Flexible DC converter station fault decision method and device based on RAG knowledge base

Abstract

The invention discloses a flexible direct current converter station fault decision-making method and device based on an RAG knowledge base, and belongs to the field of power systems. The method comprises the following steps: acquiring fault information of the flexible direct current converter station, respectively inputting the fault information into a preset rule model layer and a GNN reasoning layer, and obtaining first intermediate processing information and second intermediate processing information in combination with an RAG knowledge base; and inputting the fault information and the intermediate processing information into a large language model layer to obtain a first decision report, correcting the first decision report into a second decision report, finally sending a control instruction according to the second decision report, and updating the RAG knowledge base through a self-feedback learning method. Therefore, by implementing the method, the problem of low fault decision-making efficiency of the flexible direct current converter station in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a flexible DC converter station fault decision-making method and device based on a RAG knowledge base. Background Art

[0002] In modern power systems, the operation and maintenance of flexible DC converter stations primarily relies on SCADA systems to collect multimodal information, such as equipment status, time series data, and spatial temperature distribution. This information is then combined with historical fault records and operation and maintenance manuals to determine faults. Retrieval-augmented generation (RAG) technology builds a RAG knowledge base, semantically merging multimodal data into structured and unstructured knowledge, supporting the "retrieval-reasoning-generation" decision-making process.

[0003] In existing technologies, traditional expert systems require manual coding of numerous rules and struggle to adapt to changes in equipment topology. Large, general-purpose models often have a high error rate in understanding power terminology, and existing knowledge graphs lack sufficient coverage for flexible DC converter-specific equipment. Consequently, these systems suffer from high rule maintenance costs, poor fault diagnosis adaptability, insufficient model expertise, and incomplete knowledge coverage. In the application of RAG technology, traditional RAG knowledge bases update knowledge on a quarterly basis, resulting in knowledge update delays and insufficient real-time performance. Therefore, a fault decision-making method that can integrate multi-source heterogeneous data and support dynamic knowledge updates is urgently needed to address the low efficiency of fault decision-making in flexible DC converter stations. Summary of the Invention

[0004] The present invention provides a method and device for making a fault decision for a flexible DC converter station based on a RAG knowledge base, which can solve the problem of low fault decision efficiency of a flexible DC converter station in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a fault decision method for a flexible DC converter station based on a RAG knowledge base, comprising:

[0006] Obtaining fault information of the flexible DC converter station;

[0007] Input the fault information into a preset rule model layer, combine it with the RAG knowledge base to obtain first intermediate processing information, and input the fault information into a preset GNN reasoning layer, combine it with the RAG knowledge base to obtain second intermediate processing information;

[0008] Inputting the fault information, the first intermediate processed information, and the second intermediate processed information into a preset large language model layer to obtain a first decision report, and performing corrections based on the first decision report to obtain a second decision report;

[0009] Send a control instruction to the flexible DC converter station according to the second decision report.

[0010] The embodiment of the present application obtains the fault information of the flexible DC converter station and inputs it into the rule model layer and the GNN reasoning layer respectively, and combines it with the RAG knowledge base for hierarchical processing, which can achieve rapid response to clear threshold alarms and in-depth analysis of equipment topology-related faults, avoiding the rigidity of single rule reasoning and the lag of traditional correlation analysis. The fault information and the two-layer intermediate processing results are input into the large language model layer, which can generate a first decision report that meets the standard terminology requirements of the power industry, and form a second decision report after correction to ensure the accuracy and operability of the disposal suggestions. Finally, the fault decision is controlled based on the corrected decision report. Through this application, the fault decision of the flexible DC converter station can be constructed from data collection, multi-dimensional reasoning to standard decision output, effectively solving the problem of low efficiency of fault decision-making of flexible DC converter stations in the prior art.

[0011] As a preferred example of the first aspect, after the second decision report is obtained by amending the first decision report, the method further includes:

[0012] The RAG knowledge base is updated using a self-feedback learning method based on the first decision report and the second decision report; wherein, the RAG knowledge base is established based on the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method.

[0013] In this preferred example, the RAG knowledge base is established through multimodal data fusion and dynamic construction. This allows for the integration of multi-source data and real-time updates of the knowledge system, avoiding data silos and knowledge lags. Self-feedback learning, based on differences in decision reports, updates the knowledge base, transforming manual experience into structured knowledge and driving dynamic optimization of the RAG knowledge base.

[0014] As a preferred example of the first aspect, updating the RAG knowledge base using a self-feedback learning method according to the first decision report and the second decision report is specifically as follows:

[0015] Performing differential analysis on the first decision report and the second decision report through a preset deep twin network to obtain differential features;

[0016] The RAG knowledge base is updated using an incremental graph embedding algorithm according to the difference features.

[0017] In this preferred example, a deep twin network is used to perform differential analysis on the two versions of the decision report, accurately extract the key difference features in the correction, and then dynamically update the RAG knowledge base with the help of an incremental graph embedding algorithm. This process efficiently converts manual experience into structured knowledge, promoting the continuous evolution of the RAG knowledge base with operation and maintenance practices.

[0018] As a preferred example of the first aspect, the RAG knowledge base is established according to the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method, specifically:

[0019] The flexible DC converter station includes structured attribute data, time series data and spatial temperature distribution data;

[0020] Sequentially performing semantic fusion, structural unification, and time synchronization on the structured attribute data, the time series data, and the spatial temperature distribution data to obtain unified semantic data;

[0021] The RAG knowledge base is obtained by adopting a dual-channel knowledge injection method based on the unified semantic data.

[0022] In this preferred example, by semantically fusing, structurally unifying, and time-series synchronizing the structured attribute data, time series data, and spatial temperature distribution data of the flexible DC converter station, the heterogeneous barriers of multi-source data can be broken down, and a cross-modal, interconnected, unified semantic data system can be constructed, addressing the inefficient integration caused by data silos in traditional technologies. Using a dual-channel knowledge injection method based on this unified semantic data, structured procedural clauses and unstructured operation and maintenance experience can be simultaneously transformed into the core elements of the RAG knowledge base, providing comprehensive, real-time knowledge support for intelligent decision-making in the flexible DC converter station.

[0023] As a preferred example of the first aspect, the fault information is input into a preset rule model layer, combined with a RAG knowledge base to obtain first intermediate processing information, and the fault information is input into a preset GNN reasoning layer, combined with the RAG knowledge base to obtain second intermediate processing information, specifically:

[0024] The fault information includes quantitative threshold fault information and device identification information;

[0025] Retrieving the quantitative threshold fault information from the RAG knowledge base, generating an alarm determination result and an alarm level enhancement suggestion, and obtaining first intermediate processing information based on the alarm determination result and the alarm level enhancement suggestion;

[0026] The device identification information is retrieved from the RAG knowledge base, device topology information, fault propagation information and real-time operating condition information are generated, and the GNN message passing method is used according to the device topology information, the fault propagation information and the real-time operating condition information to obtain second intermediate processing information.

[0027] In this preferred example, by separating quantitative threshold fault information from device identification information, a multi-dimensional and precise analysis of flexible DC converter station faults can be achieved. For quantitative threshold information, real-time retrieval from the RAG knowledge base generates alarm determinations and level upgrade suggestions, quickly filtering invalid alarms and dynamically adjusting alarm priorities, avoiding the false alarm flooding and delayed response problems of traditional rule engines. For device identification information, by retrieving topological structures, fault propagation paths, and real-time operating condition data, combined with the GNN message passing mechanism to deeply explore device associations, implicit causal chains can be accurately identified, breaking through the limitation of traditional knowledge graphs that only support static association analysis, and effectively solving the problem of low fault decision-making efficiency in flexible DC converter stations.

[0028] In a second aspect, the present invention provides a fault decision-making device for a flexible DC converter station based on a RAG knowledge base, comprising: a data acquisition module, a first processing module, a second processing module, and a fault decision-making module;

[0029] The data acquisition module is used to acquire fault information of the flexible DC converter station;

[0030] The first processing module is configured to input the fault information into a preset rule model layer, combine it with the RAG knowledge base to obtain first intermediate processing information, and input the fault information into a preset GNN reasoning layer, combine it with the RAG knowledge base to obtain second intermediate processing information;

[0031] The second processing module is configured to input the fault information, the first intermediate processing information, and the second intermediate processing information into a preset large language model layer to obtain a first decision report, and to modify the first decision report to obtain a second decision report;

[0032] The fault decision module is configured to send a control instruction to the flexible DC converter station according to the second decision report.

[0033] As a preferred example of the second aspect, the second processing module further includes an updating unit;

[0034] The updating unit is used to update the RAG knowledge base using a self-feedback learning method based on the first decision report and the second decision report; wherein, the RAG knowledge base is established based on the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method.

[0035] As a preferred example of the second aspect, the updating unit includes a first updating subunit and a second updating subunit;

[0036] The first updating subunit is configured to perform differential analysis on the first decision report and the second decision report through a preset deep twin network to obtain differential features;

[0037] The second updating subunit is configured to update the RAG knowledge base using an incremental graph embedding algorithm according to the difference features.

[0038] As a preferred example of the second aspect, the updating unit includes a third updating subunit and a fourth updating subunit;

[0039] The flexible DC converter station includes structured attribute data, time series data and spatial temperature distribution data;

[0040] The third updating subunit is configured to sequentially perform semantic fusion, structural unification, and time synchronization on the structured attribute data, the time series data, and the spatial temperature distribution data to obtain unified semantic data;

[0041] The fourth updating subunit is used to obtain the RAG knowledge base by adopting a dual-channel knowledge injection method according to the unified semantic data.

[0042] As a preferred example of the second aspect, the first processing module includes a first processing unit and a second processing unit;

[0043] The fault information includes quantitative threshold fault information and device identification information;

[0044] The first processing unit is configured to retrieve the quantitative threshold fault information from the RAG knowledge base, generate an alarm determination result and an alarm level escalation suggestion, and obtain first intermediate processing information based on the alarm determination result and the alarm level escalation suggestion;

[0045] The second processing unit is used to retrieve the device identification information from the RAG knowledge base, generate device topology information, fault propagation information and real-time operating condition information, and use the GNN message passing method according to the device topology information, the fault propagation information and the real-time operating condition information to obtain second intermediate processing information.

[0046] In summary, the embodiment of the present application obtains the fault information of the flexible DC converter station and inputs it into the rule model layer and the GNN reasoning layer respectively, and combines it with the RAG knowledge base for hierarchical processing, which can achieve rapid response to clear threshold alarms and in-depth analysis of equipment topology-related faults, avoiding the rigidity of single rule reasoning and the lag of traditional correlation analysis. The fault information and the two-layer intermediate processing results are input into the large language model layer, which can generate a first decision report that meets the standard terminology requirements of the power industry, and form a second decision report after correction to ensure the accuracy and operability of the disposal suggestions. Finally, the fault decision is controlled based on the corrected decision report. Through this application, the fault decision of the flexible DC converter station can be constructed from data collection, multi-dimensional reasoning to standard decision output, effectively solving the problem of low efficiency of fault decision-making of flexible DC converter stations in the prior art.

[0047] Another embodiment of the present invention also provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps of the flexible DC converter station fault decision method based on the RAG knowledge base of the present invention.

[0048] Another embodiment of the present invention also provides a computer-readable storage medium item, including: a stored computer program, which controls the device where the computer-readable storage medium is located to execute the steps of the flexible DC converter station fault decision method based on the RAG knowledge base of the present invention when the computer program is running. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0050] Figure 1 A flow chart of an embodiment of a flexible DC converter station fault decision-making method based on a RAG knowledge base provided by the present invention;

[0051] Figure 2 A data flow diagram of an embodiment of a flexible DC converter station fault decision-making method based on a RAG knowledge base provided by the present invention;

[0052] Figure 3 This is a module structure diagram of an embodiment of a flexible DC converter station fault decision-making device based on a RAG knowledge base provided by the present invention. DETAILED DESCRIPTION

[0053] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0055] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0056] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0057] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0058] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0059] In the description of the embodiments of the present application, unless otherwise expressly specified or limited, technical terms such as "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components or interactions between two components. Those skilled in the art can understand the specific meanings of the above terms in the embodiments of the present application based on specific circumstances.

[0060] RAG (Retrieval-Augmented Generation) technology is an AI framework that combines retrieval technology with generative models. It assists the model in generating answers by retrieving relevant information from an external knowledge base, thus avoiding the "hallucination" problem of large models. In this application, the core technology of the RAG knowledge base is constructed by retrieving knowledge fragments such as the operation and maintenance procedures and fault cases of flexible direct current converter stations, thereby assisting the hybrid reasoning model in generating accurate fault decision recommendations. This solves the problems of delayed knowledge updates and lack of multimodal processing in traditional RAG.

[0061] Example 1

[0062] See also Figure 1 and Figure 2 To solve the problem of low fault decision-making efficiency in flexible DC converter stations in the prior art, an embodiment of the present invention provides a fault decision-making method for a flexible DC converter station based on a RAG knowledge base, comprising:

[0063] S1. Obtaining fault information of the flexible DC converter station;

[0064] S2. Input the fault information into a preset rule model layer, combine it with the RAG knowledge base to obtain first intermediate processing information, and input the fault information into a preset GNN reasoning layer, combine it with the RAG knowledge base to obtain second intermediate processing information;

[0065] Furthermore, in some embodiments of the present application, the fault information is input into a preset rule model layer, combined with a RAG knowledge base to obtain first intermediate processing information, and the fault information is input into a preset GNN reasoning layer, combined with the RAG knowledge base to obtain second intermediate processing information, specifically:

[0066] The fault information includes quantitative threshold fault information and device identification information;

[0067] Retrieving the quantitative threshold fault information from the RAG knowledge base, generating an alarm determination result and an alarm level enhancement suggestion, and obtaining first intermediate processing information based on the alarm determination result and the alarm level enhancement suggestion;

[0068] The device identification information is retrieved from the RAG knowledge base, device topology information, fault propagation information and real-time operating condition information are generated, and the GNN message passing method is used according to the device topology information, the fault propagation information and the real-time operating condition information to obtain second intermediate processing information.

[0069] By separating quantitative threshold fault information from device identification information, a multi-dimensional and precise analysis of flexible DC converter station faults can be achieved. For quantitative threshold information, real-time retrieval from the RAG knowledge base generates alarm determinations and level escalation recommendations, quickly filtering invalid alarms and dynamically adjusting alarm priorities, avoiding the false alarm flooding and delayed response issues of traditional rule engines. For device identification information, topology, fault propagation paths, and real-time operating condition data are retrieved, and the GNN message passing mechanism is used to deeply mine device relationships, accurately identifying implicit causal chains. This overcomes the limitation of traditional knowledge graphs that only support static association analysis and effectively addresses the low efficiency of fault decision-making in flexible DC converter stations.

[0070] Specifically, in order to fully explain the above steps, the following scheme is used as an example:

[0071] Taking the example of a flexible DC converter station monitoring a quantitative threshold alarm of "DC voltage 125kV (threshold 120kV)" and an equipment anomaly of "converter valve phase A temperature 105°C," the rule model layer first uses the Drools rule engine to match the DL / T 2664-2023 standard in the RAG knowledge base to determine that the voltage over-limit alarm is valid. If the knowledge base detects a revision of the regulation, it automatically updates the rule. It also searches for similar cases from the past three years and generates a recommendation to increase the alarm level of "emergency shutdown" based on real-time SCADA data. Parallel rule processing controls end-to-end latency to within 50ms, generating the first intermediate processing information. The GNN inference layer obtains the equipment topology map of converter valve phase A from the RAG, the fault propagation path based on the CIGRE case library, and real-time operating conditions such as an ambient temperature of 35°C and a load factor of 85%. Using a message passing mechanism, it converts the "cooling pump failure symptoms" into node attributes. Combined with SCADA vibration data, it infers the root cause of "cooling pump P1 failure leading to converter valve overheating," generating the second intermediate processing information.

[0072] S3. Input the fault information, the first intermediate processing information, and the second intermediate processing information into a preset large language model layer to obtain a first decision report, and modify the first decision report according to the first decision report to obtain a second decision report;

[0073] Furthermore, in some embodiments of the present application, after the second decision report is obtained by amending the first decision report, the method further includes:

[0074] The RAG knowledge base is updated using a self-feedback learning method based on the first decision report and the second decision report; wherein, the RAG knowledge base is established based on the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method.

[0075] Specifically, in order to fully explain the multimodal data fusion technology, the following solution is used as an example:

[0076] The multimodal data fusion technology specifically involves building a unified data representation model for power systems, designing an extended data model based on the IEEE CIM standard, and adding metadata definitions for flexible DC-specific entities (such as MMC submodules and DC circuit breakers). This achieves unified modeling and standardized representation of heterogeneous data at the semantic layer, providing structural support and semantic consistency for subsequent graph database modeling, knowledge extraction, and intelligent reasoning. A graph database (Neo4j) is used to associate and store equipment records, real-time SCADA data, and infrared images. Semantic connections are established through key fields such as device numbers and timestamps, creating a queryable, multimodal knowledge graph for the Retrieved-Augmented Generation (RAG) decision-making system. Apache Kafka is further used to build a distributed data pipeline to achieve millisecond-level real-time data (SCADA sampling values) and time alignment of multi-source data (using the PTP protocol to ensure time synchronization error <1ms). This step achieves semantic fusion, structural unification, and high-precision time synchronization of heterogeneous data at the data layer, providing a comprehensive, correlated, and computable multimodal data foundation for intelligent decision-making models.

[0077] Specifically, in order to fully explain the dynamic knowledge base construction method, the following scheme is used as an example:

[0078] First, a dual-channel knowledge infusion mechanism was established. The structured knowledge channel automatically parses XML-formatted knowledge files from the China Southern Power Grid Zhikan platform, extracting rule-based knowledge (such as "DC Voltage Fluctuation Threshold" in Article 3.2 of the "Flexible DC Operation and Maintenance Procedures") and converting it into Drools rules. The unstructured knowledge channel uses a domain-adapted BERT model (pre-trained and fine-tuned with the flexible DC corpus) to extract fault handling experience from operation and maintenance records, generating triplets of "fault phenomenon, handling measures, and effect evaluation." Incremental knowledge updates were then implemented, using a dynamic index based on a vector database (Milvus). New knowledge can be quickly embedded using few-shot learning, while existing knowledge is embedded through similarity detection.

[0079] This approach, through multimodal data fusion and dynamic construction, establishes a RAG knowledge base, integrating multi-source data and updating the knowledge system in real time, avoiding data silos and knowledge lags. Using self-feedback learning to update the knowledge base based on differences in decision reports can transform manual experience into structured knowledge, driving dynamic optimization of the RAG knowledge base.

[0080] Furthermore, in some embodiments of the present application, the RAG knowledge base is updated using a self-feedback learning method according to the first decision report and the second decision report, specifically:

[0081] Performing differential analysis on the first decision report and the second decision report through a preset deep twin network to obtain differential features;

[0082] The RAG knowledge base is updated using an incremental graph embedding algorithm according to the difference features.

[0083] Specifically, in order to fully explain the above steps, the following scheme is used as an example:

[0084] The first and second decision reports are fed into a deep twin network, which performs a differential analysis of the manually corrected solutions and the original recommendations in a semantic embedding space. For example, when an operator modifies the system-generated "load reduction to 70%" recommendation to "load reduction to 60% and activation of the backup cooling pump," the deep twin network accurately identifies the operational discrepancy of "cooling pump intervention" and adjusts the weight of the "converter valve overheating → cooling system" edge in the knowledge graph (ΔW = 0.15), driving the evolution of the knowledge base. Knowledge updates are implemented using the GraphSAGE-Stream algorithm, automatically integrating newly learned edge weights and node relationships every 24 hours. Furthermore, the system automatically generates update recommendations for high-frequency correction scenarios (e.g., a station adjusting the DC filter switching strategy three times in a row) through an expert review interface. A Git-like version management system is used to record all knowledge changes, supporting backtracking of knowledge status at any point in time.

[0085] In addition, this solution introduces an intelligent false alarm filtering model based on the LightGBM framework, improving alarm accuracy through multi-dimensional feature analysis. In the temporal dimension, the dynamic time warping (DTW) distance between alarm duration and historical baselines is calculated. In the spatial dimension, the spatial clustering of alarming devices (the Getis-Ord Gi* statistic) is calculated using GIS coordinates. In the environmental dimension, a joint feature matrix is ​​constructed that includes temperature and humidity, real-time load factor, and equipment maintenance status, enabling dynamic adjustment of alarm thresholds (for example, automatically relaxing the oil temperature alarm threshold by 5% at a high temperature of 40°C). The false alarm probability output by the model (e.g., P > 0.7) is not only used to filter invalid alarms in real time but also triggers updates to the "common false alarm patterns" index in the RAG knowledge base, forming a closed-loop optimization mechanism. The system also establishes a rigorous closed-loop validation mechanism, evaluating the module's effectiveness quarterly using three key metrics: comparing the matching rates of the response plans before and after learning using randomly sampled historical alarm data; calculating the F1-score of the false alarm filtering model; and verifying the semantic rationality of the knowledge graph edge weight adjustments using t-SNE visualization technology.

[0086] In this way, the two versions of the decision report are differentially analyzed through the deep twin network, the key difference features in the correction are accurately extracted, and then the RAG knowledge base is dynamically updated with the help of the incremental graph embedding algorithm. This process efficiently converts manual experience into structured knowledge, and promotes the continuous evolution of the RAG knowledge base with operation and maintenance practices.

[0087] Furthermore, in some embodiments of the present application, the RAG knowledge base is established based on the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method, specifically:

[0088] The flexible DC converter station includes structured attribute data, time series data and spatial temperature distribution data;

[0089] Sequentially performing semantic fusion, structural unification, and time synchronization on the structured attribute data, the time series data, and the spatial temperature distribution data to obtain unified semantic data;

[0090] The RAG knowledge base is obtained by adopting a dual-channel knowledge injection method based on the unified semantic data.

[0091] By semantically fusing, unifying, and synchronizing the structured attribute data, time series data, and spatial temperature distribution data of the flexible DC converter station, we can break down the heterogeneous barriers of multi-source data and build a cross-modal, interconnected, unified semantic data system, addressing the inefficient integration caused by data silos in traditional technologies. Using a dual-channel knowledge injection method based on this unified semantic data, we can simultaneously transform structured procedural clauses and unstructured operation and maintenance experience into the core elements of the RAG knowledge base, providing comprehensive, real-time knowledge support for intelligent decision-making in the flexible DC converter station.

[0092] S4. Send a control instruction to the flexible DC converter station according to the second decision report.

[0093] In summary, the embodiment of the present application obtains the fault information of the flexible DC converter station and inputs it into the rule model layer and the GNN reasoning layer respectively, and combines it with the RAG knowledge base for hierarchical processing, which can achieve rapid response to clear threshold alarms and in-depth analysis of equipment topology-related faults, avoiding the rigidity of single rule reasoning and the lag of traditional correlation analysis. The fault information and the two-layer intermediate processing results are input into the large language model layer, which can generate a first decision report that meets the standard terminology requirements of the power industry, and form a second decision report after correction to ensure the accuracy and operability of the disposal suggestions. Finally, the fault decision is controlled based on the corrected decision report. Through this application, the fault decision of the flexible DC converter station can be constructed from data collection, multi-dimensional reasoning to standard decision output, effectively solving the problem of low efficiency of fault decision-making of flexible DC converter stations in the prior art.

[0094] Example 2

[0095] like Figure 3 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0096] An embodiment of the present invention provides a fault decision-making device for a flexible DC converter station based on a RAG knowledge base, comprising: a data acquisition module 31, a first processing module 32, a second processing module 33, and a fault decision-making module 34;

[0097] Furthermore, in some embodiments of the present application, the data acquisition module 31 is used to obtain fault information of the flexible DC converter station; the first processing module 32 is used to input the fault information into a preset rule model layer, combine it with the RAG knowledge base to obtain first intermediate processing information, and input the fault information into a preset GNN reasoning layer, combine it with the RAG knowledge base to obtain second intermediate processing information; the second processing module 33 is used to input the fault information, the first intermediate processing information and the second intermediate processing information into a preset large language model layer to obtain a first decision report, and make corrections based on the first decision report to obtain a second decision report; the fault decision module 34 is used to send control instructions to the flexible DC converter station based on the second decision report.

[0098] Furthermore, in some embodiments of the present application, the second processing module 33 also includes an updating unit; the updating unit is used to update the RAG knowledge base using a self-feedback learning method based on the first decision report and the second decision report; wherein, the RAG knowledge base is established based on the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method.

[0099] Further, in some embodiments of the present application, the updating unit comprises a first updating subunit and a second updating subunit; the first updating subunit is configured to perform differential analysis on the first decision report and the second decision report by using a preset deep twin network to obtain a differential feature; and the second updating subunit is configured to update the RAG knowledge base by using an incremental graph embedding algorithm according to the differential feature.

[0100] Further, in some embodiments of the present application, the updating unit comprises a third updating subunit and a fourth updating subunit; the flexible HVDC converter station comprises structured attribute data, time series data and spatial temperature distribution data; the third updating subunit is configured to sequentially perform semantic fusion, structural unification and time synchronization on the structured attribute data, the time series data and the spatial temperature distribution data to obtain unified semantic data; and the fourth updating subunit is configured to obtain the RAG knowledge base by using a double-channel knowledge injection method according to the unified semantic data.

[0101] Further, in some embodiments of the present application, the first processing module 32 comprises a first processing unit and a second processing unit; the fault information comprises quantization threshold type fault information and device identification information; the first processing unit is configured to retrieve the quantization threshold type fault information from the RAG knowledge base, generate an alarm judgment result and an alarm level promotion suggestion, and obtain first intermediate processing information according to the alarm judgment result and the alarm level promotion suggestion; and the second processing unit is configured to retrieve the device identification information from the RAG knowledge base, generate device topology information, fault propagation information and real-time working condition information, and obtain second intermediate processing information by using a message passing method of GNN according to the device topology information, the fault propagation information and the real-time working condition information.

[0102] The more detailed step flow and working principle of the present embodiment can be but not limited to referring to the related description of Embodiment One.

[0103] In summary, compared with the prior art, the above embodiments have the following beneficial effects: the embodiments of the present application obtain the fault information of the flexible DC converter station and input it into the rule model layer and the GNN reasoning layer respectively, and combine it with the RAG knowledge base for hierarchical processing, which can achieve rapid response to clear threshold alarms and in-depth analysis of equipment topology-related faults, avoiding the rigidity of single rule reasoning and the lag of traditional association analysis. The fault information and the two-layer intermediate processing results are input into the large language model layer, which can generate a first decision report that meets the requirements of the power industry standard terminology, and form a second decision report after correction to ensure the accuracy and operability of the disposal suggestions. Finally, the fault decision is controlled based on the corrected decision report. Through this application, the fault decision of the flexible DC converter station can be constructed from data collection, multi-dimensional reasoning to standard decision output, effectively solving the problem of low efficiency of fault decision-making of flexible DC converter stations in the prior art.

[0104] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention, which can implement any one of the above-mentioned method embodiments of the present invention to provide a flexible DC converter station fault decision method based on the RAG knowledge base.

[0105] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Furthermore, in the drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which may be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement the present invention without inventive effort.

[0106] Example 3

[0107] Based on the above-mentioned embodiment of the flexible DC converter station fault decision method based on the RAG knowledge base, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the flexible DC converter station fault decision method based on the RAG knowledge base of any embodiment of the present invention is implemented.

[0108] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0109] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0110] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0111] Example 4

[0112] Based on the above-mentioned method embodiments, another embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the flexible DC converter station fault decision method based on the RAG knowledge base described in any one of the above-mentioned method embodiments of the present invention.

[0113] Wherein, the module / unit integrated in the device / terminal equipment, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0114] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A fault decision-making method for a flexible DC converter station based on a RAG knowledge base, characterized in that: include: Obtaining fault information of the flexible DC converter station; Input the fault information into a preset rule model layer, combine it with the RAG knowledge base to obtain first intermediate processing information, and input the fault information into a preset GNN reasoning layer, combine it with the RAG knowledge base to obtain second intermediate processing information; Inputting the fault information, the first intermediate processed information, and the second intermediate processed information into a preset large language model layer to obtain a first decision report, and performing corrections based on the first decision report to obtain a second decision report; Send a control instruction to the flexible DC converter station according to the second decision report.

2. The method for fault decision-making of a flexible DC converter station based on a RAG knowledge base according to claim 1, characterized in that: After the second decision report is obtained by amending the first decision report, the method further includes: The RAG knowledge base is updated using a self-feedback learning method based on the first decision report and the second decision report; wherein, the RAG knowledge base is established based on the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method.

3. The method for fault decision-making of a flexible DC converter station based on a RAG knowledge base according to claim 2, characterized in that: The updating of the RAG knowledge base using a self-feedback learning method according to the first decision report and the second decision report is specifically as follows: Performing differential analysis on the first decision report and the second decision report through a preset deep twin network to obtain differential features; The RAG knowledge base is updated using an incremental graph embedding algorithm according to the difference features.

4. The method for fault decision-making of a flexible DC converter station based on a RAG knowledge base according to claim 2, characterized in that: The RAG knowledge base is established based on the multimodal data fusion method and dynamic knowledge base construction method of the flexible DC converter station, specifically: The flexible DC converter station includes structured attribute data, time series data and spatial temperature distribution data; Sequentially performing semantic fusion, structural unification, and time synchronization on the structured attribute data, the time series data, and the spatial temperature distribution data to obtain unified semantic data; The RAG knowledge base is obtained by adopting a dual-channel knowledge injection method based on the unified semantic data.

5. The method for fault decision-making of a flexible DC converter station based on a RAG knowledge base according to claim 1, characterized in that: The fault information is input into a preset rule model layer, combined with the RAG knowledge base to obtain first intermediate processing information, and the fault information is input into a preset GNN reasoning layer, combined with the RAG knowledge base to obtain second intermediate processing information, specifically: The fault information includes quantitative threshold fault information and device identification information; Retrieving the quantitative threshold fault information from the RAG knowledge base, generating an alarm determination result and an alarm level enhancement suggestion, and obtaining first intermediate processing information based on the alarm determination result and the alarm level enhancement suggestion; The device identification information is retrieved from the RAG knowledge base, device topology information, fault propagation information and real-time operating condition information are generated, and the GNN message passing method is used according to the device topology information, the fault propagation information and the real-time operating condition information to obtain second intermediate processing information.

6. A fault decision-making device for a flexible DC converter station based on a RAG knowledge base, characterized in that: include: A data acquisition module, a first processing module, a second processing module, and a fault decision module; The data acquisition module is used to acquire fault information of the flexible DC converter station; The first processing module is configured to input the fault information into a preset rule model layer, combine it with the RAG knowledge base to obtain first intermediate processing information, and input the fault information into a preset GNN reasoning layer, combine it with the RAG knowledge base to obtain second intermediate processing information; The second processing module is configured to input the fault information, the first intermediate processing information, and the second intermediate processing information into a preset large language model layer to obtain a first decision report, and to modify the first decision report to obtain a second decision report; The fault decision module is configured to send a control instruction to the flexible DC converter station according to the second decision report.

7. The flexible DC converter station fault decision-making device based on RAG knowledge base according to claim 6, characterized in that: The second processing module further includes an updating unit; The updating unit is used to update the RAG knowledge base using a self-feedback learning method based on the first decision report and the second decision report; wherein, the RAG knowledge base is established based on the flexible DC converter station using a multimodal data fusion method and a dynamic knowledge base construction method.

8. The flexible DC converter station fault decision-making device based on RAG knowledge base according to claim 7, characterized in that: The updating unit includes a first updating subunit and a second updating subunit; The first updating subunit is configured to perform differential analysis on the first decision report and the second decision report through a preset deep twin network to obtain differential features; The second updating subunit is configured to update the RAG knowledge base using an incremental graph embedding algorithm according to the difference features.

9. The flexible DC converter station fault decision-making device based on RAG knowledge base according to claim 7, characterized in that: The updating unit includes a third updating subunit and a fourth updating subunit; The flexible DC converter station includes structured attribute data, time series data and spatial temperature distribution data; The third updating subunit is configured to sequentially perform semantic fusion, structural unification, and time synchronization on the structured attribute data, the time series data, and the spatial temperature distribution data to obtain unified semantic data; The fourth updating subunit is used to obtain the RAG knowledge base by adopting a dual-channel knowledge injection method according to the unified semantic data.

10. The flexible DC converter station fault decision-making device based on RAG knowledge base according to claim 6, characterized in that: The first processing module includes a first processing unit and a second processing unit; The fault information includes quantitative threshold fault information and device identification information; The first processing unit is configured to retrieve the quantitative threshold fault information from the RAG knowledge base, generate an alarm determination result and an alarm level escalation suggestion, and obtain first intermediate processing information based on the alarm determination result and the alarm level escalation suggestion; The second processing unit is used to retrieve the device identification information from the RAG knowledge base, generate device topology information, fault propagation information and real-time operating condition information, and use the GNN message passing method according to the device topology information, the fault propagation information and the real-time operating condition information to obtain second intermediate processing information.

11. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method implements the flexible DC converter station fault decision method based on the RAG knowledge base according to any one of claims 1 to 5.

12. A computer-readable storage medium, characterized in that include: A stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the flexible DC converter station fault decision method based on the RAG knowledge base according to any one of claims 1 to 5.

Images