Transmission network fault delimiting method, system and equipment based on fault model and medium

Abstract

The invention provides a transmission network fault delimiting method, system and device based on a fault model and a medium, the fault model comprises a feature coding module, a structured model and a classification model, and the method comprises the following steps: obtaining multi-dimensional index data of a transmission network; inputting the multi-dimensional index data into a feature coding module to obtain network element state data; inputting the network element state data into the structured model to obtain a network element state high-dimensional representation, the network element state high-dimensional representation being a high-dimensional feature representation of the network element state data; and inputting the network element state high-dimensional representation into the classification model to carry out fault detection, and determining a detection result. According to the method, automatic processing is performed through the fault model in which the feature coding module, the structured model and the classification model are packaged, so that the labor cost is greatly reduced, the efficiency of fault delimitation can be improved based on automatic processing of the fault model, misjudgment caused by a single data source can be avoided through the evaluation data obtained through the multi-dimensional index data, and the accuracy of fault delimitation is improved. And the accuracy of fault delimitation is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of fault demarcation, and in particular to a method, system, device and medium for transmission network fault demarcation based on a fault model. Background Art

[0002] In the field of transmission failures, link failure is a very important category. Link failures include Synchronous Digital Hierarchy / Spontaneous Protection Network Link Fault (SPN) link failures and Packet Transport Network Link Fault (PTN) link failures, which often lead to interruptions of services carried on them, trigger alarms, etc., affecting customer use. When a link failure occurs, it often leads to equipment or system alarms, which attract the attention of monitoring personnel, so as to identify and delimit the fault. Sometimes, even if no reported alarm is received, complaints from customers are received, which may also be caused by a fault.

[0003] How to identify the specific type of fault is the key and difficulty of the existing technology. In the actual network, due to the large number of network devices, complex alarm propagation relationships, and missing alarms, the same alarm may be caused by many factors. In the existing technology, when performing fault demarcation, it is necessary to combine and analyze multi-dimensional information such as alarms, and finally conclude that the link failure is caused by line interruption, equipment failure, or dynamic ring reasons, and then further locate the fault point based on the conclusion.

[0004] However, such fault demarcation has the following problems: the existing fault demarcation is generally handled manually based on expert experience and rules. Such demarcation often gives multiple reasons, and different personnel need to be sent to the site to determine the cause, which has a high labor cost. Therefore, intelligent analysis and positioning before dispatching is very important. Most of the current intelligent solutions use classification analysis methods based on alarms. Although alarms are very important analysis inputs in fault demarcation, due to the above-mentioned reasons such as complex alarm relationships and missing alarms, the conclusions based solely on alarms are not accurate enough and cannot effectively alleviate the pressure of dispatching orders, resulting in low accuracy and efficiency of fault demarcation. Summary of the invention

[0005] The embodiments of the present application provide a transmission network fault demarcation method, system, device and medium based on a fault model to solve the problems of high labor cost, high demarcation difficulty, low demarcation accuracy and efficiency in existing fault demarcation.

[0006] In order to solve the above technical problems, this application is implemented as follows:

[0007] In a first aspect, an embodiment of the present application provides a transmission network fault demarcation method based on a fault model, wherein the fault model includes a feature encoding module, a structured model, and a classification model, and the method includes:

[0008] Obtain multi-dimensional indicator data of the transmission network;

[0009] Inputting the multi-dimensional indicator data into the feature coding module to obtain network element status data;

[0010] Inputting the network element state data into the structured model to obtain a high-dimensional representation of the network element state, wherein the high-dimensional representation of the network element state is a high-dimensional feature representation of the network element state data;

[0011] The high-dimensional representation of the network element state is input into the classification model to perform fault detection and determine the detection result.

[0012] Optionally, inputting the multi-dimensional indicator data into the feature encoding module to obtain the network element status data includes:

[0013] The multi-dimensional indicator data is input into the feature coding module, wherein the feature coding module constructs features for the indicator data of each dimension respectively to obtain a feature vector corresponding to the indicator data of each dimension, and performs dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain and output the network element status data.

[0014] Optionally, the fault model is constructed, including:

[0015] Acquire a training data set, wherein the training data set includes multi-dimensional indicator data within a specified historical time period, and the multi-dimensional indicator data within the specified historical time period at least includes historical fault work order data;

[0016] Performing feature construction on the indicator data of each dimension contained in the training data set respectively, and obtaining a feature vector corresponding to the indicator data of each dimension, wherein the feature vector at least includes a historical fault delimitation label vector corresponding to the historical fault work order data;

[0017] Performing dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain historical network element status data;

[0018] Training the structured model to be trained based on the historical network element status data, and determining the trained structured model;

[0019] The classification model to be trained is trained based on the historical network element status high-dimensional representation data and the historical fault demarcation label vector to determine the trained classification model, wherein the historical network element status high-dimensional representation data is obtained based on inputting the historical network element status data into the trained structured model.

[0020] Optionally, the training a structured model to be trained based on the historical network element status data to determine the trained structured model includes:

[0021] Input the historical network element state data into the structured model to be trained, and obtain the initial network element state high-dimensional representation and the reconstructed network element state data respectively, wherein the structured model to be trained is a multi-layer structure, the initial network element state high-dimensional representation is the intermediate layer result after the historical network element state data is input into the structured model to be trained, and the reconstructed network element state data is the final layer result of the initial network element state high-dimensional representation after linear transformation and dimensional transformation;

[0022] Based on the loss between the historical network element state data and the reconstructed network element state data, the structured model to be trained is trained to obtain a trained structured model.

[0023] Optionally, the training of a classification model to be trained based on the historical network element state high-dimensional representation data and the historical fault demarcation label vector, and determining the trained classification model includes:

[0024] Inputting the historical network element status high-dimensional representation data and the historical fault demarcation label vector into the classification model to be trained to obtain an initial classification output result;

[0025] Based on the loss between the initial classification output result and the actual classification output result, the classification model to be trained is trained to obtain a trained classification model.

[0026] Optionally, the multi-dimensional indicator data includes at least one of the following: alarm data, performance data, topology data, fault ticket data, and user behavior characteristic data.

[0027] In a second aspect, an embodiment of the present application provides a transmission network fault demarcation system based on a fault model, wherein the fault model includes a feature encoding module, a structured model, and a classification model, and the system includes:

[0028] The indicator module is used to obtain multi-dimensional indicator data of the transmission network;

[0029] A data module, used for inputting the multi-dimensional indicator data into the feature coding module to obtain network element status data;

[0030] A feature module, used for inputting the network element state data into the structured model to obtain a high-dimensional representation of the network element state, wherein the high-dimensional representation of the network element state is a high-dimensional feature representation of the network element state data;

[0031] The output module is used to input the high-dimensional representation of the network element status into the classification model to perform fault detection and determine the detection result.

[0032] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the transmission network fault demarcation method based on the fault model as described in the first aspect above are implemented.

[0033] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the transmission network fault demarcation method based on the fault model as described in the first aspect above are implemented.

[0034] In a fifth aspect, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the steps of the method for fault demarcation of a transmission network based on a fault model as described in the first aspect above.

[0035] In the present application, network element status data is obtained based on multi-dimensional indicator data of the transmission network; and further, the network element status data is input into the structured model to obtain a high-dimensional feature table of the network element status data: a high-dimensional representation of the network element status, and finally the high-dimensional representation of the network element status is input into the classification model to perform fault detection and determine the detection result. In the embodiment of the present application, fault delimitation is not processed relying on expert experience and rules, but is automatically processed through a fault model that encapsulates a feature coding module, a structured model, and a classification model, which greatly reduces labor costs. Automated processing based on the fault model can also improve the efficiency of fault delimitation. In addition, the sources of indicator data in the embodiment of the present application are more diverse, and the evaluation data obtained through multi-dimensional indicator data can avoid misjudgment caused by a single data source, so that a more accurate delimitation result can be obtained, which improves the accuracy of fault delimitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0037] Figure 1It is a flow chart of a transmission network fault demarcation method based on a fault model provided in an embodiment of the present application;

[0038] Figure 2 is a schematic diagram of a fault model provided in an embodiment of the present application;

[0039] Figure 3 is a schematic diagram of network element status data provided by an embodiment of the present application;

[0040] Figure 4 It is a structural diagram of a transmission network fault demarcation system based on a fault model provided in an embodiment of the present application;

[0041] Figure 5 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0043] Please refer to Figure 1 The embodiment of the present application provides a transmission network fault demarcation method based on a fault model, wherein the fault model includes a feature encoding module, a structured model and a classification model, and the method includes:

[0044] Step 11: Obtain multi-dimensional indicator data of the transmission network;

[0045] Step 12: Input the multi-dimensional indicator data into the feature coding module to obtain network element status data;

[0046] Step 13: inputting the network element state data into the structured model to obtain a high-dimensional representation of the network element state, where the high-dimensional representation of the network element state is a high-dimensional feature representation of the network element state data;

[0047] Step 14: Input the high-dimensional representation of the network element status into the classification model to perform fault detection and determine the detection result.

[0048] The fault delimitation method in the present application is different from the prior art that only relies on alarm data for fault delimitation, but obtains network element status data based on multi-dimensional indicator data of the transmission network; and further, the network element status data is input into the structured model to obtain a high-dimensional feature table of the network element status data: a high-dimensional representation of the network element status, and finally the high-dimensional representation of the network element status is input into the classification model to perform fault detection and determine the detection result. In the embodiment of the present application, it does not rely on expert experience and rules to process fault delimitation, but automatically processes through a fault model that encapsulates a feature coding module, a structured model, and a classification model, which greatly reduces labor costs. Automated processing based on the fault model can also improve the efficiency of fault delimitation. In addition, in the embodiment of the present application, the source of indicator data is more diverse, and the evaluation data obtained through multi-dimensional indicator data can avoid misjudgment caused by a single data source, so that a more accurate delimitation result can be obtained, and the accuracy of fault delimitation is improved.

[0049] Please refer to Figure 2 Optionally, in some embodiments, the method further includes constructing the fault model, including:

[0050] Acquire a training data set, wherein the training data set includes multi-dimensional indicator data within a specified historical time period, and the multi-dimensional indicator data within the specified historical time period at least includes historical fault work order data;

[0051] Performing feature construction on the indicator data of each dimension contained in the training data set respectively, and obtaining a feature vector corresponding to the indicator data of each dimension, wherein the feature vector at least includes a historical fault delimitation label vector corresponding to the historical fault work order data;

[0052] Performing dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain historical network element status data;

[0053] Training the structured model to be trained based on the historical network element status data, and determining the trained structured model;

[0054] The classification model to be trained is trained based on the historical network element status high-dimensional representation data and the historical fault demarcation label vector to determine the trained classification model, wherein the historical network element status high-dimensional representation data is obtained based on inputting the historical network element status data into the trained structured model.

[0055] Optionally, in some embodiments, the multi-dimensional indicator data includes at least one of the following: alarm data, performance data, topology data, fault ticket data, and user behavior characteristic data.

[0056] In the process of building the fault model, the training data set includes multi-dimensional indicator data within a specified historical time. That is to say, in the data acquisition phase of building the fault model, it is necessary to collect alarm data, performance data, topology data, user behavior data and fault work order data within a specified historical time.

[0057] Performance data and user behavior data need to collect data of a certain length of history. In the embodiment of this application, the specified historical time can be set to two weeks in history. This time length can be modified according to the actual situation and the final effect of the model. Based on the actual delimited root cause, first determine the classification label of the output result, such as line failure, dynamic ring reason, equipment failure, customer reason, etc.

[0058] After collection, the fields of each type of data are screened based on the meaning of the fields and expert experience, and basic data preprocessing operations such as deduplication, padding, and format conversion are performed to facilitate feature construction in the next step. Examples of various types of data after field screening and preprocessing are as follows:

[0059] Alarm data:

[0060]

[0061]

[0062] The maximum number of intersections is the maximum number of intersections in the optical cable intersection segment information. This field can reflect the number of times the optical cable segment appears in a batch of alarm links. The maximum intersection ratio is the maximum number of intersections / total number of Los alarms. This field can reflect the proportion of the optical cable segment in a batch of alarm links.

[0063] Performance data:

[0064] Field Name Data Types Optical Power Float flow Float

[0065] Topological data:

[0066] Field Name Data Types A-side network element name Str Z-side NE name Str

[0067] User behavior data, where the fault causes are classified into customer reasons and network reasons.

[0068] Field Name Data Types Network element name Str Failure time Datetime Fault Cause Classification Str

[0069] Fault work order data:

[0070] Field Name Data Types Network element name Str Failure time Datetime Fault Cause Classification Str

[0071] The fault causes are classified into line failure, dynamic environment failure, equipment failure and customer failure, etc.

[0072] Optionally, constructing features for the indicator data of each dimension included in the training data set includes:

[0073] (1) Alarm data:

[0074] For the alarm title and alarm explanation fields in the alarm data, use the following steps to convert them into vector representation:

[0075] Word segmentation: Use Jieba word segmentation tool to segment the field. This will split the title into a list of words.

[0076] Word to Vector (Word2Vec): Use the pre-trained Word2Vec model to obtain the word vector of each word. The Word2Vec model maps each word to a fixed-length vector space that captures the semantic and contextual relationships between words.

[0077] Average pooling: For each word’s word vector, average pooling is used to aggregate them into a fixed-length vector, that is, all word vectors are added together and the average is taken.

[0078] Obtaining vector representation: Finally, the vector representation of the field is obtained. After such an encoding process, the semantic information of the words in the alarm title and alarm explanation can be captured.

[0079] For the city and region fields in the alarm data, Word2Vec is used to encode them into vectors.

[0080] For the fault occurrence time in the alarm data, after parsing, time information such as time period, whether it is a working day, whether it is a holiday, etc. is obtained.

[0081] For Int and Float type data in the alarm data, it can be directly concatenated with other features.

[0082] (2) Performance data:

[0083] Performance data can reflect whether the device has experienced abnormal conditions such as power outages and disconnections. In the event of an abnormality, data will usually drop suddenly, fluctuate greatly, or drop continuously, which can help determine the cause of the fault. For optical power and flow data, the minute-level data collected over the past two weeks constitutes a time series. The following methods are used to construct optical power and flow characteristics:

[0084] Use 10 minutes as the time window length to build sliding window features, calculate the mean, variance, and data change rate of the data in the sliding window to capture the overall trend and volatility of the data. The time window length can be adjusted.

[0085] Calculate the first-order difference of time series data, that is, the difference between the current time point and the previous time point. The differential feature can reflect the speed and trend of data changes.

[0086] (3) User behavior:

[0087] User behavior data can analyze the patterns of faults caused by customers, and reflect whether users regularly shut down their devices, which helps to eliminate the problem that faults caused by users are difficult to define. This application uses a periodic analysis method to analyze the user fault time series data of the past two weeks to obtain information such as whether customers have recently shut down their power regularly and the power-off cycle. The specific steps are as follows:

[0088] The historical data of each customer is constructed into a time series, and the time series data is processed to remove outliers and fill in missing values.

[0089] Use the Fourier transform-based periodicity analysis method to detect whether each customer's time series data has a period and the period value. The Fourier transform converts the data to the frequency domain. If there is a periodic pattern, there will be an obvious peak in the frequency domain, which can be used to determine the period.

[0090] (4) Topological data:

[0091] The topological data reflects the connection relationship between different network elements. When there is an optical cable connection between two network elements, if a line fault occurs, both network elements will report the fault. Therefore, the topological connection relationship between network elements helps to delimit the fault. This application uses the actual connection status of the network element to construct a feature vector.

[0092] (5) Fault work order data:

[0093] The fault ticket data contains the cause of the fault, which is encoded into a binary vector using One-Hot encoding for the label of the classification task.

[0094] Optionally, in the above application embodiment, the data is converted into a vector format to unify the data formats of different types and scales, allowing the interaction of different features and facilitating automated processing, thereby improving the overall analysis and modeling efficiency.

[0095] Further, the feature vector corresponding to the indicator data of each dimension of the feature vector is dimensionally transformed to obtain historical network element status data, which is network element status data converted from multi-dimensional indicator data within a specified historical time. Therefore, the conversion of the historical network element status data is consistent with the conversion step of the network element status data. Therefore, taking the conversion to the network element status data as an example, specifically:

[0096] The network element status data is obtained after dimensional conversion of the various indicator features constructed in the above embodiment, and the dimension is (batchsize, m, n, d), batchsize is the batch size, m is the number of network element states of a network element, n is the time length of a network element state, and d is the number of features of a network element state.

[0097] Please refer to Figure 3 , is the data of a network element in one day: the number of columns of data is d, which is the feature dimension after encoding the characteristics of network element alarms, topology, user behavior, performance and other indicators; there are N = 60 * 24 time slices in one day for minute-level data, and n time slices are selected as a network element state. In this application, n = 5, and the network element state is taken in a sliding window manner, that is, a network element has a total of m = N-n+1 network element states in one day, and the dimension of its network element state data is (m, n, d).

[0098] Optionally, in some embodiments, the training a structured model to be trained based on the historical network element status data to determine the trained structured model includes:

[0099] Input the historical network element state data into the structured model to be trained, and obtain the initial network element state high-dimensional representation and the reconstructed network element state data respectively, wherein the structured model to be trained is a multi-layer structure, the initial network element state high-dimensional representation is the intermediate layer result after the historical network element state data is input into the structured model to be trained, and the reconstructed network element state data is the final layer result of the initial network element state high-dimensional representation after linear transformation and dimensional transformation;

[0100] Based on the loss between the historical network element state data and the reconstructed network element state data, the structured model to be trained is trained to obtain a trained structured model.

[0101] Optionally, in some embodiments, please refer to Figure 2 , the input network element state data (batchsize, m, n, d) is recorded as X, and after operations such as Transformer Encoder (Transformer Encoder) + dimension transformation + linear layer, a high-dimensional feature representation (batchsize, m, embedding_len) is finally obtained, recorded as E. The multi-head attention mechanism is used in the Transformer structure. Each encoding layer contains sub-modules such as multi-head self-attention mechanism and feedforward neural network, which can learn the relationship between different network elements and different indicators, making the output data representation more accurate, which helps to improve the accuracy of subsequent classification. After linear transformation and dimension transformation, E is restored to the shape of the network element state and recorded as X ′, the dimension is the same as that of the input X. It is worth noting that the structured model to be trained is a multi-layer structure, the initial network element state high-dimensional representation is the intermediate layer result after the historical network element state data is input into the structured model to be trained, and the reconstructed network element state data is the final layer result after the initial network element state high-dimensional representation is linearly transformed and dimensionally transformed. After the structured model is trained, during the application process, the network element state data for fault demarcation as required by the input will be used to output the intermediate layer result of the network element state high-dimensional representation; or, based on the input historical network element state data, the historical network element state high-dimensional representation data used to train the classification model will be output.

[0102] The loss of the structured model to be trained is obtained by calculating the MSE loss between the restored network element state (the high-dimensional representation of the initial network element state) and the input network element state data. Loss = MSE (X ′ ,X). The loss value is continuously reduced through the forward propagation-backward propagation process, and finally a pre-trained structured model is obtained.

[0103] Optionally, in some embodiments, the training of a classification model to be trained based on the historical network element state high-dimensional representation data and the historical fault demarcation label vector, and determining the trained classification model includes:

[0104] Inputting the historical network element status high-dimensional representation data and the historical fault demarcation label vector into the classification model to be trained to obtain an initial classification output result;

[0105] Based on the loss between the initial classification output result and the actual classification output result, the classification model to be trained is trained to obtain a trained classification model.

[0106] Optionally, in some embodiments, please refer to Figure 2 , the data input to the classification network is (X, y), X is the high-dimensional features of the network element status obtained by inputting the network element status data into the pre-trained structured large model (in the process of training the model, it is the high-dimensional representation data of the historical network element status), and the dimension is (batchsize, m, embedding_len); y is the fault delimitation label vector obtained by encoding the fault cause into a binary vector using One-Hot encoding in the above step. Since there are five types of fault causes, including line fault, dynamic ring fault, equipment fault, customer fault and no fault, the dimension of y is (batchsize, m, c), where c = 5. The (X, y) data enters a three-layer fully connected network to obtain the classification output y ′ , the dimension is (batchsize,m,c).

[0107] The loss of the fine-tuning part of the classification model in the embodiment of the present application is the loss of the actual fault demarcation result y and the predicted fault demarcation cause y ′ Calculate the cross entropy loss, Loss = MSE (y ′ ,y). The relationship between multi-dimensional indicator data and faults is modeled through the forward propagation-backward propagation training process, and finally a trained classification model is obtained.

[0108] It is worth mentioning that transmission network fault demarcation is essentially a classification task, so the accuracy of the model can also be evaluated by comparing the actual fault causes with the predicted causes.

[0109] This application is the model building process in the embodiment. Figure 3 As shown, firstly, the transmission network data is pre-trained with the network element status data as a token to obtain a structured large model that can output a high-dimensional representation of the network element status and reconstruct the network element status, wherein the network element status data is determined based on the data set; then, a classification network is trained using the high-dimensional representation of the network element status output by the pre-training and the fault label to model the relationship between the network element status and the fault cause. The fault model constructed by the above embodiment includes three modules: feature encoding, pre-trained structured large model, and training classification model. Compared with the current method based on rules or alarm data, it can better model the relationship between various types of network element data and faults, and has a higher accuracy rate.

[0110] Optionally, to further illustrate the transmission network fault delimitation method based on the fault model in this application, please refer to the above-mentioned feature encoding module, pre-trained model, and classification model encapsulated together when used in actual scenarios to form an end-to-end fault model for transmission network fault delimitation. When the multi-dimensional indicator data of a network element is input, the model will automatically perform the three steps of feature encoding, high-dimensional feature generation, and classification result calculation to obtain the predicted fault delimitation result, providing effective data support for operation and maintenance personnel to troubleshoot and solve problems.

[0111] Optionally, in some embodiments, the multi-dimensional indicator data is input into the feature encoding module, and the network element status data is obtained, including:

[0112] The multi-dimensional indicator data is input into the feature coding module, wherein the feature coding module constructs features for the indicator data of each dimension respectively to obtain a feature vector corresponding to the indicator data of each dimension, and performs dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain and output the network element status data.

[0113] It is worth noting that, in the embodiment of the present application, the process of inputting the multi-dimensional indicator data into the feature coding module to obtain the network element status data is the same as the process of obtaining the historical network element status data in building a fault model. The difference lies in that the input data is different. In the process of building the model, it is the multi-dimensional indicator data within the historical time. In the process of applying the fault model, it is the multi-dimensional indicator data corresponding to the time period for which fault demarcation is required. The actual process is the same, so this application will not repeat it.

[0114] It is worth noting that, in some embodiments, the fault label also includes no fault, and the output result also includes: no fault. Since the indicator data in the present application includes data of multiple dimensions, in addition to alarms, there are indicator data of other dimensions. When performing fault delimitation, there may be such a result: although a certain indicator shows that there may be a fault, after an overall analysis of the fault model of the present application, the result is delimited as no fault. Compared with the manual fault delimitation in the prior art, in the present application, even if there is a false alarm caused by indicator deviation, an accurate result of the fault delimitation can be obtained based on the multi-dimensional indicator data, and even if the final result is no fault, because it is based on self-identification of the fault model, the processing speed is fast and there is no waste of manpower.

[0115] In the actual production scenarios of transmission networks, traditional rule-based methods are inefficient. Some customer shutdowns cannot be accurately located, and some devices do not have the ability to report alarms. Using more efficient methods to delimit faults and improve accuracy can help reduce alarms and dispatches, thereby reducing manpower and operation and maintenance costs, and providing data support for improving customer satisfaction.

[0116] The fault delimitation method proposed in this application combines different data such as alarm data, performance data change characteristics, user behavior characteristics, topology, etc. to assist in fault delimitation, and selects different feature construction methods for each type of data. By introducing multi-dimensional data, it can obtain more comprehensive and complete information compared to methods based only on alarm data and rule-based methods, and can achieve higher accuracy in cases of missing alarms and customer reasons.

[0117] In addition, the classification methods of the prior art require a large amount of data with fault labels, and the data acquisition cost is very high. In this application, a fault model based on a structured large model and a classification network is used. In the pre-training stage, only the data of various historical indicators of the existing network need to be obtained, and no labels are required. A relatively small amount of labeled data can be provided during the classification model training to obtain good results, and it is easier to migrate to other professional fault delimitation scenarios. The relationship between various types of data and faults of network elements is modeled using a classification network based on a structured large model to obtain the final fault delimitation result, and an analysis and processing method and feature encoding method for multi-dimensional data are proposed, which improves the accuracy of the output results of the algorithm model and solves the problem of difficulty and low accuracy in transmission network fault delimitation.

[0118] Please refer to Figure 4 The embodiment of the present application further provides a transmission network fault demarcation system 20 based on a fault model, wherein the fault model includes a feature encoding module, a structured model and a classification model, and the system includes:

[0119] The indicator module 21 is used to obtain multi-dimensional indicator data of the transmission network;

[0120] A data module 22, configured to input the multi-dimensional indicator data into the feature coding module to obtain network element status data;

[0121] A feature module 23, configured to input the network element state data into the structured model to obtain a high-dimensional representation of the network element state, wherein the high-dimensional representation of the network element state is a high-dimensional feature representation of the network element state data;

[0122] The output module 24 is used to input the high-dimensional representation of the network element status into the classification model to perform fault detection and determine the detection result.

[0123] Optionally, the data module 22 includes:

[0124] A data submodule is used to input the multi-dimensional indicator data into the feature coding module, wherein the feature coding module constructs features for the indicator data of each dimension respectively to obtain a feature vector corresponding to the indicator data of each dimension, and performs dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain and output the network element status data.

[0125] Optionally, a construction module is further included, for constructing the fault model, and the construction module includes:

[0126] A training set module, used to obtain a training data set, wherein the training data set includes multi-dimensional indicator data within a specified historical time, and the multi-dimensional indicator data within the specified historical time includes at least historical fault work order data;

[0127] A vector module, used to construct features for the indicator data of each dimension contained in the training data set, respectively, to obtain a feature vector corresponding to the indicator data of each dimension, wherein the feature vector at least includes a historical fault delimitation label vector corresponding to the historical fault work order data;

[0128] A network element status module, used to perform dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain historical network element status data;

[0129] A first training module, used to train the structured model to be trained based on the historical network element state data, and determine the trained structured model;

[0130] The second training module is used to train the classification model to be trained based on the historical network element status high-dimensional representation data and the historical fault demarcation label vector to determine the trained classification model, wherein the historical network element status high-dimensional representation data is obtained based on inputting the historical network element status data into the trained structured model.

[0131] Optionally, the first training module includes:

[0132] The first sub-training module is used to input the historical network element state data into the structured model to be trained, and obtain the initial network element state high-dimensional representation and the reconstructed network element state data respectively, wherein the structured model to be trained is a multi-layer structure, the initial network element state high-dimensional representation is the intermediate layer result after the historical network element state data is input into the structured model to be trained, and the reconstructed network element state data is the final layer result of the initial network element state high-dimensional representation after linear transformation and dimensional transformation;

[0133] Based on the loss between the historical network element state data and the reconstructed network element state data, the structured model to be trained is trained to obtain a trained structured model.

[0134] Optionally, the second training module includes:

[0135] A second sub-training module is used to input the historical network element state high-dimensional representation data and the historical fault delimitation label vector into the classification model to be trained to obtain an initial classification output result;

[0136] Based on the loss between the initial classification output result and the actual classification output result, the classification model to be trained is trained to obtain a trained classification model.

[0137] Optionally, the multi-dimensional indicator data includes at least one of the following: alarm data, performance data, topology data, fault ticket data, and user behavior characteristic data.

[0138] The transmission network fault demarcation system 20 based on the fault model provided in the embodiment of the present application can achieve Figure 1 The various processes implemented in the method embodiment achieve the same technical effect and will not be described again here to avoid repetition.

[0139] The present application embodiment provides an electronic device 30, see Figure 5 As shown, Figure 5 This is a principle block diagram of an electronic device 30 of an embodiment of the present application, including a processor 31, a memory 32, and a program or instruction stored in the memory 32 and executable on the processor 31. When the program or instruction is executed by the processor, any step in the transmission network fault delimitation method based on the fault model of the present application is implemented.

[0140] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of an embodiment of a transmission network fault delimitation method based on a fault model as described above is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0141] Computer readable media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0142] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of any of the embodiments of the transmission network fault delimitation method based on the fault model shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0143] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0144] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0145] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a service classification device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0146] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A transmission network fault demarcation method based on a fault model, characterized in that: The fault model includes a feature encoding module, a structured model and a classification model, and the method includes: Obtain multi-dimensional indicator data of the transmission network; Inputting the multi-dimensional indicator data into the feature coding module to obtain network element status data; Inputting the network element state data into the structured model to obtain a high-dimensional representation of the network element state, wherein the high-dimensional representation of the network element state is a high-dimensional feature representation of the network element state data; The high-dimensional representation of the network element state is input into the classification model to perform fault detection and determine the detection result.

2. The method according to claim 1, characterized in that The inputting the multi-dimensional indicator data into the feature coding module to obtain the network element status data comprises: The multi-dimensional indicator data is input into the feature coding module, wherein the feature coding module constructs features for the indicator data of each dimension respectively to obtain a feature vector corresponding to the indicator data of each dimension, and performs dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain and output the network element status data.

3. The method according to claim 1, characterized in that Also included is constructing the fault model, including: Acquire a training data set, wherein the training data set includes multi-dimensional indicator data within a specified historical time period, and the multi-dimensional indicator data within the specified historical time period at least includes historical fault work order data; Performing feature construction on the indicator data of each dimension contained in the training data set respectively, and obtaining a feature vector corresponding to the indicator data of each dimension, wherein the feature vector at least includes a historical fault delimitation label vector corresponding to the historical fault work order data; Performing dimension conversion on the feature vector corresponding to the indicator data of each dimension to obtain historical network element status data; Training the structured model to be trained based on the historical network element status data, and determining the trained structured model; The classification model to be trained is trained based on the historical network element status high-dimensional representation data and the historical fault demarcation label vector to determine the trained classification model, wherein the historical network element status high-dimensional representation data is obtained based on inputting the historical network element status data into the trained structured model.

4. The method according to claim 3, characterized in that The training of the structured model to be trained based on the historical network element status data to determine the trained structured model includes: Input the historical network element state data into the structured model to be trained, and obtain the initial network element state high-dimensional representation and the reconstructed network element state data respectively, wherein the structured model to be trained is a multi-layer structure, the initial network element state high-dimensional representation is the intermediate layer result after the historical network element state data is input into the structured model to be trained, and the reconstructed network element state data is the final layer result of the initial network element state high-dimensional representation after linear transformation and dimensional transformation; Based on the loss between the historical network element state data and the reconstructed network element state data, the structured model to be trained is trained to obtain a trained structured model.

5. The method according to claim 3, characterized in that: The training of the classification model to be trained based on the historical network element state high-dimensional representation data and the historical fault demarcation label vector, and determining the trained classification model includes: Inputting the historical network element status high-dimensional representation data and the historical fault demarcation label vector into the classification model to be trained to obtain an initial classification output result; Based on the loss between the initial classification output result and the actual classification output result, the classification model to be trained is trained to obtain a trained classification model.

6. The method according to claim 1, characterized in that The multi-dimensional indicator data includes at least one of the following: alarm data, performance data, topology data, fault work order data, and user behavior characteristic data.

7. A transmission network fault demarcation system based on a fault model, characterized in that: The fault model includes a feature coding module, a structured model and a classification model, and the system includes: The indicator module is used to obtain multi-dimensional indicator data of the transmission network; A data module, used for inputting the multi-dimensional indicator data into the feature coding module to obtain network element status data; A feature module, used for inputting the network element state data into the structured model to obtain a high-dimensional representation of the network element state, wherein the high-dimensional representation of the network element state is a high-dimensional feature representation of the network element state data; The output module is used to input the high-dimensional representation of the network element status into the classification model to perform fault detection and determine the detection result.

8. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the fault delimitation method for a transmission network based on a fault model are implemented as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: A computer program is stored on a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the transmission network fault demarcation method based on a fault model as described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the fault demarcation method for a transmission network based on a fault model as claimed in any one of claims 1 to 6.

Images