Intelligent measuring point matching method and system based on large model

Abstract

The invention relates to the technical field of intelligent electric power, in particular to an intelligent measuring point matching system based on a large model, which comprises a data reading unit, a data search library, a result verification unit, a cleaning and importing unit, a large model matching unit and an auditing and correcting database. According to the method, independent matching set selection is carried out through three dimensions of the measuring point name, the measuring point type and the path hierarchy, the matching calculation amount is reduced, and the preliminary matching efficiency is improved; according to the method, when the name of the measuring point is selected, the target data item can be selected when any one of the text similarity and the semantic similarity of the target data item accords with the text similarity and the semantic similarity, the error-tolerant rate of preliminary matching of the measuring point is improved, the initial text similarity and the initial semantic similarity of the data search library are adjusted, the data search library can select a more appropriate result matching set according to the actual situation, and the accuracy of matching is improved. And screening of large model matching units or measurement point name expansion is facilitated, and the measurement point matching efficiency is further improved on the basis of guaranteeing the matching accuracy.

Description

Technical Field

[0001] This invention relates to the field of smart power technology, and in particular to a smart measurement point matching method and system based on a large model. Background Technology

[0002] As a core infrastructure for the operation of modern society, the stable operation of the power system is directly related to the sustainable development of the national economy and the normal order of people's lives. However, with the increasing scale and complexity of the power grid, the management of power system equipment ledgers and measurement point data faces new challenges. In particular, when performing cross-system integration, data docking, or cross-platform analysis, there are often significant differences in equipment naming rules, measurement point numbering methods, and data structures between different systems, making it extremely difficult to establish effective data associations between these systems.

[0003] Currently, the power system faces two major challenges in the management of equipment ledgers and measurement point data: First, inconsistent naming conventions among different systems make it difficult to quickly and accurately match measurement point data; second, information integration is difficult, as measurement point data is scattered across different management systems, such as substation monitoring systems, energy management systems, and production management systems, lacking a unified data format and standard, making the effective integration and utilization of information a major challenge.

[0004] Traditional measurement point matching typically employs a combination of automated and manual methods. The process includes the following steps: Preliminary path-level screening: First, candidate measurement points are initially filtered by comparing the device hierarchy of the source and target systems, such as voltage level, bay level, and device name. Regular expression matching: Regular expressions are then written according to preset naming conventions to attempt to match measurement point names, numbers, and other fields to identify potential correspondences. Precise string comparison: The candidate results after regular expression matching are further evaluated using equals or fuzzy similarity algorithms to improve matching accuracy. Final manual confirmation: For measurement points that were not clearly identified by the automated or semi-automatic matching, technicians must verify and manually confirm each one.

[0005] While this approach offers some efficiency improvements compared to relying entirely on manual operation, it still suffers from numerous problems. For example, inconsistent path levels can lead to incorrect filtering; regular expression rules struggle to cover all naming conventions; and string matching cannot handle abbreviations, aliases, and other similar cases. Therefore, in real-world scenarios involving large-scale, multi-system operations and diverse naming conventions, traditional methods remain inefficient and error-prone, failing to meet the development needs of smart grids and digital operation and maintenance platforms. With the continuous expansion of system scale and the exponential growth in the number of measurement points, traditional manual or semi-automatic matching methods can no longer meet the demands for efficient and accurate data integration in practical engineering applications. Especially in the construction of new smart grids, digital substations, and big data analytics platforms, how to quickly and accurately complete measurement point matching between multiple systems has become a crucial aspect of improving system integration efficiency and data governance capabilities.

[0006] With the rapid development of artificial intelligence technology, advanced technologies such as knowledge graphs and large-scale model thinking chains have provided new solutions for the management and analysis of power system equipment ledgers and measurement point data. Knowledge graphs, as a structured knowledge representation method, can effectively integrate multi-source heterogeneous data to build a knowledge base covering information such as equipment ledgers, measurement point numbers, and maintenance records, providing rich background support for measurement point matching. Large-scale model thinking chains, by simulating the thinking process of human experts, propose a multi-level thinking chain of "preliminary matching - semantic understanding - contextual reasoning - final matching," guiding the system to gradually and deeply analyze measurement point characteristics and formulate scientifically reasonable matching schemes.

[0007] Therefore, developing an intelligent matching method and system for power system equipment ledgers and measurement point data based on knowledge graphs and large-scale model thinking chains is of great significance for improving power system operation and maintenance efficiency and ensuring the safe and stable operation of the power grid. This method and system aim to achieve rapid integration and in-depth analysis of measurement point information by constructing a knowledge base and knowledge graph for equipment ledgers and measurement point data, combined with multi-level thinking chain analysis technology. This provides operation and maintenance personnel with scientific and reasonable matching suggestions, thereby effectively shortening data integration time, reducing error rates, and improving the overall operational efficiency and safety of the power system. Summary of the Invention

[0008] To address this, the present invention provides an intelligent measurement point matching method and system based on a large model, which overcomes the problems of low matching efficiency and poor accuracy in existing power system measurement point matching.

[0009] To achieve the above objectives, this invention provides an intelligent measurement point matching method and system based on a large model, comprising: The data search library is used to construct several searchable target data items from the target system measurement point data, and to obtain the search object text of the original system measurement point data divided by the data reading unit. It matches and calculates real-time text similarity and real-time semantic similarity, and selects target data items whose real-time text similarity reaches the initial text similarity and / or whose real-time semantic similarity reaches the initial semantic similarity to enter the measurement point name matching set. The data search library selects target data items based on the measurement point type of the search object text, generates a measurement point type matching set, selects target data items based on the path level of the search object text, generates a path level matching set, and outputs the target data items of the intersection of the measurement point name matching set, the measurement point type matching set, and the path level matching set as the result matching set. The result verification unit is used to output the result matching set as the final result when there is only one target data item in the result matching set; the result verification unit can also adjust the initial text similarity and initial semantic similarity in the data search library according to the proportion of each type in the result matching set. The large model matching unit is used to filter target data items in the result matching set when there are multiple target data items in the result matching set; or to expand the measurement point names of the search object text when there are no target data items in the result matching set, generate search reconstructed text, and output it to the data search library for rematching.

[0010] Furthermore, the data reading unit receives the original system measurement point data and divides the original system measurement point data into several structured search object texts according to the field definitions; The data search library is used to traverse the text of each search object divided by the measurement point data of the original system. For any text of a search object, the data search library matches and calculates the real-time text similarity and real-time semantic similarity between the measurement point name of the text of the search object and the measurement point name of each target data item in the data search library. Target data items whose real-time text similarity reaches the initial text similarity and / or whose real-time semantic similarity reaches the initial semantic similarity are selected to enter the measurement point name matching set.

[0011] Furthermore, when selecting target data items based on the measurement point type of the search object text, the data search library matches the measurement point type of the search object text with the measurement point type of each target data item stored in the data search library, and selects the target data items with the same measurement point type as the measurement point type matching set. If the measurement point type of the search object text is empty, then all target data items stored in the data search database are selected as the measurement point type matching set. If the measurement point type of any target data item stored in the data search library is empty, then the target data item is determined to have the same measurement point type as the search object text.

[0012] Furthermore, the result verification unit can determine the type of the result matching set. If there is only one target data item in the result matching set, the result verification unit will output the target data item in the result matching set as the final result to the cleaning and importing unit. The cleaning and importing unit will import the measurement point ID of the target data item matched in the target system measurement point data and fill it into the corresponding position of the original system measurement point data to complete the measurement point matching. If the result matching set contains multiple target data items, the result validation unit will output the result matching set to the large model matching unit. If there is no target data item in the result matching set, the result validation unit will output the result matching set to the large model matching unit.

[0013] Furthermore, the result verification unit also sets a standard invalidity ratio. When there is no target data item in the result matching set, the search object text is determined to have no valid match. The result verification unit calculates the real-time invalidity ratio of the search object text with no valid match to the search object text output by all completed result matching sets. When the real-time invalidity ratio is greater than the standard invalidity ratio, the initial text similarity and initial semantic similarity in the data search database are adjusted and reduced.

[0014] Furthermore, the result verification unit is also set with a maximum number of matching items. When there are multiple target data items in the result matching set, the result verification unit compares the number of target data items in the result matching set with the maximum number of matching items. If the number of target data items in the result matching set is greater than the maximum number of matching items, the corresponding search object text is marked as the search object text of the super-match.

[0015] Furthermore, the result verification unit is also equipped with a standard super-item ratio. The result verification unit can calculate in real time the real-time super-item ratio of the search object text marked as super-item match to the total search object text output by the complete result matching set; and when the real-time super-item ratio is greater than the standard super-item ratio, the result verification unit adjusts and increases the initial text similarity and initial semantic similarity in the data search database.

[0016] Furthermore, the large model matching unit filters the target data items within the result matching set. The large model matching unit constructs a mapping relationship table based on the search object text. The key is the measurement point path of the target system measurement point corresponding to each target data item in the result matching set, and the value is a list composed of each target data item in the result matching set. All key-value pairs with a value list length greater than 1 are filtered out. Then, the large model in the large model matching unit is called to perform semantic-level comparison and contextual reasoning. The target data item with the highest correlation value is selected as the optimal matching result and output to the result verification unit.

[0017] Furthermore, the large model matching unit expands the measurement point names of the search object text by obtaining the measurement point names of the corresponding search object text and generating a measurement point name expansion set based on the measurement point names of the object text. The measurement point name expansion set includes the original measurement point name, synonyms of the measurement point name, abbreviations of the measurement point name, and aliases of the measurement point name. The large model matching unit replaces the measurement point names of the search object text with the measurement point name expansion set, generates the search reconstructed text, and sends it to the data search library for rematching.

[0018] This invention also provides an intelligent measurement point matching method for an intelligent measurement point matching system based on a large model, applicable to any of the above-mentioned methods, comprising: Acquire target system measurement point data and original system measurement point data; input target system measurement point data into data search library to build a searchable document; input original system measurement point data into data reading unit to split into several search object texts. A matching set is constructed based on the text similarity of the test point names, the semantic similarity of the test point names, the test point type, and the test point path hierarchy; The type of the result matching set is determined, and when there is a unique target data item in the result matching set, the matched target data item is output to the cleaning and import unit to import the original system measurement point data to complete the measurement point matching. When there is no target data item in the result matching set, the text of the search object is expanded with measurement point names and re-matched using the data search library; when there are multiple target data items in the result matching set, the optimal matching result is output through semantic-level comparison and contextual reasoning to complete the measurement point matching. Set up an audit and correction database, and feed back the audited and corrected data to the large model matching unit for model training.

[0019] Compared with existing technologies, the beneficial effects of this invention are as follows: By setting up a data search library to construct searchable documents from the target system's measurement point data, and by using a data reading unit to divide the original system's measurement point data into texts for each search object, it is easier to search and match any measurement point in the original system. Simultaneously, by selecting independent matching sets based on three dimensions—measurement point name, measurement point type, and path level—it reduces the amount of matching computation and improves the efficiency of initial matching. When selecting measurement point names, as long as either text similarity or semantic similarity meets the criteria, the corresponding target data item is selected to avoid omissions. The result verification unit directly outputs a result matching set containing only unique target data items, improving the matching efficiency of measurement points with clearly defined names, types, and path levels. Furthermore, by adjusting the initial text similarity and initial semantic similarity of the data search library, the library can select more suitable result matching sets based on actual conditions, facilitating the filtering of large model matching units or the expansion of measurement point names. This further improves the matching efficiency of measurement points while ensuring matching accuracy.

[0020] In particular, compared with traditional manual or semi-automatic matching methods, it significantly reduces manpower and time costs, effectively solves the problem of low information integration efficiency, and by introducing large model matching units, it breaks through the limitations of differences in expression forms, can more accurately identify measurement points pointing to the same physical object under different naming methods, significantly improves matching accuracy, and reduces the probability of mismatch.

[0021] Furthermore, the entire intelligent measurement point matching system achieves end-to-end automated processing, from data reading, feature extraction, preliminary matching to final result output, all without human intervention. Especially when dealing with massive amounts of measurement point data, the intelligent judgment capability of the self-trained large model can complete multiple rounds of screening and final confirmation, greatly improving matching efficiency while avoiding errors caused by human operation. At the same time, the large model matching unit can also learn and train on the data that has completed review and correction in the review and correction database, which facilitates the iterative optimization of the large model. In addition, the matching results generated by the intelligent measurement point matching system not only include successfully matched measurement point pairs, but also provide detailed explanations of the matching basis, confidence scores, and a list of unmatched items, allowing users to clearly understand the matching logic at each step. This transparency not only enhances user trust and acceptance, but also provides a solid foundation for subsequent manual review, feedback learning, and continuous model optimization. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the intelligent measurement point matching system framework based on a large model in this embodiment; Figure 2 This is a matching logic diagram of the data search library in this embodiment; Figure 3 This is the decision logic diagram of the result verification unit in this embodiment; Figure 4 This is a flowchart of the intelligent measurement point matching method based on a large model in this embodiment. Detailed Implementation

[0023] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0024] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0025] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0026] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0027] Please see Figure 1 As shown, this embodiment discloses an intelligent measurement point matching system based on a large model, including a data reading unit, a data search library, a result verification unit, a cleaning and import unit, a large model matching unit, and an audit and correction database. Specifically, this embodiment selects two types of measurement point data received from the power system in practical applications as input sources: target system measurement point data and original system measurement point data. Both types of data are provided in Excel file format and have clear field definitions and structural specifications.

[0028] The target system measurement point data includes, but is not limited to, the following key fields: Measurement point ID: A unique identifier for a measurement point, which is the core basis for filling in the original system measurement point ID after a successful match; Measurement point name: A name describing the physical meaning or function of the measurement point, used for initial semantic matching; Measurement point path: This indicates the location path of the measurement point in the system's equipment hierarchy, used to help determine contextual relevance and improve matching accuracy.

[0029] The original system's measurement point data includes the following fields. Measurement point ID: This field is initially empty and is used to fill in the measurement point ID corresponding to the target system after a successful match; Measurement point name: Used for semantic similarity analysis with the measurement point name of the target system; Measurement point path: Indicates the location path of the measurement point in the system device hierarchy, used to assist in contextual reasoning and path consistency judgment.

[0030] In this embodiment, a data reading unit is set up to receive and read the original system measurement point data; a data search library is set up to construct a searchable structured document from the target system measurement point data and store it; wherein, the structured document includes several target data items; The data reading unit divides the received original system measurement point data into several structured search object texts according to the field definitions; and transmits the read search object texts to the data search library. The data search library traverses each search object text of the original system measurement point data and extracts key field information, including measurement point name, measurement point type, and path level. Please continue reading. Figure 2 As shown, the data search library is set with initial text similarity Sr and initial semantic similarity Sa. For any search object text, the data search library calculates and compares the real-time text similarity Srs and real-time semantic similarity Sas between the measurement point names in the search object text and the measurement point names in each target data item. If the real-time text similarity Srs of the target data item is greater than or equal to the initial text similarity Sr, and / or the real-time semantic similarity Srs is greater than or equal to the initial semantic similarity Sr, then the target data item is selected to enter the measurement point name matching set. If the real-time text similarity Srs of the target data item is less than the initial text similarity Sr, and the real-time semantic similarity Srs is less than the initial semantic similarity Sr, then the target data item is determined to be a non-matching item and is not selected.

[0031] The data search library matches the measurement point types of the search object text with the measurement point types of each target data item stored in the data search library, and selects the target data items with the same measurement point type as the measurement point type matching set. If the measurement point type of the search object text is empty, then all target data items stored in the data search database are selected as the measurement point type matching set. If the measurement point type of any target data item stored in the data search library is empty, it is determined that the target data item has the same measurement point type as the search object text, and the target data item is also selected into the measurement point type matching set.

[0032] The data search library matches the path hierarchy of the search object text with the path hierarchy of each target data item stored in the data search library, and selects the target data items with the same path hierarchy as the path hierarchy matching set.

[0033] The data search library selects the target data items from the intersection of three sets: measurement point name matching set, measurement point type matching set, and path level matching set, as the result matching set, and outputs the result matching set to the result verification unit.

[0034] Please continue reading. Figure 3 As shown, the result verification unit determines the type of the result matching set. If the result matching set is a 1:1 match, that is, there is only one target data item in the result matching set, the result verification unit determines that the target data item in the result matching set is a unique and clear match, and outputs the matched target data item to the cleaning and importing unit. The cleaning and importing unit imports the measurement point ID of the matched target data item in the target system measurement point data and fills it into the corresponding position of the original system measurement point data to complete the matching. If the result matching set is a 1:n match, that is, there are multiple target data items in the result matching set, the result verification unit determines that the corresponding search object text is a multi-candidate match. The result verification unit transmits the result matching set to the large model matching unit, which then filters the target data items in the result matching set. If the result matching set is a 1:0 match, that is, there is no target data item in the result matching set, the result verification unit determines that the corresponding search object text has no valid match, and the result verification unit transmits the search object text to the large model matching unit for measurement point name expansion. The result verification unit also includes a standard invalidity ratio. When no target data item exists in the result matching set of any search object text, the result verification unit calculates the real-time invalidity ratio of the search object text with no valid match out of all search object text output from the completed result matching set, and makes a determination. When the real-time invalid percentage is less than or equal to the standard invalid percentage, the result verification unit does not adjust the initial text similarity Sr and the initial semantic similarity Sa in the data search library; When the proportion of real-time invalid data exceeds the standard invalid data proportion, the result verification unit will adjust and reduce the initial text similarity Sr and initial semantic similarity Sa in the data search database. Sr'=Sr×[1-(As-Ab) / As]; Sa'=Sa×[1-(As-Ab) / As]; In the formula, Sr' is the adjusted text similarity, Sa' is the adjusted semantic similarity, As is the real-time invalid percentage, and Ab is the standard invalid percentage; The result verification unit also includes settings for the maximum number of matching items and the percentage of standard excess items. When multiple target data items exist in the result matching set for any search object text, the result verification unit compares the number of target data items in the result matching set with the maximum number of matching items. If the number of target data items in the result matching set is less than or equal to the maximum number of matching items, the result verification unit does not mark the corresponding search object text; If the number of target data items in the result matching set is greater than the maximum number of matching items, the result verification unit will mark the corresponding search object text as the search object text of the super-item matching, and calculate the real-time super-item ratio of the search object text of the super-item matching to the search object text of all completed result matching sets. The results verification unit determines the real-time excess item percentage based on the standard excess item percentage. When the proportion of real-time excess items is less than or equal to the proportion of standard excess items, the result verification unit does not adjust the initial text similarity Sr and the initial semantic similarity Sa in the data search library; When the proportion of real-time excess items exceeds the proportion of standard excess items, the result verification unit will adjust and increase the initial text similarity Sr and initial semantic similarity Sa in the data search database. Sr'=Sr×[1+(Ar-Ae) / Ar]; Sa'=Sa×[1+(Ar-Ae) / Ar]; In the formula, Sr' is the adjusted text similarity, Sa' is the adjusted semantic similarity, Ar is the proportion of real-time super items, and Ae is the proportion of standard super items; In this embodiment, the initial text similarity Sr and initial semantic similarity Sa in the data search library are both set to 97%. These values ​​need to be adjusted based on the actual power system data. In this scheme, both the initial text similarity and initial semantic similarity can be adaptively adjusted according to the actual matching situation. Therefore, slight deviations in the initial settings have little impact on this embodiment. The standard invalidity ratio and standard excess ratio in this embodiment are both set to 0.015. Due to the data processing capacity of the large model matching unit, these values ​​need to be determined based on the actual situation during system construction. As for the maximum number of matching items in the result verification unit, in this embodiment, it is determined based on the total number of target data items in the data search library. This embodiment sets it to 0.33% of the total number of target data items. Therefore, the actual processing capacity of each functional unit and the amount of power measurement point system data processed should be considered when setting the judgment standard values. In terms of construction, a data search library with search and similarity calculation comparison functions can be built independently, or an existing search engine, such as Elasticsearch (ES), can be selected for auxiliary implementation. Specific implementation operations will not be elaborated here.

[0035] When the large model matching unit has multiple target data items in the result matching set, it will construct a mapping table based on the search object text. The key is the measurement point path of the target system measurement point corresponding to each target data item in the result matching set, and the value is a list of each target data item in the result matching set. All key-value pairs with a value list length greater than 1 are filtered out. Then, the large model in the large model matching unit is called to perform semantic-level comparison and contextual reasoning. The target data item with the highest correlation value is selected as the optimal matching result and output to the result verification unit. At the same time, the large model matching unit will also output the search object text, the corresponding result matching set, and the optimal matching result to the review and correction database. When no target data item is found in the result matching set, the large model matching unit obtains the measurement point name of the corresponding search object text. It then generates an expanded set of measurement point names based on the measurement point names of the object text. The expanded set of measurement point names includes the original measurement point name, synonyms of the measurement point name, abbreviations of the measurement point name, and aliases of the measurement point name. The expanded set of measurement point names is then used to replace the measurement point names of the search object text, generating a search reconstructed text which is sent to the data search library for secondary search matching. The data search library will then regenerate the result matching set based on the search reconstructed text. If the result set generated by the search and reconstructed text is a 1:1 match, the large model matching unit directly outputs the matching result to the result verification unit. If the result matching set generated by the search and reconstruction text is a 1:n match, the large model matching unit will filter each target data item in the result matching set and output the optimal matching result to the result verification unit. If the result set generated by the search and reconstructed text is a 1:0 match, the large model matching unit will output the search and reconstructed text to the review and correction database.

[0036] The audit correction database can also monitor whether there are audit corrections for internal data, and feed back the audit correction data to the large model matching unit for model training.

[0037] Please continue reading. Figure 4 As shown, this embodiment also provides an intelligent measurement point matching method based on a large model, including: Step S1: Obtain the target system measurement point data and the original system measurement point data. Input the target system measurement point data into the data search library to build a searchable document. Input the original system measurement point data into the data reading unit and split it into several search object texts. Step S2: The data search library constructs a result matching set based on the text similarity of the measurement point names, the semantic similarity of the measurement point names, the measurement point type, and the measurement point path hierarchy; Step S3: The result verification unit determines the type of the result matching set, and when there is a unique target data item in the result matching set, it outputs the matched target data item to the cleaning and importing unit to import the original system measurement point data to complete the measurement point matching. In step S4, when there is no target data item in the result matching set, the large model matching unit expands the measurement point name of the search object text and performs re-matching through the data search library; when there are multiple target data items in the result matching set, it outputs the optimal matching result through semantic-level comparison and contextual reasoning to complete the measurement point matching. Step S5: The data that has been reviewed and corrected is fed back to the large model matching unit for model training through the review and correction database.

[0038] In this embodiment, the large model used can be implemented by adapting a general language model to the specific domain, or it can directly use a large language model from the power industry domain, such as the Guangming Power large model. The specific training process is as follows: Data preparation: Collect equipment ledgers, historical measurement point data, operation and maintenance logs, procedures and specifications documents, and measurement point naming records from various power systems; Construct a dataset that includes synonyms, abbreviations, aliases, and typical naming patterns; A batch of measurement point pairs with clear matching relationships were labeled as training samples.

[0039] Training phase: A contrastive learning strategy is introduced to construct different expression samples of the same test point and similar named samples of different test points. Add a path context modeling module to enable the model to understand the device hierarchy and topology information in the measurement point connection path; The model is trained using labeled samples to determine whether the output matches and provides a confidence score.

[0040] Deployment of applications: The trained large model is deployed to the matching process of the intelligent test point system for semantic-level comparison and contextual reasoning of candidate test points. A continuous learning mechanism is introduced, which, together with the data feedback from the audit and correction database, continuously learns and optimizes the matching and judgment capabilities of the large model.

[0041] Through the training mechanism described above, the large model matching unit in this embodiment possesses a high sensitivity to the naming habits of power system measurement points and a deep understanding of the contextual information of measurement points, which is significantly better than the performance of the general language model in power scenarios.

[0042] Meanwhile, this embodiment, by setting up a data search library and a result verification unit, can quickly complete the preliminary matching of any measurement point in the original system's measurement point data. For measurement points with clear names, types, and path levels, it can quickly provide a unique and clear match, improving the efficiency of measurement point matching. For cases where the matching results are not unique or there are no matching results, a large model matching unit trained with power system data is used to perform accurate semantic-level comparison and contextual reasoning to provide a matching result, improving the accuracy of measurement point matching. It breaks through the limitations of traditional methods based on string similarity or general semantic models, achieving measurement point matching capabilities that are closer to business semantics. This not only improves the accuracy and coverage of matching but also significantly reduces manual intervention. It is suitable for large-scale, cross-system automatic measurement point matching needs and has good engineering application prospects and promotional value.

[0043] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent measurement point matching system based on a large model, characterized in that, include, The data search library is used to construct several searchable target data items from the target system measurement point data, and to obtain the search object text of the original system measurement point data divided by the data reading unit. It matches and calculates real-time text similarity and real-time semantic similarity, and selects target data items whose real-time text similarity reaches the initial text similarity and / or whose real-time semantic similarity reaches the initial semantic similarity to enter the measurement point name matching set. The data search library selects target data items based on the measurement point type of the search object text, generates a measurement point type matching set, selects target data items based on the path level of the search object text, generates a path level matching set, and outputs the target data items of the intersection of the measurement point name matching set, the measurement point type matching set, and the path level matching set as the result matching set. The result verification unit is used to output the result matching set as the final result when there is only one target data item in the result matching set. The result verification unit can also adjust the initial text similarity and initial semantic similarity in the data search library according to the proportion of each type in the result matching set; The large model matching unit is used to filter the target data items in the result matching set when there are multiple target data items in the result matching set; or to expand the measurement point name of the search object text and generate the search reconstructed text before outputting it to the data search library for rematching when there are no target data items in the result matching set.

2. The intelligent measurement point matching system based on a large model according to claim 1, characterized in that, The data reading unit receives the original system measurement point data and divides the original system measurement point data into several structured search object texts according to the field definitions. The data search library is used to traverse the text of each search object divided by the measurement point data of the original system. For any text of a search object, the data search library matches and calculates the real-time text similarity and real-time semantic similarity between the measurement point name of the text of the search object and the measurement point name of each target data item in the data search library. Target data items whose real-time text similarity reaches the initial text similarity and / or whose real-time semantic similarity reaches the initial semantic similarity are selected to enter the measurement point name matching set.

3. The intelligent measurement point matching system based on a large model according to claim 1, characterized in that, When selecting target data items based on the measurement point type of the search object text, the data search library matches the measurement point type of the search object text with the measurement point type of each target data item stored in the data search library, and selects each target data item with the same measurement point type as the measurement point type matching set. If the measurement point type of the search object text is empty, then all target data items stored in the data search database are selected as the measurement point type matching set. If the measurement point type of any target data item stored in the data search library is empty, then the target data item is determined to have the same measurement point type as the search object text.

4. The intelligent measurement point matching system based on a large model according to claim 1, characterized in that, The result verification unit can determine the type of the result matching set. If there is only one target data item in the result matching set, the result verification unit outputs the target data item in the result matching set as the final result to the cleaning and importing unit. The cleaning and importing unit imports the measurement point ID of the target data item matched in the target system measurement point data and fills it into the corresponding position of the original system measurement point data to complete the measurement point matching. If the result matching set contains multiple target data items, the result verification unit will output the result matching set to the large model matching unit; If there is no target data item in the result matching set, the result verification unit outputs the result matching set to the large model matching unit.

5. The intelligent measurement point matching system based on a large model according to claim 4, characterized in that, The result verification unit also includes a standard invalidity ratio. When there is no target data item in the result matching set, the search object text is determined to have no valid match. The result verification unit calculates the real-time invalidity ratio of the search object text with no valid match to the search object text output by all completed result matching sets. When the real-time invalidity ratio is greater than the standard invalidity ratio, the initial text similarity and initial semantic similarity in the data search library are adjusted and reduced.

6. The intelligent measurement point matching system based on a large model according to claim 4, characterized in that, The result verification unit is also set with a maximum number of matching items. When there are multiple target data items in the result matching set, the result verification unit compares the number of target data items in the result matching set with the maximum number of matching items. If the number of target data items in the result matching set is greater than the maximum number of matching items, the corresponding search object text is marked as the search object text of the super-match.

7. The intelligent measurement point matching system based on a large model according to claim 6, characterized in that, The result verification unit is also equipped with a standard super-item ratio. The result verification unit can calculate in real time the real-time super-item ratio of the search object text marked as super-item match to the total search object text output by the complete result matching set. When the proportion of real-time excess items is greater than the proportion of standard excess items, the result verification unit adjusts and increases the initial text similarity and initial semantic similarity in the data search library.

8. The intelligent measurement point matching system based on a large model according to claim 1, characterized in that, The large model matching unit filters target data items within the result matching set by constructing a mapping table based on the search object text. The key is the measurement point path of the target system measurement point corresponding to each target data item in the result matching set, and the value is a list composed of each target data item in the result matching set. All key-value pairs with a value list length greater than 1 are filtered out. Then, the large model in the large model matching unit is called to perform semantic-level comparison and contextual reasoning. The target data item with the highest correlation value is selected as the optimal matching result and output to the result verification unit.

9. The intelligent measurement point matching system based on a large model according to claim 1, characterized in that, The large model matching unit expands the measurement point names of the search object text by obtaining the measurement point names of the corresponding search object text and generating a measurement point name expansion set based on the measurement point names of the object text. The measurement point name expansion set includes the original measurement point name, synonyms of the measurement point name, abbreviations of the measurement point name, and aliases of the measurement point name. The large model matching unit replaces the measurement point names of the search object text with the measurement point name expansion set, generates the search reconstructed text, and sends it to the data search library for rematching.

10. An intelligent measurement point matching method applied to the intelligent measurement point matching system based on a large model as described in any one of claims 1-9, characterized in that, include, Acquire target system measurement point data and original system measurement point data; input target system measurement point data into data search library to build a searchable document; input original system measurement point data into data reading unit to split into several search object texts. A matching set is constructed based on the text similarity of the test point names, the semantic similarity of the test point names, the test point type, and the test point path hierarchy; The type of the result matching set is determined, and when there is a unique target data item in the result matching set, the matched target data item is output to the cleaning and import unit to import the original system measurement point data to complete the measurement point matching. When there is no target data item in the result matching set, the text of the search object is expanded with measurement point names and re-matched using the data search library; when there are multiple target data items in the result matching set, the optimal matching result is output through semantic-level comparison and contextual reasoning to complete the measurement point matching. Set up an audit and correction database, and feed back the audited and corrected data to the large model matching unit for model training.

Images