Multi-source heterogeneous data semantic alignment method and device, equipment and medium

Abstract

According to the multi-source heterogeneous data semantic alignment method and device, the equipment and the medium disclosed by the invention, the multi-source heterogeneous data is obtained and preprocessed to obtain a standardized data unit set; constructing sample pairs according to the standardized data unit set in combination with comparative learning, performing training optimization on the distance between the sample pairs, identifying data consistency in unit semantic features, and outputting a consistency label; performing unsupervised feature extraction on each unit of the standardized data unit set to obtain unit potential semantic features; and performing semantic alignment on the multi-source heterogeneous data based on the consistency tag and the unit potential semantic features, and constructing a semantic space. According to the scheme, semantic standardized alignment of the multi-source heterogeneous data can be realized, and accurate mapping and fusion of the multi-source heterogeneous data of the equipment in the knowledge graph are ensured.

Description

Technical Field

[0001] This invention relates to the field of power equipment operation and maintenance technology, and in particular to a method, apparatus, device and medium for semantic alignment of multi-source heterogeneous data. Background Technology

[0002] Knowledge graphs, as an advanced data organization and representation method, can effectively integrate and link power equipment operation and maintenance knowledge scattered across different systems, providing a new approach to solving problems such as the complexity and variability of power equipment operation and maintenance databases and the high maintenance costs of power equipment fault diagnosis models. The knowledge graph-based external knowledge base technology aims to build a dynamic, flexible, and easily maintainable knowledge management system. By combining the static capabilities of model decision-making with the dynamic updates of the knowledge base, it enhances the reliability and response speed of power equipment operation and maintenance decisions. The core characteristics of knowledge graphs lie in their powerful semantic representation capabilities and efficient query and retrieval mechanisms.

[0003] In the context of power equipment operation and maintenance, knowledge graphs can clearly represent heterogeneous data from multiple sources, such as equipment attributes, operating status, fault modes, and maintenance history, forming an organically connected knowledge network. However, power equipment operation and maintenance data comes from a wide range of sources (including sensor monitoring data, maintenance records, technical procedures, etc.) and contains a mixture of structured numerical data and unstructured text / image data. Traditional methods struggle to automatically eliminate semantic ambiguities between different data sources, leading to problems such as chaotic data mapping and difficulties in data fusion during knowledge graph construction. Summary of the Invention

[0004] To address the aforementioned shortcomings, this invention provides a method, apparatus, device, and medium for semantic alignment of multi-source heterogeneous data, which can achieve semantic standardization alignment of multi-source heterogeneous data and ensure accurate mapping and fusion of multi-source heterogeneous data in knowledge graphs.

[0005] This invention provides a method for semantic alignment of multi-source heterogeneous data, the method comprising: Acquire multi-source heterogeneous data and preprocess the multi-source heterogeneous data to obtain a standardized data unit set; Based on the standardized data unit set, sample pairs are constructed using contrastive learning. The distance between the sample pairs is trained and optimized to identify data consistency in the semantic features of the units and output consistency labels. Unsupervised feature extraction is performed on each unit of the standardized data unit set to obtain the unit's latent semantic features; Based on the consistency label and the unit's latent semantic features, the multi-source heterogeneous data is semantically aligned to construct a semantic space.

[0006] Preferably, the multi-source heterogeneous data includes equipment ledgers recording basic equipment parameters and maintenance reports consisting of unstructured text data.

[0007] Preferably, the sample pairs include similar sample pairs and dissimilar sample pairs; The optimization of the distance between the sample pairs, the identification of data consistency in the semantic features of the units, and the output of consistency labels include: Using a deep neural network as the basic architecture, the data samples of the standardized data units are input into the model, and feature vector representations are obtained through feature extraction by a multi-layer neural network. The contrastive loss function is used to calculate the sample distance between sample pairs, and the backpropagation algorithm is used to adjust the neural network parameters to bring the sample distance of positive sample pairs closer and push the sample distance of negative sample pairs further apart in the feature space. The optimized sample distance is judged by a preset distance threshold to identify the consistency between units and output a consistency label.

[0008] Preferably, the step of performing unsupervised feature extraction on each unit of the standardized data unit set to obtain the unit's latent semantic features includes: Text data features are extracted using word vector models or autoencoders; Numerical data features are extracted using dimensionality reduction algorithms; Image data features are extracted using multi-layer convolution and pooling operations of a convolutional neural network; The text data features, the numerical data features, and the image data features are used as the latent semantic features of the unit.

[0009] Preferably, the step of semantically aligning the multi-source heterogeneous data based on the consistency label and the unit latent semantic features to construct a semantic space includes: The consistency label is used to associate semantically consistent units in the unit's latent semantic features to determine the feature vector; Clustering algorithms are used to group units with similar feature vectors into the same semantic group; All the potential semantic features of the units are mapped to the same low-dimensional space, and entity nodes are determined according to the determined semantic groups to form a semantic network containing relation edges, thus obtaining the semantic space.

[0010] Preferably, the method further includes; The semantic space is updated by adding new entity nodes and their corresponding relationships to the knowledge graph.

[0011] Preferably, the step of updating the semantic space by adding new entity nodes and corresponding relationships to the knowledge graph includes: Identify semantic feature representations in novel multi-source heterogeneous data; Based on the semantic features, new entity nodes are added to the semantic network of the semantic space, and relationship edges are established between the entity nodes and related device nodes, fault cause nodes, and handling measure nodes. The attributes of the entities and the relationship weights are updated, and the semantic space is updated.

[0012] This invention also provides a semantic alignment device for multi-source heterogeneous data, the device comprising: The data acquisition module is used to acquire multi-source heterogeneous data and preprocess the multi-source heterogeneous data to obtain a standardized data unit set; The labeling module is used to construct sample pairs based on the standardized data unit set and comparative learning, train and optimize the distance between the sample pairs, identify data consistency in the semantic features of the units, and output consistency labels. The feature module is used to perform unsupervised feature extraction on each unit of the standardized data unit set to obtain the unit's potential semantic features; A construction module is used to perform semantic alignment on the multi-source heterogeneous data based on the consistency label and the unit latent semantic features to construct a semantic space.

[0013] This invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the multi-source heterogeneous data semantic alignment method as described in any of the above embodiments.

[0014] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the multi-source heterogeneous data semantic alignment method as described in any of the above embodiments.

[0015] The present invention provides a method, apparatus, device, and medium for semantic alignment of multi-source heterogeneous data. This method acquires multi-source heterogeneous data and preprocesses it to obtain a standardized data unit set. Based on the standardized data unit set, it constructs sample pairs using contrastive learning, trains and optimizes the distance between the sample pairs, identifies data consistency in the semantic features of the units, and outputs consistency labels. Unsupervised feature extraction is performed on each unit of the standardized data unit set to obtain the unit's latent semantic features. Semantic alignment of the multi-source heterogeneous data is then performed based on the consistency labels and the unit's latent semantic features to construct a semantic space. This solution enables semantic standardization alignment of multi-source heterogeneous data, ensuring accurate mapping and fusion of multi-source heterogeneous data in a knowledge graph. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for semantic alignment of multi-source heterogeneous data provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the multi-source heterogeneous data semantic alignment device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 This is a flowchart illustrating a semantic alignment method for multi-source heterogeneous data provided in an embodiment of the present invention. The method includes steps S1 to S4: Step S1: Obtain multi-source heterogeneous data and preprocess the multi-source heterogeneous data to obtain a standardized data unit set; Step S2: Construct sample pairs based on the standardized data unit set and contrastive learning, train and optimize the distance between the sample pairs, identify data consistency in the semantic features of the units, and output consistency labels; Step S3: Perform unsupervised feature extraction on each unit of the standardized data unit set to obtain the unit's latent semantic features; Step S4: Based on the consistency label and the unit latent semantic features, perform semantic alignment on the multi-source heterogeneous data to construct a semantic space.

[0019] In the specific implementation of this embodiment, the core differences of multi-source heterogeneous data (such as database tables, text, images, etc.) are reflected in the format, structure and representation, but in essence they may contain related semantic information, such as user ID=123, where records on different platforms point to the same user.

[0020] First, multi-source data is aggregated through acquisition technology to break down data silos. Then, through preprocessing, cleaning, transformation, and standardization to eliminate format differences, heterogeneous data is mapped into unified standardized data units. Each unit is a basic semantic particle, such as a record or a piece of text, giving the data a comparable basic format and providing structured input for subsequent semantic analysis.

[0021] By using contrastive learning to optimize distances to identify data consistency, we can uncover the mathematical expression of "semantic associations." Semantic consistency can be quantified by distances in the feature space; the more semantically consistent the units, the smaller the distance between their feature vectors should be, and vice versa.

[0022] Contrastive learning builds sample pairs, trains the model to learn distance metric rules, and optimizes the model parameters through a loss function (such as InfoNCE) to bring positive sample pairs closer together and negative sample pairs further apart.

[0023] The trained model can automatically identify which units are semantically consistent by calculating the distance between them, thus achieving mathematical recognition of implicit semantic associations.

[0024] The surface features of data units, such as the literal content of text or the pixel values ​​of images, may not directly reflect their core semantics, and implicit information needs to be mined through unsupervised learning.

[0025] Unsupervised feature extraction does not rely on manual labels. Instead, it automatically learns the inherent patterns in the data through the model, such as the contextual relationships in text and the combination of local features in images, and transforms the original features (such as words and pixels) into latent semantic feature vectors.

[0026] For example, text captures contextual semantics through unsupervised pre-training of BERT, while images learn deep features unaffected by lighting / angle through self-supervised models (such as SimCLR), so that feature vectors directly correspond to the essential semantics of the unit, rather than its surface form.

[0027] The core of semantic alignment is to map units from different sources and of different types to the same space based on their semantic associations, so that the "distance" in the space directly corresponds to "semantic similarity".

[0028] By using data consistency as a strong constraint and combining latent semantic features, i.e. implicit semantic information, as a weak constraint (such as units with similar features tending to be close together), the feature vectors of all units are adjusted to the same space through clustering, mapping, or network optimization.

[0029] The constructed semantic space is a low-dimensional vector space, where each point represents a data unit. The distance between points intuitively reflects semantic similarity. The closer the distance, the more consistent the semantics, thus achieving unified semantic understanding of cross-source and heterogeneous data.

[0030] Transforming raw heterogeneous data into quantifiable and comparable semantic representations essentially uses distance relationships in mathematical space to simulate human perception of semantic similarity, providing a foundation for tasks such as cross-source data fusion, retrieval, and analysis. This involves achieving semantic standardization and alignment of multi-source heterogeneous data, thereby constructing a knowledge network with dynamic association capabilities. This ensures accurate mapping and fusion of data such as equipment operating status, fault history, and maintenance records within the knowledge graph. Furthermore, training the embedded representations of multi-source heterogeneous data ensures the automatic identification and elimination of semantic conflicts during knowledge base construction, guaranteeing semantic standardization and consistency of data before it enters the knowledge base.

[0031] In another embodiment provided by the present invention, the multi-source heterogeneous data includes equipment ledgers that record basic equipment parameters and maintenance reports of unstructured text data.

[0032] In this specific implementation, acquiring multi-source heterogeneous data includes using a graph engine to search based on the association characteristics of nodes and edges in a knowledge graph.

[0033] Specifically, in the operation of power systems, data sources are widespread and diverse in structure. Common data sources include equipment ledgers and maintenance reports. Equipment ledgers are presented in structured tabular form, recording basic equipment parameters; maintenance reports, on the other hand, are unstructured text data. This data is stored in a scattered manner, with inconsistent formats and semantic standards. Constructing equipment ledgers and maintenance reports into a knowledge graph organizes knowledge in a graph structure, where nodes represent entities and edges represent relationships. Graph search engines leverage the associative characteristics of knowledge graphs. When receiving a semantic query request, such as a query for information related to a transformer fault, the query statement is first semantically parsed and transformed into instructions recognizable by the graph database. Subsequently, using efficient graph traversal algorithms from graph databases (such as Neo4j), relevant data is quickly located and extracted from different data sources, aggregating the scattered data and providing a comprehensive data foundation for subsequent processing. This is a crucial step in achieving initial integration of multi-source data.

[0034] In another embodiment provided by the present invention, the sample pairs include similar sample pairs and dissimilar sample pairs; The optimization of the distance between the sample pairs, the identification of data consistency in the semantic features of the units, and the output of consistency labels include: Using a deep neural network as the basic architecture, the data samples of the standardized data units are input into the model, and feature vector representations are obtained through feature extraction by a multi-layer neural network. The contrastive loss function is used to calculate the sample distance between sample pairs, and the backpropagation algorithm is used to adjust the neural network parameters to bring the sample distance of positive sample pairs closer and push the sample distance of negative sample pairs further apart in the feature space. The optimized sample distance is judged by a preset distance threshold to identify the consistency between units and output a consistency label.

[0035] In the specific implementation of this embodiment, constructing sample pairs includes constructing similar sample pairs, i.e., positive sample pairs, and dissimilar sample pairs, i.e., negative sample pairs.

[0036] Furthermore, the training optimization includes using a deep neural network as the basic architecture, inputting data samples of standardized data units into the model, extracting feature vectors through a multi-layer neural network, calculating the distance between sample pairs using a contrastive loss function, and adjusting the neural network parameters through a backpropagation algorithm to bring positive sample pairs closer and push negative sample pairs further apart in the feature space.

[0037] Positive sample pairs: semantically consistent unit combinations (such as records of the same user on different platforms, synonymous text, similar images), which can be generated through rules (such as ID association) or preliminary clustering (such as K-means).

[0038] Negative sample pairs: semantically inconsistent unit combinations (such as records from different users, irrelevant text), which can be generated by random sampling or through hard negative example mining (such as units that are similar to but different from positive samples).

[0039] Euclidean distance, cosine similarity, and Manhattan distance are used to measure the semantic distance between sample pairs (the smaller the distance, the higher the consistency).

[0040] The goal of using contrastive loss functions (such as InfoNCE and Triplet Loss) is to reduce the distance between positive sample pairs and increase the distance between negative sample pairs. For example, neural networks (such as Transformer and CNN) map units to a feature space, making positive sample pairs closer together and negative sample pairs more distant in the space.

[0041] After training, the consistency between units is judged by the optimized distance threshold (e.g., setting cosine similarity > 0.8 as consistency), and a "consistency label" is output (e.g., 1 indicates consistency, 0 indicates inconsistency).

[0042] Specifically, the semantic representation of power equipment data is rich and diverse, with multiple ways to express the same semantic meaning, and different sources may have different definitions of the same concept. Contrastive learning training aims to enable the model to deeply learn the semantic similarity and differences between data. Its core principle is to construct similar and dissimilar sample pairs. Training optimization specifically involves optimizing the loss function. By optimizing the loss function, the distance between positive sample pairs and the distance between negative sample pairs in the feature space are reduced. In implementation, deep neural networks (such as convolutional neural networks, recurrent neural networks, and their variants) are often used as the basic architecture. Data samples are input into the model, and feature vectors are extracted through multi-layer neural networks. The distance between sample pairs is calculated using a contrastive loss function (such as the InfoNCE loss function), and the neural network parameters are adjusted through backpropagation. For example, when processing text descriptions of power equipment faults, texts describing the same fault type but with different expressions are input into the model as positive sample pairs. After multiple iterations of training, the model can accurately identify the semantic consistency between texts, effectively solving the problem of inconsistent semantic understanding.

[0043] In another embodiment of the present invention, the step of performing unsupervised feature extraction on each unit of the standardized data unit set to obtain the unit's latent semantic features includes: Text data features are extracted using word vector models or autoencoders; Numerical data features are extracted using dimensionality reduction algorithms; Image data features are extracted using multi-layer convolution and pooling operations of a convolutional neural network; The text data features, the numerical data features, and the image data features are used as the latent semantic features of the unit.

[0044] In this specific implementation, unsupervised feature extraction includes extracting data features from text data, numerical data, and image data; Text data features are extracted using word vector models or autoencoders; numerical data features are extracted using dimensionality reduction algorithms; and image data features are extracted using multi-layer convolutional neural networks with pooling operations.

[0045] Specifically, the multi-source heterogeneous data of power equipment contains a wealth of valuable semantic features, but these features are difficult to directly utilize by models. Unsupervised feature extraction techniques eliminate the need for manual annotation, automatically mining potential features from raw data.

[0046] For text data, word vector models (such as Word2Vec and GloVe) are often used to map words into low-dimensional vectors to capture semantic similarity, or autoencoders are used to learn compressed feature representations of text. For numerical data, dimensionality reduction algorithms such as principal component analysis (PCA) and independent component analysis (ICA) can remove noise and redundancy, extract principal components, and reveal the underlying structure of the data. For image data, features such as edges, textures, and shapes are automatically extracted through multi-layer convolution and pooling operations of convolutional neural networks.

[0047] For example, when analyzing transformer operating parameters, key feature dimensions are extracted; when processing infrared thermal imaging images of power equipment, features such as the shape and size of temperature anomaly areas are extracted, adding rich information to the semantic representation of the data.

[0048] Use an autoencoder or a pre-trained language model (such as the unsupervised pre-training of BERT) to extract contextual semantic vectors and capture implicit semantics.

[0049] Self-supervised models (such as MoCo and SimCLR) are used to learn latent features, such as deep features that are not affected by surface differences (such as brightness and volume) through tasks like image rotation prediction and audio mask recovery.

[0050] Use variational autoencoders (VAEs) or graph neural networks (GNNs, which handle field associations) to extract latent patterns (such as latent preferences in user behavior).

[0051] The text data features, the numerical data features, and the image data features are used as the latent semantic features of the unit.

[0052] The "latent semantic feature vector" (low-dimensional dense vector) of each unit serves as the basis for semantic alignment.

[0053] In another embodiment of the present invention, the step of semantically aligning the multi-source heterogeneous data based on the consistency label and the unit latent semantic features to construct a semantic space includes: The consistency label is used to associate semantically consistent units in the unit's latent semantic features to determine the feature vector; Clustering algorithms are used to group units with similar feature vectors into the same semantic group; All the potential semantic features of the units are mapped to the same low-dimensional space, and entity nodes are determined according to the determined semantic groups to form a semantic network containing relation edges, thus obtaining the semantic space.

[0054] In this specific implementation, a semantic space is constructed based on data consistency and semantic alignment of potential semantic features.

[0055] Based on data consistency, semantically consistent units are associated using the identified consistency labels, correcting the inconsistencies in the feature space, such as forcing positive sample pairs to overlap in space.

[0056] Clustering or matching algorithms are used to group units with similar feature vectors into the same semantic group. For example, the category of electronic products includes text / image units such as mobile phones and computers.

[0057] Mapping the latent semantic feature vectors of all units to the same low-dimensional space, such as through principal component analysis (PCA) dimensionality reduction, allows the distance between units in the space to directly reflect semantic similarity; the closer the distance, the more consistent the semantics.

[0058] Entity nodes are determined based on the defined semantic groups. Semantic nodes and relation edges can be defined in the space to form a structured semantic network that supports subsequent tasks.

[0059] Transforming multi-source heterogeneous data into a unified semantic space enables semantic understanding and association of cross-source data.

[0060] In yet another embodiment of the present invention, the method further includes; The semantic space is updated by adding new entity nodes and their corresponding relationships to the knowledge graph.

[0061] In this specific implementation, the semantic space is updated by adding new entity nodes and corresponding relationships to the knowledge graph. Through continuous updates, the semantic space can reflect new knowledge and situations in the operation and maintenance of power equipment in real time, enabling the fault diagnosis model based on this semantic space to continuously learn and adapt to new changes. This provides more accurate and comprehensive semantic support for power equipment fault diagnosis decisions, ensuring that the accuracy and reliability of diagnosis decisions are continuously improved.

[0062] In another embodiment of the present invention, the step of updating the semantic space by adding new entity nodes and corresponding relationships to the knowledge graph includes: Identify semantic feature representations in novel multi-source heterogeneous data; Based on the semantic features, new entity nodes are added to the semantic network of the semantic space, and relationship edges are established between the entity nodes and related device nodes, fault cause nodes, and handling measure nodes. The attributes of the entities and the relationship weights are updated, and the semantic space is updated.

[0063] In this specific implementation, after graph engine search, comparative learning training, and unsupervised feature extraction, the semantic space constructed based on the knowledge graph already contains numerous entities, relationships, and their semantic representations. However, as new data is continuously generated from the operation of power equipment, such as new fault cases and new equipment parameter standards, the semantic space needs continuous optimization and expansion. When new data is added, it is first processed according to the previous process to obtain its semantic feature representation. Then, the new semantic features are integrated into the existing semantic space. Specific operations include adding new entity nodes (such as new fault type nodes) to the knowledge graph, establishing their relationships with related equipment, fault causes, and handling measures, and updating information such as entity attributes and relationship weights.

[0064] This invention also provides a semantic alignment device for multi-source heterogeneous data, see [link to relevant documentation]. Figure 2 This is a schematic diagram of the structure of a multi-source heterogeneous data semantic alignment device provided in an embodiment of the present invention. The device includes: The data acquisition module is used to acquire multi-source heterogeneous data and preprocess the multi-source heterogeneous data to obtain a standardized data unit set; The labeling module is used to construct sample pairs based on the standardized data unit set and comparative learning, train and optimize the distance between the sample pairs, identify data consistency in the semantic features of the units, and output consistency labels. The feature module is used to perform unsupervised feature extraction on each unit of the standardized data unit set to obtain the unit's potential semantic features; A construction module is used to perform semantic alignment on the multi-source heterogeneous data based on the consistency label and the unit latent semantic features to construct a semantic space.

[0065] It should be noted that the multi-source heterogeneous data semantic alignment device provided in the embodiments of the present invention can execute the multi-source heterogeneous data semantic alignment method described in any of the above embodiments. The specific functions of the multi-source heterogeneous data semantic alignment device will not be elaborated here.

[0066] See Figure 3 This is a schematic diagram of the structure of a terminal device provided in an embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a multi-source heterogeneous data semantic alignment program. When the processor executes the computer program, it implements the steps in the various embodiments of the multi-source heterogeneous data semantic alignment method described above, for example... Figure 1 The steps S1 to S4 are shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments.

[0067] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing functions, which describe the execution process of the computer program in the terminal device. For example, the computer program can be divided into various modules, the specific functions of which will not be elaborated again.

[0068] The terminal device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the schematic diagram is merely an example of a terminal device and does not constitute a limitation on the terminal device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0069] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0070] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0071] Wherein, if the modules / units integrated in the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in code form, object code form, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0072] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A semantic alignment method for multi-source heterogeneous data, characterized in that, The method includes: Acquire multi-source heterogeneous data and preprocess the multi-source heterogeneous data to obtain a standardized data unit set; Based on the standardized data unit set, sample pairs are constructed using contrastive learning. The distance between the sample pairs is trained and optimized to identify data consistency in the semantic features of the units and output consistency labels. Unsupervised feature extraction is performed on each unit of the standardized data unit set to obtain the unit's latent semantic features; Based on the consistency label and the unit's latent semantic features, the multi-source heterogeneous data is semantically aligned to construct a semantic space.

2. The semantic alignment method for multi-source heterogeneous data as described in claim 1, characterized in that, The multi-source heterogeneous data includes equipment ledgers that record basic equipment parameters and maintenance reports consisting of unstructured text data.

3. The semantic alignment method for multi-source heterogeneous data as described in claim 1, characterized in that, The sample pairs include similar sample pairs and dissimilar sample pairs; The optimization of the distance between the sample pairs, the identification of data consistency in the semantic features of the units, and the output of consistency labels include: Using a deep neural network as the basic architecture, the data samples of the standardized data units are input into the model, and feature vector representations are obtained through feature extraction by a multi-layer neural network. The contrastive loss function is used to calculate the sample distance between sample pairs, and the backpropagation algorithm is used to adjust the neural network parameters to bring the sample distance of positive sample pairs closer and push the sample distance of negative sample pairs further apart in the feature space. The optimized sample distance is judged by a preset distance threshold to identify the consistency between units and output a consistency label.

4. The semantic alignment method for multi-source heterogeneous data as described in claim 1, characterized in that, The step of performing unsupervised feature extraction on each unit of the standardized data unit set to obtain the unit's latent semantic features includes: Text data features are extracted using word vector models or autoencoders; Numerical data features are extracted using dimensionality reduction algorithms; Image data features are extracted using multi-layer convolution and pooling operations of a convolutional neural network; The text data features, the numerical data features, and the image data features are used as the latent semantic features of the unit.

5. The semantic alignment method for multi-source heterogeneous data as described in claim 1, characterized in that, The step of semantically aligning the multi-source heterogeneous data based on the consistency label and the unit latent semantic features to construct a semantic space includes: The consistency label is used to associate semantically consistent units in the unit's latent semantic features to determine the feature vector; Clustering algorithms are used to group units with similar feature vectors into the same semantic group; All the potential semantic features of the units are mapped to the same low-dimensional space, and entity nodes are determined according to the determined semantic groups to form a semantic network containing relation edges, thus obtaining the semantic space.

6. The semantic alignment method for multi-source heterogeneous data as described in claim 1, characterized in that, The method further includes; The semantic space is updated by adding new entity nodes and their corresponding relationships to the knowledge graph.

7. The semantic alignment method for multi-source heterogeneous data as described in claim 6, characterized in that, The step of updating the semantic space by adding new entity nodes and corresponding relationships to the knowledge graph includes: Identify semantic feature representations in novel multi-source heterogeneous data; Based on the semantic features, new entity nodes are added to the semantic network of the semantic space, and relationship edges are established between the entity nodes and related device nodes, fault cause nodes, and handling measure nodes. The attributes of the entities and the relationship weights are updated, and the semantic space is updated.

8. A semantic alignment device for multi-source heterogeneous data, characterized in that, The device includes: The data acquisition module is used to acquire multi-source heterogeneous data and preprocess the multi-source heterogeneous data to obtain a standardized data unit set; The labeling module is used to construct sample pairs based on the standardized data unit set and comparative learning, train and optimize the distance between the sample pairs, identify data consistency in the semantic features of the units, and output consistency labels. The feature module is used to perform unsupervised feature extraction on each unit of the standardized data unit set to obtain the unit's potential semantic features; A construction module is used to perform semantic alignment on the multi-source heterogeneous data based on the consistency label and the unit latent semantic features to construct a semantic space.

9. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the multi-source heterogeneous data semantic alignment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the multi-source heterogeneous data semantic alignment method as described in any one of claims 1 to 7.

Images