An optical fiber jumper relationship intelligent identification method, program product, device and storage medium

Abstract

The application discloses a kind of optical fiber jumper relationship intelligent identification method, program product, equipment and storage medium, the method of the present application includes: collecting optical box-in splitter information, the physical location of each port and the information on label, for each splitter service class label, extract the splitter information in its label information, with pre-stored splitter information is matched, if consistent, then the optical box port corresponding to current label and the splitter matched form jumper relationship;For each customer service class label, calculate communication equipment name similarity, address similarity, semantic similarity, and form similar candidate pool with itself with several labels with high similarity to current label, and according to physical location is matched, the port corresponding to the label matched forms jumper relationship;For each route service class label, according to the route information in its label information is matched, and the two ports matched form jumper relationship.The present application has high degree of automation.

Description

Technical Field

[0001] This invention relates to optical fibers, and more particularly to a method, program product, device, and storage medium for intelligent identification of optical fiber patch relationships. Background Technology

[0002] In optical communication networks such as Fiber to the Home (FTTH), operators deploy a large number of optical distribution boxes (ODCs). These ODCs are outdoor port devices used to connect upstream and downstream optical cables. Specifically, ODCs have two types of ports: the first type connects to the upstream optical cable, and the second type connects to the downstream optical cable or a splitter. For example, a backbone ODC connects the backbone optical cable at one end and the distribution optical cable at the other, while a distribution ODC connects the distribution optical cable at one end and the drop cable at the other end via a splitter. Figure 1 As shown, the Type I and Type II ports in the optical distribution box may be ports located on the splice tray after splicing the backbone optical cable or distribution optical cable, or they may be specific ports of the optical splitter. When a patch cord is used to connect one end to the corresponding port of the backbone optical cable and the other end to the corresponding port of the distribution optical cable, the backbone optical cable and the distribution optical cable are connected. Similarly, a patch cord can be used to connect the distribution optical cable and the optical splitter. During network operation, maintenance personnel will use a patch cord to connect one end to the Type I port and the other end to the Type II port to perform fiber optic patching, thereby connecting the Type I and Type II ports and enabling the client to connect to the network. Usually, when connecting patch cords, maintenance personnel will attach a label to each end of the patch cord. The label may be marked with the customer's address, the name of the device to which the patch cord is connected (e.g., a certain optical splitter), or routing information (i.e., information about where it connects from), or several types of information. However, because the service labels are manually operated by maintenance personnel, there may be cases where the label information at the two ends of the same patch cord is different.

[0003] Based on the above, the existing technology has the following problems when organizing fiber optic patch connections:

[0004] 1. Disorganized physical wiring: The optical distribution box has a large number of ports. An optical distribution box with two 24-panel 12-port optical distribution boxes may have 12*24*2=576 ports. After multiple jumpers, expansions and rectifications, the jumpers in the optical distribution box are complicated and it is difficult to find out where the two ends of the jumpers are connected to. Manually reading and understanding a large number of text labels and matching them is very costly, inefficient and prone to errors.

[0005] 2. Unstructured business label information: The label information pasted on the optical distribution box is not uniform in format, not written in a standardized way, and the information is incomplete. It is a non-standard text type of label, which usually cannot be directly understood and used by traditional systems.

[0006] 3. Reliance on manual experience: The sorting and verification of fiber jumper relationships are highly dependent on the experience of maintenance personnel, which is inefficient, prone to errors, and cannot be automated.

[0007] Therefore, there is an urgent need for a technical solution that can automatically identify and analyze disordered physical ports and service information, and intelligently deduce the correct fiber optic patching relationships. Summary of the Invention

[0008] To address the problems existing in the prior art, the purpose of this invention is to provide an intelligent identification method, program product, device, and storage medium for fiber optic patching relationships that can intelligently identify fiber optic patching relationships.

[0009] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0010] A method for intelligent identification of fiber optic patch connection relationships includes:

[0011] Collect the physical location of each port in the target optical distribution box, the label information on the label affixed to the nearest end of the jumper connected to each port, and the information of the splitter and port in the target optical distribution box, and match the nearest port of each jumper with the label;

[0012] Using a pre-trained first semantic recognition model, each tag information is identified as belonging to the splitter service category, customer service category, or routing service category.

[0013] For each optical splitter service category tag, extract the optical splitter and port information from its tag information and match it with the optical splitter and port information of each optical splitter in the pre-stored target optical distribution box. If they match, then form a jumper relationship between the optical distribution box port corresponding to the current tag and the matched optical splitter port and store it in the first jumper relationship pool.

[0014] For each customer business category tag, a pre-trained communication device recognition model is used to identify communication device keywords in the tag information. Tags with the same communication device keywords are divided into a communication device candidate tag set, wherein the communication device keywords include the communication device name and adjacent sequence number.

[0015] For each tag in the candidate tag set of each communication device, calculate its semantic similarity with each other tag in the same set, and add several tags with similarity higher than the threshold to the current tag, along with the current tag itself and the current tag, to the similarity candidate pool of the current tag;

[0016] For each customer business category tag for which no communication device keywords are identified, its tag information is segmented and each segmented word is identified as belonging to the address category. Based on the segmentation category, the address similarity and semantic similarity between the current tag and other customer business category tags are calculated in turn. Several tags with similarity to the current tag higher than the threshold, as well as the current tag, are added to the similarity candidate pool of the current tag.

[0017] For all tags in each similar candidate pool, pairwise matching is performed based on the physical location of the port corresponding to the tag. The two ports bound to the matched tags form a jumper relationship and are stored in the second jumper relationship pool.

[0018] For each routing service class label, match the two ports in the routing information according to the routing information in the label information, form a jump fiber relationship between the two matched ports, and store them in the third jump fiber relationship pool.

[0019] The first, second, and third jumper relationship pools are merged and processed to form the final jumper relationship pool.

[0020] Furthermore, the step of sequentially calculating the address similarity and semantic similarity between the current tag and other customer business category tags according to the word segmentation category, and adding the current tag and several tags with similarity higher than a threshold to the current tag's similarity candidate pool, specifically includes:

[0021] For each customer business category tag for which no communication device keywords are identified, its address similarity with other tags is calculated based on its address-based and non-address-based word segmentation. Several tags with similarity higher than the threshold are added to the similarity candidate pool of the current tag along with the tag itself.

[0022] For each remaining customer business category tag, calculate its semantic similarity to other tags, and add several tags with similarity higher than the threshold, along with the tag itself, to the similarity candidate pool of the current tag.

[0023] Furthermore, for each customer service category tag for which no communication device keywords were identified, the address similarity between it and other tags is calculated based on its address-based and non-address-based word segmentation, specifically including:

[0024] For each customer business category tag that has not identified communication device keywords and has not been added to the similar candidate pool, obtain the original word segmentation set formed after word segmentation; extract each address category word and the non-address category words adjacent to the address category word from the original word segmentation set, and merge the word segments that belong to numbers or letters in the adjacent non-address category words with the corresponding address category words into a key address word to form the updated word segmentation set of customer business category tags;

[0025] All customer business category tags that did not identify communication device keywords and were not added to the similar candidate pool were paired up to form several pairs of customer business category tags.

[0026] The updated word segmentation sets of each pair of customer business category tags are merged and deduplicated to obtain the common word segmentation set of each pair of customer business category tags;

[0027] For each pair of customer business category tags, weights are assigned based on each word in the common word segmentation set, resulting in a word segmentation weight sequence for each customer business category tag. When assigning weights, the weights of key address words, address-only words, and non-address-only words decrease sequentially, and the weight of words that appear only in the common word segmentation set but not in the corresponding updated word segmentation set is 0.

[0028] For each pair of customer business category tags, the initial address similarity is calculated based on the word segmentation weight sequence, wherein the initial address similarity is the quotient of the dot product of the word segmentation weight sequences of the two corresponding customer business category tags and the sum of their moduli.

[0029] Determine whether the address keywords of each pair of customer business category tags are consistent. If they are inconsistent, multiply the initial address similarity by a preset penalty factor to obtain the final address similarity.

[0030] Furthermore, the semantic similarity is obtained through a pre-trained second semantic recognition model, which is used to identify the semantic similarity of the label information of any two labels.

[0031] Furthermore, for all tags in each similar candidate pool, pairwise matching is performed based on the physical location of the port corresponding to the tag. The two ports bound to the matched tags form a jumper relationship and are stored in the second jumper relationship pool. Specifically, this includes:

[0032] For each similar candidate pool, if the number of tags in it is odd, then delete the tag with the lowest similarity.

[0033] Extract the physical location of the port corresponding to each remaining tag;

[0034] Based on the common connection method of patch cord and port physical location, identify two ports connected by the same patch cord, form a patch cord relationship between these two ports, and store it in the second patch cord relationship pool.

[0035] Furthermore, after collecting tag information and spectrometer information, the information is preprocessed to remove meaningless text information.

[0036] Furthermore, the merging and processing of the first, second, and third jumper relationship pools to form the final jumper relationship pool specifically includes:

[0037] The first, second, and third fiber optic jumper relationship pools are merged, and conflict verification and rationality filtering are performed to form the final fiber optic jumper relationship pool.

[0038] A computer program product includes a computer program that, when executed by a processor, implements the above-described method.

[0039] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described above.

[0040] A computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the above-described method.

[0041] Compared with the prior art, the beneficial effects of this invention are:

[0042] 1. High efficiency and accuracy: This invention is an intelligent machine automatic identification, which significantly improves the accuracy and efficiency of identifying the correct fiber jumper relationship from chaotic information compared with manual identification, and reduces labor costs.

[0043] 2. High level of intelligence: It uses a pre-trained communication device recognition model and semantic similarity to identify the similarity of tags containing communication devices. It uses word segmentation, part-of-speech recognition, and pre-trained semantic model to identify the address similarity of tags, thereby understanding unstructured and diverse tag text. This breaks through the limitations of traditional rule-based matching and further improves the accuracy and efficiency of recognition.

[0044] 3. Automated closed loop: From information collection, analysis, and matching to final execution (automatic fiber jump or work order generation), a complete automated operation and maintenance closed loop is formed, which greatly improves the speed and quality of operation and maintenance response.

[0045] 4. High robustness: Multiple strategies complement each other, so even if one type of information is missing or incorrect, other strategies may still yield the correct result, improving the system's fault tolerance and universality;

[0046] 5. When calculating address similarity, different weights are assigned to different word segments. The higher the importance, the greater the weight. The address similarity is then calculated in combination with the weights, thereby improving the accuracy of address similarity calculation. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the optical distribution box;

[0048] Figure 2 This is a flowchart illustrating the intelligent identification method for fiber optic patching relationships provided in an embodiment of the present invention.

[0049] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0050] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0051] Example 1

[0052] This invention provides an intelligent identification method for fiber optic patch connection relationships, such as... Figure 1 As shown, it includes the following steps:

[0053] S101. Collect the physical location of each port in the target optical distribution box and the tag information on the nearest end of the jumper connected to each port, as well as the splitter and port information of the target optical distribution box, and bind the nearest port of each jumper to the tag.

[0054] The ports within the target optical distribution box include the ports of the optical splitters and the ports on the fusion splicers. For example, the third input port on the second optical splitter within the distribution box can be physically defined as POS02-IN3, where POS is an abbreviation for Passive Optical Splitter. The sixth port on the second fusion splice tray in the second fusion frame from top to bottom on either side of the distribution box (distribution boxes typically have multiple sides) can be physically defined as 2-2-2-6. For a two-sided distribution box with 10 fusion frames per side, 6 fusion splices per frame, and 12 ports per fusion splice tray, the first port would be 1-1-1-1, and the last port would be 2-10-6-12.

[0055] Because there are so many and they are scattered around, for ease of maintenance, theoretically, maintenance personnel would affix a label to each jumper near a port when inserting it, indicating that the two ports are connected by the same jumper and have a jumper relationship. Understandably, some labels might be missed. Therefore, when collecting labels, the port on the jumper closest to the label will be matched with that label.

[0056] The tag information and the splitter and port information are entered through OCR recognition or manual input. The tag information is the text information on the tag, which is generally the tag's corresponding information. The splitter and port information specifically includes the splitter's name, unique identifier (for example, POS2 can be the unique identifier of the second splitter), and the port number on each splitter. For example, for a splitter with 1 input port and 4 output ports, they are numbered POS01-IN1, POS01-OUT1~POS01-OUT4, respectively.

[0057] In other embodiments, after collecting the label information and splitter information, preprocessing of the information can also be performed to eliminate meaningless text information.

[0058] S102. Use a pre-trained first semantic recognition model to identify whether each label information belongs to the splitter service category, customer service category, or routing service category.

[0059] Among them, the first semantic recognition model can be a pre-trained language model based on the Transformer architecture. Specifically, the hfl / chinese-roberta-wwm-ext pre-trained model released by HIT can be used as the base model. This model has been pre-trained on the Chinese Whole Word Masking task and has excellent understanding capabilities for Chinese vocabulary, entities, and context relationships. In the data preparation stage, the model can be lightly fine-tuned according to the labeled label information samples (splitter service category labels, customer service category labels, routing service category labels) so that it can more accurately classify the label information into the aforementioned three major categories: Splitter service category labels: usually include the name, unique identification number, port information, etc. of the splitter connected to the port. For example, a certain splitter service category label is "Dadao Wansheng Street China Bai塘 Park HWOLTO1-3-5 China Xinghu Street Modern Avenue GJ9210POS03-OUT1", which means that the port corresponding to this label is connected to the OUT 1 port of the 3rd splitter; Customer service category labels: usually include customer address, customer name, device name, etc. For example, a certain customer service category label information can be "Jiangsu Wenchao Construction Engineering Co., Ltd. Xiangcheng RT-Mart". If the label information on another port is also this, it means that these 2 ports are connected; Business routing information category labels: usually include optical path directions, upstream / downstream location information, etc. For example, a certain routing information category label information can be "FR:1-1-2-7 / 8:TO:1-1-19-5 / 6 Data Center Bare Fiber Suzhou Bank Co., Ltd.-Guoke". From this routing information, the meanings of from and to can be seen. The above represents that 1 side 1 frame 2 disks 7 ports and 1 side 1 frame 2 disks 8 ports are respectively connected to 1 side 1 frame 19 disks 5 ports and 1 side 1 frame 19 disks 6 ports. It can be understood that the first semantic recognition model can also be any other machine model that can implement label information classification.

[0060] S103. For each splitter service category label, extract the splitter and port information in its label information, and match it with the splitter and port information of each splitter in the pre-stored target optical cross-connect box. If they are consistent, form a fiber jump relationship between the optical cross-connect box port corresponding to the current label and the matched splitter port, and store it in the first fiber jump relationship pool.

[0061] Among them, the splitter information of each splitter in the pre-stored target optical cross-connect box can be a splitter information table, which stores the splitter information of each splitter in the target optical cross-connect box, that is, information such as the splitter name, unique identification number, and ports.

[0062] During matching, first poll all splitter service class tags and use keywords for regular matching: look for keywords such as "POS" and "splitter" in the tags, and extract the following numbers after finding them, which represent the splitter serial number; then look for keywords such as "OUT", "out", "IN", and "in", which represent the port type; after finding the port type, extract the following numbers, which represent the port serial number. Finally, extract the splitter serial number, splitter port type, and splitter port serial number in the tag information. For example, for the splitter service class tag "Dadao Wansheng Street China White Pond Park HWOLTO1-3-5 China Xinghu Street Modern Avenue GJ9210POS03-OUT1", the identified splitter and port information is "POS03-OUT1", that is, port 2 of the OUT type of splitter No. 3. After extraction, match with the pre-stored splitter information table. If there is port 2 of the OUT type of splitter No. 3 in the splitter information table, then form a fiber jump relationship between the port corresponding to the tag and the splitter port, and store it in the first fiber jump relationship pool.

[0063] S104. For each customer service class tag, use a pre-trained communication device recognition model to recognize the communication device keywords in the tag information, and divide the tags with the same communication device keywords into a communication device candidate tag set respectively.

[0064] The communication device keywords specifically include the communication device name and adjacent serial numbers (e.g., numbers or letters). The communication device identification model can be a pre-trained language model based on Transformer. Specifically, the bert-base-Chinese pre-trained model released by Google can be used as the base model, and then fine-tuned to obtain the communication device identification model. This model has excellent performance in identifying specific targets in text. During the data preparation stage, Label Studio can be used to label the large amount of prepared data with device information, mainly labeling communication devices and their serial numbers. After preparing the data, the BERT-base-chinese basic model is used for training to obtain the final fine-tuned model, referred to here as the "communication device identification model." This model can accurately label the device name and device serial number in the business information string. For example, a business category tag containing communication devices is "Donghu Dajun Phase 3 Lighting Box 1 Suzhou Avenue East 5G". The model can label it with: {'entity': 'B-DEVICE', 'score': 0.99, 'word': 'lighting box'} and {'entity':'B-NUM', 'score': 0.99, 'word': '1'}. This identifies the communication device name as "lighting box," the device serial number as 1, and the communication device keyword as "lighting box 1." It's understandable that the communication device identification model can also be any other machine model capable of identifying communication device keywords.

[0065] For example, by identifying model annotations through communication devices, the following labels can be processed as follows:

[0066] The label information for port 5 (1-2-10-5) of 1 side, 2 frames, 10 panels is: Donghu Dajun Phase 3 Lighting Box 1 Suzhou Avenue East 5G; the model annotation results are: {'entity': 'B-DEVICE', 'score': 0.99, 'word': 'lighting box'} and {'entity': 'B-NUM', 'score': 0.99, 'word': '1'}; the identified communication device keyword is Lighting Box 1;

[0067] The label information for port 6 (1-2-11-6) of 1 side, 2 frames, 11 trays is: Donghu Dajun Phase 3 Lighting Box 2 Suzhou Avenue East 5G; The model annotation results are: {'entity': 'B-DEVICE', 'score': 0.99, 'word': 'lighting box'} and {'entity': 'B-NUM', 'score': 0.99, 'word': '2'}, and the identified communication device keyword is Lighting Box 2;

[0068] The label information for port 11 (2-5-16-11) on 2 sides, 5 frames, and 16 panels is: Donghu Dajun Phase 3 Lighting Box 1 5G; the model annotation results are: {'entity': 'B-DEVICE', 'score': 0.99, 'word': 'lighting box'} and {'entity': 'B-NUM', 'score': 0.99, 'word': '1'}, and the identified communication device keyword is Lighting Box 1;

[0069] The label information for port 12 (2-5-16-12) on the 2-sided, 5-frame, 16-disk system is: Donghu Dajun Phase 3 Lighting Box 2 5G; the model annotation results are: {'entity': 'B-DEVICE', 'score': 0.99, 'word': 'lighting box'} and {'entity': 'B-NUM', 'score': 0.99, 'word': '2'}, and the identified communication device keyword is "lighting box 2".

[0070] The communication equipment keywords for 1-2-10-5 and 2-5-16-11 are both "light-collecting box 1", so they are stored in the communication equipment candidate tag set 1. Similarly, the communication equipment keywords for 1-2-11-6 and 2-5-16-12 are both "light-collecting box 2", so they are stored in the communication equipment candidate tag set 2.

[0071] S105. For each tag in the candidate tag set of each communication device, calculate its semantic similarity with each other tag in the same set, and add several tags with similarity higher than the threshold to the current tag, the current tag itself, and the current tag to the similarity candidate pool of the current tag.

[0072] For candidate tag set 1 of communication devices, use hanlp to calculate the semantic similarity between 1-2-10-5 and 2-5-16-11. If the semantic similarity between the two exceeds the threshold, then 1-2-10-5 and 2-5-16-11 are stored in the similarity candidate pool of 1-2-10-5. Similarly, if the semantic similarity between 1-2-11-6 and 2-5-16-12 is higher than the threshold, then 1-2-11-6 and 2-5-16-12 are stored in the similarity candidate pool of 1-2-11-6.

[0073] It should be noted that if the keywords of the communication device are not identified and only the similarity is calculated, then 1-2-10-5 and 1-2-11-6 differ by only one number, and the similarity is much higher than the threshold. They will be wrongly added to the similarity candidate pool. Therefore, communication device identification is a necessary process.

[0074] S106. For each customer service class label for which the communication device keyword is not recognized, tokenize its label information and identify whether each token belongs to the address category. Calculate the address similarity and semantic similarity between the current label and other customer service class labels in sequence according to the token category, and add several labels with similarity higher than the threshold to the current label and the current label to the similarity candidate pool of the current label.

[0075] Among them, the label information may be a relatively long sentence or contain multiple words. Therefore, first, perform preprocessing, remove symbols, and convert all uppercase letters to lowercase to reduce the later matching workload. Second, perform tokenization. Use the HanLP tool to divide the label information into multiple words. For example, for "pointing to Room 101, Building A, Happy Community", it can be tokenized into the token set {"pointing to", "Happy Community", "A", "Building", "101", "Room"}.

[0076] After that, identify whether each token is an address token or a non-address name. A zero-shot classification method based on PaddleNLP can be used. The token categories include address category and non-address category. Address tokens can be "Happy Community", "Building", etc., and non-address tokens can be "pointing to", "A", "101", etc., including numbers, letters, and meaningless words.

[0077] For each customer service class label for which the communication device keyword is not recognized, calculate its address similarity with other labels according to its address tokens and non-address tokens, and add several labels with similarity higher than the threshold and itself to the similarity candidate pool of the current label.

[0078] Specifically, the steps for calculating address similarity include: for each customer business category tag for which no communication device keywords were identified and which was not added to the similarity candidate pool, obtaining the original word segmentation set after word segmentation; extracting each address-type word and adjacent non-address-type words from the original word segmentation set, and merging the adjacent non-address-type words that belong to numbers or letters with the corresponding address-type words into a key address word, forming an updated word segmentation set for the customer business category tag. For example, for the word segmentation set { The keywords "pointing to", "Happy Community", "A", "Building", "101", and "Room" are used to define the key address words "Building A" and "Room 101". All customer service tags that did not identify communication device keywords and were not added to the similarity candidate pool are paired to form several pairs of customer service tags. The updated word sets of each pair of customer service tags are merged and deduplicated to obtain a common word set for each pair. For each pair of customer service tags, weights are assigned based on each word in the common word set to obtain the weights for each customer service tag. The word segmentation weight sequence for customer business category tags is as follows: when assigning weights, the weights of key address segmentation, address-only segmentation, and non-address-only segmentation decrease sequentially, with a weight of 0 for segmentation that appears only in the common segmentation set but not in the corresponding updated segmentation set. For each pair of customer business category tags, an initial address similarity is calculated based on the word segmentation weight sequence, where the initial address similarity is the quotient of the dot product of the word segmentation weight sequences of the two corresponding customer business category tags and the sum of their moduli. It is then determined whether the address keywords of each pair of customer business category tags are consistent. If they are inconsistent, the initial address similarity is multiplied by a preset penalty factor to obtain the final address similarity. The specific formula for calculating the initial address similarity is as follows:

[0079]

[0080] in, The word segmentation weight sequence is represented as A = {A} i} and B={B i The initial address similarity of two customer business class tags, A i B i These represent the weights of the i-th word in the word segmentation weight sequences A and B, respectively.

[0081] For example, suppose the original word set in a pair of tags is as follows:

[0082] The original word segmentation set for customer business category tag 1 is: {“pointing to”, “Happy Community”, “a”, “Building”, “101”, “Room”}; among them, “Happy Community”, “Building”, and “Room” are address-related words, while “pointing to”, “a”, and “101” are non-address-related words.

[0083] The original word segmentation set of customer business label 2 is {"point to", "Happy Community", "b", "building", "101", "room"}. Among them, "Happy Community", "building", and "room" are address-related word segmentations, and "point to", "b", and "101" are non-address-related word segmentations.

[0084] In customer business label 1, the letter adjacent to the address-related word segmentation "building" is "a", and the number adjacent to the address-related word segmentation "room" is 101. So the address keywords are "a building" and "101 room", and the updated word segmentation set is {"point to", "Happy Community", "a building", "101 room"}.

[0085] In customer business label 2, the letter adjacent to the address-related word segmentation "building" is "b", and the number adjacent to the address-related word segmentation "room" is 101. So the address keywords are "b building" and "101 room", and the updated word segmentation set is {"point to", "Happy Community", "b building", "101 room"}.

[0086] Merge the updated word segmentation sets of the two labels to get the common word segmentation set {"point to", "Happy Community", "a building", "b building", "101 room"}.

[0087] Set the weights of key address word segmentations, only address-related word segmentations, and only non-address-related word segmentations to 6, 3, and 1 respectively, and the weight of the word segmentation that only appears in the common word segmentation set but not in the corresponding updated word segmentation set is 0.

[0088] For customer business label 1, the types of the word segmentations "point to", "Happy Community", "a building", "b building", and "101 room" in the common word segmentation set are only non-address-related word segmentation, only address-related word segmentation, key address word segmentation, the word segmentation that only appears in the common word segmentation set but not in the updated word segmentation set of this label, and key address word segmentation respectively. Therefore, the weights are assigned as 1, 3, 6, 0, and 6 respectively, and the word segmentation weight sequence A = {1, 3, 6, 0, 6}.

[0089] For customer business label 2, the types of the word segmentations "point to", "Happy Community", "a building", "b building", and "101 room" in the common word segmentation set are only non-address-related word segmentation, only address-related word segmentation, the word segmentation that only appears in the common word segmentation set but not in the corresponding updated word segmentation set, key address word segmentation, and key address word segmentation respectively. Therefore, the weights are assigned as 1, 3, 0, 6, and 6 respectively, and the word segmentation weight sequence B = {1, 3, 0, 6, 6}.

[0090] Calculate the initial address similarity according to the word segmentation weight sequence. The specific calculation formula is:

[0091]

[0092] Then, address keyword penalty is applied: First, the number of mismatches is calculated. Building A and Building B do not match, so the number of mismatches is n=1. Then, a preset penalty factor q is used, which is assumed to be q=0.75. The final address similarity is: = 0.56 * 0.75 ≈ 0.42. This can be considered highly dissimilar.

[0093] Finally, several tags with address similarity higher than a threshold (e.g., 0.8) are added to the similarity candidate pool of the current tag along with themselves.

[0094] For each customer business category tag that has not yet been added to the similarity candidate pool, its semantic similarity with other tags is calculated, and several tags with similarity higher than a threshold are added to the similarity candidate pool of the current tag along with the tag itself. The semantic similarity is obtained through a pre-trained second semantic recognition model, which is used to identify the semantic similarity of tag information between any two tags.

[0095] The aforementioned pre-trained second semantic recognition model can be the paraphrase-multilingual-MiniLM-L12-v2 pre-trained model developed by the sentence-transformers team based on the MiniLM architecture. It is a lightweight, multilingual pre-trained language model that demonstrates excellent performance in calculating the similarity between two sentences based on semantic vectors. Understandably, the second semantic recognition model can also be any other machine model capable of achieving semantic similarity between two texts.

[0096] S107. For all tags in each similar candidate pool, perform pairwise matching based on the physical location of the port corresponding to the tag, form a jumper relationship between the two ports bound to the matched tags, and store it in the second jumper relationship pool.

[0097] This step specifically includes: for each similar candidate pool, if the number of tags in it is odd, delete the tag with the lowest similarity; extract the physical location of the port corresponding to each remaining tag; according to the general connection method between jumpers and port physical locations, identify two ports connected by the same jumper, form a jumper relationship between these two ports, and store them in the second jumper relationship pool.

[0098] For example, for a similar candidate pool S1={b1, b3, b4, b5}, the physical locations of the ports bound to b1, b3, b4, and b5 are 1-1-1-1, 1-1-1-5, 1-2-2-2, and 1-2-2-8, respectively. In a general connection method, ports on the same fusion splice cannot be jump-connected; that is, 1-1-1-1 and 1-1-1-5 cannot be jump-connected, and 1-2-2-2 and 1-2-2-8 cannot be jump-connected. Moreover, in a general connection method, maintenance personnel typically jump-connect in sequential order, i.e., 1- ... 1 is placed before 1-1-5, and 1-2-2-2 is placed before 1-2-2-8. Typically, the port in front is connected to the port in front, and the port behind is connected to the port behind. That is, 1-1-1 is connected to 1-2-2-2, and 1-1-1-5 is connected to 1-2-2-8, thus establishing a jumper relationship between 1-1-1-1 and 1-2-2-2, and between 1-1-1-5 and 1-2-2-8. These relationships are then stored in the second jumper relationship pool. However, it's possible that maintenance personnel might not jump in the correct order, such as 1-1-1-1 being connected to 1-2-2-8 and 1-1-1-5 being connected to 1-2-2-2. In this case, misidentification of jumper relationships may occur. Therefore, the method of this invention cannot guarantee 100% accuracy, but it can maximize automatic identification and minimize the workload of personnel. Manual verification is still required later.

[0099] S108. For each routing service class label, match the two ports in the routing information according to the routing information in the label information, form a jump fiber relationship between the two matched ports, and store them in the third jump fiber relationship pool.

[0100] For example, given the tag "Wangzhou City, Sancheng District, Beiqiao Middle School 23-21726-Beiqiao-34-5FR:1-1-18-7 / 8 TO:1-1-21-8 / 9", semantic analysis using a semantic model extracts "FR:1-1-1-18-7 / 8". Combine "TO:1-1-21-8 / 9" with "fr" and "to" or other "pointing words", and extract the following "number + '-' + number + '-'..." combination. By analyzing this combination, we can obtain the port position that the combination points to. For example, 1-1-18-7 / 8 refers to ports 7 and 8 of panel 1, frame 1, and 18 disks. 1-1-21-8 / 9 refers to ports 8 and 9 of panel 1, frame 21. Establish a jumper relationship between port 1-1-18-7 and port 1-1-21-8, and establish a jumper relationship between 1-1-18-8 and port 1-1-21-9. Store these relationships in the third jumper relationship pool.

[0101] S109. Merge and process the first, second, and third jumper relationship pools to form the final jumper relationship pool.

[0102] Specifically, the first, second, and third patch cord relationship pools are merged, and conflict verification and rationality filtering are performed to form the final patch cord relationship pool. Rationality filtering includes filtering out patch cord pairs with ports located on the same splice tray, and filtering out patch cord pairs with ports at both ends belonging to the same optical cable (which usually do not require patching). The verified final patch cord relationship pool is then output to the network management system, or directly drives an automatic patching robot / equipment to complete the automatic connection of ports, thereby achieving intelligent and automated management of fiber optic patching.

[0103] Example 2

[0104] This invention also provides a computer program product, such as an app on a mobile phone or tablet, or an installer on a computer. This product includes a computer program / instructions that, when executed by a processor, implement the method described in Embodiment 1. The code for the computer-executable program used to perform the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0105] Example 3

[0106] Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. The embodiments of the present invention provide services for the implementation of the method described in Embodiment 1 of the present invention. Figure 2 As shown, the device may include: a memory 301 storing a computer-executable program; a processor 302 coupled to the memory 301; the processor 302 calls the computer-executable program stored in the memory 301 to perform the steps in the method described in Embodiment 1.

[0107] Memory 301 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, memory 301 may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as a "hard disk drive"). A program / utility having a set (at least one) of program modules may be stored, for example, in memory 301. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The computer-executable program of the program modules typically performs the functions and / or methods described in the embodiments of the present invention.

[0108] The processor 302 executes various functional applications and data processing by running programs stored in the memory 301, such as implementing the method provided in Embodiment 1 of the present invention.

[0109] The code of a computer executable program can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages.

[0110] Example 4

[0111] This invention provides a storage medium containing a computer-executable program, which, when executed by a computer processor, is used to perform the method of Embodiment 1.

[0112] The storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0113] Of course, the computer-executable program provided in the embodiments of the present invention is not limited to the above-described method operations, but can also perform related operations in the methods provided in any embodiment of the present invention.

[0114] It should be understood that the embodiments and descriptions above are only the principles, main features and advantages of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the invention, and all such changes and modifications fall within the protection scope of the present invention.

Claims

1. A method for intelligent identification of fiber optic patch connection relationships, characterized in that, include: Collect the physical location of each port in the target optical distribution box, the label information on the label affixed to the nearest end of the jumper connected to each port, and the information of the splitter and port in the target optical distribution box, and match the nearest port of each jumper with the label; Using a pre-trained first semantic recognition model, each tag information is identified as belonging to the splitter service category, customer service category, or routing service category. For each optical splitter service category tag, extract the optical splitter and port information from its tag information and match it with the optical splitter and port information of each optical splitter in the pre-stored target optical distribution box. If they match, then form a jumper relationship between the optical distribution box port corresponding to the current tag and the matched optical splitter port and store it in the first jumper relationship pool. For each customer business category tag, a pre-trained communication device recognition model is used to identify communication device keywords in the tag information. Tags with the same communication device keywords are divided into a communication device candidate tag set, wherein the communication device keywords include the communication device name and adjacent sequence number. For each tag in the candidate tag set of each communication device, calculate its semantic similarity with each other tag in the same set, and add several tags with similarity higher than the threshold to the current tag, along with the current tag itself and the current tag, to the similarity candidate pool of the current tag; For each customer business category tag for which no communication device keywords are identified, its tag information is segmented and each segmented word is identified as belonging to the address category. Based on the segmentation category, the address similarity and semantic similarity between the current tag and other customer business category tags are calculated in turn. Several tags with similarity to the current tag higher than the threshold, as well as the current tag, are added to the similarity candidate pool of the current tag. For all tags in each similar candidate pool, pairwise matching is performed based on the physical location of the port corresponding to the tag. The two ports bound to the matched tags form a jumper relationship and are stored in the second jumper relationship pool. For each routing service class label, match the two ports in the routing information according to the routing information in the label information, form a jump fiber relationship between the two matched ports, and store them in the third jump fiber relationship pool. The first, second, and third jumper relationship pools are merged and processed to form the final jumper relationship pool.

2. The intelligent identification method for fiber optic patch connection relationships according to claim 1, characterized in that, The step of calculating the address similarity and semantic similarity between the current tag and other customer business category tags according to the word segmentation category, and adding the current tag and several tags with similarity higher than the threshold to the similarity candidate pool of the current tag, specifically includes: For each customer business category tag for which no communication device keywords are identified, its address similarity with other tags is calculated based on its address-based and non-address-based word segmentation. Several tags with similarity higher than the threshold are added to the similarity candidate pool of the current tag along with the tag itself. For each remaining customer business category tag, calculate its semantic similarity to other tags, and add several tags with similarity higher than the threshold, along with the tag itself, to the similarity candidate pool of the current tag.

3. The intelligent identification method for fiber optic patch connection relationships according to claim 2, characterized in that, For each customer service category tag for which no communication device keywords were identified, its address similarity to other tags is calculated based on its address-based and non-address-based word segmentation, specifically including: For each customer business category tag that has not identified communication device keywords and has not been added to the similar candidate pool, obtain the original word segmentation set formed after word segmentation; extract each address category word and the non-address category words adjacent to the address category word from the original word segmentation set, and merge the word segments that belong to numbers or letters in the adjacent non-address category words with the corresponding address category words into a key address word to form the updated word segmentation set of customer business category tags; All customer business category tags that did not identify communication device keywords and were not added to the similar candidate pool were paired up to form several pairs of customer business category tags. The updated word segmentation sets of each pair of customer business category tags are merged and deduplicated to obtain the common word segmentation set of each pair of customer business category tags; For each pair of customer business category tags, weights are assigned based on each word in the common word segmentation set, resulting in a word segmentation weight sequence for each customer business category tag. When assigning weights, the weights of key address words, address-only words, and non-address-only words decrease sequentially, and the weight of words that appear only in the common word segmentation set but not in the corresponding updated word segmentation set is 0. For each pair of customer business category tags, the initial address similarity is calculated based on the word segmentation weight sequence, wherein the initial address similarity is the quotient of the dot product of the word segmentation weight sequences of the two corresponding customer business category tags and the sum of their moduli. Determine whether the address keywords of each pair of customer business category tags are consistent. If they are inconsistent, multiply the initial address similarity by a preset penalty factor to obtain the final address similarity.

4. The intelligent identification method for fiber optic patch connection relationships according to claim 2, characterized in that, The semantic similarity is obtained through a pre-trained second semantic recognition model, which is used to identify the semantic similarity of the label information of any two labels.

5. The intelligent identification method for fiber optic patch connection relationships according to claim 1, characterized in that, The process of matching all tags in each similar candidate pool based on the physical location of the port corresponding to the tag, forming a jumper relationship between the two ports bound to the matched tags, and storing it in the second jumper relationship pool, specifically includes: For each similar candidate pool, if the number of tags in it is odd, then delete the tag with the lowest similarity. Extract the physical location of the port corresponding to each remaining tag; Based on the common connection method of patch cord and port physical location, identify two ports connected by the same patch cord, form a patch cord relationship between these two ports, and store it in the second patch cord relationship pool.

6. The intelligent identification method for fiber optic patch connection relationships according to claim 1, characterized in that, After collecting tag information and spectrometer information, the information is preprocessed to remove meaningless text information.

7. The intelligent identification method for fiber optic patch connection relationships according to claim 1, characterized in that, The process of merging and processing the first, second, and third jumper relationship pools to form the final jumper relationship pool specifically includes: The first, second, and third fiber optic jumper relationship pools are merged, and conflict verification and rationality filtering are performed to form the final fiber optic jumper relationship pool.

8. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, it implements the method of any one of claims 1-7.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that: The computer program / instructions, when executed by a processor, implement the method of any one of claims 1-7.

Images