Method and system for constructing device health profile library based on unified semantic representation of data

Abstract

The application discloses a device health record library construction method and system based on data unified semantic representation, and comprises the following steps: inputting device state data of different sources respectively; extracting health features from the device state data, and establishing device state health sub-records of each transformer substation device respectively; uniformly representing and fusing each device state health sub-record into a complete device state health record; representing the health features in the complete device state health record into feature vectors by vectorization; and constructing a classification model based on the feature vectors to realize classification and evaluation of the device state of the transformer substation. The application can solve the problem that, in the prior art, the multi-source heterogeneous data is not fully fused and uniformly represented, leading to complex construction of the transformer substation device health record library and low accuracy, and the application has less dependence on experts and experience data, and automatically constructs the record library in a data-driven form combined with a natural language processing method, so that the record library is more easily connected with a device analysis method.

Description

Technical Field

[0001] This invention relates to power grid equipment assessment technology, specifically to a method and system for constructing an equipment health record database based on unified semantic representation of data. Background Technology

[0002] With the rapid development of the domestic power grid, the number of power grid equipment has increased exponentially. The traditional scheduled maintenance model fails to allocate maintenance cycles based on the condition of each piece of equipment, leading to over-maintenance of equipment in good condition and under-maintenance of equipment in poor condition. Adopting refined, standardized, and more scientific maintenance strategies for power grid equipment operation and maintenance, while minimizing resource consumption under the premise of ensuring continuous, stable, safe, and reliable operation of the power grid, has become an inevitable trend in the development of power grid enterprise equipment management. Currently, equipment monitoring stations have accumulated a large amount of equipment data, including infrared monitoring data, online monitoring data, and defect data. This data is currently scattered across various systems and offline tables at the grassroots level, making data exchange, sharing, and timely updates between different departments difficult and hindering a comprehensive evaluation of equipment condition. Utilizing existing equipment historical defects, repair and testing records, condition evaluations, historical infrared thermograms of offline equipment, ledger data, and online monitoring data to create a multi-dimensional, integrated equipment health record, intelligently linked with maintenance plans, enables differentiated operation and maintenance, improving the quality and efficiency of operation and maintenance. Substation anomaly information exists in various data types with different structures, posing a challenge to direct status classification of equipment. Traditionally, text classification methods are used, employing different status description texts as input to train classification models. However, this approach suffers from drawbacks such as data redundancy and inconsistencies between different equipment data tables. It often requires manual, experience-based fusion of different databases, significantly hindering the generation and analysis of health records and resulting in low efficiency. Therefore, how to integrate multi-source heterogeneous data and perform unified semantic representation to construct health records for power grid substation equipment has become an urgent technical problem to be solved. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method and system for constructing a device health record database based on unified semantic representation of data, addressing the aforementioned problems in the prior art. This invention aims to solve the problem that the construction of substation device health record databases is complex and inaccurate due to the lack of sufficient integration and unified representation of multi-source heterogeneous data in the prior art. This invention relies less on expert and experience data, and mainly uses a data-driven approach combined with natural language processing methods to automatically construct the database. The resulting health record database is also easier to interface with equipment analysis methods.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] A method for constructing a device health archive based on unified semantic representation of data includes:

[0006] S1, input device status data from different sources respectively;

[0007] S2, extract health characteristics from device status data;

[0008] S3, establish equipment status health sub-files for each substation device based on health characteristics;

[0009] S4, perform unified semantic representation on each device status health sub-file and merge them into a complete device status health file;

[0010] S5, vectorizes the health features in the complete equipment status health record into feature vectors;

[0011] S6. Construct a feature vector-based classification model to classify and evaluate the status of substation equipment.

[0012] Optionally, the sources in step S1 include the substation production management system, the substation online monitoring system, and the substation equipment ledger. The equipment status data from different sources include main transformer equipment data, main transformer evaluation data, main transformer alarm data, main transformer repair and testing data, main transformer defect data, and substation data.

[0013] Optionally, the health characteristics in step S2 include direct characteristics and indirect characteristics. Direct characteristics refer to the static data characteristics of the substation equipment, which are data characteristics that do not change over time. Indirect characteristics refer to the time-series monitoring data corresponding to the substation equipment, which change over time.

[0014] Optionally, the direct features include equipment ID, equipment name, equipment type, voltage level, equipment model, manufacturer, commissioning date, manufacturing date, substation ID, substation name, substation voltage level, city, district / county, substation type, and production management system evaluation; the indirect features include defect frequency, defect trend, defect nature, defect elimination status, acceptance evaluation, hidden danger status, deduction value, alarm frequency, alarm trend, alarm level, alarm type, and test conclusion.

[0015] Optionally, step S4 includes:

[0016] S4.1 Extract semantic information from the health status sub-files of each device;

[0017] S4.2 Align and link the semantic information of each device status health sub-file to form a unified semantic representation system for device status analysis;

[0018] S4.3 integrates all equipment status health sub-files of the unified semantic representation system for equipment status analysis to establish a complete equipment status health file for classifying and assessing the health status of substation equipment.

[0019] Optionally, the alignment of device state entities in step S4.2 refers to using a matching function to map and align entities from different device state sub-archives at the vector space level. The matching function is a multi-layer neural network, whose inputs are labeled device state entities and the device state entity to be judged, and whose output is a vector in the vector space. The matching function is designed and trained using a manually labeled entity alignment dataset S to ensure semantic consistency and minimize mapping loss from data from different sources. The expression for calculating semantic consistency and mapping loss is as follows:

[0020] L = L merge (CD1,CD2)+L base (ρ,S),

[0021] In the above formula, L represents semantic consistency and mapping loss, L merge (CD1, CD2) represents the sum of the absolute differences between each term in distributed vector representations V1 and V2, where distributed vector representation V1 is the distributed vector representation of the manually labeled entity alignment dataset S based on the device status health sub-file CD1, and distributed vector representation V2 is the distributed vector representation of the manually labeled entity alignment dataset S based on the device status health sub-file CD2. base (ρ,S) represents the probability that the matching function ρ maps the features of the device status health sub-files CD1 and CD2 to the correct device status entities based on the manually labeled entity alignment dataset S.

[0022] Optionally, in step S5, when vectorizing the health features in the complete device status health record into feature vectors, the indirect features are normalized to the [0,1] interval using a normalization mapping function; the direct features are encoded using multi-head sequence encoding: first, the feature values ​​of the direct features are sorted and a unique queue is constructed, and each feature value in the unique queue is mapped to a different sequence number; then, each feature value in the unique queue is encoded as the multi-head sequence encoding result by dividing its corresponding sequence number by the sum of the sequence numbers.

[0023] Optionally, in step S6, the classification model constructed based on feature vectors is a multilayer perceptron classification network. The multilayer perceptron classification network includes an input layer, a hidden layer, and an output layer connected in sequence. When training the multilayer perceptron classification network, the feature vector is used as the input of the multilayer perceptron classification network, and the output of the multilayer perceptron classification network is compared with the known labels. The weights of the network parameters in the multilayer perceptron classification network are adjusted accordingly until the maximum number of allowed iterations is reached or the accuracy of the multilayer perceptron classification network is greater than a set value.

[0024] Furthermore, the present invention also provides a device health record construction system based on unified semantic representation of data, including a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the steps of the device health record construction method based on unified semantic representation of data.

[0025] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program for being programmed or configured by a microprocessor to perform the steps of the device health record construction method based on unified semantic representation of data.

[0026] Compared with existing technologies, the present invention has the following advantages: The present invention includes inputting equipment status data from different sources; extracting health features from the equipment status data; establishing equipment status health sub-files for each substation equipment based on the health features; performing unified semantic representation on each equipment status health sub-file and merging them into a complete equipment status health file; vectorizing the health features in the complete equipment status health file into feature vectors; and constructing a classification model based on feature vectors to achieve the classification and evaluation of substation equipment status. The present invention can solve the problem in existing technologies where the construction of substation equipment health file databases is complex and the accuracy is low due to the lack of sufficient integration and unified representation of multi-source heterogeneous data. The present invention relies less on expert and experience data, and mainly uses a data-driven approach combined with natural language processing methods to automatically construct the database. The resulting health file database is also easier to interface with equipment analysis methods. This invention processes equipment status data from different sources separately, identifying semantic information for each type of data to establish sub-files of the equipment health record. The semantic information from these sub-files is then aligned and linked to form a unified semantic representation system for equipment status analysis, establishing a complete equipment health record library. A distributed learning method is then used to transform the equipment status information into vectors, combined with an autoencoder for feature compression and extraction, and a fully connected deep neural network is used for status classification. Based on the status classification results, the equipment health level is calculated, and the substation equipment health record library is updated. Compared to traditional methods, this invention, based on a unified semantic representation construction method for multi-source heterogeneous data, fully considers multi-source monitoring data related to substation equipment, enabling more comprehensive, accurate, dynamic, and comprehensive monitoring of substation equipment health, providing data support for differentiated equipment operation and maintenance. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention. Detailed Implementation

[0028] like Figure 1 As shown, the method for constructing a device health record database based on unified semantic representation of data in this embodiment includes:

[0029] S1, input device status data from different sources respectively;

[0030] S2, extract health characteristics from device status data;

[0031] S3, establish equipment status health sub-files for each substation device based on health characteristics;

[0032] S4, perform unified semantic representation on each device status health sub-file and merge them into a complete device status health file;

[0033] S5, vectorizes the health features in the complete equipment status health record into feature vectors;

[0034] S6. Construct a feature vector-based classification model to classify and evaluate the status of substation equipment.

[0035] In this embodiment, the data sources in step S1 include the substation production management system, the substation online monitoring system, and the substation equipment ledger. The equipment status data from different sources include main transformer equipment data, main transformer evaluation data, main transformer alarm data, main transformer repair and testing data, main transformer defect data, and substation data. It should be noted that the substation production management system, the substation online monitoring system, and the substation equipment ledger are all well-known data sources for substations and will not be detailed here. The main transformer equipment data, main transformer evaluation data, main transformer alarm data, main transformer repair and testing data, main transformer defect data, and substation data are also basic data for substation management. The main transformer refers to the main transformer, which is also a basic piece of substation equipment and will not be detailed here.

[0036] In this embodiment, the health characteristics in step S2 include direct characteristics and indirect characteristics. Direct characteristics refer to the static data characteristics of the substation equipment, which are data characteristics that do not change over time. Indirect characteristics refer to the time-series monitoring data corresponding to the substation equipment, which change over time.

[0037] In this embodiment, direct features include equipment ID, equipment name, equipment type, voltage level, equipment model, manufacturer, commissioning date, manufacturing date, substation ID, substation name, substation voltage level, city, district / county, substation type, and production management system evaluation (PMS evaluation); indirect features include defect frequency, defect trend, defect nature, defect elimination status, acceptance evaluation, hidden danger status, deduction value, alarm frequency, alarm trend, alarm level, alarm type, and test conclusion. Direct and indirect features together constitute 27 health features.

[0038] In step S3, when establishing equipment status health sub-files for each substation device based on health features, the equipment status health sub-file is a specific source of the equipment health file. The equipment status health sub-file is also represented by 27 health feature values, expressed using data types such as characters, numbers, and text, forming structured feature data. For example, the equipment status health sub-file of a certain substation equipment is: "dca1d13e31ff8080814d7c4ca3014ddca0b38940c0, Dinggong Substation #1 Main Transformer A Phase, Main Transformer, AC 500kV, ODFS-334000 / 500, Shandong Electric Power Equipment Co., Ltd., 2015 / 6 / 28, 2014 / 11 / 1, 500kV, Dinggong Substation, d748881f91ff8080814d7c4ca3014dd747785d2d4c, 35, Changsha, Changsha, 4, 0.19, 0.20, 0.28, 0.33, 0.33, 0, 0.06, 0, 0, 0, 0, 0.33".

[0039] In this embodiment, step S4 includes:

[0040] S4.1 Extract semantic information from the health status sub-files of each device; it should be noted that the semantic information can be extracted according to the existing semantic information extraction model as needed. In this embodiment, the semantic information is only used and does not involve the improvement of the semantic information extraction method. Therefore, the detailed implementation of extracting semantic information will not be described in detail.

[0041] S4.2 aligns and links the semantic information of each equipment status health sub-file into equipment status entities, forming a unified semantic representation system for equipment status analysis. An equipment status entity refers to a text keyword representing a characteristic of equipment health status. For example, "Dinggong Substation #1 Main Transformer A Phase" is a status description of the transformer name, which can be used as an equipment status entity to represent this equipment, specifically the main transformer of phase A in Dinggong Substation No. 1. Similarly, "main transformer" is a type feature description of equipment health status; therefore, "main transformer" can be used as an equipment status entity to represent a value of the equipment type. Based on equipment status entities, entity alignment methods can be used to merge and unify entity objects representing the same actual object.

[0042] S4.3 integrates all equipment status health sub-files of the unified semantic representation system for equipment status analysis to establish a complete equipment status health file for classifying and assessing the health status of substation equipment.

[0043] In this embodiment, the alignment of device state entities in step S4.2 refers to using a matching function to map and align entities from different device state sub-archives at the vector space level. The matching function is a multi-layer neural network, whose input is the labeled device state entity and the device state entity to be judged, and whose output is a vector in the vector space. The matching function is designed and trained using a manually labeled entity alignment dataset S to ensure semantic consistency of data from different sources and minimize mapping loss. The expression for calculating semantic consistency and mapping loss is as follows:

[0044] L = L merge (CD1,CD2)+L base (ρ,S),

[0045] In the above formula, L represents semantic consistency and mapping loss, L merge (CD1, CD2) represents the sum of the absolute differences between each term in distributed vector representations V1 and V2, where distributed vector representation V1 is the distributed vector representation of the manually labeled entity alignment dataset S based on the device status health sub-file CD1, and distributed vector representation V2 is the distributed vector representation of the manually labeled entity alignment dataset S based on the device status health sub-file CD2. base (ρ,S) represents the probability that the matching function ρ maps the features of the equipment status health sub-files CD1 and CD2 to the correct equipment status entities based on the manually labeled entity alignment dataset S. Here, equipment status health sub-files CD1 and CD2 refer to two equipment status health sub-files from different sources for the same substation equipment. By analyzing and fusing the different features in equipment status health sub-files CD1 and CD2, the files are merged into a complete equipment health file.

[0046] A matching function is used to map and align entities from different device state sub-archives at the vector space level. The key is to design and train the vector matching function using labeled data to ensure semantic consistency and minimize mapping loss across different data sources. The manually labeled entity alignment dataset S is a set of manually labeled entities containing natural language entity objects related to device archives, representing different device states. The format is one entity per row, with each row containing a sequence number and entity name. The labeling process involves users using experience to identify keywords representing states, labeling them as entities, and adding them to the set. The entity alignment process uses training data to adjust the encoding scheme (one-hot encoding, distributed encoding, etc., are possible; distributed encoding is used in this embodiment) and matching function for different device state sub-archives to achieve semantic consistency and minimize mapping loss.

[0047] The semantic information in the sub-archives mainly consists of entity information. Different representations of the same entity in different sub-archives are clearly defined and mapped to the same entity. Therefore, in this embodiment, all equipment status health sub-archives of the unified semantic representation system for equipment status analysis are merged to establish a complete equipment status health archive for classifying and assessing the health status of substation equipment. A unified semantic representation is used to abstractly represent the semantic relationships inherent in all data types.

[0048] In this embodiment, when the health features in the complete device status health record are vectorized into feature vectors in step S5, the indirect features are normalized to the [0,1] interval using a normalization mapping function; the direct features are encoded using multi-head sequence encoding: first, the feature values ​​of the direct features are sorted and a unique queue is constructed, and each feature value in the unique queue is mapped to a different sequence number; then, each feature value in the unique queue is encoded as the multi-head sequence encoding result by dividing its corresponding sequence number by the sum of the sequence numbers.

[0049] Indirect features normalize different data ranges to the [0,1] interval through a series of normalization mapping functions. These normalization mapping functions can be the same or different for different indirect features. Taking the alarm trend of substation equipment as an example, the alarm trend is a feature that measures the time elapsed since an alarm occurred. The normalization mapping function for the alarm trend is as follows:

[0050]

[0051] The formula defines F as having a value range of [0,1]. The alarm trend feature represents the magnitude of the trend in device alarm occurrence; a larger alarm trend indicates a higher probability of device alarm occurrence, meaning a larger alarm trend F indicates a lower device health level. The direct feature data type is string, specifically divided into different heterogeneous representations such as text, time, and numerical values. This embodiment uses a unified and universal vectorized representation encoding method for different heterogeneous data: Multi-Head Sequence encoding, which effectively solves the problem of different heterogeneous representations such as text, time, and numerical values.

[0052] Each feature value in the unique queue is encoded as a multi-head sequence by dividing its corresponding sequence number by the sum of the sequence numbers. This can be represented as:

[0053]

[0054] Because of the uniqueness of the sequence, the sequence number can also guarantee the uniqueness of the encoding, thus enabling different direct features to be encoded into vector representations.

[0055] In this embodiment, step S6 constructs a feature vector-based classification model as a multilayer perceptron (MPP) classification network. The MPP classification network includes an input layer, a hidden layer, and an output layer connected sequentially. During operation, the input layer receives the instance feature vectors from the training set, which are then passed to the next layer via the weights of the connecting nodes. The output of the previous layer becomes the input of the next layer, and the final result is output through the output layer. When training the MPP classification network, the feature vectors are used as input, and the output of the MPP classification network is compared with known labels. The weights of the network parameters in the MPP classification network are adjusted accordingly (weights are usually initialized with random values) until the maximum allowed number of iterations is reached or the accuracy of the MPP classification network exceeds a set value. In this embodiment, the MPP classification network calculates the substation equipment status level into four levels: levels 1 to 4 correspond to normal state, alert state, abnormal state, and critical state, respectively. If any feature in the equipment health database changes, steps S1 to S6 can be used to recalculate the equipment health level and update the substation equipment health database.

[0056] In summary, this embodiment's method for constructing an equipment health archive based on unified semantic representation of data processes equipment status data from different sources separately, identifies semantic information for each type of equipment status data, and establishes sub-archives for the equipment health archive. The semantic information of different sub-archives is aligned and linked to form a unified semantic representation system for equipment status analysis, establishing a complete equipment health archive. Then, a distributed learning method is used to transform the equipment status information into vectors, combined with an autoencoder for feature compression and extraction, and a fully connected deep neural network is used to achieve status classification. Based on the status classification results, the equipment health level is calculated, and the substation equipment health archive is updated. Compared to traditional methods, this invention, based on a unified semantic representation construction method for multi-source heterogeneous data, fully considers multi-source monitoring data and time-series operation data related to substation equipment, enabling more comprehensive, accurate, dynamic, and comprehensive monitoring of the health status of substation equipment, providing data support for differentiated equipment operation and maintenance.

[0057] Furthermore, this embodiment also provides a device health record construction system based on unified semantic representation of data, including a microprocessor and a memory interconnected thereto. The microprocessor is programmed or configured to execute the steps of the aforementioned device health record construction method based on unified semantic representation of data. Additionally, this embodiment also provides a computer-readable storage medium storing a computer program for being programmed or configured by the microprocessor to execute the steps of the aforementioned device health record construction method based on unified semantic representation of data.

[0058] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for constructing a device health record database based on unified semantic representation of data, characterized in that, include: S1, input device status data from different sources respectively; S2, extract health features from equipment status data; the health features include direct features and indirect features. Direct features refer to the static data features of substation equipment, which are data features that do not change over time; indirect features refer to the time-series monitoring data corresponding to the substation equipment, which change over time. S3, establish equipment status health sub-files for each substation device based on health characteristics; S4, perform unified semantic representation on each equipment status health sub-file and merge it into a complete equipment status health file, including: S4.1, extract the semantic information of each equipment status health sub-file; S4.2, align and link the semantic information of each equipment status health sub-file to form a unified semantic representation system for equipment status analysis; S4.3, merge all equipment status health sub-files of the unified semantic representation system for equipment status analysis to establish a complete equipment status health file for the classification and evaluation of substation equipment health status; the alignment of equipment status entities in step S4.2 refers to using a matching function to map and align entities of different equipment status sub-file libraries at the vector space level. The matching function is a multi-layer neural network, whose input is the labeled equipment status entity and the equipment status entity to be judged, and whose output is a vector in the vector space. The matching function is designed and trained using a manually labeled entity alignment dataset S to ensure semantic consistency of data from different sources and minimize mapping loss. The expression for the calculation function of semantic consistency and mapping loss is: ， In the above formula, L represents semantic consistency and mapping loss. This represents the sum of the absolute differences between each term in distributed vector representations V1 and V2, where distributed vector representation V1 is the distributed vector representation of the manually labeled entity alignment dataset S based on the device status health sub-file CD1, and distributed vector representation V2 is the distributed vector representation of the manually labeled entity alignment dataset S based on the device status health sub-file CD2. Represents the matching function The probability of mapping the features of device status health sub-files CD1 and CD2 to the correct device status entities based on the manually annotated entity alignment dataset S. S5, the health features in the complete equipment status health record are vectorized into feature vectors. For indirect features, a normalization mapping function is used to normalize them to the [0,1] interval; for direct features, multi-head sequence encoding is used: first, the feature values ​​of the direct feature are sorted and a unique queue is constructed, and each feature value in the unique queue is mapped to a different sequence number; then, each feature value in the unique queue is encoded as the multi-head sequence encoding result by dividing its corresponding sequence number by the sum of the sequence numbers. S6. Construct a feature vector-based classification model to classify and evaluate the status of substation equipment.

2. The method for constructing a device health record database based on unified semantic representation of data according to claim 1, characterized in that, The sources in step S1 include the substation production management system, the substation online monitoring system, and the substation equipment ledger. The equipment status data from different sources include main transformer equipment data, main transformer evaluation data, main transformer alarm data, main transformer repair and testing data, main transformer defect data, and substation data.

3. The method for constructing a device health record database based on unified semantic representation of data according to claim 1, characterized in that, The direct features include equipment ID, equipment name, equipment type, voltage level, equipment model, manufacturer, commissioning date, manufacturing date, substation ID, substation name, substation voltage level, city, district / county, substation type, and production management system evaluation; the indirect features include defect frequency, defect trend, defect nature, defect elimination status, acceptance evaluation, hidden danger status, deduction value, alarm frequency, alarm trend, alarm level, alarm type, and test conclusion.

4. The method for constructing a device health record database based on unified semantic representation of data according to claim 1, characterized in that, In step S6, a classification model based on feature vectors is constructed as a multilayer perceptron classification network. The multilayer perceptron classification network includes an input layer, a hidden layer, and an output layer connected in sequence. When training the multilayer perceptron classification network, the feature vector is used as the input of the multilayer perceptron classification network, and the output of the multilayer perceptron classification network is compared with the known labels. The weights of the network parameters in the multilayer perceptron classification network are adjusted accordingly until the maximum number of allowed iterations is reached or the accuracy of the multilayer perceptron classification network is greater than a set value.

5. A system for constructing a device health record database based on unified semantic representation of data, comprising interconnected microprocessors and memory, characterized in that, The microprocessor is programmed or configured to perform the steps of the device health record construction method based on unified semantic representation of data as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing a computer program, characterized in that, The computer program is used to be programmed or configured by a microprocessor to perform the steps of the device health record construction method based on unified semantic representation of data as described in any one of claims 1 to 4.

Images