Intelligent operation and maintenance method for radio and television trunk optical cable transmission system

Abstract

The present invention discloses an intelligent operation and maintenance method for a radio and television trunk optical cable transmission system, comprising the following steps: S1, construction of a knowledge graph for optical cable transmission system fault handling, by analyzing the text content of the optical cable transmission system fault handling strategy, designing the pattern layer of the knowledge graph in a top-down manner, and then constructing the data layer in a bottom-up manner; S2, decision-making on the operation and maintenance plan of the optical cable transmission system based on the knowledge graph, first mapping the fault description to the computer room operation and maintenance knowledge graph to form a fault description subgraph; then, further mapping the user's operation and maintenance requirements to the knowledge graph; finally, performing a graph search in the knowledge graph to match the user's requirements with the fault description to obtain an operation and maintenance plan; S3, optical cable transmission system fault handling based on deep learning, based on the big data of computer room operation and maintenance, designing a method for computer room fault feature modeling and fault handling strategy generation based on deep learning.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent control and decision-making, and specifically relates to an intelligent operation and maintenance method for a radio and television trunk optical cable transmission system. Background Art

[0002] The technical architecture of Zhejiang Radio and Television’s backbone network is complex. The province’s 1,800-kilometer links are divided into sections with different structures. The transmission equipment system is complex, and various integrated network management systems generate massive amounts of data every day, making information management for operation, maintenance, and operation difficult. In addition, due to industry and regional characteristics, routine maintenance, cutover, relocation, and even interruption of the backbone optical cable transmission network in Zhejiang Province are very frequent, causing the geographical characteristics and security of each section of the network to change periodically, resulting in a large amount of valuable data, but lacking scientific and effective means of collection, analysis, and utilization. With the rapid development and promotion of media integration by the Provincial Radio and Television Group, the service guarantee objects of the provincial backbone network have been extended to more county-level units in the province, with higher requirements for network maintenance, emergency command, and rapid response, and the contradiction of lagging behind in data-based operation and maintenance has become increasingly prominent.

[0003] In recent years, Zhejiang Guanglian Cable TV Transmission Center has successively completed the expansion and transformation of the provincial backbone network, the transformation and relocation of the computer room system, and the construction of provincial network radio and television security transmission projects. In order to better respond to the needs of media integration development, safe broadcasting, high-bandwidth and high-quality transmission of the backbone network under the background of smart radio and television construction, it is necessary to intelligently analyze and handle faults including the provincial backbone network transmission system. At present, the faults of optical cable transmission are collected and exported by the software system, but there is a lack of effective analysis and processing. It mainly relies on the subjective experience judgment of the on-duty technicians to locate and analyze the faults, and further carry out fault handling. This handling method requires relevant staff to repeatedly review and memorize a large amount of fault handling information in structured and unstructured forms, such as system stability requirements, fault operation characteristics, and fault handling points. Although the current optical cable transmission system guarantees good performance and effects, the manual method is prone to negligence and omissions, which reduces the efficiency of fault calculation. With the development of the province's optical cable line system, the line structure is more complex, the data is more huge, and the difficulty of fault analysis and processing is constantly increasing. When an emergency occurs, such as data interruption, signal instability, line failure, etc., it is difficult to quickly detect, analyze and handle it manually, which can easily cause major problems such as broadcast accidents and information security. Optical cable transmission systems require intelligent technology to achieve efficient operation and maintenance, organize information such as fault handling methods and system alarms into a structured and visual form, so as to help technicians quickly analyze the cause of the fault when a system failure occurs, fully grasp the key information of fault handling, and make auxiliary decisions to improve the emergency handling capabilities of trunk optical cable transmission systems.

[0004] In recent years, artificial intelligence technology has begun to be applied to fields such as industrial automatic control and decision-making, but few studies have focused on improving the intelligent level of optical cable transmission system operation and maintenance. Aiming at the specific characteristics of radio and television trunk optical cable transmission system operation and maintenance, this paper proposes a method based on knowledge graph and deep learning, and based on the big data of optical cable transmission, processing experience and other data, realizes intelligent detection, analysis and decision-making of optical cable faults. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an intelligent operation and maintenance method for a radio and television trunk optical cable transmission system. The method uses a knowledge graph and a deep learning method to realize intelligent detection, analysis and decision-making processing of optical cable faults based on big data, processing experience and other data of optical cable transmission.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] An intelligent operation and maintenance method for a radio and television trunk optical cable transmission system comprises the following steps:

[0008] S1, construction of knowledge graph for optical cable transmission system fault handling. By analyzing the text content of optical cable transmission system fault handling strategy, the model layer of the knowledge graph is designed in a top-down manner, and then the data layer is constructed in a bottom-up manner;

[0009] S2, the decision of the operation and maintenance plan of the optical cable transmission system based on the knowledge graph, firstly maps the fault description to the operation and maintenance knowledge graph of the computer room to form a fault description subgraph; then, further maps the user's operation and maintenance requirements to the knowledge graph; finally, a graph search is performed in the knowledge graph to match the user's requirements and the fault description to obtain the operation and maintenance plan;

[0010] S3, Fault handling of optical cable transmission system based on deep learning, based on the big data of computer room operation and maintenance, a method for modeling computer room fault features and generating fault handling strategies based on deep learning is designed, which specifically includes: after the optical cable fault occurs, refer to the operation manual and historical record files of the trunk optical cable transmission system, make decisions according to the suggestions of technicians and experts to handle the fault, and the effective decision is used as the label quantity in supervised learning; for the training of the fault model, first pre-process the data to generate corresponding features and labels, where the data includes equipment operation history data, alarm information, and mining the fault-related data of optical cable interruption and data service error; then use the features and labels as the input of the multi-layer perceptron model, so as to establish a fault training model under the deep learning framework; the upper and lower neurons in the constructed neural network model are fully connected, so as to make more full use of more features for training; based on this model, the back propagation algorithm is used to train the multi-layer artificial neural network, and the weights and bias values ​​of the network are learned through continuous iteration, and the output value of the network is made as close to the true target value as possible through the stochastic gradient descent optimizer.

[0011] Preferably, in S1, the model layer is a data model that describes entities, relationships between entities and attributes in the field of optical cable transmission systems, including a brief description of the fault, system stability requirements, post-fault processing methods and the relationships between them.

[0012] Preferably, in S1, the construction of the data layer includes text classification, named entity recognition and extraction of relations between entities.

[0013] Preferably, text classification is performed using a BERT method based on the latest proposed bidirectional encoder representation from a converter, including: (1) preprocessing long texts; (2) selecting the most effective layer for text classification according to requirements; and (3) dealing with overfitting problems.

[0014] Preferably, the implementation scheme of named entity recognition is as follows: given a sequence of N words {w1, w2, ... N}, the predefined BEAT model first generates a set of vectors {x1(NER),x2(NER),…,x N (NER)}, and then used as the input of the Feed Forward Neural Network (FFNN) for classification, and then the BIOES-based NER labels of the training data are obtained, and the number of labels is |C (NER) |, during the training process, a cross entropy loss is used as the objective function of NER, which is defined as follows:

[0015]

[0016] Among them, S n (NER) Indicates the score of the nth word belonging to the correct category, S n,c (NER) represents the score of the nth word belonging to category c.

[0017] Preferably, the implementation scheme for extracting the relationship between entities is as follows: First, the score vector {S i (NER)} to obtain the maximum value and get the predicted entity category of each word; then input it into the embedding layer to get a continuous vector of fixed length {e i (NER)}, and finally concatenate each word with the output of the BERT model x i (RE) =x i (NER) ||

[0018] e i (NER) . Then each x i(RE) Mapped to head and tail vectors h i (head) =FFNN head (x i (RE) ),h i (tail) =FFNN tail (x i (RE) ), and then input the head and tail vectors into a biaffifine classifier:

[0019]

[0020]

[0021] Among them, U, W, and b are the parameters of the biaffifine classifier, and the loss function of RE is also the cross entropy loss function:

[0022]

[0023] Among them, S n (RE) Indicates the score of the nth word belonging to the correct category, S n,c (RE) It represents the score of the nth word belonging to category c. The final loss function is defined as follows:

[0024] Γ=Γ NER +λΓ RE

[0025] In the formula, the parameter λ gradually increases from 0 to 1 during the training process, so that the NER model can quickly achieve a better effect.

[0026] The present invention has the following beneficial effects:

[0027] (1) The knowledge graph construction of the optical cable transmission system fault processing described in the embodiment of the present invention realizes the knowledge extraction, representation and management of the optical cable transmission system fault processing information, which can be used to assist technical personnel in fault processing and improve the emergency processing capability and intelligent level of the backbone optical cable transmission system.

[0028] (2) The knowledge graph-based operation and maintenance solution decision for the optical cable transmission system described in the embodiment of the present invention can match accurate operation and maintenance solutions based on the system operation and maintenance requirements and the operation and maintenance knowledge graph information of the optical cable transmission system, thereby providing solution support for operation and maintenance personnel.

[0029] (3) The optical cable transmission system fault processing based on deep learning described in the embodiment of the present invention adopts the deep learning framework of multi-layer perceptron to perform supervised learning on the fault environment information of the optical cable transmission system, and can automatically generate correct and feasible fault processing decision plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The present invention is a schematic diagram of the processing flow of an intelligent operation and maintenance method for a radio and television trunk optical cable transmission system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Reference Figure 1 , shown is an intelligent operation and maintenance method for a radio and television trunk optical cable transmission system according to an embodiment of the present invention, comprising the following steps:

[0033] S1, construction of knowledge graph for optical cable transmission system fault handling.

[0034] First, by analyzing the content of the optical cable transmission system fault handling plan, the model layer is designed in a top-down manner, and then the data layer is constructed in a bottom-up manner. Among them, the model layer is a data model that describes the entities, relationships between entities and attributes in the field of optical cable transmission systems, including a brief description of the fault, system stability requirements, post-fault processing methods and their interrelationships.

[0035] The construction of the data layer includes text classification, named entity recognition, and extraction of relations between entities, as follows.

[0036] Text classification: Based on the latest Bidirectional Encoder Representation from Transformers (BERT) method, text classification is performed. In order to better apply the BERT method to the intelligent operation and maintenance field of the present invention, a fine-tuning strategy for BERT is proposed. Three aspects are considered at this time: (1) Preprocessing long texts to adapt to the maximum length limit of BERT; (2) The native BERT model contains 1 embedding layer, 12 encoding layers and 1 pooling layer. The most effective layer for text classification needs to be selected according to needs; (3) Dealing with overfitting problems. In addition, since the lower layers of the BERT model include more general information, different learning rates can be used for fine-tuning. First, the BERT model parameters θ are decomposed into {θ 1 ,θ 2 ,…,θ L ,}, for the parameters of the lth layer (the value range of the subscript l is 1 to L) with a time step of t, the stochastic gradient descent method is used to update, which is defined as follows:

[0037]

[0038] Among them, η l represents the learning rate of the lth layer, θ L is the preset initial value, q is the decay system, and k is the subscript in the learning rate iteration process. Represents the gradient of the model objective equation.

[0039] Named Entity Recognition (NER) and Relationship of Entity (RE) extraction. The implementation scheme of NER is as follows: Given a sequence of N words {w1, w2, ... w N}, the predefined BEAT model first generates a set of vectors {x1(NER),x2(NER),…,x N (NER)}, and then used as the input of the Feed Forward Neural Network (FFNN) for classification. Then the NER labels based on BIOES (Begin, Intermediate, Other, End, Single) of the training data are obtained, and the number of labels is |C (NER) |. During the training process, a cross entropy loss is used as the objective function of NER, which is defined as follows:

[0040]

[0041] Among them, S n (NER) Indicates the score of the nth word belonging to the correct category, S n,c (NER) represents the score of the nth word belonging to category c.

[0042] Implementation plan for entity relationship (RE) extraction: First, the score vector {S i (NER)} to obtain the maximum value and get the predicted entity category of each word. Then input it into the embedding layer to get a fixed-length continuous vector {e i (NER)}, and finally concatenate each word with the output of the BERT model x i (RE) =x i (NER) ||e i (NER) . Then each x i (RE) Mapped to head and tail vectors h i (head) =FFNN head (x i (RE) ),h i (tail) =FFNN tail (x i (RE) ), and then input the head and tail vectors into a biaffifine classifier:

[0043]

[0044]

[0045] Among them, U, W, and b are the parameters of the biaffifine classifier. The loss function of RE is also the cross entropy loss function:

[0046]

[0047] Among them, S n (RE) Indicates the score of the nth word belonging to the correct category, S n,c (RE) represents the score of the nth word belonging to category c. The final loss function is defined as follows:

[0048] Γ=Γ NER +λΓ RE

[0049] In the formula, the parameter λ gradually increases from 0 to 1 during the training process, so that the NER model can quickly achieve a better effect.

[0050] S2, implementation plan for the operation and maintenance of the optical cable transmission system based on the knowledge graph. The operation and maintenance plan decision of the optical cable transmission system based on the knowledge graph first maps the fault description to the computer room operation and maintenance knowledge graph to form a fault description subgraph; then, the user's operation and maintenance requirements are further mapped to the knowledge graph; finally, a graph search is performed in the knowledge graph to match the user requirements and the fault description to obtain an operation and maintenance plan.

[0051] In the specific application example, the operation and maintenance requirements and solutions for a single fault are matched. For the operation and maintenance requirement X of the optical cable transmission system, all the entities and entity relationships contained therein can be mapped to a single acyclic subgraph in the knowledge graph. At this time, if there is a fault description Q in the solution library, and the subgraph mapped by this description in the knowledge graph completely overlaps with the subgraph of X, then Q and X are successfully matched, and the operation and maintenance solution is found. In order to handle the "synonym" relationship in the text, the "synonym" relationship in the knowledge graph and the word vector are used to jointly construct a synonym table, and the CBOW (Continuous Bag of Words) model is used for judgment, and the synonyms are judged by cosine similarity. The specific matching process is as follows: (1) Input the operation and maintenance demand X and the fault description Q, and mark the entity set T = {n1, n2, n3, n4, ...} contained in Q; (2) Select any entity from the entities {n1', n2', n3', n4', ...} contained in the operation and maintenance demand X, corresponding to the node n' in the graph; (3) If the node n' has been marked, proceed to the next step; if not, return to the previous step; (4) Determine whether there is an unselected entity in X. If so, repeat the above steps; if not, proceed to the next step; (5) Determine whether there is any unselected node in the set T. If not, it means the match is successful; otherwise, it means success.

[0052] S3, Fault handling of optical cable transmission system based on deep learning.

[0053] Generally speaking, after an optical cable fault occurs, it is usually referred to the operation manual, historical records and other documents of the trunk optical cable transmission system, and decisions are made according to the suggestions of technicians and experts to handle the fault. These effective decisions are the label quantity in supervised learning. For the training of the fault model, the data is first preprocessed to generate corresponding features and labels, where the data includes equipment operation history data, alarm information, and mining of fault-related data such as optical cable interruption and data service errors. The features and labels are then used as inputs of the multi-layer perceptron model to establish a fault training model under the deep learning framework. In the neural network model constructed by the present invention, the upper and lower neurons are fully connected, so that more features can be more fully utilized for training. Based on this model, the back propagation algorithm is used to train the multi-layer artificial neural network, and the weights and bias values ​​of the network are learned in a continuous iterative manner, and the output value of the network is made as close to the true target value as possible through the stochastic gradient descent optimizer.

[0054] The intelligent operation and maintenance method of the radio and television trunk optical cable transmission system in the above embodiment can effectively improve the intelligent level of the operation and maintenance of the optical cable transmission system by making full use of the existing operation and maintenance experience and big data, through the knowledge graph and deep learning. The technical solution of the present invention can be further applied to the operation and maintenance of the optical cable transmission system in the radio and television, telecommunications, and Internet industries. Through experience organization and big data collection, it can effectively improve the operation and maintenance level of the optical cable transmission system, reduce errors, and improve work efficiency.

[0055] It should be understood that the exemplary embodiments described herein are illustrative rather than restrictive. Although one or more embodiments of the present invention have been described in conjunction with the accompanying drawings, it should be understood by those skilled in the art that various changes in form and detail may be made without departing from the spirit and scope of the present invention as defined by the appended claims.

Claims

1. An intelligent operation and maintenance method for a radio and television trunk optical cable transmission system, characterized in that: The following steps are involved: S1, construction of knowledge graph for optical cable transmission system fault handling. By analyzing the text content of optical cable transmission system fault handling strategy, the model layer of the knowledge graph is designed in a top-down manner, and then the data layer is constructed in a bottom-up manner; S2, the decision of the operation and maintenance plan of the optical cable transmission system based on the knowledge graph, first maps the fault description to the operation and maintenance knowledge graph of the computer room to form a fault description subgraph; Then, the user's operation and maintenance requirements are further mapped into the knowledge graph. Finally, a graph search is performed in the knowledge graph to match the user's requirements with the fault description to obtain an operation and maintenance plan. S3, Fault handling of optical cable transmission system based on deep learning, based on the big data of computer room operation and maintenance, design a method for modeling computer room fault features and generating fault handling strategies based on deep learning, specifically including: after the optical cable fault occurs, refer to the operation manual and historical record files of the trunk optical cable transmission system, make decisions according to the suggestions of technicians and experts to handle the fault, and the effective decision is used as the label quantity in supervised learning; for the training of the fault model, first pre-process the data to generate corresponding features and labels, where the data includes equipment operation history data, alarm information, and mining the fault-related data of optical cable interruption and data service error; then use the features and labels as the input of the multi-layer perceptron model, so as to establish a fault training model under the deep learning framework; the upper and lower neurons in the constructed neural network model are fully connected, so as to make more full use of more features for training; based on this model, the back propagation algorithm is used to train the multi-layer artificial neural network, and the weights and bias values ​​of the network are learned through continuous iteration, and the output value of the network is made as close to the real target value as possible through the stochastic gradient descent optimizer; In S1, the construction of the data layer includes text classification, named entity recognition, and extraction of relations between entities; Text classification uses the BERT method based on the latest proposed bidirectional encoder representation from the converter to perform text classification, including: (1) preprocessing long texts; (2) selecting the most effective layer for text classification according to requirements; (3) dealing with the overfitting problem; The implementation scheme of named entity recognition is as follows: given a sequence of N words {w1,w2,…w N }, the predefined BEAT model first generates a set of vectors {x1(NER),x2(NER),…,x N (NER)}, and then used as the input of the FeedForward Neural Network (FFNN) for classification, and then the BIOES-based NER labels of the training data are obtained, and the number of labels is |C (NER) |, during the training process, a cross entropy loss is used as the objective function of NER, which is defined as follows: Among them, S n (NER) Indicates the score of the nth word belonging to the correct category, S n,c (NER) represents the score of the nth word belonging to category c; The implementation scheme for extracting the relationship between entities is as follows: First, the score vector {S i (NER) } to obtain the maximum value and get the predicted entity category of each word; then input it into the embedding layer to get a continuous vector of fixed length {e i (NER) }, and finally concatenate each word with the output of the BERT model x i (RE) =x i (NER) ||e i (NER) ; Then each x i (RE) Mapped to head and tail vectors h i (head) =FFNN head (x i (RE) ),h i (tail) =FFNN tail (x i (RE) ), and then input the head and tail vectors into a biaffifine classifier: Among them, U, W, and b are the parameters of the biaffifine classifier, and the loss function of RE is also the cross entropy loss function: Among them, S n (RE) Indicates the score of the nth word belonging to the correct category, S n,c (RE) It represents the score of the nth word belonging to category c. The final loss function is defined as follows: C=C NER +λΓ RE In the formula, the parameter λ gradually increases from 0 to 1 during the training process, so that the NER model can quickly achieve a better effect; In S2, for the maintenance demand X of the optical cable transmission system, all entities and entity relationships contained therein are mapped to a single acyclic subgraph in the knowledge graph. At this time, if there is a fault description Q in the solution library, and the subgraph mapped by this description in the knowledge graph completely overlaps with the subgraph of X, then Q and X are successfully matched, and the maintenance solution is found. In order to handle the "synonym" relationship in the text, the "synonym" relationship in the knowledge graph and the word vector are used to jointly construct a synonym table, and the continuous bag of words CBOW model is used to judge, and the synonyms are judged by cosine similarity. The specific matching process is as follows: (1) Input the maintenance demand Find X and fault description Q, and mark the entity set T = {n1, n2, n3, n4, ...} contained in Q; (2) Select any entity from the entities {n1', n2', n3', n4', ...} contained in the operation and maintenance request X, corresponding to the node n' in the graph; (3) If node n' has been marked, continue to the next step; if not, return to the previous step; (4) Determine whether there is an unselected entity in X. If so, repeat the above steps; if not, proceed to the next step; (5) Determine whether there is any unselected node in the set T. If not, it means the match is successful; otherwise, it means success.

2. The intelligent operation and maintenance method of the radio and television trunk optical cable transmission system according to claim 1, characterized in that: In S1, the model layer is a data model that describes the entities, relationships between entities and attributes in the field of optical cable transmission systems, including a brief description of the fault, system stability requirements, post-fault processing methods and the relationships between them.

Images