Search method and device based on energy storage professional knowledge, equipment and storage medium

Abstract

The application provides a retrieval method and device based on energy storage professional knowledge, equipment and storage medium, relates to the technical field of retrieval, and includes the following steps: receiving an energy storage professional retrieval request; generating semantic features corresponding to the Chinese text content of the retrieval request, and extracting scalar conditions from the retrieval request; retrieving a first standardized knowledge unit from a plurality of standardized knowledge units in the knowledge base of the energy storage professional, wherein the vector of the first standardized knowledge unit matches the semantic features, and the scalar in the first standardized knowledge unit matches the scalar conditions; and returning a retrieval result to the user based on the first standardized knowledge unit. In this way, when the electronic device retrieves according to the retrieval request, the retrieval result can be more accurate.

Description

Technical Field

[0001] This application relates to the field of retrieval technology, and in particular to a retrieval method, apparatus, device and storage medium based on energy storage expertise. Background Technology

[0002] Currently, electronic devices can use keyword-based matching for retrieval; however, this method struggles to understand semantics, potentially leading to incomplete results. Alternatively, vector-based semantic retrieval can capture semantic similarity, but in energy storage, its ambiguity in the vector space can result in inaccurate results. For example, it may be difficult to precisely distinguish the subtle semantic differences between "greater than 100V" and "less than 100V," making it challenging to pinpoint accurate technical monitoring indicators. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a retrieval method, apparatus, device, and storage medium based on energy storage expertise, which can be used for retrieval in the energy storage field and improve the accuracy of retrieval results.

[0004] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: Firstly, this application provides a retrieval method based on energy storage expertise, comprising the following steps: receiving a retrieval request related to energy storage expertise; generating semantic features corresponding to the text content in the retrieval request, and extracting scalar conditions from the retrieval request; retrieving a first standardized knowledge unit from multiple standardized knowledge units in an energy storage expertise knowledge base, wherein the vector and semantic features of the text content in the first standardized knowledge unit are matched, and the scalars and scalar conditions in the first standardized knowledge unit are matched; and returning retrieval results to the user based on the first standardized knowledge unit. In other words, by combining semantic retrieval and scalar filtering, results that are semantically similar to the retrieval request and satisfy the scalar conditions in the retrieval request can be retrieved, thereby improving the accuracy of the retrieval results.

[0005] Furthermore, the aforementioned standardized knowledge units were obtained through preprocessing of knowledge data in the energy storage field. This preprocessing included mapping terms from the energy storage knowledge data to standard terms and associated metadata, including parameter ranges and / or applicable scenarios. In other words, the standardized knowledge units in the knowledge base all contain standard terms, along with their parameter ranges and / or applicable scenarios. This aligns with the composite expressions in energy storage that include both technical terms and specific numerical values / applicable scenarios, such as "battery state of charge (SOC) maintained at 30%-70%". Subsequently, when a search request contains similar expressions, accurate results can be retrieved from the knowledge base.

[0006] In one possible design approach of the first aspect, the knowledge base includes vector indexes and scalar indexes. The vector indexes include the correspondence between vectors of text content in each standardized knowledge unit and the identifiers (e.g., ids) of the standardized knowledge units. The scalar indexes include the correspondence between scalars within standardized knowledge units and the identifiers of the standardized knowledge units. Thus, after matching the vectors and scalars, the standardized knowledge units can be located based on their corresponding identifiers.

[0007] In one possible design approach of the first aspect, retrieving a first standardized knowledge unit from multiple standardized knowledge units in an energy storage knowledge base includes: matching semantic features with vectors in a vector index to obtain a matching first vector. In the vector index, the first vector corresponds to a first identifier; thus, matching the first vector is equivalent to finding a standardized knowledge unit whose identifier matches the semantic features, designated as the first identifier. In the scalar index, the first identifier corresponds to a first scalar. Next, matching the scalar conditions with the first scalar in the scalar index yields a matching second scalar. In the scalar index, the second scalar corresponds to a second identifier; thus, matching the second scalar is equivalent to finding a standardized knowledge unit among the standardized knowledge units indicated by the first identifier that further matches the scalar conditions, designated as the second identifier. Therefore, obtaining the first standardized knowledge unit indicated by the second identifier from multiple standardized knowledge units yields a standardized knowledge unit that matches both the semantic features and scalar conditions of the retrieval request.

[0008] In one possible design approach of the first aspect, the knowledge data includes inspection report table data. Preprocessing further includes converting the inspection report table data into report text and establishing semantic relationship mappings between table headers and data units within the report text. This way, when faced with complex table data in energy storage, not only can the text of the table data be obtained, but the logical structure of the table can also be reconstructed by establishing semantic relationships between table headers and data units. This facilitates more accurate retrieval results based on the logical structure of the table during subsequent searches.

[0009] In one possible design approach of the first aspect, after converting the test report table data into report text, the method further includes: verifying the report text based on a knowledge constraint framework of energy storage, resulting in a verified report text; wherein, the knowledge constraint framework includes at least one of the following: the unit of each data item and the standard range of each data item. This ensures that the verified report text conforms to the knowledge constraint framework of energy storage, such as correct units and correct numerical units. Subsequently, a semantic association mapping between the table header and data units in the report text is established, including: establishing a semantic association mapping between the table header and data units in the verified text, thereby improving the correctness of the established semantic association.

[0010] In one possible design approach of the first aspect, the knowledge data includes operation and maintenance work order image data. Preprocessing further includes: recognizing the operation and maintenance work order image data to obtain the work order text. For example, Optical Character Recognition (OCR) can be used to extract the text from the operation and maintenance work order image data. Based on a work order knowledge graph specific to energy storage, missing content in the work order text is supplemented and / or errors are corrected. For example, the work order knowledge graph includes the relationships between entities involved in the work order, and missing content in the work order text can be supplemented based on these relationships. This ensures the accuracy of the obtained work order text and avoids missing or incorrect content.

[0011] In one possible design approach of the first aspect, knowledge data also includes technical specification documents.

[0012] Secondly, this application provides a retrieval device based on energy storage expertise, comprising: a receiving module for receiving a retrieval request related to energy storage expertise; a retrieval module for generating semantic features corresponding to the text content in the retrieval request, and extracting scalar conditions from the retrieval request; and a retrieval module for obtaining a first standardized knowledge unit from multiple standardized knowledge units in a knowledge base of energy storage expertise, wherein the vector and semantic features of the text content in the first standardized knowledge unit are matched, and the scalars and scalar conditions in the first standardized knowledge unit are matched; and an output module for returning retrieval results to the user based on the first standardized knowledge unit; wherein the multiple standardized knowledge units are obtained by preprocessing knowledge data related to energy storage expertise, the preprocessing including: mapping terms in the knowledge data of energy storage expertise into standard terms and metadata associated with the standard terms, the metadata including parameter ranges and / or applicable scenarios.

[0013] Thirdly, this application provides an electronic device, including: a memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the electronic device performs a knowledge base-based retrieval method as described in any of the first aspects above.

[0014] Fourthly, this application provides a computer-readable storage medium storing computer instructions thereon, which, when executed on an electronic device, cause the electronic device to perform a knowledge base-based retrieval method as described in any of the first aspects above.

[0015] Fifthly, this application provides a computer program product, including computer instructions, which, when run on an electronic device, cause the electronic device to execute a knowledge base-based retrieval method as described in any of the first aspects above.

[0016] It is understood that the beneficial effects achieved by the apparatus described in the second aspect, the electronic device described in the third aspect, the computer-readable storage medium described in the fourth aspect, and the computer program product described in the fifth aspect can be referred to the beneficial effects in the first aspect and any of its possible design embodiments, which will not be repeated here. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a retrieval method based on energy storage expertise provided in an embodiment of this application; Figure 2 A schematic diagram illustrating a process for constructing a knowledge base, provided as an embodiment of this application; Figure 3 A flowchart illustrating a terminology enhancement method provided for an embodiment of this application; Figure 4 A schematic diagram of a data parsing process provided in an embodiment of this application; Figure 5 A schematic diagram illustrating an information completion process provided in an embodiment of this application; Figure 6 A complete flowchart for constructing and retrieving a knowledge base is provided in this application embodiment; Figure 7 A schematic diagram of a retrieval device based on energy storage expertise provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0018] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0020] In the technical solutions provided in this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved are all information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision and disclosure of the above information and data all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0021] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0022] In retrieval scenarios targeting energy storage knowledge bases, the sheer volume and variety of content, coupled with frequent updates, make keyword matching a reliance on incomplete results. While semantic vector retrieval can capture semantic relationships between texts and improve recall, its inherent semantic ambiguity can lead to less precise results.

[0023] For example, a user's search request is "safe lithium iron phosphate battery, voltage 3.2V, cycle life over 5000 cycles", and the knowledge base contains the following documents: Document 1: (Model: Lithium Iron Phosphate (LFP)-280Ah, Material: Lithium Iron Phosphate, Voltage: 3.2V, Lifespan: 6000 cycles, Safety: High, Description: LFP-280Ah lithium iron phosphate battery, excellent safety). Document 2: (Model: LFP-300Ah, Material: LFP, Voltage: 3.3V, Lifespan: 5500 cycles, Safety: High, Description: High-capacity LFP battery with good cycle performance). Document 3: (Model: NCM-100Ah, Material: Ternary, Voltage: 3.7V, Lifespan: 2000 cycles, Safety: Medium, Description: High energy density ternary battery). Document 4: (Model: LTO-20Ah, Material: Lithium titanate, Voltage: 2.3V, Lifespan: 4800 cycles, Safety: Extremely High, Description: Lithium titanate battery, ultra-long lifespan, extremely safe). Document 5: (Model: LFP-280Ah-Pro, Material: Lithium Iron Phosphate, Voltage: 3.2V, Lifespan: 5800 cycles, Safety: High, Description: Improved LFP battery with enhanced performance).

[0024] Electronic devices can find document 1 by using keyword matching search, such as "lithium iron phosphate battery", "3.2V" and "5000 times", but document 2, which uses the English letters "LFP" to represent "lithium iron phosphate battery", cannot be found.

[0025] Electronic devices can use semantic vector retrieval to transform a user's search request "safe lithium iron phosphate battery, voltage 3.2V, cycle life of more than 5000 times" into semantic features representing the text content "safe lithium iron phosphate battery, voltage, long life". Then, through the semantic association of the text content, they can find documents 1, 2, 4 and 5 in the knowledge base that are related to the text content "safe lithium iron phosphate battery, voltage, long life". However, the search results are obviously not accurate. For example, the "life" in document 4 does not meet the requirement of "more than 5000 times".

[0026] To address the aforementioned issue of inaccurate search results, such as Figure 1 As shown in the figure, this application provides a retrieval method based on energy storage expertise, including the following steps: S101. Receive the retrieval request from the energy storage field, generate the semantic features corresponding to the text content in the retrieval request, and extract scalar conditions from the retrieval request.

[0027] The search request is the search content entered by the user. For example, "Find the insulation testing standards for energy storage systems above 1000V".

[0028] Among these, generating semantic features is equivalent to generating vectors, meaning that retrieval requests can be converted into vectors. Furthermore, some retrieval requests may involve scalar conditions, such as values ​​above 1000V or cycle life above 5000 cycles. In these cases, scalar conditions must be extracted from the retrieval request so that subsequent searches yield results that satisfy these conditions.

[0029] For example, if the search request is "safe lithium iron phosphate battery, voltage 3.2V, cycle life over 5000 times", then a semantic vector can be generated based on the text content of the search request "safe lithium iron phosphate battery, voltage 3.2V, cycle life over 5000 times", and the scalar conditions "voltage 3.2V" and "cycle life > 5000 times" in the search request can be extracted.

[0030] For example, if the search request is "find the insulation testing standards for energy storage systems above 1000V", then a semantic vector can be generated based on the text content of the search request "find the insulation testing standards for energy storage systems above 1000V", and the scalar condition "voltage ≥ 1000V" in the search request can be extracted.

[0031] In one specific implementation, electronic devices can employ the Bidirectional Encoder Representations from Transformers (BERT) model to transform retrieval requests into semantic vectors. The BERT model can be trained deeply bidirectionally using text content, enabling it to simultaneously understand the contextual information on both sides of a word, thereby generating vectors.

[0032] In some embodiments, electronic devices can also infer the true search intent by analyzing historical search requests and contextual information, thereby making the generated semantic vectors and extracted scalar conditions more accurate.

[0033] For example, if the search request is "abnormal temperature of equipment in compartment 3", the electronic device can infer the true search intent as "fault handling solution for abnormal temperature of equipment in compartment 3" based on previous search requests and context information. Then, it can generate a semantic vector and extract scalar conditions based on the search intent.

[0034] S102. The first standardized knowledge unit is retrieved from multiple standardized knowledge units in the knowledge base of the energy storage profession. The vector and semantic features of the text content in the first standardized knowledge unit are matched, and the scalar and scalar conditions in the first standardized knowledge unit are matched.

[0035] In other words, the knowledge base includes multiple standardized knowledge units in the field of energy storage. These standardized knowledge units are obtained by processing knowledge data in the field of energy storage. For details on the specific implementation of obtaining multiple standardized knowledge units, please refer to the following introduction, which will not be elaborated on here.

[0036] Electronic devices can retrieve the first standardized knowledge unit that satisfies the search request from multiple standardized knowledge units by combining semantic retrieval and scalar filtering. Specifically, the semantic vector of the first standardized knowledge unit matches the semantic features (i.e., semantic vector) corresponding to the search request, and the scalar conditions of the first standardized knowledge unit also match the scalar conditions in the search request.

[0037] Taking the search request "safe lithium iron phosphate battery, voltage 3.2V, cycle life over 5000 cycles" as an example, the electronic device can calculate the similarity between the semantic vector of this search request and the semantic vectors of documents 1-5, respectively, and determine that the similarity with the semantic vectors of documents 1, 2, 4, and 5 is higher than the similarity threshold. Furthermore, the electronic device can filter out documents 2 and 4 from documents 1, 2, 4, and 5 that do not meet the scalar conditions "voltage 3.2V" and "cycle life over 5000 cycles" of the search request, thus obtaining the first standard knowledge unit including documents 1 and 5.

[0038] S103. Return the search results to the user based on the first standardized knowledge unit.

[0039] Electronic devices can generate search results, or responses, from the first standardized knowledge unit and display them to the user.

[0040] In some embodiments, the electronic device may also display a portion of the content of the first standardized knowledge unit to the user, as well as a jump link.

[0041] In some embodiments, the search results include core answers, knowledge displays, reasoning explanations, and source information. The knowledge displays include key data and information from the search results. The reasoning explanations demonstrate the reasoning process that generated the response to the user. The source information includes links to the search results; in response to a user clicking a link, the electronic device can jump to display the complete content of the first standardized knowledge unit.

[0042] In one specific implementation, electronic devices can employ Retrieval Augmented Generation (RAG) technology to construct a dual-path processing architecture: recalling the first standardized knowledge unit through the retrieval path; and generating a standardized response by performing knowledge reasoning in an isolated sandbox environment based on the first standardized knowledge unit and context through the generation path, using a large language model fine-tuned by the corpus supervised by energy storage technology.

[0043] In one specific implementation, the electronic device can also intelligently sort and combine the first standardized knowledge units according to their relevance to the retrieval request.

[0044] In some embodiments, the electronic device may record the search request and the user's click behavior as feedback data to optimize subsequent search results.

[0045] In addition to the aforementioned standardized knowledge units, the knowledge base may also include vector indexes and scalar indexes. The vector index contains the correspondence between the vectors of the text content in each standardized knowledge unit and the identifiers of the standardized knowledge units. The scalar index contains the correspondence between the scalars in the standardized knowledge units and their identifiers. Thus, the corresponding standardized knowledge unit can be found using the vectors in the vector index and the scalars in the scalar index.

[0046] Taking five standardized knowledge units as an example, the vector indexes in the knowledge base can be shown in Table 1 below: Table 1

[0047] For example, if the semantic features of the retrieval request match the vector x1, the electronic device can determine that the standardized knowledge unit indicated by identifier 111 matches the retrieval request.

[0048] Taking 5 standardized knowledge units as an example, the scalar index in the knowledge base can be shown in Table 2 below: Table 2

[0049] For example, if a scalar condition in a search request matches a scalar x1, the electronic device can determine that the standardized knowledge unit indicated by identifier 111 matches the search request.

[0050] In some embodiments, electronic devices may employ a similarity search library (Facebook AI SimilaritySearch, FASS) to build a vector index, thereby improving retrieval efficiency.

[0051] After constructing the vector index and scalar index, in some embodiments, in the above S102, the electronic device can perform parallel retrieval and then take the intersection. Specifically, semantic features are retrieved from the vector index and scalar conditions are retrieved from the scalar conditions, and then the intersection of the two is taken to obtain the first standardized knowledge unit.

[0052] In other embodiments, during S102 above, the electronic device can also perform sequential retrieval. Specifically, it first retrieves semantic features from the vector index, and then, based on the retrieval results, further retrieves scalar conditions from the scalar index to obtain the first standardized knowledge unit. Using sequential retrieval allows for a gradual narrowing of the retrieval scope and reduces the amount of data queried. This paper primarily uses this implementation method as an example.

[0053] In the sequential retrieval method, semantic features are matched with vectors in the vector index to obtain a first matching vector. In the vector index, the first vector corresponds to a first identifier. Therefore, matching the first vector is equivalent to finding a standardized knowledge unit (SBU) that matches the semantic features and is indicated by the first identifier. In the scalar index, the first identifier corresponds to a first scalar. Matching the scalar conditions with the first scalar in the scalar index yields a second matching scalar. In the scalar index, the second scalar corresponds to a second identifier. Therefore, matching the second scalar is equivalent to finding a standardized knowledge unit (SBU) that further matches the scalar conditions and is indicated by the second identifier within the SBU indicated by the first identifier. Finally, the first standardized knowledge unit indicated by the second identifier is retrieved from multiple standardized knowledge units.

[0054] Taking Tables 1 and 2 as examples, the semantic features of the retrieval request are matched with the vectors in the vector index shown in Table 1 to obtain the first matching vectors, such as vectors x1, x3, x4, and x5. The first identifiers are then 111, 333, 444, and 555, respectively. In the scalar index shown in Table 2, the first scalars corresponding to the above first identifiers are scalars y1, y3, y4, and y5, respectively. Therefore, the scalar conditions in the retrieval request can be further matched with scalars y1, y3, y4, and y5, respectively, to obtain the second matching scalars, such as scalars y1 and y3. Furthermore, in the scalar index shown in Table 2, the second identifiers corresponding to the above second scalars are 111 and 333, respectively. Therefore, the first standardized knowledge unit is the standardized knowledge unit indicated by identifiers 111 and 333, respectively.

[0055] In some implementations, vector retrieval can employ deep learning-based retrieval models, including Deep Structured Semantic Models (DSSM) and cross-encoders, combined with efficient vector indexing techniques to achieve fast retrieval. In scalar retrieval, Text-to-Structured Query Language (Text-to-SQL) technology can be used to parse natural language and extract constraints, enabling precise comparison of the original numerical values.

[0056] In some embodiments, the electronic device may also sort the retrieved first standardized knowledge units according to their relevance to the search request, and then generate search results.

[0057] In some embodiments, the electronic device can also count the frequency of query terms in the retrieval request, use precise indexing for high-frequency query terms, approximate indexing for medium-frequency query terms, and dynamic indexing for low-frequency query terms, so as to achieve the optimal balance between query efficiency and resource consumption.

[0058] For example, electronic devices can continuously count the frequency of query terms, place high-frequency terms in the fastest in-memory hash table to create an exact index, place mid-frequency terms in a solid-state drive (SSD) Bloom filter to create an approximate index, and place low-frequency terms in a database to create an approximate index, and dynamically adjust the index.

[0059] In some embodiments, the electronic device may also continuously collect and statistically analyze scalars, calculate the reasonable range of each scalar under different conditions, and thereby replace the fixed scalar parameter value with the reasonable range of the scalar when determining the second identifier.

[0060] The above methods can significantly improve the accuracy of retrieval, achieving a success rate of 83.5%, and greatly shorten the retrieval time, thereby significantly improving the work efficiency of technical personnel.

[0061] This application provides an exemplary embodiment of the specific implementation for constructing the aforementioned knowledge base, such as... Figure 2 As shown: S201. Obtain knowledge data related to energy storage.

[0062] Electronic devices acquire data from various data sources in different formats, including PDF documents, scans, images, and tables. For example, knowledge data includes real-time data from monitoring systems (such as inspection report forms), forms from work order systems (such as maintenance work order images), and operation manuals, national standards, industry standards, and technical specifications (collectively referred to as technical specification documents).

[0063] In some specific implementations, electronic devices can store acquired data in a mixed manner based on data type.

[0064] Specifically, structured data with a fixed format is stored in a relational database, unstructured data is stored in a distributed file system, and vector data is stored in a dedicated vector database, thus achieving the optimal storage solution for different types of data.

[0065] In some specific implementations, electronic devices can also be configured with different data access and operation permissions based on different user roles. For example, a user with the role of an administrator can access and operate all data, while a user with the role of an operations and maintenance personnel can only access and operate some data.

[0066] In some embodiments, the electronic device also supports multi-terminal access, allowing users to obtain relevant knowledge data in real time on-site via other mobile devices, thereby improving the efficiency and accuracy of on-site problem handling.

[0067] S202. Preprocess the knowledge data to obtain multiple standardized knowledge units.

[0068] Standardized knowledge units are obtained through preprocessing knowledge data in the energy storage field. Preprocessing yields logically clear, structurally complete, and terminologically consistent standardized knowledge units, which improves the efficiency of subsequent retrieval.

[0069] Standardized knowledge units refer to complete knowledge encapsulations containing text content, structured data, metadata, and unique identifiers. They are the smallest knowledge units in a knowledge base that can be independently stored, retrieved, and reused.

[0070] For example, the knowledge base contains raw data: Data 1 and Data 2. Data 1 comes from inspection records and is in tabular format, while Data 2 comes from the monitoring system and is in JavaScript Object Notation (JSON) format. Electronic devices can transform Data 1 and Data 2 into standardized knowledge units. The processed Data 1 and Data 2 have the same structure but different content.

[0071] In some embodiments, the preprocessing of knowledge data includes data parsing, information completion, and terminology enhancement, which will be described separately below.

[0072] First, terminology enhancement.

[0073] Terminology enhancement can standardize the terminology, correct typos, and unify synonyms in the text content of knowledge data, and associate its parameter range and applicable scenarios, thereby making the text content of knowledge data more reliable. The text content in the knowledge data can be text extracted or identified from the knowledge data itself.

[0074] For example, the knowledge data includes the text content "state of charge". Electronic devices can use terminology enhancement to map this text content to the standard term "State of Charge (SOC)" and associate it with the standard measurement range (0%-100%).

[0075] For example, the knowledge data includes the text content “LFP battery”, “lithium iron phosphate battery”, and “LiFePO4”. The electronic device maps these three data to the standard term “lithium iron phosphate battery” through terminology enhancement and associates the parameter range voltage (2.5, 3.65).

[0076] In some embodiments, such as Figure 3 As shown, term enhancement includes the following steps: S301. Construct the terminology feature space.

[0077] Electronic devices can construct a terminology feature space based on a professional terminology database of energy storage expertise. The terminology feature space includes text features, parameter ranges, applicable scenarios, etc.

[0078] For example, the constructed term feature space is shown in Table 3: Table 3

[0079] In some specific implementations, electronic devices can also recognize changes in terminology across different periods and standards. For example, the standard term "lithium iron phosphate battery" was used as "lithium iron battery" from 2010-2015, "LFP battery" from 2015-2020, and "lithium iron phosphate battery" from 2020 to the present. As another example, the standard term "lithium iron phosphate battery" might be used as "lithium iron phosphate battery" in one standard document and "lithium iron battery" in another.

[0080] Electronic devices can establish time-dimensional mappings and standard-dimensional mappings based on the changes in the expression of terms in different periods and standards, and construct a term similarity matrix to enrich the term feature space, providing richer semantic associations and mapping relationships for subsequent dynamic matching.

[0081] In some embodiments, the electronic device employs a multi-level caching strategy, storing high-frequency terms in memory, mid-frequency terms in SSDs, and low-frequency terms in a database, thereby achieving the best balance between processing efficiency and resource consumption through hierarchical storage.

[0082] S302. Dynamically match the text content in the knowledge data with the terminology feature space.

[0083] For example, the text content is "The voltage of the lithium iron phosphate battery is too high". Electronic devices can decompose the text content into the words "lithium iron phosphate battery", "voltage", and "too high", and then dynamically compare each word with the terminology feature space. Taking the word "lithium iron phosphate battery" as an example, candidate terms in the terminology feature space include "lithium iron phosphate battery", "ternary lithium battery", and "lithium titanate battery".

[0084] In some specific implementations, before dynamically matching the text content with the terminology feature space, electronic devices can also use Retrieval Augmented Generation (RAG) technology to construct a dual-path processing architecture. The retrieval path acquires knowledge related to the text content, while the generation path standardizes the fuzzy text fields, providing more accurate candidate terms for subsequent dynamic matching.

[0085] S303, Sort the candidate terms.

[0086] Candidate terms are scored using a weighted average of edit distance and semantic similarity, and then ranked according to their scores. Edit distance measures the similarity between two words, while semantic similarity measures the closeness of meaning between two words.

[0087] Continuing with the example above, the candidate terms "lithium iron phosphate battery," "ternary lithium battery," and "lithium titanate battery" are scored based on edit distance and semantic similarity. For example, "lithium iron phosphate battery" scores 0.98, "ternary lithium battery" scores 0.76, and "lithium titanate battery" scores 0.45. Ranking these three candidate terms according to their scores, "lithium iron phosphate battery" is first, "ternary lithium battery" is second, and "lithium titanate battery" is third. The best match is "lithium iron phosphate battery," which has the highest score.

[0088] In some embodiments, after weighted scoring of candidate terms but before ranking, when a term has different definitions in different documents, such as the candidate term "lithium iron phosphate battery," which has a voltage range of 2.5~3.65V in one standard document and 2.8~4.0V in another, the electronic device can determine the most appropriate definition based on factors such as the usage scenario and the version of the standard document. If the electronic device cannot determine the definition based on these factors, it can mark the term to prompt manual confirmation.

[0089] S304. Generate enhanced knowledge data.

[0090] Specifically, words in the text can be mapped to the best matching results in the terminology feature space, along with their associated parameter ranges and applicable scenarios.

[0091] Continuing with the examples in S302 and S303, the electronic device can map "Lithium iron phosphate battery" in "Lithium iron phosphate battery voltage is too high" to "Lithium iron phosphate battery, voltage (V): 2.5-3.65, temperature (°C): -20-60, cycle life (cycles): 4000-8000, applicable scenarios: energy storage power station, electric vehicle, backup power supply", thus obtaining the enhanced knowledge data as "Lithium iron phosphate battery (voltage (V): 2.5-3.65, temperature (°C): -20-60, cycle life (cycles): 4000-8000, applicable scenarios: energy storage power station, electric vehicle, backup power supply) voltage is too high".

[0092] Second, data analysis.

[0093] In the field of energy storage, technical supervision reports and testing reports contain a large amount of complex tabular data. These tables often have multi-level headers, nested structures, or parallel structures. In some implementation methods, only the text content can be extracted, making it difficult to accurately reconstruct the logical structure of the table. This results in the loss of semantic connections between the headers and data units, failing to meet subsequent retrieval needs.

[0094] Through data parsing, electronic devices can not only obtain the text in complex tables, but also reconstruct the logical structure of the tables, which is beneficial for obtaining more accurate search results based on the logical structure of the tables during subsequent searches.

[0095] In some embodiments, such as Figure 4 As shown, data parsing includes the following steps: S401. Convert the test report table data into report text.

[0096] For example, the test report form is shown in Table 4 below: Table 4

[0097] Electronic devices use Optical Character Recognition (OCR) technology to convert Table 4 into text.

[0098] S402. Establish a semantic association mapping between table headers and data units in the report text.

[0099] In some embodiments, electronic devices can establish semantic association mappings using deep learning-based recognition models. This model can incorporate attention-enhanced Transformer networks to improve the electronic device's ability to recognize text regions, layout structures, and semantic relationships in table images. During model training, data augmentation strategies can be employed to simulate various complex table styles, thereby enhancing the model's generalization ability.

[0100] For example, the data in row 5 of Table 4 above (i.e., the data in row B001) can be mapped to the table header in the following semantic relationship: The measured voltage value of the test items for battery number B001 is 3.21V, the standard limit is 3.0-3.4V, the measured internal resistance value is 0.52mΩ, and the standard limit is <0.8mΩ.

[0101] In some embodiments, after S501, the electronic device can further verify the report text based on a knowledge constraint framework specific to energy storage, obtaining a verified report text. The knowledge constraint framework includes at least one of the following: the units of each data item and the standard range of each data item. This ensures that the verified report text conforms to the knowledge constraint framework of energy storage, such as correct units and correct numerical units. Subsequently, a semantic association mapping between the header and data units in the verified text can be established, thereby improving the correctness of the established semantic association.

[0102] For example, the knowledge constraint framework includes the following predefined set of rules: {Rule 1: Detection items must correspond to valid units of measurement; Rule 2: Standard limits should be consistent with industry standards.......}, thereby ensuring that data processing complies with domain standards and requirements.

[0103] After identifying data in the table that exceeds the normal range or violates logical relationships, the data can be marked to remind humans to review it in a timely manner.

[0104] Third, information completion.

[0105] Issues with the quality and completeness of work order data. In actual operation and maintenance processes, work order information often suffers from problems such as missing fields, ambiguous terminology, OCR recognition misalignment, or spelling errors. Traditional processing methods rely on manual verification and supplementation, which is inefficient and prone to errors.

[0106] By completing the information, electronic devices can correct incomplete or non-standard information in the text content extracted from the image data of maintenance work orders.

[0107] Among them, electronic devices can use OCR or other technologies to extract text content from image data of maintenance work orders.

[0108] In some embodiments, such as Figure 5 As shown, information completion includes the following steps: S501, Construct a knowledge graph.

[0109] Electronic devices construct knowledge graphs based on the acquired knowledge data.

[0110] For example, the knowledge graph in the field of energy storage includes entities such as "energy storage equipment", "testing items", "operation and maintenance personnel", and "standards and specifications", and the relationships between entities are such as "energy storage equipment - execution - testing items" and "operation and maintenance personnel - responsible for - energy storage equipment".

[0111] In one possible implementation, electronic devices can also fuse knowledge data in multiple formats before constructing a knowledge graph, providing more comprehensive contextual support for information completion.

[0112] S502, Entity Link Matching.

[0113] Electronic devices use contextual information to detect and identify fields in knowledge data that have incomplete or erroneous text content, then match them with entities in the knowledge graph, and select the appropriate entity based on the context.

[0114] For example, if the extracted text includes "perform insulation testing on a certain lithium iron phosphate battery pack" but does not specify the battery pack number, the electronic device can automatically infer and complete the correct battery pack number based on contextual information, such as the work order location and the area the operator is responsible for, combined with the device location relationships in the knowledge graph. For instance, the correct battery pack number can be determined based on the area the maintenance personnel are responsible for and the battery pack numbers included in that area, such as determining the correct battery number as 003.

[0115] S503, Information Completion.

[0116] For example, after an electronic device completes the statement "perform insulation test on a certain lithium iron phosphate battery pack", the completed result can be "perform insulation test on lithium iron phosphate battery pack 003".

[0117] Furthermore, electronic devices can also verify the rationality of supplementary knowledge data, ensuring the accuracy of the supplemented knowledge data.

[0118] Practice has shown that by completing information, the completeness of key data field identification can be improved by 42.8%, and the omission rate of important entities can be reduced to 4.6%.

[0119] In one possible implementation, to improve the accuracy of the completion results, the electronic device can also adopt a confidence assessment mechanism to score the completion results. Completion results with low confidence scores are marked as "pending review" and then manually reviewed before being included in the knowledge base.

[0120] In some specific implementations, electronic devices can also preprocess the acquired data in real time before storing it in a knowledge base.

[0121] For example, electronic devices can use a format conversion engine to convert data in different formats into a unified JSON structured intermediate format. Then, Natural Language Processing (NLP) technology is applied to standardize the terminology used in the acquired data. For instance, different expressions such as "Battery Management System," "BMS System," and "Battery Management Unit" can be unified into the standard term "Battery Management System (BMS)." Simultaneously, electronic devices can also standardize numerical units, such as unifying "kWh," "kilowatt-hour," and "degree" into "kWh."

[0122] In some embodiments, the preprocessed data can also be manually verified and corrected using electronic devices. The electronic devices record these manual modifications to optimize the algorithms used for automated processing, thereby improving the reliability of the data in the knowledge base.

[0123] In some embodiments, electronic devices can record the source, processing, and version evolution of data in the knowledge base, thereby improving the reliability of the data in the knowledge base.

[0124] Furthermore, electronic devices can automatically identify updated content in the knowledge base, establish version difference comparisons and correlation mappings for the data, and return the latest version of the data during subsequent queries.

[0125] For example, in the knowledge base, version 1 is "insulation test on lithium iron phosphate battery pack 003" and version 2 is "insulation test on lithium iron phosphate battery pack 003". Electronic devices can automatically recognize that the name of the battery pack has changed from "lithium iron phosphate battery" to "lithium iron phosphate battery", establish a relationship between version 1 and version 2, and finally return the latest version 2 data in subsequent queries.

[0126] In some embodiments, after data is stored in the knowledge base, the electronic device can also automatically evaluate the data in the knowledge base and combine it with manual sampling to ensure the reliability of the data in the knowledge base.

[0127] In some embodiments, the electronic device can also adjust the weight and sorting of data in the knowledge base based on subsequent user query conditions, clicked query results, and feedback evaluations, thereby improving subsequent retrieval efficiency and retrieval quality.

[0128] For example, electronic devices assign higher weights to data that users frequently query and that receive high ratings, and lower weights to data that users rarely query and that receive low ratings. Then, these data are sorted according to their weights, so that in subsequent searches, the data that ranks higher can be retrieved first.

[0129] In some embodiments, when the same information exists in different data in the knowledge base, the electronic device can mark these information and prompt human judgment to ensure the reliability of the data in the knowledge base.

[0130] For example, the information "temperature abnormality" is expressed as "high temperature" in data 1, "temperature abnormality" in data 2, and "overheating" in data 3. Electronic devices mark this information to prompt human judgment and select a uniform term such as "temperature abnormality".

[0131] In some embodiments, electronic devices store high-activity (high-frequency use + high timeliness) data in the main storage space and migrate low-activity (low-frequency use + low timeliness) data to the secondary storage space based on the two dimensions of data usage frequency and timeliness, and all data can still be retrieved.

[0132] It should be noted that the above-mentioned "first, terminology enhancement," "second, data parsing," and "third, information completion" do not have a strict execution order; they can be executed in parallel or sequentially. For example, an electronic device can first execute data parsing and information completion in parallel, and then, based on the results of parsing and completion, further perform terminology enhancement. As another example, an electronic device can first perform data parsing, then perform information completion, and finally perform terminology enhancement.

[0133] In addition, technical documents from different power plants and departments, including PDF specifications, Excel reports, Word reports, and other formats, can be uniformly converted into a structured intermediate format by electronic devices through a format conversion engine. Then, Natural Language Processing (NLP) technology is used for terminology standardization: identifying professional terms in the documents, comparing them with a standard terminology database, and unifying inconsistent expressions.

[0134] S203. Store multiple standardized knowledge units into the knowledge base.

[0135] Through the above preprocessing, electronic devices can obtain multiple standardized knowledge units, such as one standardized data unit corresponding to one row in the test report table, or one standardized knowledge unit corresponding to one maintenance work order.

[0136] In some embodiments, the electronic device can also store the aforementioned vector indexes, scalar indexes, and knowledge graphs in a private knowledge base, thereby achieving secure storage of knowledge data.

[0137] See Figure 6Using the method of this application embodiment, the electronic device can preprocess knowledge data related to energy storage, such as test report forms, maintenance work order images, and technical specification documents, through processes such as data parsing, information completion (where the knowledge graph used can be stored in a knowledge base), and terminology enhancement, to obtain multiple standardized knowledge units. These standardized knowledge units, along with vector and scalar indexes, are then stored in the knowledge base. Subsequently, upon receiving a retrieval request, the electronic device can retrieve the first matching standardized knowledge unit from the knowledge base using a hybrid vector / scalar retrieval method and generate retrieval results.

[0138] The method provided in this application effectively solves the problems of numerous standards and specifications and difficulties in locating knowledge points in energy storage technology supervision, enabling rapid location and intelligent management of knowledge points in different types of standards and specifications. Through the precise parsing technology of structured test report data, the structured extraction accuracy of complex table data is ≥90%, while fully preserving the semantic relationships between data. Through the intelligent completion technology of multimodal work order information, the completeness of key field identification is improved by ≥40%, and the omission rate of important business entities is reduced to below 5%. Through domain terminology enhancement and vector / scalar hybrid retrieval mechanism, the success rate of knowledge base retrieval tools is ≥80%. Overall, it improves the efficiency and accuracy of energy storage professional operation and maintenance technology supervision, and provides comprehensive technical support for rapid knowledge retrieval, reasoning, tracing, display, and click-to-jump.

[0139] In a pilot application at a power grid company, the system successfully integrated technical documents from 12 energy storage power stations across 8 different types, covering over 3,000 technical standards and specifications. It processed over 2,000 query requests daily, achieving a user satisfaction rate of 95.6%. System performance testing showed that, at a scale of hundreds of millions of knowledge units, the average retrieval response time was no more than 800 milliseconds, supporting over 200 concurrent user accesses and meeting enterprise-level application requirements. Through in-depth analysis of knowledge usage behavior, the system also identified potential conflicts and knowledge gaps between several standard documents, providing data support for improving the technical standard system. Furthermore, the system significantly reduced new employee training costs, shortened the learning curve for technical personnel, and improved overall operation and maintenance quality and efficiency.

[0140] The following describes a knowledge base retrieval device provided by an embodiment of this application. The knowledge base retrieval device described below corresponds to and can be referred to in conjunction with the knowledge base retrieval method described above.

[0141] See Figure 7 The retrieval device 700 includes: a receiving module 701, a retrieval module 702, an output module 703, and a preprocessing module 704.

[0142] Receiver module 701 is used to receive retrieval requests from the energy storage field; The retrieval module 702 is used to generate semantic features corresponding to the text content in the retrieval request, and to extract scalar conditions from the retrieval request; and to retrieve a first standardized knowledge unit from multiple standardized knowledge units in the knowledge base of the energy storage profession, wherein the vector and semantic features of the text content in the first standardized knowledge unit are matched, and the scalar and scalar conditions in the first standardized knowledge unit are matched. Output module 703 is used to return search results to the user based on the first standardized knowledge unit; The preprocessing module 704 is used to preprocess the knowledge data of the energy storage profession to obtain multiple standardized knowledge units. The preprocessing includes mapping the terms in the knowledge data into standard terms and the metadata associated with the standard terms. The metadata includes parameter ranges and / or applicable scenarios.

[0143] In some embodiments, the knowledge base includes a vector index and a scalar index. The vector index includes a correspondence between the vectors of text content in each standardized knowledge unit and the identifiers of the standardized knowledge units, while the scalar index includes a correspondence between the scalars in the standardized knowledge units and the identifiers of the standardized knowledge units.

[0144] In some embodiments, the retrieval module 702 is further configured to match semantic features with vectors in the vector index to obtain a matched first vector; wherein, in the vector index, the first vector corresponds to a first identifier, and in the scalar index, the first identifier corresponds to a first scalar. The scalar condition is matched with the first scalar in the scalar index to obtain a matched second scalar; wherein, in the scalar index, the second scalar corresponds to a second identifier. The first standardized knowledge unit indicated by the second identifier is obtained from multiple standardized knowledge units.

[0145] In some embodiments, the knowledge data includes detection report table data. The preprocessing module 704 described above is also used to convert the detection report table data into report text and establish a semantic association mapping between the table header and data units in the report text.

[0146] In some embodiments, the preprocessing module 704 is further configured to verify the report text based on a knowledge constraint framework specific to energy storage, thereby obtaining a verified report text; wherein the knowledge constraint framework includes at least one of the following: the unit of each data item and the standard range of each data item. A semantic association mapping between the table header and data units in the verified text is established.

[0147] In some embodiments, the knowledge data includes operation and maintenance work order image data. The preprocessing module 704 described above is further used to identify the operation and maintenance work order image data to obtain work order text. Based on the work order knowledge graph of the energy storage profession, it completes the missing content in the work order text and / or corrects any errors in the work order text.

[0148] In some embodiments, knowledge data also includes technical specification documents.

[0149] The energy storage-based retrieval device provided in this application can execute the method shown in the above-described method embodiments. Its implementation principle and beneficial effects can be found in the relevant descriptions in the method embodiments, and will not be repeated here. Furthermore, each module in the above-described energy storage-based retrieval device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0150] In other embodiments, a computer device is provided, which may be the electronic device in the above method embodiments, for executing the method steps performed by the electronic device in the above method flow. The internal structure diagram of this computer device may be as follows: Figure 8 As shown, it includes a processor, memory, input / output interface, communication interface, display unit, and input device.

[0151] The computer device comprises a processor, internal memory, and input / output interfaces connected via a system bus. A communication interface, display unit, and input devices are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external electronic devices; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a retrieval method based on energy storage expertise. The display unit is used to create a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0152] Those skilled in the art will understand that Figure 8The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0153] In some embodiments, a computer device includes a memory, a processor, and a communication interface. The communication interface is used to interact with other devices to send and receive data. For example, in this embodiment, the communication interface can specifically be used to receive data. The memory is used to store computer program code, which includes computer instructions. These computer instructions run in the computer device to implement the method shown in the above-described method embodiments. For example, the memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, portable hard drive, read-only memory, disk, or optical disk, etc.

[0154] The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a network processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The processor can also be other general-purpose processors. A general-purpose processor can be a microprocessor or any conventional processor.

[0155] Memory, communication interfaces, and processor communication connections. For example, memory and communication interfaces can connect to the processor via the system bus and communicate with each other. The system bus can be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, an Industry Standard Architecture (ISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0156] Alternatively, the memory can be either standalone or integrated with the processor. When the memory is set up independently, it is connected to the processor via the system bus.

[0157] This application also provides a chip for executing instructions, which is used to execute the technical solution of the retrieval method based on energy storage expertise in the above embodiments.

[0158] This application also provides a computer-readable storage medium storing computer instructions. When these computer instructions are executed by a processor, they are used to implement the technical solution of the knowledge base-oriented retrieval method described in the above embodiments. Specifically, when the computer instructions are executed by a processor, the computer device can execute the technical solution of the energy storage-based professional knowledge retrieval method described in the above embodiments.

[0159] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the retrieval method based on energy storage expertise described in the above embodiments.

[0160] The aforementioned computer-readable storage media can be implemented using any type of volatile or non-volatile storage device or a combination thereof. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random-access memory (ReRAM), magnetic random-access memory (MRAM), ferroelectric random-access memory (FRAM), phase-change memory (PCM), graphene memory, electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), etc.

[0161] Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take many forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). Computer-readable storage media may be any available medium accessible to general-purpose or special-purpose computers.

[0162] An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic control unit or main control device; this application does not limit this.

[0163] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0164] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0165] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0166] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0167] It should be understood that the steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.

[0168] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0169] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

[0170] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A retrieval method based on energy storage expertise, characterized in that, include: Receive search requests related to energy storage. Generate semantic features corresponding to the text content in the search request, and extract scalar conditions from the search request; The first standardized knowledge unit is obtained by retrieving multiple standardized knowledge units from the knowledge base of the energy storage specialty; The search results are returned to the user based on the first standardized knowledge unit; The multiple standardized knowledge units are obtained by preprocessing the knowledge data of the energy storage profession. The preprocessing includes mapping the terms in the knowledge data of the energy storage profession into standard terms and the metadata associated with the standard terms. The metadata includes parameter ranges and / or applicable scenarios. The knowledge base includes a vector index and a scalar index; the vector index includes the correspondence between the vectors of the text content in each standardized knowledge unit and the identifier of the standardized knowledge unit, and the scalar index includes the correspondence between the scalars in the standardized knowledge unit and the identifier of the standardized knowledge unit. The step of retrieving the first standardized knowledge unit from multiple standardized knowledge units in the knowledge base of the energy storage specialty includes: matching the semantic features with vectors in the vector index to obtain a matched first vector; wherein, in the vector index, the first vector corresponds to a first identifier, and in the scalar index, the first identifier corresponds to a first scalar; matching the scalar condition with the first scalar in the scalar index to obtain a matched second scalar; wherein, in the scalar index, the second scalar corresponds to a second identifier; and obtaining the first standardized knowledge unit indicated by the second identifier from the multiple standardized knowledge units.

2. The method according to claim 1, characterized in that, The knowledge data includes test report form data; The preprocessing further includes: converting the detection report table data into report text, and establishing a semantic association mapping between the table header and data units in the report text.

3. The method according to claim 2, characterized in that, After converting the test report form data into report text, the method further includes: Based on the knowledge constraint framework of the energy storage profession, the report text is verified to obtain the verified report text; wherein, the knowledge constraint framework includes at least one of the following: the unit of each data and the standard range of each data; The step of establishing a semantic association mapping between the table header and data units in the report text includes: establishing a semantic association mapping between the table header and data units in the verified text.

4. The method according to claim 1, characterized in that, The knowledge data includes operation and maintenance work order image data; The preprocessing further includes: The operation and maintenance work order image data is identified to obtain the work order text; Based on the work order knowledge graph of the energy storage profession, the missing content in the work order text is supplemented and / or the erroneous content in the work order text is corrected.

5. The method according to claim 1, characterized in that, The knowledge data also includes technical specification documents.

6. A retrieval device based on energy storage expertise, characterized in that, include: The receiving module is used to receive retrieval requests from the energy storage field. The retrieval module is used to generate semantic features corresponding to the text content in the retrieval request, and to extract scalar conditions from the retrieval request; and to retrieve the first standardized knowledge unit from multiple standardized knowledge units in the knowledge base of the energy storage profession. The output module returns the search results to the user based on the first standardized knowledge unit; The multiple standardized knowledge units are obtained by preprocessing the knowledge data of the energy storage profession. The preprocessing includes mapping the terms in the knowledge data of the energy storage profession into standard terms and the metadata associated with the standard terms. The metadata includes parameter ranges and / or applicable scenarios. The knowledge base includes a vector index and a scalar index; the vector index includes the correspondence between the vectors of the text content in each standardized knowledge unit and the identifier of the standardized knowledge unit, and the scalar index includes the correspondence between the scalars in the standardized knowledge unit and the identifier of the standardized knowledge unit. The retrieval module is further configured to: match the semantic features with the vectors in the vector index to obtain a matched first vector; wherein, in the vector index, the first vector corresponds to a first identifier, and in the scalar index, the first identifier corresponds to a first scalar; match the scalar condition with the first scalar in the scalar index to obtain a matched second scalar; wherein, in the scalar index, the second scalar corresponds to a second identifier; and obtain the first standardized knowledge unit indicated by the second identifier from the plurality of standardized knowledge units.

7. An electronic device, characterized in that, include: A memory and at least one processor; the memory is communicatively connected to the processor; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the electronic device performs the method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, are used to implement the method as described in any one of claims 1-5.

Images