Optical network intelligent operation and maintenance method and device based on knowledge graph and digital twin

Abstract

The present invention belongs to the field of communication operation and maintenance technology, and more specifically, it relates to a method and device for intelligent operation and maintenance of optical networks based on knowledge graphs and digital twins. It includes: extracting the feature sequence of each operation and maintenance scenario in the digital twin network, and composing the extracted feature sequence into a feature data set; encoding the image corresponding to each operation and maintenance rule in the knowledge graph, and composing the code into a rule data set; using the feature data set as a training set and the rule data set as output data to train a deep neural network; when a fault occurs, extracting the feature sequence of the operation and maintenance scenario corresponding to the fault, inputting the feature sequence into the trained deep neural network to obtain the corresponding code in the rule data set, and obtaining the operation and maintenance rules in the corresponding knowledge graph according to the code. The present invention greatly improves the feedback rate of the intelligent operation and maintenance system by constructing a fast mapping relationship between scenes between digital twins and knowledge graphs.

Description

Technical Field

[0001] The present invention belongs to the technical field of communication operation and maintenance, and more specifically, relates to a method and device for intelligent operation and maintenance of optical networks based on knowledge graphs and digital twins. Background Art

[0002] There are many problems in the operation and maintenance of the telecommunications industry. Operators maintain a large number of isolated operation support systems (OSS). The data and tools of business systems in specialties such as mobile networks, fixed networks, and bearers are fragmented from each other, and the relevance of resources and data is poor, lacking end-to-end operation and maintenance capabilities. Moreover, the current operation and maintenance models are all manual operation and maintenance. Operation and maintenance personnel spend 90% of their time locating faults, resulting in low operation and maintenance efficiency. Summary of the Invention

[0003] In view of the above defects or improvement requirements of the prior art, the present invention provides a solution for intelligent operation and maintenance of optical networks based on knowledge graphs and digital twins. The digital twin body established by the digital twin system solves the problem of digital representation of network-wide resources. By constructing an intelligent fault operation and maintenance knowledge graph for optical networks, experience methods such as network-wide fault location, fault analysis, and troubleshooting solutions are integrated to achieve intelligent fault tracing and operation and maintenance.

[0004] To achieve the above object, according to one aspect of the present invention, a method for intelligent operation and maintenance of optical networks based on knowledge graphs and digital twins is provided, including: extracting the feature sequences of each operation and maintenance scenario in the digital twin network, and forming the extracted feature sequences into a feature data set; encoding the images corresponding to each operation and maintenance rule in the knowledge graph, and forming the encoded data into a rule data set; using the feature data set as a training set and the rule data set as output data to train a deep neural network; when a fault occurs, extracting the feature sequence of the operation and maintenance scenario corresponding to the fault, inputting the feature sequence into the trained deep neural network to obtain the corresponding encoding in the rule data set, and obtaining the operation and maintenance rules in the corresponding knowledge graph according to the encoding.

[0005] Preferably, the extracting the feature sequences of each operation and maintenance scenario in the digital twin network and forming the extracted feature sequences into a feature data set specifically includes: digitally simulating the operation and maintenance scenarios of at least one type of fault in the digital twin network, and forming a digital library of operation and maintenance scenarios; performing multimodal fusion and feature extraction on each operation and maintenance scenario in the digital library, and forming the extracted feature sequences into a feature data set of operation and maintenance scenarios.

[0006] Preferably, the multi-modal fusion and feature extraction for each operation and maintenance scenario in the digital library specifically include: extracting features for each link related to faults in the digital twin network to obtain a set of eigenvalue for each link; performing multi-modal fusion on all the eigenvalues to obtain the multi-modal joint distribution of the eigenvalues, and extracting the feature sequence of the corresponding scenario according to the multi-modal joint distribution.

[0007] Preferably, the extracting features for each link related to faults in the digital twin network specifically include: recording the network data of each link when a fault occurs as a string, and then converting the string into a digital representation to obtain a network data array; designing a unified convolution kernel to perform convolution operation on the network data array to obtain a feature sequence, where the convolution kernel is a square matrix of different orders and is used to perform weighted average on the target array and then extract features from the target array.

[0008] Preferably, the encoding the image corresponding to each operation and maintenance rule in the knowledge graph and forming a rule data set specifically includes: for each operation and maintenance scenario in the digital library, searching for the operation and maintenance rules for processing the corresponding faults in the knowledge graph, and forming an operation and maintenance rule library with the searched operation and maintenance rules; generating an image in pixel format for each operation and maintenance rule in the operation and maintenance rule library, encoding each image, and forming a rule data set with the encodings corresponding to the operation and maintenance rules.

[0009] Preferably, the generating an image in pixel format for each operation and maintenance rule in the operation and maintenance rule library and encoding each image specifically includes: visually representing the knowledge graph corresponding to each operation and maintenance rule, and generating an image in a specified pixel format for the visually represented knowledge graph; using a hash algorithm to generate the hash value of each image, and using the hash value as the encoding of the corresponding image.

[0010] Preferably, the generating an image in a specified pixel format for the visually represented knowledge graph specifically includes: solidifying and outputting the knowledge graph in each operation and maintenance scenario as an original image in the pixel format of m*m, and calculating the mean value μ of all pixel points in the original image; the standardized image is as follows: where x i is the pixel point of the original image, μ is the mean value of all pixel points in the original image, σ is the standard deviation of all pixel points in the original image, N is the number of pixel points in the original image, and m is a preset value.

[0011] Preferably, the using a hash algorithm to generate the hash value of each picture specifically includes: arranging the pixel points in the standardized image in a zigzag scanning manner to obtain the standardized pixel sequence of each image; calculating the standardized pixel sequence using a hash algorithm to obtain the hash value of each image.

[0012] Preferably, inputting the feature sequence into the trained deep neural network to obtain the corresponding encoding in the rule dataset, and obtaining the operation and maintenance rules in the corresponding knowledge graph according to the encoding specifically includes: inputting all feature sequences into the trained deep neural network to obtain the encoding output by the deep neural network, and obtaining the image corresponding to the encoding in the rule dataset; obtaining the operation and maintenance rules of the corresponding operation and maintenance scenario in the knowledge graph according to the image, and obtaining the corresponding fault handling method according to the operation and maintenance rules.

[0013] According to another aspect of the present invention, there is also provided an optical network intelligent operation and maintenance device based on a knowledge graph and digital twin, which is used to complete the optical network intelligent operation and maintenance method based on a knowledge graph and digital twin provided in the first aspect, including a knowledge graph standardization and encryption unit, a deep neural network training unit, and a network fault operation and maintenance unit, wherein: the knowledge graph standardization and encryption unit is used to format the knowledge graph into a corresponding image and encode the image to obtain a rule dataset; the deep neural network training unit is used to extract features of the links related to faults in the digital twin network to obtain relevant feature sequences, and input all feature sequences as training data into the deep neural network for training to obtain a trained deep neural network; the network fault operation and maintenance unit is used to extract features of the links related to faults in the digital twin network with faults to obtain relevant feature sequences, input all feature sequences into the trained deep neural network, and obtain a fault handling method according to the output of the deep neural network.

[0014] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the following beneficial effects are obtained:

[0015] (1) By constructing a fast scene mapping relationship between the digital twin and the knowledge graph, the feedback rate of the intelligent operation and maintenance system is greatly improved;

[0016] (2) By training the professional knowledge graph with examples and verifying the reliability of the rules by experts, the problem of the artificial intelligence black box is partially solved, and the interpretability of the system is enhanced;

[0017] (3) Using the hash algorithm to encode the standardized knowledge graph makes the knowledge graph modular, ensures the security of the knowledge graph, and enhances the portability and generality of the knowledge graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of the operation and maintenance system architecture based on a knowledge graph and digital twin in an embodiment of the present invention;

[0019] Figure 2 is a schematic flowchart of an optical network intelligent operation and maintenance method based on a knowledge graph and digital twin in an embodiment of the present invention;

[0020] Figure 3 It is a schematic diagram of the knowledge graph hashing encoding principle in an embodiment of the present invention;

[0021] Figure 4 It is a schematic flowchart of another optical network intelligent operation and maintenance method based on a knowledge graph and digital twin in an embodiment of the present invention;

[0022] Figure 5 It is a schematic diagram of the digital twin multimodal representation principle in an embodiment of the present invention;

[0023] Figure 6 It is a schematic flowchart of another optical network intelligent operation and maintenance method based on a knowledge graph and digital twin in an embodiment of the present invention;

[0024] Figure 7 It is a schematic diagram of the digital twin - knowledge graph fast mapping principle in an embodiment of the present invention;

[0025] Figure 8 It is a schematic structural diagram of an optical network intelligent operation and maintenance device based on a knowledge graph and digital twin in an embodiment of the present invention. Detailed implementation manners

[0026] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0027] Embodiment 1:

[0028] The knowledge graph (abbreviated as KG) is an optical network operation and maintenance knowledge graph, and its covered content includes all network data such as network device types, network connection types, network fault types, network node states, network environmental factors, etc. and the relationships between them. In theory, it encompasses all types of optical network faults and the network states and causes when they occur.

[0029] In the prior art, both the knowledge graph and the digital twin (abbreviated as DT) network are emerging technologies in the telecommunications industry. In the existing network operation and maintenance system architecture based on the knowledge graph and digital twin, such as Figure 1As shown. However, there is currently no mature solution for the interconnection between the knowledge graph and the digital twin. In the scenario of optical network intelligent operation and maintenance, there are problems with the coordination between the digital twin and the knowledge graph. Specifically, in the intelligent operation and maintenance system, when a network failure occurs, it is necessary to input the fault digital scenario in the digital twin network into the knowledge graph to seek solutions. Currently, there is no effective data transmission method between the knowledge graph and the digital twin network, mainly relying on manual input.

[0030] In the method provided in this embodiment, by introducing the typical fault operation and maintenance feature data set and the standard operation and maintenance rule data set of the existing network, a pre-trained deep neural network is used between the two data sets to realize the real-time and fast mapping between the knowledge graph and the digital twin network.

[0031] As Figure 2 shown, the present invention provides an optical network intelligent operation and maintenance method based on a knowledge graph and a digital twin, including:

[0032] Step 101: Extract the feature sequence of each operation and maintenance scenario in the digital twin network, and form the extracted feature sequences into a feature data set.

[0033] To complete the mapping, first, key features need to be extracted from the full-network digital twin by using multi-modal representation learning, and the feature sequences of the key features are formed into a feature data set. To obtain the feature data of the faults in the digital twin network, it is necessary to extract the features of N links related to the faults in the digital twin to obtain N relevant feature sequences.

[0034] Step 102: Encode the images corresponding to each operation and maintenance rule in the knowledge graph, and form the encodings into a rule data set.

[0035] When directly retrieving the operation and maintenance rules in the knowledge graph using ordinary data retrieval methods, it is necessary to retrieve in a large amount of data and may perform multiple associated retrievals or branch judgments. Therefore, in this embodiment, the knowledge graph is visualized, each operation rule is converted into a corresponding image according to the specified rules, and then the image is encoded, and the encoding is used as rule data to form a rule data set.

[0036] Step 103: Use the feature data set as the training set and the rule data set as the output data to train the deep neural network.

[0037] After obtaining the feature data set and the rule data set, the deep neural network can be trained. Through the deep neural network, the feature data of the faults in the digital twin network are paired with the corresponding operation and maintenance rules in the knowledge graph, realizing the fast mapping from the digital twin network to the knowledge graph in optical network intelligent operation and maintenance, and solving the data interconnection problem in the overall intelligent operation and maintenance system.

[0038] During the training phase, the input vector of the deep neural network is the multi-modal feature vector of the digital twin network when each fault occurs, that is, the feature sequence in the feature dataset; the output vector is the knowledge graph encoding value corresponding to the fault, that is, the operation and maintenance rule encoding in the rule dataset. Taking the feature sequences of all faults as training data and inputting them into the deep neural network, and using the corresponding operation and maintenance rule encoding values as labels to train the deep neural network, the mapping relationship between the feature sequence and the encoding value can be obtained.

[0039] Step 104: When a fault occurs, extract the feature sequence of the operation and maintenance scenario corresponding to the fault, input the feature sequence into the trained deep neural network to obtain the corresponding encoding in the rule dataset, and obtain the operation and maintenance rules in the corresponding knowledge graph according to the encoding.

[0040] After the deep neural network is trained, the fault features in the digital twin network can be input into the trained network to obtain the corresponding encoding values. In specific implementation, all feature sequences can be input into the trained deep neural network to obtain the encoding output by the deep neural network, and the image corresponding to the encoding can be obtained in the rule dataset; the operation and maintenance rules of the corresponding operation and maintenance scenario in the knowledge graph can be obtained according to the image, and the corresponding fault handling method can be obtained according to the operation and maintenance rules. Since the encoding values and operation and maintenance solutions in the knowledge graph are in one-to-one correspondence, when the encoding values are obtained, the corresponding operation and maintenance solutions in the knowledge graph can also be obtained in reverse.

[0041] In specific implementation, as Figure 3 shown, feature extraction can be performed on N links related to the fault in the digital twin network of the target fault network to obtain N relevant feature sequences, input all feature sequences into the trained deep neural network, obtain the corresponding knowledge graph encoding values, and output the corresponding operation and maintenance methods according to the encoding values.

[0042] After steps 101 - 104 provided in this embodiment, the feature sequence of the fault can be mapped to the operation rules in the knowledge graph through the deep neural network, realizing the intelligent operation and maintenance of the optical network.

[0043] Furthermore, in actual implementation, different training strategies can be used when training the deep neural network in step 103. For relatively stable and mature optical networks, offline training can be performed based on a large amount of historical data, and the trained deep neural network can be directly used to obtain the operation and maintenance solutions. For newly erected networks or frequently changing networks, the fault features and operation and maintenance rules after each fault handling can be added to the training set for real-time online training, and the deep neural network can be updated with new data to avoid the problem of too few samples and adapt to the changes of the optical network.

[0044] To obtain the feature dataset, at least one type of operation and maintenance scenario of a fault can be digitally simulated in the digital twin network, and the simulation results are combined into a digital library of operation and maintenance scenarios; for each operation and maintenance scenario in the digital library, multimodal fusion and feature extraction are performed, and the extracted feature sequences are combined into the feature dataset of the operation and maintenance scenarios.

[0045] As Figure 4 shown, the multimodal fusion and feature extraction of each operation and maintenance scenario in the digital library can be performed in the following manner.

[0046] Step 201: Extract features from each link related to the fault in the digital twin network to obtain a set of feature values for each link.

[0047] As Figure 5 shown, when a fault occurs, record the network data in the digital twin network and extract the feature data therefrom.

[0048] In specific implementation, a neural network can be used for feature extraction. Specifically: record the network data of each link when a fault occurs as a string, and then convert the string into a digital representation to obtain a network data array. Then design a unified convolution kernel to perform convolution operation on the network data array to obtain a feature sequence, where the convolution kernel is a square matrix of different orders and is used to perform weighted averaging on the target array and then extract features from the target array.

[0049] In an actual scenario, the network data includes: network alarms, configurations, performances, topologies, operation logs, virtual IPs, working hours of network devices, and factory configurations of network devices. First, record the 8-dimensional data when a fault occurs as a string, and then convert the string into a digital representation to obtain an 8-dimensional array; design a unified convolution kernel to perform convolution operation on the 8-dimensional array to obtain an 8-dimensional feature sequence, where the convolution kernel is a square matrix of different orders and is used to perform weighted averaging on the target array and then extract features from the target array.

[0050] Step 202: Perform multimodal fusion on all the feature values to obtain the multimodal joint distribution of the feature values, and extract the feature sequence of the corresponding scenario according to the multimodal joint distribution.

[0051] Represent the information of different modalities in the extracted feature values in a unified manner, fuse the unified represented data together and perform feature extraction to perform multimodal representation fusion feature extraction on all the data, and obtain the feature sequence of the digital twin network fault in this state.

[0052] After steps 201 - 202 provided in this embodiment, the feature sequence in the digital twin network can be obtained.

[0053] To obtain a rule dataset, for each operation and maintenance scenario in the digital library, search for operation and maintenance rules for handling corresponding faults in the knowledge graph, and form an operation and maintenance rule library with the searched operation and maintenance rules; generate images in pixel format for the knowledge graph corresponding to each operation and maintenance rule in the operation and maintenance rule library, encode each image, and form a rule dataset with the encodings corresponding to the operation and maintenance rules. In specific implementation, an appropriate encoding method can be selected according to needs. For example: specified serial numbers, unique identification codes of image files generated by the system, crc32 check codes of image files, and MD5 hash values of image files, etc. As long as the encoding used can correspond one-to-one with the images of the operation and maintenance rules.

[0054] The following is an implementation scheme using hash values as the encoding method. It can be understood that the following scheme is only a specific implementation scheme of the method provided in this embodiment in a specific scenario and does not limit the protection scope. Using hash values to represent operation and maintenance solutions has the following advantages: 1. The method provided in this embodiment visualizes the knowledge graph, and in order to accurately represent the knowledge graph, the image data volume will be huge. Hash values will effectively compress the image data initially; 2. Hash values have good confidentiality effects. Using hash values as the output results of the deep neural network instead of the pixel values of the image itself will effectively protect the specific content of the knowledge graph. Here, the operation and maintenance methods corresponding to the hash values have been one-to-one corresponded and solidified during the previous training, enabling users to directly obtain the operation and maintenance methods through the hash values during the usage stage without perceiving the specific content of the knowledge graph, achieving the effect of keeping the knowledge graph itself confidential.

[0055] As Figure 6 shown, the following method can be used to generate images in pixel format for the knowledge graph corresponding to each operation and maintenance rule in the operation and maintenance rule library and encode each image.

[0056] Step 301: Visually represent the knowledge graph corresponding to each operation and maintenance rule, and generate an image in a specified pixel format for the visually represented knowledge graph.

[0057] The knowledge graph stores operation and maintenance related data nodes and the association relationships between data nodes. Each type of operation and maintenance rule corresponds to one or more data nodes or association relationships. In order to intuitively reflect the operation and maintenance rules, the data nodes and association relationships related to the operation and maintenance rules can be graphically represented according to specified visualization rules. For example: use specified geometric shapes to represent data nodes, use specified lines to represent the association relationships between data nodes, and use specified colors or filling methods to represent the attributes of data nodes.

[0058] After completing the graphical representation, then use the specified display size to convert the visually represented graph into an image in the corresponding pixel format. Specifically, as Figure 7As shown, the knowledge graph in each operation and maintenance scenario can be solidified and output as an original image in the pixel format of m*m, and the mean value μ of all pixel points in the original image is calculated. The standardized image is as follows:

[0059]

[0060] Where: x i is the pixel point of the original image, μ is the mean value of all pixel points in the original image, σ is the standard deviation of all pixel points in the original image, N is the number of pixel points in the original image, and m is a preset value. In actual implementation, the value of m can be determined according to the data volume of the knowledge graph and the required resolution. For example, m = 256.

[0061] Furthermore, after obtaining the image in pixel format, for the convenience of subsequent use, the image can be saved in a picture file with a standard format, such as JPEG, PNG, GIF, TIF, and BMP, etc.

[0062] Step 302: Use the hash algorithm to generate the hash value of each image, and use the hash value as the encoding of the corresponding image.

[0063] In specific implementation, the pixel points in the standardized image can be arranged in a zigzag scanning manner to obtain the standardized pixel sequence of each image; the hash algorithm is used to calculate the standardized pixel sequence to obtain the hash value of each image.

[0064] In a specific scenario, the MD5 hash algorithm is used to encode the image to obtain the hash value of the knowledge graph corresponding to different network states. Specifically: the pixel points of the standardized image are arranged in a zigzag scanning (ZigTag) manner to obtain the standardized pixel sequence of each picture, and then the MD5 algorithm is used for this sequence to obtain the MD5 hash value of each picture.

[0065] After steps 301 - step 302 provided in this embodiment, the hash value of each image can be calculated so that the hash value can be used as the encoding of the operation and maintenance solution.

[0066] The optical network intelligent operation and maintenance method based on knowledge graph and digital twin provided in this embodiment effectively integrates the two technologies of digital twin network and knowledge graph in the intelligent operation and maintenance scenario with knowledge graph and digital twin as the key technologies, proposes the visualization of knowledge graph and the multi-modal representation of digital twin system, and quickly and accurately corresponds the results of the two technologies by training the deep neural network for the feature data set and the rule data set, ensuring the accuracy and security of the operation and maintenance solution.

[0067] Embodiment 2:

[0068] Based on the optical network intelligent operation and maintenance method based on knowledge graph and digital twin provided in Embodiment 1, in some specific embodiments, it can be implemented through the specific embodiments in this embodiment. It can be understood that the specific embodiments provided in this embodiment are only used to illustrate the specific implementation process of the method in Embodiment 1 in some specific scenarios, and do not limit the scope of protection.

[0069] A: Preparation stage.

[0070] A1: Build an in-network experimental scenario and construct a corresponding digital twin network (DT) according to the physical network environment.

[0071] A2: Construct a complete optical network operation and maintenance professional knowledge graph (KG) based on expert experience and a large number of basic experiences such as operation and maintenance manuals, logs, and cases.

[0072] B: Training stage.

[0073] B1: Manually sort out N typical operation and maintenance scenarios, perform digital simulation in the digital twin network, and obtain a digital library of typical fault operation and maintenance scenarios in the in-network: S = {s 1 , s 2 ,..., s n}.

[0074] B2: Use multi-modal representation learning to extract features of the links related to faults, and perform modal fusion on each digital scenario in S to obtain a dataset of typical fault operation and maintenance features in the in-network: D = {d 1 , d 2 , …, d n}.

[0075] B3: For different operation and maintenance scenarios in S, use KG to automatically search for corresponding operation and maintenance rules. At this time, the input data of KG is manually input. Obtain a standard operation and maintenance rule database: R = {r 1 , r 2 ,..., r n}, and at the same time standardize the KG corresponding to R into standard pictures, and use the MD5 hash algorithm to calculate the hash value of each picture to obtain a rule dataset of standard operation and maintenance solutions: H = {h 1 , h 2 ,..., h n}.

[0076] B4: Build a deep neural network N, and use D = {d 1 , d 2 ,..., d n} as the training set, and H = {h 1 , h 2 ,..., h nUsing the labeled data as the output, train N to obtain the trained deep neural network, i.e., the KG-DT mapping network N*.

[0077] C: Usage phase.

[0078] C1: When a fault occurs in the live network, extract the features of the links related to the fault of the DT at this time to obtain d'.

[0079] C2: Input d' into N* to obtain h', and output the corresponding operation and maintenance rule r' according to the correspondence rule between H and R.

[0080] It can be seen from the specific examples provided in this embodiment that by using the optical network intelligent operation and maintenance method based on knowledge graph and digital twin provided in Embodiment 1, it is possible to establish a mapping relationship between the faults in the digital twin network and the operation and maintenance rules in the knowledge graph, so as to achieve a rapid mapping between faults and operation and maintenance rules.

[0081] Embodiment 3:

[0082] Based on the optical network intelligent operation and maintenance method based on knowledge graph and digital twin provided in the above Embodiment 1 to Embodiment 2, the present invention also provides an optical network intelligent operation and maintenance device based on knowledge graph and digital twin that can be used to implement the above method, which is used to complete the optical network intelligent operation and maintenance method based on knowledge graph and digital twin provided in Embodiment 1 and Embodiment 2.

[0083] As Figure 8 shown, it is a schematic diagram of the device architecture of an embodiment of the present invention. It includes: a knowledge graph standardization and encryption unit, a deep neural network training unit, and a network fault operation and maintenance unit.

[0084] The knowledge graph standardization and encryption unit is used to format the knowledge graph into a corresponding image and encode the image to obtain a rule data set. In a specific implementation, the knowledge graph standardization and encryption unit formats the complete knowledge graph in the field of optical network intelligent operation and maintenance into a standard image, and encodes the standard image using the MD5 hash algorithm to obtain the knowledge graph hash values corresponding to different network states.

[0085] The deep neural network training unit is used to extract the features of the links related to the fault in the digital twin network to obtain relevant feature sequences, and input all the feature sequences into the deep neural network as training data to obtain the trained deep neural network. In a specific implementation, the deep neural network training unit extracts the features of N links related to the fault in the digital twin to obtain N relevant feature sequences, inputs all the feature sequences into the deep neural network as training data, and trains the deep neural network with the corresponding knowledge graph hash value as the label.

[0086] The network fault operation and maintenance unit is used to extract the features of the links related to the fault in the digital twin network with a fault, obtain the relevant feature sequences, input all the feature sequences into the trained deep neural network, and obtain the fault handling method according to the output of the deep neural network. In specific implementation, the network fault operation and maintenance unit is used in the usage stage, mainly for automatically extracting the digital twin network link features and quickly mapping the KG-DT to output the operation and maintenance plan. Specifically, the network fault operation and maintenance unit extracts the features of N links related to the fault in the digital twin network of the target fault network, obtains N relevant feature sequences, inputs all the feature sequences into the trained deep neural network, obtains the corresponding knowledge graph hash value, and outputs the corresponding operation and maintenance method according to the hash value.

[0087] Further, the deep neural network training unit extracts the features of the links related to the fault. Specifically, when the fault occurs, it records the network data in the digital twin and performs multi-modal representation fusion feature extraction on all the data to obtain the digital twin feature sequence in this state.

[0088] Further, the knowledge graph standardization and encryption unit formats the complete knowledge graph in the field of optical network intelligent operation and maintenance into a standard image. Specifically, it solidifies and outputs the knowledge graph in a certain determined state in the image size of m*m, and then calculates the mean μ of all pixel points of this image. The standardized image is as follows where x i is the pixel point of the original image, σ is the standard deviation of all pixel points of the original image, N is the number of pixel points of the original image, and m is a preset value.

[0089] Further, the knowledge graph standardization and encryption unit encodes the image using the MD5 hash algorithm to obtain the knowledge graph hash value corresponding to different network states. Specifically, it arranges the pixel points of the standardized image in a zigzag scanning manner to obtain the standardized pixel sequence of each picture, and then uses the MD5 algorithm on this sequence to obtain the hash value of each picture.

[0090] Those skilled in the art can easily understand that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An optical network intelligent operation and maintenance method based on knowledge graph and digital twin, characterized in that: include: Extract the feature sequence of each operation and maintenance scenario in the digital twin network, and form a feature data set with the extracted feature sequences; Encode the image corresponding to each operation and maintenance rule in the knowledge graph, and form the code into a rule data set; specifically, for each operation and maintenance scenario in the digital library, search the knowledge graph for the operation and maintenance rules for handling the corresponding fault, and form the searched operation and maintenance rules into an operation and maintenance rule library; visualize the knowledge graph corresponding to each operation and maintenance rule, and generate an image of a specified pixel format from the visualized knowledge graph; use a hash algorithm to generate a hash value for each image, use the hash value as the code for the corresponding image, and form the code corresponding to the operation and maintenance rule into a rule data set; Use the feature dataset as the training set and the rule dataset as the output data to train the deep neural network; When a fault occurs, the feature sequence of the operation and maintenance scenario corresponding to the fault is extracted, and the feature sequence is input into the trained deep neural network to obtain the corresponding code in the rule data set. The operation and maintenance rules in the corresponding knowledge graph are obtained according to the code, which specifically includes: inputting all feature sequences into the trained deep neural network to obtain the code output by the deep neural network, and obtaining the image corresponding to the code in the rule data set; obtaining the operation and maintenance rules of the corresponding operation and maintenance scenario in the knowledge graph according to the image, and obtaining the corresponding fault handling method according to the operation and maintenance rules.

2. The optical network intelligent operation and maintenance method based on knowledge graph and digital twin according to claim 1, characterized in that: The feature sequence of each operation and maintenance scenario in the digital twin network is extracted, and the extracted feature sequence is formed into a feature data set, which specifically includes: Digitally simulate at least one operation and maintenance scenario of a fault in the digital twin network, and compose a digital library of operation and maintenance scenarios with the simulation results; Multimodal fusion and feature extraction are performed on each operation and maintenance scenario in the digital library, and the extracted feature sequences are combined into a feature data set of the operation and maintenance scenario.

3. The optical network intelligent operation and maintenance method based on knowledge graph and digital twin according to claim 2, characterized in that: The multimodal fusion and feature extraction for each operation and maintenance scenario in the digital library specifically includes: Extract features of each link related to the fault in the digital twin network to obtain a set of feature values ​​for each link; All eigenvalues ​​are multimodally fused to obtain a multimodal joint distribution of eigenvalues, and the feature sequence of the corresponding scene is extracted according to the multimodal joint distribution.

4. The optical network intelligent operation and maintenance method based on knowledge graph and digital twin according to claim 3 is characterized in that: The feature extraction of each link related to the fault in the digital twin network specifically includes: The network data of each link when the fault occurs is recorded as a character string, and then the character string is converted into a digital representation to obtain a network data array; A unified convolution kernel is designed to perform convolution operation on the network data array to obtain a feature sequence, wherein the convolution kernel is a square matrix of different orders, which is used to perform weighted averaging on the target array and then extract features from the target array.

5. The optical network intelligent operation and maintenance method based on knowledge graph and digital twin according to claim 1, characterized in that: The step of generating an image of a specified pixel format from the visually represented knowledge graph specifically includes: Solidify the knowledge graph in each operation and maintenance scenario into an m*m pixel format and output it as an original image, and calculate the mean μ of all pixels in the original image; The normalized image is as follows: where x i is the pixel of the original image, μ is the mean of all pixels in the original image, σ is the standard deviation of all pixels in the original image, N is the number of pixels in the original image, and m is the preset value.

6. The optical network intelligent operation and maintenance method based on knowledge graph and digital twin according to claim 1, characterized in that: The use of a hash algorithm to generate a hash value for each image specifically includes: Arrange the pixels in the standardized image in a zigzag scanning manner to obtain a standardized pixel sequence for each image; The hash algorithm is used to calculate the standardized pixel sequence to obtain the hash value of each image.

7. An optical network intelligent operation and maintenance device based on knowledge graph and digital twin, used to complete the optical network intelligent operation and maintenance method based on knowledge graph and digital twin as described in any one of claims 1-6, characterized in that: It includes a knowledge graph standardization encryption unit, a deep neural network training unit, and a network fault operation and maintenance unit, among which: The knowledge graph standardization encryption unit is used to format the knowledge graph into a corresponding image and encode the image to obtain a rule data set; The deep neural network training unit is used to extract features of fault-related links in the digital twin network to obtain relevant feature sequences, and input all feature sequences as training data into the deep neural network for training to obtain a trained deep neural network; The network fault operation and maintenance unit is used to extract features of links related to the fault in the digital twin network where the fault occurs, obtain relevant feature sequences, input all feature sequences into the trained deep neural network, and obtain a fault handling method based on the output of the deep neural network.

Images