth characteristic variable, and outputting the confidence of the downstream optical network unit through the output layer.
[0086] like Figure 6 As shown, the output of the second tree model layer is weighted and optimized, and the confidence of the downband optical network unit is obtained through the output layer. The weighted weight is used as the target parameter of the model training, which is the same as the training target parameter of the model itself, and is finally obtained through model training.
[0087] S507: Determine a fourth number of downband optical network units whose confidence level is greater than or equal to a preset confidence threshold.
[0088] In the embodiment of the present disclosure, the confidence threshold may be 0.5. In other implementations, the confidence threshold may also be other values, such as 0.6 or 0.7, and the present disclosure does not impose any limitation on this.
[0089] S508: Obtain a ratio between the fourth quantity and the total quantity of the downband optical network units.
[0090] In the disclosed embodiment, based on the ratio of the fourth number to the total number of down-band optical network units, the proportion of down-band optical network units having a confidence level greater than or equal to a preset confidence threshold relative to all down-band optical network units is obtained.
[0091] S509: Determine the predicted sample type of the light classification bin samples based on the quantity ratio.
[0092] In some embodiments, if the quantity ratio is greater than or equal to a preset ratio threshold, the predicted sample type of the optical classification box sample is determined to be a hidden danger box sample; if the quantity ratio is greater than the preset ratio threshold, the predicted sample type of the optical classification box sample is determined to be a hidden danger box sample; if the quantity ratio is less than the preset ratio threshold, the predicted sample type of the optical classification box sample is determined to be a non-hidden danger box sample.
[0093] S510, reversely adjust and iteratively train the parameters of the candidate box recognition model based on the predicted sample type and the reference sample type of the optical classification box sample until the training is completed to obtain a target box recognition model.
[0094] In some embodiments, based on the predicted sample type and the reference sample type of the optical bin sample, a loss function is obtained, and based on the loss function, the parameters of the candidate bin recognition model are reversely adjusted and iteratively trained until the training is completed to obtain the target bin recognition model. The reference sample type is a pre-set or specified sample type of the optical bin sample.
[0095] In the disclosed embodiments, the problem that weak light hopping is difficult to analyze and process can be located and solved, and hidden dangers of optical sub-boxes can be mined. Cluster analysis of hopping characteristics in special scenarios can be performed, focusing on hidden danger points of optical sub-box resources that are most likely to cause hopping. This can serve as data support to accurately investigate the causes of weak light hopping. Through scenario-based data analysis and mining, users who are hopping in different scenarios can be obtained, thereby accelerating the location of weak light problems and reducing the proportion of weak light hopping, reducing the average interruption duration of a single user, the duration of faults and the proportion of weak light ONUs, and avoiding waste of resources.
[0096] Figure 7 FIG. 1 is a flow chart of a method for identifying hidden dangers of an optical distribution box according to an embodiment of the present disclosure. Figure 7 As shown, the method comprises the following steps:
[0097] S701, obtaining attribute parameters of a target feature type of a light distribution box to be identified.
[0098] In some embodiments, the attribute parameter of the target feature type may be an attribute parameter of a specified feature type. In some embodiments, the target feature type is obtained by screening multiple feature types using the training method of the above-mentioned box recognition model.
[0099] Optionally, the target feature type may include at least two of the following: weather, temperature, relative humidity, geographical environment, cabinet capacity, port utilization, manufacturer, model, installation method, ONU transmit optical power, ONU receive optical power, ONU upstream rate, ONU downstream rate, whether XGPON, PON port transmit optical power, and PON receive optical power.
[0100] S702, feature encoding is performed on the attribute parameters of the target feature type to obtain a target feature variable of the target feature type.
[0101] In some implementations, feature encoding may be performed based on the parameter type of the attribute parameter to obtain a target feature variable of the target feature type.
[0102] S703: Input the target characteristic variable into the target cabinet recognition model to obtain the confidence of multiple downlink optical network units included in the optical sub-cabinet to be recognized.
[0103] In the disclosed embodiment, the target box recognition model is trained by the above-mentioned box recognition model training method.
[0104] S704: Determine whether the optical distribution box to be identified is a hidden danger box based on the confidence of the downband optical network unit.
[0105] The ratio of the number of downband optical network units whose confidence exceeds the confidence threshold to the total number of downband optical network units is obtained. If the ratio exceeds the preset ratio threshold, the optical sub-box to be identified is determined to be a hidden danger box; otherwise, the optical sub-box to be identified is determined to be a non-hidden danger box.
[0106] In the disclosed embodiment, the target characteristic variable is input into the target box identification model, the confidence of multiple down-band optical network units included in the optical sub-box to be identified is obtained, and whether the optical sub-box to be identified is a hidden danger box is judged based on the confidence of the down-band optical network unit. This can speed up the positioning of weak light problems and reduce the proportion of jumping weak light, reduce the average interruption time of a single user, the fault duration and the proportion of weak light ONU, and avoid wasting resources.
[0107] Figure 8 is a structural block diagram of a training device for a box recognition model according to an embodiment of the present disclosure. Figure 8 As shown, the training device 800 of the box recognition model includes:
[0108] The first acquisition module 810 is used to acquire a plurality of attribute parameters of a first feature type of a preset optical distribution box sample, and perform feature encoding and feature derivation on the attribute parameters to obtain candidate feature variables of a second feature type, wherein the attribute parameters of the first feature type include box attribute parameters, environment attribute parameters, attribute parameters of a downlink optical network unit included in the optical distribution box, and attribute parameters of an uplink optical line terminal;
[0109] A feature screening module 820 is used to perform feature screening on candidate feature variables of the second feature type based on a preset random forest model to obtain target feature variables of the target feature type;
[0110] The second acquisition module 830 is used to input the target feature variable into a preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box sample;
[0111] The training module 840 is used to train the candidate box recognition model based on the confidence of the downband optical network unit until the training is completed to obtain the target box recognition model.
[0112] In some implementations, the first feature type includes a specified third feature type and a fourth feature type, and the first acquisition module 810 is further configured to:
[0113] Based on the parameter type of the attribute parameter, feature encoding is performed on the attribute parameter to obtain an initial feature variable of the first feature type;
[0114] The initial feature variables of the third feature type and the initial feature variables of the fourth feature type are subjected to second-order cross-feature derivation to obtain candidate feature variables of the second feature type.
[0115] In some embodiments, the feature screening module 820 is further configured to:
[0116] Inputting candidate feature variables of the second feature type into a random forest model for pre-training to obtain importance parameters of each second feature type;
[0117] Based on the importance parameter of the second feature type, feature screening is performed on the second feature type and its candidate feature variables to obtain a target feature variable of the target feature type.
[0118] In some embodiments, the reference sample types of the optical classification bin samples include hidden danger bin samples and non-hidden danger bin samples, and the bin recognition model training device 800 further includes a training set division module 850 for:
[0119] Obtain a first number of potential hazard box samples and a second number of non-potential hazard box samples;
[0120] In response to the absolute value of the difference between the first quantity and the second quantity being greater than a preset quantity threshold, determining a larger quantity of target box samples among the hidden danger box samples and the non-hidden danger box samples;
[0121] Obtaining a third number of downband optical network units included in the target cabinet sample;
[0122] Based on the third quantity, clustering the target box samples to obtain multiple cluster categories;
[0123] Based on clustering categories, the hidden danger box samples are divided into training sets.
[0124] In some implementations, the candidate box recognition model includes N tree model layers and an output layer, each tree model layer includes a plurality of different types of tree models, and the second acquisition module 830 is further used to:
[0125] Input the target feature variable into the first tree model layer of the candidate box recognition model, and each tree model of the first tree model layer predicts the target feature variable to obtain multiple first feature variables;
[0126] The multiple first feature variables are concatenated and input into the next tree model layer for prediction, until multiple N-th feature variables outputted by the N-th layer are obtained, where N is an integer greater than 1;
[0127] The Nth characteristic variable is weighted, and the confidence of the downstream optical network unit is output by the output layer.
[0128] In some implementations, the training module 840 is further configured to:
[0129] Determining a fourth number of downband optical network units having a confidence level greater than or equal to a preset confidence threshold;
[0130] Obtaining a ratio of the fourth quantity to the total quantity of downband optical network units;
[0131] Determine the predicted sample type of the light bin sample based on the quantity ratio;
[0132] Based on the predicted sample type and the reference sample type of the optical classification bin sample, the parameters of the candidate bin recognition model are reversely adjusted and iteratively trained.
[0133] In the disclosed embodiment, the attribute parameters are feature encoded and feature derived to obtain candidate feature variables of the second feature type. The candidate feature variables of the second feature type are feature screened based on a preset random forest model to obtain target feature variables of the target feature type. Feature types with higher importance can be screened out, and the influence of attribute parameters of different feature types on the prediction results can be comprehensively learned. The target feature variables are input into the preset candidate cabinet recognition model to obtain the confidence of multiple down-band optical network units included in the optical distribution cabinet sample. The candidate cabinet recognition model is trained based on the confidence of the down-band optical network unit to obtain the target cabinet recognition model. The target cabinet recognition model can be used to accurately judge the cabinet with hidden dangers, improve the efficiency of locating weak light problems, and provide data support for installation or line maintenance, thereby reducing the proportion of jump weak light.
[0134] Fig. 9 FIG. 1 is a structural block diagram of a hidden danger identification device for an optical sub-box according to an embodiment of the present disclosure. Fig. 9 As shown, the hidden danger identification device 900 of the optical sub-box includes:
[0135] A first acquisition module 910 is used to acquire attribute parameters of a target feature type of a light separation box to be identified;
[0136] A feature encoding module 920 is used to perform feature encoding on the attribute parameters of the target feature type to obtain a target feature variable of the target feature type;
[0137] The second acquisition module 930 is used to input the target feature variable into a preset candidate cabinet recognition model to obtain the confidence of multiple downlink optical network units included in the optical sub-cabinet to be identified;
[0138] A judgment module 940 is used to judge whether the optical sub-box to be identified is a hidden danger box based on the confidence of the downband optical network unit;
[0139] The target box recognition model is obtained by training with the training device of claim 8.
[0140] In the disclosed embodiment, the target characteristic variable is input into the target box identification model, the confidence of multiple down-band optical network units included in the optical sub-box to be identified is obtained, and whether the optical sub-box to be identified is a hidden danger box is judged based on the confidence of the down-band optical network unit. This can speed up the positioning of weak light problems and reduce the proportion of jumping weak light, reduce the average interruption time of a single user, the fault duration and the proportion of weak light ONU, and avoid wasting resources.
[0141] The acquisition, transmission, storage, use, and processing of data in the technical solution disclosed in this disclosure are in compliance with the relevant provisions of national laws and regulations.
[0142] It should be noted that in the embodiments of the present disclosure, certain software, components, models and other existing solutions in the industry may be mentioned, which should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.
[0143] Fig.10 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure.
[0144] like Fig.10 As shown, the electronic device 1000 includes:
[0145] The memory 1001 and the processor 1002, a bus 1003 connecting different components (including the memory 1001 and the processor 1002), the memory 1001 stores a computer program, and when the processor 1002 executes the program, the training method of the box recognition model or the hidden danger recognition method of the optical separation box of the embodiment of the present disclosure is implemented.
[0146] Bus 1003 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnect (PCI) bus.
[0147] The electronic device 1000 typically includes a variety of electronic device readable media. These media can be any available media that can be accessed by the electronic device 1000, including volatile and non-volatile media, removable and non-removable media.
[0148] The memory 1001 may also include computer system readable media in the form of volatile memory, such as random access memory (RAM) 1004 and / or cache memory 1005. The electronic device 1000 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 1006 may be used to read and write non-removable, non-volatile magnetic media ( Fig.10 not shown, usually called a "hard drive"). Although Fig.10 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 1003 via one or more data medium interfaces. The memory 1001 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present disclosure.
[0149] A program / utility 1008 having a set (at least one) of program modules 1007 may be stored, for example, in the memory 1001, such program modules 1007 including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 1007 generally perform the functions and / or methods of the embodiments described in the present disclosure.
[0150] The electronic device 1000 may also communicate with one or more external devices 1009 (e.g., keyboard, pointing device, display 1011, etc.), one or more devices that enable a user to interact with the electronic device 1000, and / or any device that enables the electronic device 1000 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed through an input / output (I / O) interface 1012. Furthermore, the electronic device 1000 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 1013. Fig.10 As shown, the network adapter 1013 communicates with other modules of the electronic device 1000 via the bus 1003. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0151] The processor 1002 executes various functional applications and data processing by running the programs stored in the memory 1001 .
[0152] It should be noted that the implementation process and technical principles of the electronic device of this embodiment refer to the aforementioned explanation of the training method of the cabinet recognition model of the embodiment of the present disclosure or the hidden danger recognition method of the optical separation cabinet, and will not be repeated here.
[0153] In order to implement the above embodiments, the present disclosure also proposes a computer-readable storage medium.
[0154] When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device can execute the training method of the box recognition model or the hidden danger recognition method of the optical box. Optionally, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0155] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0156] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A training method for a box recognition model, It is characterized in that include: Acquire multiple attribute parameters of the first feature type of the preset optical distribution box sample, and perform feature encoding and feature derivation on the attribute parameters to obtain candidate feature variables of the second feature type, wherein the attribute parameters of the first feature type include box attribute parameters, environment attribute parameters, attribute parameters of the downband optical network unit included in the optical distribution box, and attribute parameters of the uplink optical line terminal; Performing feature screening on the candidate feature variables of the second feature type based on a preset random forest model to obtain a target feature variable of a target feature type; Inputting the target feature variable into a preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box sample; The candidate cabinet recognition model is trained based on the confidence of the downband optical network unit until the training is completed to obtain a target cabinet recognition model.
2. The method according to claim 1, It is characterized in that The first feature type includes a specified third feature type and a fourth feature type, and the feature encoding and feature derivation of the attribute parameters to obtain candidate feature variables of the second feature type include: Based on the parameter type of the attribute parameter, feature encoding the attribute parameter to obtain an initial feature variable of the first feature type; The initial feature variables of the third feature type and the initial feature variables of the fourth feature type are subjected to second-order cross-feature derivation to obtain the candidate feature variables of the second feature type.
3. The method according to claim 1, It is characterized in that The feature screening of the candidate feature variables of the second feature type based on the preset random forest model to obtain the target feature variables of the target feature type includes: Inputting the candidate feature variables of the second feature type into the random forest model for pre-training to obtain an importance parameter of each second feature type; Based on the importance parameter of the second feature type, feature screening is performed on the second feature type and its candidate feature variables to obtain the target feature variable of the target feature type.
4. The method according to claim 1, It is characterized in that The reference sample types of the optical classification cabinet samples include hidden danger cabinet samples and non-hidden danger cabinet samples, and the method further includes: Acquire a first quantity of the potential danger box samples and a second quantity of the non-potential danger box samples; In response to an absolute value of a difference between the first quantity and the second quantity being greater than a preset quantity threshold, determining a larger number of target box samples between the hidden danger box samples and the non-hidden danger box samples; Obtaining a third number of downband optical network units included in the target cabinet sample; Based on the third quantity, clustering the target box samples to obtain multiple cluster categories; Based on the clustering categories, the hidden danger box samples are divided into training sets.
5. The method according to any one of claims 1 to 4, It is characterized in that The candidate box recognition model includes N tree model layers and an output layer, each tree model layer includes multiple tree models of different types, and the target feature variable is input into the preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box sample, including: Input the target feature variable into the first tree model layer of the candidate box recognition model, and each tree model of the first tree model layer predicts the target feature variable to obtain a plurality of first feature variables; splicing the plurality of the first feature variables and inputting the plurality of the first feature variables into the next tree model layer for prediction, until a plurality of N-th feature variables outputted by the N-th layer are obtained, where N is an integer greater than 1; The Nth characteristic variable is weighted, and the confidence of the downband optical network unit is output by the output layer.
6. The method according to any one of claims 1 to 4, It is characterized in that The training of the candidate cabinet recognition model based on the confidence of the downband optical network unit includes: Determine a fourth number of downband optical network units whose confidence is greater than or equal to a preset confidence threshold; Obtaining a ratio of the fourth quantity to the total quantity of the downband optical network units; Determining a predicted sample type of the light bin sample based on the quantity ratio; The parameters of the candidate box recognition model are reversely adjusted and iteratively trained based on the predicted sample type and the reference sample type of the optical classification box sample.
7. A method for identifying hidden dangers of optical distribution boxes, It is characterized in that include: Obtaining attribute parameters of the target feature type of the optical sub-box to be identified; Performing feature encoding on the attribute parameters of the target feature type to obtain a target feature variable of the target feature type; Inputting the target characteristic variable into the target box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box to be identified; Determining whether the optical sub-box to be identified is a hidden danger box based on the confidence of the lower-band optical network unit; The target box recognition model is trained by the training method described in any one of claims 1-6.
8. A training device for a box recognition model, It is characterized in that include: A first acquisition module is used to acquire multiple attribute parameters of a first feature type of a preset optical distribution box sample, and perform feature encoding and feature derivation on the attribute parameters to obtain candidate feature variables of a second feature type, wherein the attribute parameters of the first feature type include box attribute parameters, environment attribute parameters, attribute parameters of a downband optical network unit included in the optical distribution box, and attribute parameters of an uplink optical line terminal; A feature screening module, used for performing feature screening on the candidate feature variables of the second feature type based on a preset random forest model to obtain a target feature variable of a target feature type; A second acquisition module is used to input the target feature variable into a preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box sample; A training module is used to train the candidate box recognition model based on the confidence of the downband optical network unit until the training is completed to obtain a target box recognition model.
9. A device for identifying hidden dangers of optical distribution boxes, It is characterized in that include: A first acquisition module is used to acquire attribute parameters of a target feature type of a light separation box to be identified; A feature encoding module, used for performing feature encoding on the attribute parameters of the target feature type to obtain a target feature variable of the target feature type; A second acquisition module is used to input the target feature variable into a preset candidate box recognition model to obtain the confidence of multiple down-band optical network units included in the optical sub-box to be identified; A judgment module, used for judging whether the optical sub-box to be identified is a hidden danger box based on the confidence of the downband optical network unit; The target box recognition model is obtained by training the training device described in claim 8.
10. An electronic device, include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method of any one of claims 1 to 6, or execute the method of claim 7.
Citation Information
Cited By
Baseboard management controller platform identification method, product, equipment and medium
CN120561757A
A method, product, device, and medium for identifying a baseboard management controller platform.
CN120561757B
