Power equipment information retrieval method and system

Abstract

The invention relates to a power equipment information retrieval method and system, and belongs to the technical field of data retrieval, and the method comprises the following steps: obtaining query information of a user for power equipment, and analyzing the query information into a query semantic vector; based on the query semantic vector, performing semantic projection on the equipment embedding vector, so that the query semantic vector and the equipment embedding vector are aligned in the same semantic space; outputting a first set based on the query semantic vector, the device embedding vector after semantic projection and a target function; calculating the similarity between the equipment embedding vector after each semantic projection in the first set and the query semantic vector; and determining a retrieval result based on a similarity calculation result. The retrieval mode has a high semantic perception capability, can realize accurate and stable power equipment matching under query conditions of different types and different expression modes, effectively improves the adaptability and overall accuracy in a complex retrieval scene, and enhances the semantic reasoning capability of a system at the same time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data retrieval, and in particular to a power equipment information retrieval method and system. BACKGROUND

[0002] Power equipment is a general term for all equipment in the power system for power generation, power transmission, power transformation, power distribution and power use, covering the whole process from energy production to terminal use.

[0003] Due to the large number of power equipment and the complex types, the traditional manual management method is inefficient, so currently it mainly depends on the retrieval system.

[0004] It is found in research that the traditional power equipment information retrieval method mainly depends on keyword matching, such as that the user can input a text keyword to query related equipment information. The existing method lacks the ability of deep modeling of semantic information, especially in the background of a large number of heterogeneous data and complex semantic relationships in the power field, the existing method is difficult to effectively understand the query intention, resulting in low retrieval accuracy and poor retrieval effect. SUMMARY

[0005] To solve the above technical problems, the present application provides a power equipment information retrieval method and system.

[0006] In a first aspect, the present application provides a power equipment information retrieval method, comprising: acquiring query information of a user for power equipment, and parsing the query information into a query semantic vector; wherein the dimension of the query semantic vector is the same as that of a device embedding vector; each power equipment entity in a pre-constructed target graph corresponds to a device embedding vector; and each device embedding vector has the same dimension; the target graph represents the association relationship between power equipment entities; based on the query semantic vector, performing semantic projection on the device embedding vector to align the query semantic vector and the device embedding vector in the same semantic space; based on the query semantic vector, the device embedding vector after semantic projection, and a target function, outputting a first set; wherein the target function is used to determine the device embedding vector after semantic projection associated with the query semantic vector; calculating the similarity between each device embedding vector after semantic projection in the first set and the query semantic vector; and determining a retrieval result based on the similarity calculation result.

[0007] Optionally, the target graph is constructed by the following steps: obtaining power equipment data; wherein the power equipment data comprises power equipment entities; determining semantic relationship strength between the power equipment entities based on an attribute set of the power equipment entities; calculating information entropy of each power equipment entity; determining quantified semantic correlation degree between the power equipment entities based on the semantic relationship strength between the power equipment entities and the information entropy of each power equipment entity; and generating the target graph; wherein a node set of the target graph comprises the power equipment entities, and an edge set of the target graph comprises the quantified semantic correlation degree between the power equipment entities.

[0008] Optionally, the generating the target graph comprises: calculating semantic weight of each power equipment entity based on occurrence frequency of each power equipment entity in the data set; and generating the target graph based on the semantic weight of each power equipment entity.

[0009] Optionally, the outputting the first set based on the query semantic vector, the device embedding vector after semantic projection, and the target function comprises: constructing a regularization term based on the quantified semantic correlation degree between the power equipment entities; constructing the target function based on a preset loss function and the regularization term; and outputting the first set based on the query semantic vector, the device embedding vector after semantic projection, and the target function.

[0010] Optionally, the obtaining the power equipment data comprises: obtaining initial power equipment data; in response to the existence of missing data of one power equipment entity in one dimension in the initial power equipment data, calculating mean and variance based on non-missing data of the dimension; determining a filling value of the power equipment entity in the dimension based on the mean and variance; and completing data of the power equipment entity based on the filling value to obtain the power equipment data.

[0011] Optionally, the obtaining the power equipment data comprises: obtaining initial power equipment data; and performing normalization processing on data of the same dimension of the power equipment entities in the initial power equipment data to obtain the power equipment data.

[0012] Optionally, the obtaining the power equipment data comprises: obtaining initial power equipment data; and removing duplicate data in the initial power equipment data based on a hash signature and cosine similarity judgment mode to obtain the power equipment data.

[0013] Optionally, the determining the retrieval result based on the similarity calculation result comprises: determining a to-be-output power equipment entity based on the similarity calculation result; determining an adjacent device set based on the to-be-output power equipment entity; constructing a directed subgraph based on the adjacent device set; and outputting the directed subgraph; wherein the directed subgraph represents the retrieval result.

[0014] Optionally, the method further comprises: obtaining a feedback score of the user in the interaction process; constructing a user aggregated feedback vector based on the feedback score; updating the semantic projection matrix based on the user aggregated feedback vector, or adjusting the objective function based on the user aggregated feedback vector.

[0015] In a second aspect, the present application provides a power equipment information retrieval system, comprising: a query acquisition module, configured to acquire query information of a user for power equipment, and parse the query information into a query semantic vector; wherein the dimension of the query semantic vector is the same as that of a device embedding vector; each power equipment entity in a pre-constructed target graph corresponds to a device embedding vector; and each device embedding vector has the same dimension; the target graph represents the association relationship between power equipment entities; a vector optimization module, configured to perform semantic projection on the device embedding vector based on the query semantic vector, so that the query semantic vector and the device embedding vector are aligned in the same semantic space; a screening module, configured to output a first set based on the query semantic vector, the device embedding vector after semantic projection, and an objective function; wherein the objective function is used to determine the device embedding vector after semantic projection associated with the query semantic vector; a similarity calculation module, configured to calculate the similarity between each device embedding vector after semantic projection in the first set and the query semantic vector; and a retrieval output module, configured to determine a retrieval result based on the similarity calculation result.

[0016] The beneficial effects of the present application include: the power equipment information retrieval method provided by the embodiments of the present application first acquires query information of a user for power equipment, and parses the query information into a query semantic vector; performs semantic projection on a device embedding vector based on the query semantic vector, so that the query semantic vector and the device embedding vector are aligned in the same semantic space; then outputs a first set based on the query semantic vector, the device embedding vector after semantic projection, and an objective function; calculates the similarity between each device embedding vector after semantic projection in the first set and the query semantic vector; and determines a retrieval result based on the similarity calculation result. That is, a retrieval model with semantic perception ability and dynamic adjustment mechanism is established in combination with the query condition input by the user, so that adaptive matching of the query target in a high-dimensional complex power equipment data space is realized. The core of this method is to construct a semantic embedding representation of power equipment information, and to establish a projection mapping based on the query intention, so as to accurately identify the power equipment entity most relevant to the query condition through similarity calculation. This retrieval method has high semantic perception ability, can realize accurate and stable power equipment matching under different types and different expression modes of query conditions, effectively improves the adaptability and overall accuracy in complex retrieval scenarios, enhances the system semantic reasoning ability, and is suitable for long-term deployment and application in large-scale power data systems. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of a method for retrieving power equipment information provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the steps of another method for retrieving power equipment information provided in an embodiment of the present invention. Figure 3 This is a block diagram of a power equipment information retrieval system provided in an embodiment of the present invention; Figure 4 This is a module block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0019] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0020] The study found that traditional power equipment information retrieval methods mainly rely on keyword matching, such as allowing users to input text keywords to query relevant equipment information. However, existing methods lack the ability to deeply model semantic information, especially in the context of a large amount of heterogeneous data and complex semantic relationships in the power sector. This makes it difficult for existing methods to effectively understand the query intent, resulting in low retrieval accuracy and poor retrieval performance.

[0021] In view of the above problems, this application proposes the following embodiments to solve the above technical problems.

[0022] Please see Figure 1 This application provides a method for retrieving information on power equipment, including steps 101 to 105.

[0023] Step 101: Obtain the user's query information for power equipment and parse the query information into a query semantic vector.

[0024] The query semantic vector has the same dimension as the device embedding vector; each power equipment entity in the pre-constructed target graph corresponds to a device embedding vector; and each device embedding vector has the same dimension; the target graph represents the association between power equipment entities.

[0025] That is, users can output a query description for power equipment to obtain the query information, and then the query information is parsed into a query semantic vector q; where, . Let represent a d-dimensional real vector space. The query semantic vector q represents the user's expected combination of attributes for the search target.

[0026] Assume the set V of power equipment entities in the target graph is: Each electrical equipment entity Generate a d-dimensional vector representation based on its attributes, relationships, and semantic context. This representation serves as a building block of the device semantic matrix. Let the overall device semantic matrix X be denoted as... Where n is the number of entities, d is the embedding dimension, and T represents transpose; Let represent a real vector space of dimension n×d.

[0027] In the embodiments of this application, power equipment includes power generation equipment, power transmission equipment, power transformation equipment, power distribution equipment, power consumption equipment, and / or auxiliary monitoring equipment.

[0028] The equipment includes, among others: power generation equipment (generators, steam turbines, turbines, blades, photovoltaic panels, inverters, etc.); transmission equipment (conductors, cables, cable trays, capacitor banks, etc.); transformation equipment (transformers, circuit breakers, relays, fuses, busbars, disconnect switches, etc.); distribution equipment (distribution cabinets, instrument transformers, voltage stabilizers, power cabinets, etc.); and power consumption equipment (motors, frequency converters, displays, charging piles, etc.). Auxiliary monitoring equipment includes PLC (Programmable Logic Controller) controllers, smart meters, sensors, grounding devices, etc.

[0029] Specific examples of the above query information include "No-load loss standard for type A oil-immersed transformer", "Dimensional diagram of type B fuse switch", and "Temperature resistance comparison between type C and type D cable".

[0030] Step 102: Based on the query semantic vector, perform semantic projection on the device embedding vector so that the query semantic vector and the device embedding vector are aligned in the same semantic space.

[0031] In this embodiment, semantic projection can be understood as a linear mapping of the transformation matrix. Specifically, to ensure that the query semantics and the device embedding vector are aligned in the same semantic space, a linear transformation matrix is ​​introduced. The device embedding vector is semantically projected, and the transformed device semantic matrix is ​​represented as follows: ; Rd denotes a real number vector space of dxd dimensionality (the parameter meaning of R is not described again later).

[0032] The plurality of projected device embedding vectors are included in the middle (all are described in the i-th).

[0033] It should be noted that the user's query information, such as "high-temperature-resistant 10kV cable", and the power device entities (such as cable model, parameters) in the target graph may exist in different semantic representation spaces, and the direct matching effect is poor. Therefore, in this step, the power device vector is mapped to the same semantic space as the query semantic vector q through the projection matrix P, so that similar intentions and devices are closer in space.

[0034] Step 103: Based on the query semantic vector, the semantic projected device embedding vector, and the objective function, output a first set.

[0035] Among them, the objective function is used to determine the semantic projected device embedding vector associated with the query semantic vector.

[0036] Step 104: Calculate the similarity between each semantic projected device embedding vector in the first set and the query semantic vector.

[0037] For each semantic projected device embedding vector in the first set , the similarity between it and the query semantic vector q can be calculated , and the specific formula can be: ; Among them, the numerator is the vector dot product, and the denominator is the product of the respective norms, that is, the cosine similarity can be calculated in this way.

[0038] Step 105: Determine the search result based on the similarity calculation result.

[0039] In the embodiments of the present application, all power device entities can be sorted according to the value of , and the top k power device entities with the highest scores are taken as the final search result, realizing adaptive matching and filtering of power device information.

[0040] In summary, the power equipment information retrieval method provided by the embodiment of the application first acquires the query information of the user for the power equipment, and parses the query information into a query semantic vector; based on the query semantic vector, the device embedding vector is projected in semantics, so that the query semantic vector and the device embedding vector are aligned in the same semantic space; then, based on the query semantic vector, the device embedding vector after semantic projection, and a preset loss function, a first set is output; the similarity between each semantic projection device embedding vector in the first set and the query semantic vector is calculated; based on the similarity calculation result, a retrieval result is determined. That is, a retrieval model with semantic perception ability and dynamic adjustment mechanism is established combined with the query condition input by the user, and adaptive matching of the query target in the high-dimensional complex power equipment data space is realized. The core of this method is to construct a semantic embedding representation of power equipment information, and to establish a projection mapping based on the query intention, and to accurately identify the power equipment entity most related to the query condition through similarity calculation. This retrieval method has high semantic perception ability, can accurately and stably match the power equipment under different types and different expression modes of query conditions, effectively improves the adaptability and overall accuracy in complex retrieval scenarios, enhances the system semantic reasoning ability, and is suitable for long-term deployment and application in large-scale power data systems.

[0041] Before retrieval, a target graph needs to be constructed. The construction process of the target graph provided by the embodiment of the application is described below.

[0042] Optionally, referring to Figure 2 The target graph can be constructed by the following steps, including steps 201-205.

[0043] Step 201: Acquire power equipment data.

[0044] The power equipment data includes power equipment entities.

[0045] Step 202: Determine the semantic relationship strength between power equipment entities based on the attribute set of the power equipment entities.

[0046] Specifically, let any two power equipment entities be and , and the corresponding attribute sets are and . Wherein, represents the attribute set corresponding to the power equipment entity ; represents the i-th attribute in the set. represents the attribute set corresponding to the power equipment entity ; represents the i-th attribute in the set, and m represents the number of attributes. The attribute is a dimension describing static characteristics or dynamic characteristics, and can specifically include technical parameters, physical characteristics, management information, operating states, and the like.

[0047] Defining semantic relationship strength between power equipment entities is defined as: ; wherein, represents the weight of the k-th attribute in the semantic relationship, which is set according to a domain prior or an information gain method; is an attribute matching function, which is 1 when the attribute semantics are consistent, and 0 otherwise; is a semantic score based on rule matching, for example, whether two power equipment entities belong to the same transformer unit, or whether there is a common maintenance cycle; is a coefficient for adjusting the influence degree of the rule score, used to balance the contribution between explicit attributes and implicit rules.

[0048] Step 203: Calculate the information entropy of each power equipment entity.

[0049] Suppose and respectively represent the information entropy of the equipment and . The calculation formula for the information entropy is as follows: ; wherein, is the normalized probability of the power equipment entity taking a value on the k-th dimensional attribute. The calculation formula for the information entropy can refer to the above formula, which is not described herein; represents a smoothing factor.

[0050] The purpose of introducing the information entropy is to normalize and differentiate the semantic correlation degree between different power equipment entities in the process of constructing the semantic graph, so as to improve the expression accuracy of the graph. Specifically, the information entropy of the equipment entity measures the diversity and uncertainty of its attribute values. If the attribute values of a certain entity are single or change little, the information entropy of the entity is low, indicating that the entity has limited contribution to the semantic modeling of the graph; and the entity with high information entropy usually has more rich and complex attribute combinations, and contains stronger information carrying capacity.

[0051] Step 204: Based on the semantic relationship strength between the power equipment entities and the information entropy of each power equipment entity, the semantic correlation degree between the power equipment entities is determined quantitatively.

[0052] That is, in order to quantify the strength of semantic relationship to build the graph edge weight, the correlation degree calculation mechanism is introduced. The calculation formula of the semantic correlation degree between two power equipment entities is: ; wherein, represents the semantic correlation degree between the power equipment entity and the power equipment entity , is the initial correlation score derived from the aforementioned relationship strength function.

[0053] In the semantic correlation degree calculation formula , the denominator item plays a role in dynamically adjusting the semantic edge weight. When both equipment entities have high information entropy (i.e., large amount of information), their semantic correlation degree remains at a high level, indicating that the semantic connection is more reliable and effective; on the contrary, if the information entropy of an entity is too low, even if the initial semantic similarity is high, it will be appropriately compressed through denominator normalization , thereby reducing the connection strength of low-information entities in the graph, effectively suppressing unnecessary high-strength connections of single-attribute and weak-information entities in the graph, improving the overall stability of the semantic graph and the reliability of the reasoning path, and helping the subsequent retrieval model to more accurately identify high-value semantic paths.

[0054] wherein, the calculation formula of the initial correlation score can be represented as: ; wherein, represents the weight of the kth attribute, m represents the total number of attributes, which is set according to information gain or domain knowledge, represents 1 when the semantics of two attributes and are consistent, otherwise 0, represents the rule matching score (such as whether the substation is consistent, whether it shares the maintenance period, etc.), represents the hyperparameter for adjusting the rule score weight.

[0055] Step 205: generating the target graph.

[0056] wherein, the node set of the target graph includes power equipment entities, and the edge set of the target graph includes the quantified semantic correlation degree between power equipment entities.

[0057] Based on the above calculation results, the target graph can be constructed, which can be interpreted as a power equipment semantic knowledge graph, specifically represented as wherein, the node set represents all identified power equipment entities, and the edge set represents a weighted directed semantic connection between entities. The graph has structural traversability and semantic interpretability, and can be used not only to represent direct connections between devices, but also to support information reasoning across devices and types, providing a high-dimensional semantic basis for query intent-oriented retrieval.

[0058] That is, the present application constructs a structured graph with semantic hierarchy between power equipment entities to express the complex correlation between various power equipment in the power system in terms of operating state, functional attributes, spatial layout, and maintenance events. The target graph can comprehensively reflect various semantic connections between power equipment entities, support device linkage reasoning, maintenance path analysis, similar device clustering, and provide a solid structural support for semantic retrieval.

[0059] Optionally, the step of generating the target graph comprises: calculating a semantic weight of each power equipment entity based on the frequency of occurrence of each power equipment entity in the data set; and generating the target graph based on the semantic weight of each power equipment entity.

[0060] Suppose the power equipment data set is , wherein represents the data vector of the i-th power equipment, including p feature dimensions such as type, operating state, installation location, etc. To identify key entities and measure their semantic importance in the overall system, a weighted frequency method is introduced to calculate the semantic weight of each power equipment entity . The semantic weight is defined as: ; , wherein is the frequency of occurrence of the i-th power equipment entity in the data set, and N is the total number of entities in the data set. The logarithmic term in the denominator can prevent high-frequency entities from dominating the weight, while improving the information expression ability of low-frequency high-value devices.

[0061] That is, the above method can identify key entity nodes in the construction process of the graph, and high-weight devices occupy a core position in the graph.

[0062] Optionally, the step of outputting the first set based on the query semantic vector, the device embedding vector after semantic projection, and the target function comprises: constructing a regularization term based on the quantified semantic association degree between power equipment entities; constructing a target function based on a preset loss function and the regularization term; and outputting the first set based on the query semantic vector, the device embedding vector after semantic projection, and the target function.

[0063] To introduce the auxiliary role of the graph structure in the retrieval process, a semantic association degree matrix between power equipment entities is further introduced , wherein is the device and The semantic correlation degree in the target graph. Based on this, a regularization term is constructed So that entities with similar semantics remain close in the embedding space, defined as: ; This regularization term Can preserve the structural information in the graph and improve semantic consistency; here, Indicates the projected i-th device embedding vector; Indicates the projected j-th device embedding vector.

[0064] The final constructed target function Is the weighted sum of the preset loss function And the regularization term : ; The preset loss function Is: ; Where, Indicates the Euclidean norm (the same symbol is not repeated here), Indicates the label indicator variable, When the power device entity Meets the query condition, , otherwise 0, Is the boundary interval parameter, used to limit the minimum distance between the negative sample and the query vector; max indicates maximization. This preset loss function aims to minimize the distance between the power device entity that meets the condition and the query semantic vector, while maximizing the distance of the unrelated power device entity, thereby improving the matching accuracy. That is, the above preset loss function can be understood as the projected device embedding vector Of the positive sample device Try to approach the query semantic vector q, while the projected device embedding vector Of the negative sample device Is at least greater than

[0065] . Where, Is the regularization coefficient, used to balance the weight relationship between matching accuracy and graph structure preservation. In the training process, by gradient descent solving of the target function, the optimal semantic projection matrix P can be obtained. In practical applications, through the target function, the first set

[0066] It should be noted that the introduction of the semantic correlation degree matrix between power device entities can enhance the logicality of the retrieval results, The edge weight in the target graph corresponds to the rules, such as device affiliation, physical connection, etc., to ensure that the search returns results consistent with the power system operation logic.

[0067] Considering that the current power equipment operating state, model, installation location and maintenance record and other information are usually collected by multiple sensors, and due to the complex distribution environment, non-uniform sampling frequency, communication interruption and other reasons, the data generated by these sensors often have problems such as missing, discontinuity, abnormal fluctuations and repeated records. Therefore, in order to ensure the accuracy of the target graph construction and the effectiveness and accuracy of the subsequent retrieval process, the above power equipment data is preprocessed data. That is, after obtaining the initial power equipment data, the initial power equipment data is preprocessed.

[0068] The data preprocessing method provided by the embodiment of the application will be described below.

[0069] In an embodiment, the data preprocessing process includes: obtaining initial power equipment data; in response to the existence of missing data of one power equipment entity in one dimension in the initial power equipment data, calculating the mean and variance based on the non-missing data of the dimension; determining the filling value of the power equipment entity in the dimension based on the mean and variance; based on the filling value, completing the data of the power equipment entity to obtain the power equipment data.

[0070] First, for the problem of data missing, the application proposes a method suitable for handling the case where there is missing data in a certain dimension. Let the initial power equipment data set collected be where each is a multi-dimensional data vector containing P features. If the jth dimension is missing, first calculate the mean of all non-missing samples in the dimension where , m represents the number of missing samples, and n represents the total number, and then calculate the sample variance to measure the fluctuation of the dimension. Then, the missing value is completed using the following formula: ; where represents the filled value, is an adjustable coefficient (with a value range of 0 to 1), used to control the distance between the completion result and the mean, so as to retain part of the original data distribution trend. The above completion method maintains the data center position while considering the overall dispersion, which can improve the stability.

[0071] In one embodiment, the data preprocessing process includes: acquiring initial power equipment data; and performing normalization processing on data of the same dimension of power equipment entities in the initial power equipment data to obtain power equipment data.

[0072] Features of any dimension in each data vector For example, its standardized result is denoted as : ; in, This is the mean of the feature in this dimension. The standard deviation is given. This transformation maps all features to a normal distribution with a mean of 0 and a standard deviation of 1, effectively solving the model bias problem caused by different feature dimensions, and providing a unified numerical basis for calculating semantic weights and device correlation.

[0073] In one embodiment, the data preprocessing process includes: obtaining initial power equipment data; and removing duplicate data from the initial power equipment data based on hash signature and cosine similarity determination method to obtain power equipment data.

[0074] To eliminate interference from duplicate records during device information collection, a structured deduplication method based on hash signature and cosine similarity discrimination is used to deduplicat each record. Mapped to a compact hash signature This is used for fast indexing and identifying potential duplicates. The hash value generation process involves first weighting the data and then calculating... ,in, The weight coefficient for the k-th feature; express The weighted mapping can be determined based on criteria such as information gain. For the value of this dimension; then use the hash function. Generate hash signature When the hash value distance between two data records is less than a set threshold... Furthermore, their cosine similarity in the original feature space is higher than the set value. That is, satisfying: ; If a record is identified as a duplicate, only the one with the higher confidence level is retained. This algorithm combines structural similarity and numerical similarity as dual criteria, significantly improving deduplication accuracy.

[0075] In summary, data preprocessing can complete, standardize, and remove redundancy from the collected multi-source heterogeneous power equipment data, ultimately forming a high-quality, standardized, and non-duplicated dataset. This provides a solid data foundation for subsequent semantic modeling and improves the system's robustness to complex input data.

[0076] It is considered that in the real application scenario of power equipment information retrieval, users often not only pay attention to basic fields such as equipment number or state, but also hope to obtain multi-dimensional deep information including equipment operation history, fault frequency, associated equipment state and maintenance record, and therefore the result presentation process must have the ability of semantic layering, attribute aggregation and structure expansion.

[0077] Further in an embodiment, the step of determining the retrieval result based on the similarity calculation result can specifically include: determining a to-be-output power equipment entity based on the similarity calculation result; determining an adjacent equipment set with the to-be-output power equipment entity; constructing a directed subgraph based on the adjacent equipment set; outputting the directed subgraph; wherein the directed subgraph represents the retrieval result.

[0078] Firstly, for the top k power equipment entities output by retrieval , the corresponding embedding vector set is , and each vector represents the comprehensive characteristics of an equipment in the semantic space. Let the current target graph be , wherein V is the node set, E is the semantic edge set, and the weight matrix of the edge is . In order to improve the structural readability of the result display, a local graph substructure extraction mechanism is introduced to extract the r-order adjacent equipment set of each target equipment , that is: ; , wherein represents the shortest path length between nodes and in the graph G, and r is the adjacent range radius. Each power equipment entity and its semantic adjacent equipment are uniformly constructed into a directed subgraph , wherein , is a subset of edges between the above entities in the graph G. This structure is used to show the context information associated with the retrieval target to the user, and to assist in understanding the semantic position of the equipment.

[0079] It should be noted that the directed subgraph is a substructure constructed from the target equipment node according to the entity relationship edges in the semantic graph. Its directionality rule is derived from the semantic dependence between entities, for example: From "operation state" to "fault record", it means that the state affects the fault; From "equipment" to "maintenance record", it means that the equipment contains maintenance behavior; From "equipment" to "installation location", it means the geographical affiliation relationship.

[0080] ​The directionality is based on the direction of the edge weight of the entity defined in the semantic graph modeling process.

[0081] The output directed subgraph contains the information of the target device itself and exhibits the upstream and downstream devices and their semantic paths that have strong semantic correlations with the target device, thereby presenting the search results in a structured manner. This structure is more helpful for users to understand the context and make decisions than a simple list of results, and is used to show the user the context information associated with the search target to assist in understanding the semantic position of the device.

[0082] Further, the directed subgraph not only reflects the relationship between the device and its upstream and downstream semantic entities in structure, but also reflects the logical link of the search results in semantics. When generating search results, the system takes the query target device as the center node, determines the directionality and hierarchy of its associated devices according to the size of the semantic correlation degree, and forms a top-down semantic influence path. For users, the output directed subgraph is a semantic expression form of the search results: The nodes represent the target device and its associated devices related to the query condition; The direction of the edge represents the flow direction of the semantic dependence (such as the running state causing a fault, the device containing a maintenance event, etc.); The edge weight reflects the semantic strength.

[0083] Therefore, the directed subgraph can intuitively display the semantic association logic of the search results in the form of a graph structure, so that the search results not only contain “which devices are searched”, but also contain “why they are related and how they affect each other”. This structured output method enhances the explainability and practicality of the system search results, and users can use it for device fault tracing, maintenance relationship analysis, or running state reasoning.

[0084] Optionally, the method further comprises: obtaining a feedback score of the user in the interaction process; constructing a user aggregated feedback vector based on the feedback score; updating the semantic projection matrix based on the user aggregated feedback vector, or adjusting the objective function based on the user aggregated feedback vector.

[0085] To achieve personalized output, the embodiment constructs a feedback matrix based on the user interaction behavior , wherein, n is the total number of users, and n is the number of device entities, represents the feedback score of user i to power device j, and the value range is [0, 1], and the higher the value, the more satisfied the user is with the device search result. The user aggregated feedback vector is constructed as , the jth dimension of which is: ; , wherein, A set of users who produce feedback on the power equipment j. The feedback vector is used to adjust the weight update strategy of the semantic projection matrix P, so that the semantic distance of the user high feedback equipment is further close to the query intent vector q in the training process. A feedback positive incentive term is introduced Supplement to the original objective function, defined as: ; This objective is to minimize the semantic distance between the user positive feedback equipment and the query vector, and to strengthen the learning ability of the model to high satisfaction results. Finally, the overall optimization objective function is adjusted as: ; Wherein, Indicates the optimized objective function, The original matching loss term (preset loss function), The regularization coefficient, The regularization term, The adjustment coefficient of the user feedback incentive term is used to control the response degree of the feedback preference.

[0086] It can be seen that the semantic expansion graph, attribute aggregation table and historical behavior summary are provided on the result display, and a continuous optimization feedback loop is constructed based on user behavior data, so that the system has self-learning ability. In the practical application of power equipment information retrieval, the user's understanding ability of the retrieval content and the credibility of the system output are significantly improved, and the system realizes the enhancement of semantic matching ability and the continuous evolution of result precision in long-term operation.

[0087] In summary, compared with the existing power equipment information retrieval technology which mainly relies on keyword matching, field rule screening or simple semantic model, the present application realizes high-precision response and structured output to complex query requirements by constructing a retrieval system oriented to semantic understanding and graph reasoning, and has the following significant technical advantages and effects: 1、Data integrity and quality improvement ability is significantly enhanced: the existing technology generally lacks perfect data preprocessing mechanism, and the data missing, discontinuity and redundancy generated by sensor are not fully handled, resulting in unstable retrieval accuracy. In the data acquisition and cleaning module of the present application, the missing data completion strategy with variance adjustment and the standardization and hash-similarity joint deduplication mechanism are introduced, which not only makes the equipment data consistent in statistical distribution, but also eliminates potential redundant interference terms at the structure level. Thus, a stable and high-credibility data foundation is provided for subsequent semantic modeling, which fundamentally improves the robustness and result accuracy of the retrieval system.

[0088] 2. Multi-dimensional semantic relationship modeling and graph expression of device information are realized: traditional methods only perform planar field matching on power device information, ignoring the multi-dimensional relationships such as function, space and time between devices, and are difficult to support joint queries under complex conditions. The application constructs a device relationship extraction model based on attribute similarity and rule driving, and combines an information entropy mechanism to weight the correlation between entities, so that the semantic knowledge graph has traversable and computable structural expression capabilities. On this basis, the query system can understand the semantic path and reasoning chain between devices, and realize accurate information backtracking and semantic association under multi-attribute conditions.

[0089] 3. The retrieval process realizes a dynamic projection and matching mechanism based on a semantic space: existing retrieval mechanisms mostly rely on static rules or fixed vector matching, and cannot adapt to changes in semantic direction caused by different query targets. The application constructs a semantic projection model integrated with embedding and transformation, maps device information to a unified low-dimensional semantic space, and learns an optimal semantic alignment matrix according to the query conditions, so that the retrieval model has self-adaptive adjustment capability. At the same time, by introducing a graph regularization term, the local consistency of devices with similar semantics in the embedding space is maintained, thereby improving the generalization ability and stability of the model when facing diversified retrieval scenarios.

[0090] 4. Model iteration optimization driven by user feedback is realized: traditional information retrieval methods lack effective user feedback loops and cannot dynamically adjust retrieval strategies, and the system is prone to performance degradation after long-term operation. In the result output and feedback module, the application introduces a user rating vector modeling mechanism, converts positive feedback into gradient information, and injects it into the projection model training process in the reverse direction, forming a personalized model update path driven by supervision. By constraining the matching loss function with a feedback incentive term, the model tends to match device semantic features with higher user satisfaction in the continuous operation process, realizing dynamic enhancement of system effect and self-correction of result deviation.

[0091] Please refer to Figure 3 , based on the same inventive concept, the application also provides a power device information retrieval system 300, comprising: A query acquisition module 301 is configured to acquire query information of a user for a power device, and parse the query information into a query semantic vector; wherein the dimension of the query semantic vector is the same as that of a device embedding vector; each power device entity in a pre-constructed target graph corresponds to a device embedding vector; and each device embedding vector has the same dimension; the target graph represents the association relationship between power device entities; A vector optimization module 302 is configured to perform semantic projection on the device embedding vector based on the query semantic vector, so that the query semantic vector and the device embedding vector are aligned in the same semantic space; The screening module 303 is configured to output a first set based on the query semantic vector, the semantic projected device embedding vector, and a target function, where the target function is used to determine the semantic projected device embedding vector associated with the query semantic vector. The similarity calculation module 304 is configured to calculate the similarity between each semantic projected device embedding vector in the first set and the query semantic vector. The retrieval output module 305 is configured to determine a retrieval result based on the similarity calculation result.

[0092] Please refer to Figure 4 , based on the same inventive concept, the embodiment of the present application provides a module frame of an electronic device 400 applying the above method. The electronic device 400 comprises: at least one processor 401 (only one is shown in the figure) Figure 4 , a memory 402, a computer program 403 stored in the memory 402 and executable on the at least one processor 401, and the processor 401 implements the steps of the method in any of the preceding embodiments when executing the computer program 403.

[0093] The electronic device 400 can be a server, a personal computer, a notebook computer, etc.

[0094] Those skilled in the art can understand that Figure 4 The electronic device 400 is only an example and does not constitute a limitation on the electronic device 400, which can include more or fewer components than shown, or combine certain components, or different components.

[0095] The processor 401 can be a central processing unit (CPU), and the processor 401 can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0096] The memory 402 may, in some embodiments, be an internal storage unit of the electronic device 400, such as a hard disk or memory of the electronic device 400. The memory 402 may, in other embodiments, also be an external storage device of the electronic device 400, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 400. Further, the memory 402 may, in addition, include both an internal storage unit and an external storage device of the electronic device 400.

[0097] It should be noted that the above system, device, etc. are based on the same concept as the method embodiments of the present application, and the modules designed by the system and the steps performed by the device and the resulting technical effects can be found in the method embodiments, which will not be described here.

[0098] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of functional units and modules is exemplified, and in actual applications, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or in the form of software functional unit. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the present application. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0099] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps in each of the above method embodiments.

[0100] The embodiments of the present application provide a computer program product. When the computer program product is run on a mobile terminal, the mobile terminal is caused to implement the steps in each of the above method embodiments.

[0101] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the present application can implement all or part of the processes in the above-mentioned embodiment methods through a computer program to instruct relevant hardware to complete, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable medium at least includes any entity or device capable of carrying the computer program code to the photographing device / electronic device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium. For example, U disk, mobile hard disk, magnetic disk or optical disk, etc.

[0102] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0103] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0104] In the embodiments provided by the present application, it should be understood that the disclosed apparatus / network device and method can be implemented in other ways. For example, the above-described apparatus / network device embodiments are merely schematic, for example, the division of the modules or units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0105] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may also be distributed to multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0106] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit them; 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 still 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, and should be included in the protection scope of the present application.

Claims

1. A method for retrieving information on power equipment, characterized in that, include: The system acquires user query information for power equipment and parses the query information into query semantic vectors; wherein, the dimension of the query semantic vectors is the same as that of the device embedding vectors; each power equipment entity in the pre-constructed target graph corresponds to a device embedding vector; and each device embedding vector has the same dimension; the target graph represents the association relationship between power equipment entities. Based on the query semantic vector, the device embedding vector is semantically projected so that the query semantic vector and the device embedding vector are aligned in the same semantic space; Based on the query semantic vector, the semantically projected device embedding vector, and the objective function, a first set is output; wherein, the objective function is used to determine the semantically projected device embedding vector associated with the query semantic vector; Calculate the similarity between the semantically projected device embedding vector and the query semantic vector for each semantically projected element in the first set; Based on the similarity calculation results, the search results are determined.

2. The method for retrieving power equipment information according to claim 1, characterized in that, The target map is constructed through the following steps: Acquire power equipment data; wherein, the power equipment data includes power equipment entities; Based on the attribute set of power equipment entities, determine the strength of semantic relationships between power equipment entities; Calculate the information entropy of each power equipment entity; Based on the semantic relationship strength between power equipment entities and the information entropy of each power equipment entity, the quantified semantic association degree between power equipment entities is determined. Generate the target graph; wherein the node set of the target graph includes power equipment entities, and the edge set of the target graph includes quantified semantic association between power equipment entities.

3. The method for retrieving power equipment information according to claim 2, characterized in that, The generation of the target map includes: The semantic weight of each power equipment entity is calculated based on the frequency of its occurrence in the dataset. The target graph is generated based on the semantic weight of each power equipment entity.

4. The method for retrieving power equipment information according to claim 2, characterized in that, The first set, based on the query semantic vector, the semantically projected device embedding vector, and the objective function, is output, including: Based on the quantified semantic association between power equipment entities, a regularization term is constructed; The objective function is constructed based on the preset loss function and the regularization term; Based on the query semantic vector, the semantically projected device embedding vector, and the objective function, the first set is output.

5. The method for retrieving power equipment information according to claim 2, characterized in that, The acquisition of power equipment data includes: Obtain initial power equipment data; In response to the presence of a missing data item in one dimension of the initial power equipment data, the mean and variance are calculated based on the non-missing data in that dimension. Based on the mean and variance, determine the filler value for this power equipment entity in this dimension; Based on the fill-in values, the data of the power equipment entity is completed to obtain the power equipment data.

6. The method for retrieving power equipment information according to claim 2, characterized in that, The acquisition of power equipment data includes: Obtain initial power equipment data; The power equipment data is obtained by normalizing the data of the same dimension of the power equipment entities in the initial power equipment data.

7. The method for retrieving power equipment information according to claim 2, characterized in that, The acquisition of power equipment data includes: Obtain initial power equipment data; Based on the hash signature and cosine similarity determination method, duplicate data in the initial power equipment data is removed to obtain the power equipment data.

8. The method for retrieving power equipment information according to claim 2, characterized in that, The process of determining the search results based on similarity calculation results includes: Based on the similarity calculation results, the entities of the power equipment to be output are determined; Based on the entity of the power equipment to be output, determine the set of adjacent devices; Construct a directed subgraph based on the set of connection devices; Output the directed subgraph; wherein the directed subgraph represents the retrieval result.

9. The method for retrieving power equipment information according to claim 1, characterized in that, The method further includes: Obtain user feedback scores during the interaction process; Based on the feedback score, construct the user aggregated feedback vector; The semantic projection matrix is ​​updated based on the user aggregated feedback vector, or the objective function is adjusted based on the user aggregated feedback vector.

10. A power equipment information retrieval system, characterized in that, include: The query acquisition module is used to acquire user query information for power equipment and parse the query information into query semantic vectors; wherein, the dimension of the query semantic vector is the same as that of the device embedding vector; each power equipment entity in the pre-constructed target graph corresponds to a device embedding vector; and each device embedding vector has the same dimension; the target graph represents the association relationship between power equipment entities; The vector optimization module is used to perform semantic projection on the device embedding vector based on the query semantic vector, so that the query semantic vector and the device embedding vector are aligned in the same semantic space; A filtering module is used to output a first set based on the query semantic vector, the semantically projected device embedding vector, and a target function; wherein the target function is used to determine the semantically projected device embedding vector associated with the query semantic vector. A similarity calculation module is used to calculate the similarity between each semantically projected device embedding vector in the first set and the query semantic vector; The retrieval output module is used to determine the retrieval results based on the similarity calculation results.

Images