Method, system and medium for power equipment health management based on machine learning

Abstract

The application discloses a machine learning-based power equipment health management method and system and a storage medium, and the method comprises the following steps: acquiring operation condition data of power equipment to be analyzed collected by an intelligent terminal; identifying type parameter information of the equipment and components; calling a fault sample model of the same type in advance according to the identified type parameter information of the equipment and components; pre-processing the operation data; inputting the processed data into a health state perception model based on a neural network time sequence, combining the fault sample model to judge the health state of the equipment data; identifying and extracting the equipment and component models of the input data and obtaining the health state. The application can solve the problems that there are few fault samples of power equipment in the current power grid, the health state perception method of the system is lacking, and the predictive maintenance and repair knowledge points are disordered, and provide new technical support for power grid power equipment fault diagnosis and prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment health state knowledge, and in particular relates to a power equipment health management method and system based on machine learning and a medium. BACKGROUND

[0002] Power equipment has a large and disordered text knowledge, and different types of equipment face a variety of fault problems, involving a wide range. Usually, the fault judgment of power grid equipment and the operation and maintenance knowledge of the equipment rely on expert experience, and maintenance personnel also need a large amount of professional training before they can handle equipment abnormal conditions. And the traditional power equipment related knowledge base usually only contains the most original text information, without a mature and comprehensive fault library. Therefore, the current power equipment health degree is faced with high labor cost, fragmented work condition data, and one-sided and single fault sample technical problems. SUMMARY

[0003] The power equipment health management method and system based on machine learning and the medium provided by the present application can solve the technical problems in the background art.

[0004] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0005] A power equipment health management method based on machine learning, comprising the following steps: acquiring operation condition data of a power equipment to be analyzed collected by an intelligent terminal;

[0006] Identifying the type parameter information of the equipment and components;

[0007] According to the identified equipment component type parameter information, a pre-constructed fault sample model of the same type is called, and the sample model is obtained by fusing multiple algorithms;

[0008] Pretreating the operation data;

[0009] Inputting the processed data into a health state perception algorithm based on neural network time series, and combining the fault sample model to judge the health state of the equipment data;

[0010] If the judgment result reaches a preset threshold, the real-time condition data is labeled as the health state corresponding to the threshold;

[0011] Identifying and extracting the equipment and component model of the input data and the obtained health state.

[0012] Further comprising:

[0013] According to the identified power equipment name and attribute and the corresponding health state, searching a pre-constructed knowledge graph in a search module, and searching for the corresponding maintenance strategy in the classified knowledge graph.

[0014] Further, the fault sample model includes three kinds, and the training steps are as follows,

[0015] (1) The convolutional neural network training step of obtaining actual samples is as follows:

[0016] The existing operation data of power equipment of known fault types are collected, the collected fault sample data are preprocessed, the fault sample data are subjected to wavelet transform smoothing and denoising processing, and the denoised data are input into different algorithm models to convert the fault data into fault samples that can be expressed by features.

[0017] (2) The logical steps of obtaining a migration fault sample are as follows:

[0018] The features and feature structures of the existing fault samples of the same type of equipment or components are extracted;

[0019] The same features of the same type of equipment are kept unchanged, and specific features and corresponding parameters are retrained according to different working conditions;

[0020] The retrained specific features and parameters are added to the original same features to construct a new fault sample, and in order to reduce the influence of different units of the extracted features on the classification results, all the features are normalized to obtain a migration sample model.

[0021] (3) The logical steps of simulating a theoretical fault model are as follows:

[0022] The theoretical fault characteristics of equipment or components without fault records are determined;

[0023] The signal parameter set meeting the requirements is constructed according to the theoretical fault characteristics, the parameter set is input into the model constructed by the basic function signal, the simulated signal meeting the fault characteristics is output, and the theoretical fault sample is obtained.

[0024] Further, the specific health state recognition model logical steps are as follows:

[0025] The operation condition data are preprocessed and subjected to denoising and smoothing processing;

[0026] A nonlinear autoregressive neural network time series model is input, a feedforward neural network is used to reduce the simulation error, a Bayesian regularization algorithm is used as the training function to prevent overfitting, and a closed-loop network is used for multi-step prediction, that is, the data in the future multi-step time are predicted according to the historical data;

[0027] The historical data and the predicted data are jointly input into the corresponding fault sample model, when the comparison and classification results reach a preset first threshold value, it is judged that the equipment or component has determined to occur the fault; when the comparison and classification results reach a preset second threshold value, it is judged that the equipment or component has a possibility of occurring the fault in the future period of time;

[0028] The multi-step prediction is characterized in that the multi-step refers to a plurality of time steps, and is flexibly set according to the type of the device predicted;

[0029] The first threshold refers to historical data of the working condition being judged as meeting the characteristics of the fault sample model after the historical value of the working condition data is input into the fault sample model;

[0030] The second threshold refers to predicted data being judged as meeting the characteristics of the fault sample model after the predicted value of the working condition data is input into the fault sample model.

[0031] Further, the construction process of the knowledge graph is as follows:

[0032] A small amount of manual annotation is performed on the knowledge point text set, the general name of the device or component in the knowledge point text is annotated as the first type of entity, and the attribute name of the power device is annotated as the second type of entity, to construct an entity word table;

[0033] Unannotated text is input into a KNN clustering model, and the most representative text is automatically annotated with the first and second types of entities by using a preset entity word table;

[0034] The automatically annotated representative text and the manually annotated text are used as features, and all entities contained in the knowledge point text are annotated according to a CRF entity recognition algorithm.

[0035] On the other hand, the application also discloses a power grid power device health management system based on machine learning, and the corresponding system is as follows:

[0036] The first part is a state sensing unit:

[0037] The state sensing unit includes a health state sensing module and a fault sample module.

[0038] The state sensing unit is used for acquiring and identifying the running condition data related to the power device, and sensing the health state of the device or component by analyzing and processing the data.

[0039] The fault sample module includes a pre-constructed actual fault sample, a migrated fault sample, and a theoretical fault sample.

[0040] The second part is a strategy unit:

[0041] The strategy unit includes a search module and a fault knowledge text module.

[0042] The search module can perform text segmentation retrieval on the device and component model, the fault type, the fault reason, the fault corresponding measure, and the predictive fault measure.

[0043] The search content is pre-classified knowledge base content, and the knowledge graph refers to a plurality of power grid power device fault knowledge texts collected and annotated according to a named entity recognition algorithm.

[0044] In a third aspect, the present application also discloses a computer readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the above method.

[0045] According to the above technical solution, the method, system and storage medium for health management of power equipment based on machine learning can construct a structured knowledge model by collecting existing power equipment information, train a fault sample library by existing fault sample data, and judge the health state of equipment or components by establishing a state perception model based on working condition data and knowledge points, thereby providing comprehensive and systematic technical support for the field of power equipment knowledge.

[0046] The present application not only includes various types of equipment fault sample models, but also includes a training method for fault sample models, thereby providing a relatively complete fault sample library for power equipment.

[0047] Compared with the traditional recognition model which can only judge whether the equipment is currently faulty, the health state perception model of the present application can predict the trend of the equipment data and judge whether it meets the characteristics of the fault sample, thereby solving the problem of difficult predictive maintenance of power equipment.

[0048] The present application classifies power knowledge point texts by extracting the first entity equipment or component name and the second entity attribute name, forms a systematic knowledge graph, facilitates searching and classification, and can find corresponding maintenance or repair knowledge for different types of power equipment.

[0049] Specifically, the method for health management of power equipment based on machine learning provided by the present application first trains existing power grid power equipment and component fault data, simulates abnormal data and migrates similar faults, obtains a comprehensive and updatable fault sample library, then compares real-time equipment operation data with the fault sample library to obtain the current health state of the equipment and components, and finally provides corresponding predictive maintenance strategies according to the obtained health state of the equipment and components, thereby solving the problems of large and disordered power grid power equipment data and difficult predictive maintenance, and providing a new technical support for power grid power equipment fault diagnosis and prediction. For example, it can be used for equipment or component fault diagnosis, equipment health state evaluation and equipment fault repair strategy. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 is a flowchart of the method of the present application;

[0051] Figure 2 is a flowchart of the method of the present application;

[0052] Figure 3 A specific step schematic diagram of the named entity recognition algorithm of the present application;

[0053] Figure 4 A training flowchart of the three fault samples of the present application;

[0054] Figure 5 A specific state perception calculation model example provided by the embodiment of the present application;

[0055] Figure 6 A system schematic diagram for the power equipment knowledge base provided by the embodiment of the present application. DETAILED DESCRIPTION

[0056] To make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be described clearly and completely below in combination with the drawings in the embodiment of the present application. Obviously, the described embodiment is a part of the embodiments of the present application, rather than all the embodiments.

[0057] As shown in Figure 1 the power equipment health management method based on machine learning described in the embodiment, first, a knowledge graph is constructed according to a named entity recognition algorithm, then three fault sample models are constructed according to equipment names and attributes, then a health state perception model is constructed for each fault sample, and finally, an open source retrieval system is added to the system;

[0058] which includes the following steps,

[0059] acquiring operation condition data of a power equipment to be analyzed collected by an intelligent terminal;

[0060] identifying type parameter information of the equipment and components;

[0061] According to the identified equipment component type parameter information, the same type of fault sample model constructed in advance is called;

[0062] preprocessing the operation data;

[0063] inputting the processed data into a health state perception model based on neural network time series, and combining the fault sample model to judge the health state of the equipment data;

[0064] If the judgment result reaches a preset threshold value, the real-time condition data is marked as the health state corresponding to the threshold value;

[0065] identifying and extracting the equipment and component model of the input data and the obtained health state;

[0066] Further comprising searching the pre-constructed knowledge graph according to the identified power equipment name and attribute and the corresponding health state, and searching for the corresponding maintenance strategy in the classified knowledge graph.

[0067] The following is a specific description:

[0068] As Figure 2 The machine learning-based power equipment health management method of the embodiment includes:

[0069] S101: Obtain power equipment knowledge point text, and label all first-class entities and second-class entities and corresponding information contained in the text according to a named entity recognition algorithm, wherein the first-class entities refer to equipment or component standard names, and the second-class entities refer to equipment attribute names;

[0070] The attribute names include four aspects: electrical properties, physical and chemical properties, working environment properties, and appearance properties. Specific examples are “current”, “load”, “temperature”, “deformation”, “damage”, and “thunder”. The entity naming annotation format of the attribute names is set as: [attribute category: attribute name], such as: [electrical: current], [physical and chemical: temperature], [appearance: deformation], and [environment: thunder].

[0071] The power equipment knowledge point text refers to open-source unstructured or structured power equipment knowledge points, including general names, standard names, component names, attribute knowledge, and fault knowledge of power equipment in the power industry.

[0072] The equipment or component standard name refers to a specific power equipment name, including common names and general names (for example, 110KV oil-immersed transformer and 35KV switch cabinet).

[0073] S102: Collect corresponding power equipment or component fault working condition data for the first and second-class entities, and construct a fault sample library with entity name annotation.

[0074] The fault working condition data can be running data generated when a device has a certain attribute fault, or running data generated when a certain component of a device has a running fault.

[0075] S103: According to all sample models of the fault sample library, establish a state perception model of the same type of equipment or component corresponding to the sample.

[0076] S104: Store the power equipment knowledge point text, fault sample, and state perception method corresponding to the first and second-class entities according to the named entity classification, and construct the knowledge base.

[0077] The Figure 3 The specific steps of the named entity recognition algorithm include:

[0078] S201: pre-labeling original information set; a small amount of information set is pre-labeled manually, the standard name of the power equipment or component is labeled as the first class entity, and the attribute name of the power equipment is labeled as the second class entity;

[0079] S202: based on the first and second class entities labeled manually, inputting the unlabeled set into a KNN clustering model, selecting the most representative text to automatically label all entities, and extracting features;

[0080] S203: taking the automatically labeled representative text and the manually labeled text as features, performing normalization processing, inputting the CRF entity recognition model, and completing automatic sequence labeling of all texts.

[0081] Figure 4 A flowchart of three kinds of fault samples provided by the embodiment of the application is shown, including:

[0082] Actual samples, migration samples and theoretical samples;

[0083] The following is a construction embodiment of a switch cabinet switch gate deformation and damage identification model according to an actual sample flow:

[0084] The switch cabinet switch gate damage identification model based on the gradient descent method-convolutional neural network includes a convolutional layer, a down-sampling layer and a fully connected layer;

[0085] The first convolutional layer extracts features of a plurality of damaged switch valve grayscale images;

[0086] In order to reduce the influence of different features on the classification results, all features are normalized, and the extracted features are w1, w2, w3, w4, w5, w6,..., wn, all features form a set W, W includes the features of the switch valve grayscale image, such as centroid, contrast, entropy, inverse difference moment, inertia moment and box dimension, and after normalization, each feature:

[0087]

[0088] x1, x2 are specified scaling intervals [x1, x2];

[0089] The first pooling layer of the second layer removes a large number of edge features extracted, reduces parameters and prevents overfitting;

[0090] After the picture data is processed by the convolutional layer and the second pooling layer of the activation function RE LU to the maximum extent, the training result is finally output by the fully connected layer, and the feature is that the bias function of each layer is calculated by the batch gradient descent method, and the recognition rate is improved by continuously updating the parameters.

[0091] After the optimization of the parameters, a trained switch cabinet switch damage identification model for judging whether the switch exists damage phenomenon is obtained.

[0092] The trained switch cabinet switch damage identification model is obtained, and then the collected switch cabinet switch picture is input into the damage identification model to obtain the result of whether the switch exists damage phenomenon.

[0093] Attached Figure 5 An example of a specific state-aware computing model provided by the embodiment of the application is shown, which includes:

[0094] S401: Identify the first and second types of entities to be input into real-time power equipment working condition data;

[0095] S402: Denoising the real-time working condition data, extracting the input working condition data features after inputting the working condition data into the feature extraction model based on the machine learning algorithm, and calling the same type of equipment or component fault features according to the equipment or component general name in the fault sample library;

[0096] S403: Normalize all features and compare the similarity and compare with the threshold value;

[0097] S404: Output the health status corresponding to the first threshold value and the second threshold value.

[0098] The following is an actual example of transformer winding temperature anomaly awareness

[0099] Obtain the type and model of the transformer and the various data monitored by the intelligent terminal, and input the data after preprocessing into NAR (nonlinear neural network) for multi-step prediction; obtain the winding temperature anomaly sample model of the transformer based on the SVR algorithm training, input the historical data and the winding temperature data after multi-step prediction into the SVR model which can be optimized, judge whether the winding temperature reaches the first and second threshold values, and no warning and maintenance strategy is needed. If the winding temperature state reaches the second threshold value, the winding will reach an abnormal state in a certain period of time, and a predictive maintenance strategy is needed.

[0100] The training process of the optimized Gaussian kernel SVR temperature model:

[0101] The training process of the optimized Gaussian kernel SVR temperature model includes:

[0102] The multi-dimensional historical monitoring data of the transformer are acquired, the data including transformer oil temperature, ambient temperature, current and voltage value, power and load, the data having correlation with transformer winding hot spot temperature, the nonlinearity of the optimized Gaussian kernel support vector machine regression algorithm, and the regression relationship between multiple features and winding temperature can be better described; all dimensions of data are randomly combined and input into the optimized nonlinear Gaussian kernel SVR model, wherein the semi-positive definite kernel function is set as:

[0103]

[0104] The Gaussian kernel function is used to find the optimal function f (temperature value).

[0105] By comparing the RMSE, it is found that the oil temperature, the load rate, the indoor temperature and humidity, and the reactive power are the most influential characteristic factors on the winding temperature anomaly, after the characteristic factors are determined, each characteristic parameter under the optimal recall rate is found through 1000 iterations, and the winding temperature anomaly judgment model of the transformer is constructed.

[0106] In addition, the application can perform text segmentation on fault information, extract keywords through the established knowledge base information, and perform fault classification and association with knowledge base information. The application uses a traditional relational database MYSQL to store fault information and knowledge base information, and finally stores the data in a storage medium ROM (Read-Only Memory). When performing fault strategy analysis, the stored associated data is loaded into the storage medium RAM (RANDOM Access Memory) through database indexing, so that the corresponding device knowledge base data is retrieved through the fault information.

[0107] The following is a search example of the search module of the application for the knowledge points of an inverter cabinet of a certain power grid

[0108] The input is the model of a certain inverter cabinet, the fault type is selected as battery fault-voltage low, and the search is performed, the possible fault reason is obtained as unbalanced battery voltage, overcharging, and large current charging, and the corresponding maintenance measures are to try multiple shallow discharges, and prevent large current high voltage power charging.

[0109] Figure 5 A system schematic diagram for the power equipment knowledge base provided by the embodiment of the application is provided, S501: a state perception unit;

[0110] The health state perception module and the fault sample module are included.

[0111] The health state perception module is used to acquire and identify the operation condition data related to the power equipment, and the health state of the equipment or component is perceived by analyzing and processing the data.

[0112] The fault sample module includes pre-constructed actual fault samples, migrated fault samples, and theoretical fault samples.

[0113] S502: a policy unit;

[0114] The search module, the fault knowledge text module

[0115] The search module can perform text segmentation retrieval on equipment and component models, fault types, fault causes, fault countermeasures, and predictive fault measures.

[0116] The search content is pre-classified knowledge base content, and the knowledge graph refers to a plurality of power grid power equipment fault knowledge texts collected according to a named entity recognition algorithm.

[0117] In summary, the power equipment health management method based on machine learning provided by the embodiment of the application first trains the existing power grid power equipment and component fault data, simulates abnormal data and migrates the same type of faults, obtains a comprehensive and updatable fault sample library, then compares the real-time obtained equipment operation data with the fault sample library to obtain the current health status of the equipment and components, finally provides corresponding predictive maintenance strategies according to the obtained health status of the equipment and components, can solve the problems of large and disordered power grid power equipment data and difficulty in predictive maintenance, and provides a new technical thought for power grid power equipment fault diagnosis and prediction. For example, it can be used for equipment or component fault diagnosis, equipment health status evaluation, and equipment fault maintenance strategy.

[0118] In another aspect, the application further discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to make the processor execute the steps of any of the above methods.

[0119] In another aspect, the application further discloses a computer device, which comprises a memory and a processor, and the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of any of the above methods.

[0120] In another embodiment provided in the application, a computer program product containing instructions is also provided, which makes a computer execute the steps of any of the above methods when the computer program product is run on the computer.

[0121] It can be understood that the system provided by the embodiment of the application corresponds to the method provided by the embodiment of the application, and the explanation, examples and beneficial effects of the related content can refer to the corresponding part in the above method.

[0122] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory.

[0123] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for power equipment health management based on machine learning, characterized in that Comprise the following steps, Obtain the operation condition data of the power equipment to be analyzed collected by the intelligent terminal; Identify the type parameter information of the equipment and components; According to the identified equipment component type parameter information, call the previously constructed fault sample model of the same type; Preprocess the operation data; Input the processed data into the health state perception model based on neural network time series, and judge the health state of the equipment data combined with the fault sample model; If the judgment result reaches the preset threshold value, the operation condition data to be analyzed is labeled as the health state corresponding to the threshold value; Identify and extract the equipment and component model of the input data and the health state obtained by judgment; The training steps of the fault sample model include, a. The actual sample training step is: Collect the existing operation data of the power equipment with known fault types, preprocess the collected fault sample data, and smooth the denoising processing of the fault sample data; input the denoised data into the model of multiple algorithm fusion to train the fault sample with feature representation; b. The logic steps of obtaining the migration fault sample are: Extract the features and feature structures of the existing fault samples of the same type of equipment or components; Keep the same features of the same type of equipment unchanged, and retrain specific features and corresponding parameters according to different working conditions; Add the retrained specific features and parameters to the original same features to construct a new fault sample, and normalize all features as a migration sample model; c. The logic steps of simulating the theoretical fault model are: Determine the theoretical fault characteristics of the equipment or components without fault records; According to the theoretical fault characteristics, construct a specified standard signal parameter set, input the parameter set into the model constructed by the basic function signal, output the simulated signal conforming to the fault characteristics, and obtain the theoretical fault sample. 2.The machine learning based power equipment health management method of claim 1, wherein: It also includes searching the pre-constructed knowledge graph according to the identified power equipment name and attributes and the corresponding health state, and searching for the corresponding maintenance strategy in the classified knowledge graph. 3.The machine learning based power equipment health management method of claim 1, wherein: The health state perception model construction steps are as follows, Preprocess the operation condition data and perform denoising and smoothing processing; Input the nonlinear autoregressive neural network time series model, reduce the simulation error by using feedforward neural network, the training function is Bayesian regularization algorithm to prevent overfitting, and closed loop network is used for multi-step prediction, that is, according to historical data, predict the data in the future multiple time steps; Input the historical data and predicted data into the corresponding fault sample model, when the comparison and classification results reach the preset first threshold value, it is judged that the equipment or component has occurred this type of fault; when the comparison and classification results reach the preset second threshold value, it is judged that the equipment or component has the possibility of occurring this type of fault in the future; Wherein, the multi-step prediction is that the multi-step refers to multiple time steps, and is set according to the predicted equipment type; The first threshold value refers to that after the historical value of the working condition data is input into the fault sample model, the historical data is judged to conform to the characteristics of the fault sample model; The second threshold value refers to that after the predicted value of the working condition data is input into the fault sample model, the predicted data is judged to conform to the characteristics of the fault sample model. 4.The machine learning based power equipment health management method of claim 2, wherein: The construction process of the knowledge graph comprises, The knowledge point text set is manually annotated, the general name of the equipment or component in the knowledge point text is annotated as the first entity, and the attribute name of the power equipment is annotated as the second entity, so as to construct an entity word table; Unannotated text is input into a KNN clustering model, and the most representative text is automatically annotated with the first and second entities by using a preset entity word table; The automatically annotated representative text and the manually annotated text are taken as features, and all entities contained in the knowledge point text are annotated according to a CRF entity recognition algorithm. 5.The machine learning based power equipment health management method of claim 1, wherein: The fault sample model further comprises a winding temperature anomaly perception model for a certain type of transformer, and the construction steps are as follows: The type and model of the transformer and the various data monitored by the intelligent terminal are obtained, and all the data are preprocessed and input into a nonlinear neural network for multi-step prediction; The winding temperature anomaly perception model of the transformer based on the SVR algorithm is obtained, the historical data and the winding temperature data after multi-step prediction are input into the optimizable SVR model, it is judged that the winding temperature does not reach the first and second thresholds, and no early warning and maintenance strategy is needed; if the winding temperature state reaches the second threshold, it is warned that the winding will reach an anomaly in a certain period of time, and a predictive maintenance strategy is needed; The training process of the optimizable SVR model comprises: Multi-dimensional historical monitoring data of the transformer are obtained, the data comprising the transformer oil temperature, the ambient temperature, the current and voltage values, the power, and the load, all the data in various dimensions are randomly combined and input into the optimizable optimizable SVR model, wherein the semi-positive definite kernel function is set as: Using a Gaussian kernel function can find the optimal function i.e. temperature values; By comparing the RMSE, it is found that the oil temperature, the load rate, the indoor temperature and humidity, and the reactive power are the most influential characteristic factors on the winding temperature anomaly, after the characteristic factors are determined, the characteristic parameters under the optimal recall rate are found again through 1000 iterations, and the winding temperature anomaly perception model of the transformer of this type is constructed. 6.A power grid power equipment health management system based on machine learning, used to implement the machine learning-based power equipment health management method of any one of claims 1-5, characterized in that: The state perception unit comprises: The state perception unit comprises a health state perception module and a fault sample module; The health state perception module is used for acquiring and identifying the running condition data related to the power equipment, and perceiving the health state of the equipment or component by analyzing and processing the data; The fault sample module comprises actual fault samples, migrated fault samples, and theoretical fault samples.

7. The machine learning based power grid power equipment health management system of claim 6, wherein: The strategy unit further comprises a search module and a fault knowledge text module; The search module can perform text segmentation retrieval on the equipment and component model, the fault type, the fault cause, the corresponding measures for the fault, and the predictive fault measures; The search content is the pre-classified knowledge base content, and the knowledge graph refers to a plurality of power grid power equipment fault knowledge texts collected and classified according to a named entity recognition algorithm. 8.A computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to make the processor perform the steps of the method according to any one of claims 1 to 5. ​

Images