A method and device for constructing an interactive power material knowledge unit

Abstract

This invention discloses a method and apparatus for constructing interactive power material knowledge units. The method includes: acquiring power material data and quantifying it to obtain raw data units; based on the raw data units, considering reliability factors and time series factors, performing dynamic weighted aggregation to provide basic knowledge units; based on the basic knowledge units, acquiring spatial, topological, and temporal relationships between power materials, and combining fault propagation mechanisms to provide comprehensive relationships; fusing the basic knowledge units and comprehensive relationships to generate enhanced knowledge units, and constructing a power material knowledge unit library; using a pre-trained language model, generating responses based on the interactive units and constraints from the power rule base, updating the power material knowledge unit library in response to user feedback, and iteratively optimizing the pre-trained language model until convergence, thereby realizing the construction of interactive power material knowledge units. This method accurately integrates power material data, mines deep relationships, and achieves real-time, intelligent management and control.

Description

Technical Field

[0001] This invention belongs to the field of power material management and knowledge engineering technology, specifically relating to a method and apparatus for constructing interactive power material knowledge units. Background Technology

[0002] Power material management is a core aspect of power equipment operation, involving the entire lifecycle management of basic attributes, procurement, warehousing, operation and maintenance, and IoT data. With the intelligent development of the power industry, the sources of material data are becoming increasingly complex, including material management devices, ERP devices, warehouse management devices, operation and maintenance work order devices, and equipment monitoring platforms. Data formats encompass structured tables, unstructured text, and real-time streaming data. To effectively integrate this multi-source, heterogeneous data, the industry urgently needs to build power material knowledge units covering business processes such as procurement, warehousing, and operation and maintenance to support intelligent decision-making and question-and-answer services, meeting the real-time, compliance, and intelligent requirements of power material management.

[0003] Patent application CN118035463A discloses a method and apparatus for constructing a multimodal knowledge graph for power grid dispatching. The method includes: collecting diverse and heterogeneous data from the power grid dispatching field; preprocessing the data according to the characteristics of multimodal data types; labeling the preprocessed multimodal data and constructing a knowledge sample library; performing unified feature fusion on the knowledge sample library data obtained after labeling the multimodal data, and extracting multimodal power material knowledge for power grid dispatching using a deep learning model; establishing relationships between multimodal control power materials based on the extracted multimodal power material knowledge, and fusing and linking multimodal control knowledge to form a multimodal knowledge graph for power grid dispatching. By establishing this multimodal knowledge graph, the method enables the co-construction and sharing of multimodal knowledge across devices and services, improves the level of multimodal knowledge reconstruction and fusion in power grid dispatching, and provides multimodal data support for control services. However, in this method: the data fusion mechanism is not adapted to the dynamic characteristics of power materials and lacks real-time linkage with business stages, material status, and environmental factors; the power material association modeling is only based on topological connection relationships, which does not fully explore the deep associations between power materials, and the real-time early warning effect of fault propagation is poor; and there is a lack of interaction mechanism, which does not support intelligent decision-making and question-and-answer services.

[0004] Therefore, how to accurately integrate multi-source heterogeneous data throughout the entire life cycle of power materials, deeply explore the real correlations between power materials, and introduce interactive mechanisms to achieve real-time and intelligent management and control of power materials throughout their entire life cycle is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and apparatus for constructing interactive power material knowledge units. The method includes: acquiring and quantifying power material data to obtain raw data units; based on the raw data units, considering reliability factors and time-series factors, performing dynamic weighted aggregation to generate basic knowledge units; based on the basic knowledge units, acquiring spatial, topological, and temporal relationships between power materials, and combining fault propagation mechanisms to generate comprehensive relationships; fusing the basic knowledge units and comprehensive relationships to generate enhanced knowledge units and constructing a power material knowledge unit library; using a pre-trained language model, generating responses based on interactive units and power rule base constraints, updating the power material knowledge unit library in response to user feedback, and iteratively optimizing the pre-trained language model until convergence, thereby realizing the construction of interactive power material knowledge units. This method accurately integrates multi-source heterogeneous data throughout the entire lifecycle of power materials, deeply mines the real relationships between power materials, and introduces an interactive mechanism to achieve real-time, intelligent management and control of the entire lifecycle of power materials.

[0006] In a first aspect, the present invention provides a method for constructing interactive power material knowledge units, specifically including the following steps:

[0007] Acquire power material data and quantify it to obtain raw data units;

[0008] Based on the original data units, taking into account the credibility factor and the time series factor, dynamic weighted aggregation is performed to give the basic knowledge units;

[0009] Based on basic knowledge units, spatial, topological, and temporal relationships among power materials are obtained, and comprehensive relationships are given in combination with fault propagation mechanisms.

[0010] By integrating basic knowledge units and comprehensive connections, enhanced knowledge units are generated, and a knowledge unit library for power materials is constructed.

[0011] A pre-trained language model is used to generate answers based on interactive units and constraints from the power rule base. In response to user feedback, the power material knowledge unit base is updated, and the pre-trained language model is iteratively optimized until convergence, thus realizing the construction of interactive power material knowledge units.

[0012] Furthermore, the credibility factor includes a first credibility factor and a second credibility factor. Based on the original data unit, taking into account the credibility factor and the time series factor, dynamic weighted aggregation is performed to give the basic knowledge unit, which specifically includes:

[0013] The system acquires the business stage, material status, and environmental factors of power materials, and uses pre-set weighting coefficients to weight the business stage, material status, and environmental factors to provide the first reliable factor for power material data.

[0014] Obtain the rated parameters and historical baselines of power materials, verify them, and provide a second credibility factor for the power material data;

[0015] Based on the lifetime-aware decay mechanism, the time-series factor of power material data is obtained;

[0016] By integrating the first credibility factor, the second credibility factor, and the time series factor, a comprehensive factor is obtained. This factor is then weighted and aggregated with the original data units to give the basic knowledge units.

[0017] Furthermore, the rated parameters and historical baselines of power equipment are obtained, verified, and a second reliability factor for the power equipment data is provided, specifically including:

[0018] Based on the type of power equipment, select key operating parameters and determine the rated values ​​and ranges of the parameters;

[0019] For each key operating parameter, check whether the current value meets the parameter's rated range. If it does not, filter the data. If it does, perform historical deviation verification to obtain the deviation between the current value and the historical average of the key operating parameter. Combine the deviation adjustment coefficient and the parameter's rated value to give a second confidence factor.

[0020] Based on the lifetime-aware decay mechanism, the time-series factors for obtaining power material data specifically include:

[0021] Determine the basic time-series attenuation coefficient based on the type of power equipment;

[0022] Based on the basic timing decay coefficient and combined with the running time, a lifetime-perceived decay coefficient is given;

[0023] Based on the lifetime-aware decay coefficient, and combined with the current time and the key timestamps of power material data business, a time series factor is given.

[0024] Furthermore, the basic knowledge unit is specifically represented as follows:

[0025]

[0026] In the formula, K t Let i be the basic knowledge unit at time t, i be the data category, N be the number of data categories, and D be the data type. i (t) represents the original data unit of the i-th type of data. The first confidence factor for the i-th type of data. The second confidence factor for the i-th type of data. Let i be the time series factor of the i-th type of data. This represents the comprehensive factor, where t is the current time;

[0027] The first credibility factor is specifically expressed as:

[0028]

[0029] In the formula, The first confidence factor for the i-th type of data. For the business phase, Regarding the status of supplies, As environmental factors, , , These are the weighting coefficients;

[0030] The second credibility factor is specifically expressed as:

[0031]

[0032] In the formula, Let P be the second reliability factor for the i-th type of data, P be the set of key operating parameters for power materials, and p be the key operating parameters for power materials. This represents the current value of p, a key operating parameter for power equipment. This represents the historical average value of p, a key operating parameter for power equipment. This represents the lower limit of the rated range of the key operating parameter p for power equipment. Where p is the upper limit of the rated range of the key operating parameter p for power materials, and k is the deviation adjustment coefficient. t represents the rated value of the key operating parameter p for power equipment, and t represents the current time.

[0033] The time series factor is specifically expressed as:

[0034]

[0035] In the formula, Let i be the time series factor of the i-th type of data. Here, l is the base decay coefficient, l is the lifetime weight, and Age is the runtime. T represents the perceived lifespan degradation coefficient. i Let t be the business-critical timestamp for the i-th type of data, and t be the current time.

[0036] Furthermore, based on basic knowledge units, spatial, topological, and temporal relationships among power materials are obtained, and combined with fault propagation mechanisms, a comprehensive correlation is provided, specifically including:

[0037] Obtain the spatial range of each power material, calculate the overlapping area and total area of ​​the spatial range, obtain the spatial intersection-union ratio, set spatial weights according to environmental factors, and combine the maximum value of the spatial range to give the spatial correlation.

[0038] Based on the location of each power material in the topology, the topological correlation between power materials is matched according to the topology relation library and predefined topology hierarchical rules, and the topological weight is set according to the business stage to give the topological correlation.

[0039] Obtain the key business timestamps for each power material, calculate the time correlation, set time weights, and provide the time correlation.

[0040] Obtain the fault type, fault severity, and fault propagation distance from the faulty power equipment to the target power equipment. Combine this with the distance scale to calculate the fault distance attenuation. Set the fault propagation gain according to the equipment status and provide the fault propagation correlation.

[0041] By integrating spatial correlation, topological correlation, temporal correlation, and fault propagation correlation, a comprehensive correlation is presented.

[0042] Furthermore, considering the overall relationship, it can be specifically expressed as follows:

[0043]

[0044] In the formula, E represents electrical materials a and power materials E b The comprehensive correlation, , , These are spatial weights, topological weights, and temporal weights, respectively. For fault conduction gain, For power materials E a and power materials E b Spatial intersection-union ratio, S a For power materials E a Spatial range, S b For power materials E b spatial range, E represents electrical materials a and power materials E b The maximum value of the spatial range, For power materials E a and power materials E b The degree of topological association, T is the time correlation coefficient. a T b E, respectively, power materials a Power materials E b The key timestamps for the business are: Sev (severity of the fault), Dis (fault propagation distance), Scl (distance scale), t (current time), Env(t) (environmental factor), Phs(t) (business stage), and Stat(t) (material status).

[0045] The intersection-union ratio of spaces is specifically expressed as:

[0046]

[0047] In the formula, E represents electrical materials a and power materials E b Spatial intersection and union ratio, E represents electrical materials a and power materials E b The intersection of spatial ranges, E represents electrical materials a and power materials E b The union of spatial ranges.

[0048] Furthermore, by integrating basic knowledge units and comprehensive relationships, enhanced knowledge units are generated, and a power material knowledge unit library is constructed, specifically including:

[0049] An attention mechanism is used to integrate basic knowledge units with comprehensive connections, and combined with power rule verification, to generate enhanced knowledge units;

[0050] Metadata is associated with each enhanced knowledge unit, an index is built, and a dynamic update mechanism is set up to form a knowledge unit library for power materials.

[0051] Furthermore, a pre-trained language model is used to generate answers based on the interaction units and constraints from the power rule base, specifically including:

[0052] Input the interactive unit into the search engine to retrieve relevant enhanced knowledge units from the power materials knowledge unit database;

[0053] By combining relevant enhanced knowledge units, interaction units, and constraints in the power rule base, the context of the pre-trained language model is constructed;

[0054] The constructed context is input into the pre-trained language model, and the RLAIF mechanism is used to generate candidate answers. Combined with the compliance verification of the power rule base, the final answer is given.

[0055] Furthermore, in response to user feedback, the power material knowledge unit base is updated, and the pre-trained language model is iteratively optimized until convergence, specifically including:

[0056] Analyze user responses, extract feedback values, and locate relevant augmented knowledge units;

[0057] The validity of the feedback value is verified using pre-defined verification rules. Based on the verified feedback value, the knowledge unit is modified and enhanced, and the power material knowledge unit database is updated synchronously.

[0058] A training set is built based on user feedback, and the pre-trained language model is trained on the training set to fine-tune the model parameters.

[0059] The model performance was evaluated using a validation set, which included the model’s answer accuracy on the validation set, the proportion of the model’s generated answers that conformed to the electricity rule base, and the error rate reported by users.

[0060] Repeat the training and validation process of the pre-trained language model until the termination condition is met.

[0061] Secondly, the present invention also provides an apparatus for constructing interactive power material knowledge units, employing the aforementioned method for constructing interactive power material knowledge units, specifically including:

[0062] The data acquisition module is used to acquire power material data and quantify it to obtain raw data units;

[0063] The knowledge construction module is used to perform dynamic weighted aggregation based on the original data units, taking into account the reliability factor and the time series factor, to give the basic knowledge units; based on the basic knowledge units, it obtains the spatial, topological and temporal correlations between power materials, and gives the comprehensive correlations in combination with the fault propagation mechanism; it integrates the basic knowledge units and the comprehensive correlations to generate enhanced knowledge units, and builds a power material knowledge unit library.

[0064] The interactive question-and-answer module uses a pre-trained language model to generate answers based on interactive units and constraints from the power rule base. In response to user feedback, it updates the power material knowledge unit base and iteratively optimizes the pre-trained language model until convergence, thereby realizing the construction of interactive power material knowledge units.

[0065] The method and apparatus for constructing interactive power material knowledge units provided by this invention have at least the following beneficial effects:

[0066] (1) By integrating multi-source heterogeneous data of the entire life cycle of power materials, a basic knowledge unit with high timeliness is generated; through comprehensive association, the deep relationship between power materials is reflected in a true and comprehensive manner; reliable answers are generated based on pre-trained language models and compliance constraints, and knowledge unit correction and model iteration are driven by user feedback, thus realizing the intelligent and efficient management of the entire life cycle of power materials.

[0067] (2) Based on the first reliable factor of business stage, equipment status and environmental factors, key data is prioritized for fusion; outliers are filtered by combining power rated parameters and historical baselines to reduce false alarm rate; the lifespan-aware decay mechanism effectively distinguishes between new and old power materials. Dynamic weighted aggregation significantly improves the quality and timeliness of power material data.

[0068] (3) By integrating spatial correlation, topological correlation, temporal correlation and fault propagation correlation, we can deeply explore the real correlation between power materials, analyze the impact of fault dynamic diffusion, accurately locate related power materials, and achieve deep adaptation of power business.

[0069] (4) Through language models and interactive question and answer, reliable answers that conform to the constraints of power rules are generated, and knowledge unit correction and model iteration are driven by user feedback to support intelligent decision-making and management of power materials. Attached Figure Description

[0070] Figure 1 A flowchart illustrating a method for constructing an interactive power material knowledge unit provided by the present invention;

[0071] Figure 2 A flowchart illustrating the basic knowledge unit is provided for one embodiment of the present invention;

[0072] Figure 3 A schematic diagram of the integrated association process is provided for one embodiment of the present invention;

[0073] Figure 4 A schematic diagram of the process for generating an answer according to one embodiment of the present invention;

[0074] Figure 5 A schematic diagram illustrating the logical relationship for generating answers in one embodiment of the present invention;

[0075] Figure 6 A schematic diagram illustrating the process of updating the power material knowledge unit library and iteratively optimizing the pre-trained language model according to one embodiment of the present invention;

[0076] Figure 7 This invention provides a schematic diagram of the structure of an interactive power material knowledge unit construction device. Detailed Implementation

[0077] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0078] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0079] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0080] In the field of power material management, integrating multi-source and dispersed power material data to form a structured and dynamic knowledge carrier is essential. This knowledge carrier must adapt to different business scenarios, accurately reflect the status of power materials, and be updated promptly according to environmental changes. Interactive access to compliant power material knowledge, coupled with continuous optimization through feedback, enables dynamic, scenario-based, compliant, and interactive management of power materials, thereby improving the intelligence level and decision-making efficiency of power material management.

[0081] Firstly, such as Figure 1 As shown, this invention provides a method for constructing interactive power material knowledge units, specifically including the following steps:

[0082] S101: Acquire power material data and quantify it to obtain raw data units.

[0083] Specifically, power material data includes basic attribute data, procurement data, warehousing data, operation and maintenance data, and IoT data. Basic attribute data includes power material model, rated capacity, and manufacturer; procurement data includes purchase contract number, purchase time, and supplier information; warehousing data includes storage warehouse location, entry time, and inventory quantity; operation and maintenance data includes maintenance records, fault type, and maintenance personnel; and IoT data includes real-time operating temperature, load rate, and vibration frequency. Basic attribute data and procurement data can be obtained by calling the open API interfaces between the power material management device and the ERP (Enterprise Resource Planning) device; warehousing data can be obtained by establishing a connection with the database of the warehousing management device through database synchronization tools; operation and maintenance data can be obtained by establishing a connection with the database of the operation and maintenance work order device through database synchronization tools; and IoT data can be obtained by subscribing to the data stream of the power equipment online monitoring platform through MQTT message queues.

[0084] After acquiring power material data, the process involves cleaning (removing duplicate data and correcting errors), standardization (e.g., unifying the naming format of material models), and structuring (converting unstructured descriptive text into structured fields) to ensure consistent data format and accurate content across different sources. For structured data, numerical encoding or one-hot encoding is used for vectorization. For unstructured text such as fault descriptions and maintenance records, the Sentence-BERT model can be used to convert the text into semantic vectors. For time-series data from IoT, sliding windows and feature engineering (e.g., mean, variance, peak value) can be used to convert the time series into fixed-length vectors. For spatial data such as storage locations and installation sites, GeoHash encoding can be used to convert latitude and longitude into string vectors, which are then converted into low-dimensional numerical vectors using Word2Vec. After vectorization, the original data vector corresponding to each power material is obtained—the vectorization of the original data unit. The vector dimension can be adjusted according to the data type. The original data unit is a digital and standardized representation of power material data, enabling machines to uniformly process data from different sources and in different formats. It retains the core information of the data and serves as the core basis for subsequent knowledge unit generation and interactive question answering.

[0085] S102: Based on the original data units, taking into account the reliability factor and the time series factor, perform dynamic weighted aggregation to give the basic knowledge units.

[0086] In a specific embodiment, the credibility factor includes a first credibility factor and a second credibility factor. For example... Figure 2 As shown, based on the original data units, and taking into account the reliability factor and the time series factor, dynamic weighted aggregation is performed to give the basic knowledge units, which specifically include:

[0087] S1021: Obtain the business stage, material status, and environmental factors of power materials, and use pre-set weighting coefficients to weight the business stage, material status, and environmental factors to provide the first reliable factor of power material data.

[0088] It is understandable that business phases refer to the key process stages in the entire lifecycle of power materials, including the emergency phase (such as fault repair and typhoon disaster relief, with the highest business priority and data requiring real-time accuracy), the operation and maintenance phase (such as routine maintenance and equipment inspection, with medium business priority and data requiring timeliness and completeness), the procurement phase (such as material bidding and contract signing, with lower business priority and data requiring standardization and traceability), and the decommissioning phase (such as equipment scrapping and material disposal, with the lowest business priority and data requiring archiving and compliance). Data in different business phases can be quantified with different numerical values ​​due to their varying business priorities. In this example, the emergency phase is 1.0, the operation and maintenance phase is 0.7, the procurement phase is 0.4, and the decommissioning phase is 0.1.

[0089] Material status refers to the current health or usage status of electrical materials, reflecting the validity of the data. Material status includes fault status, defect status, normal status, and decommissioned status. In this example, different numerical values ​​are used to quantify material status: fault status is 1.0, defect status is 0.8, normal status is 0.5, and decommissioned status is 0.2.

[0090] Environmental factors refer to external environmental factors that affect the operation of power equipment, reflecting the scenario adaptability of the data. Environmental factors include natural environment, load environment, and human environment, and different numerical values ​​can be used to represent the degree of environmental impact on power equipment. In this example, the environmental impact is represented by a value of 1 ± Δ (Δ is the environmental deviation; the larger the value, the greater the environmental impact). For example, typhoon weather is set to 1.2 (Δ=0.2), high load is set to 1.1 (Δ=0.1), and normal environment is set to 1.0 (Δ=0).

[0091] The first credibility factor is obtained by weighting the business stage, material status, and environmental factors using pre-set weighting coefficients. These weighting coefficients can be flexibly set or dynamically adjusted according to actual needs. The first credibility factor quantifies the credibility of data in the current business scenario. Through the weighted fusion of business stage, material status, and environmental factors, data priority ranking is achieved, ensuring that highly credible data is displayed first, aggregated first into basic knowledge units, and used first to generate candidate answers, thus meeting the actual needs of power material management.

[0092] Furthermore, the first credibility factor is specifically expressed as:

[0093]

[0094] In the formula, The first confidence factor for the i-th type of data. For the business phase, Regarding the status of supplies, As environmental factors, , , t represents the weighting coefficient, and t represents the current time.

[0095] S1022: Obtain the rated parameters and historical baselines of power materials, verify them, and provide a second reliability factor for the power material data.

[0096] Specifically, first select key operating parameters based on the type of power equipment, and determine the rated values ​​and ranges of the parameters;

[0097] For each key operating parameter, check whether the current value meets the parameter's rated range. If it does not, filter the data (i.e., set the second confidence factor to 0). If it does, perform historical deviation verification to obtain the deviation between the current value of the key operating parameter and the historical average. Combine the deviation adjustment coefficient and the parameter's rated value to give the second confidence factor.

[0098] Understandably, different types of electrical equipment require different operating parameters reflecting their core operating status. For example, transformers require key operating parameters such as oil temperature, insulation resistance, and load rate; cables require key operating parameters such as partial discharge, laying temperature, and conductor resistance. The appropriate key operating parameters should be selected based on the type of electrical equipment. Rated parameter values ​​refer to the design target values ​​of the equipment (e.g., rated oil temperature of a transformer = 85℃), which can be read from the nameplate of the electrical equipment. Rated parameter ranges refer to the permissible operating range of the equipment, and can be referenced from industry standards.

[0099] The historical baseline refers to the historical statistical value of a certain operating parameter under normal operating conditions of power equipment, reflecting the long-term stable operating level of the equipment. In this example, the historical average is selected as the historical baseline. Then, through historical deviation calculation, the degree of data anomaly is quantified. The quantification results of individual key operating parameters are combined to obtain the quality reliability of all key operating parameters, i.e., the second reliability factor. The second reliability factor quantifies the quality reliability of the data, i.e., whether the data conforms to the normal operating conditions of power equipment (compared to the historical baseline) and design limits. Abnormal data is filtered to prevent erroneous data from entering the knowledge unit; high-reliability data is prioritized for aggregation into basic knowledge units to improve the data accuracy of the knowledge units; and high-reliability data is prioritized for generating candidate answers to improve the reliability of the answers.

[0100] Furthermore, the second credibility factor is specifically expressed as:

[0101]

[0102] In the formula, Let P be the second reliability factor for the i-th type of data, P be the set of key operating parameters for power materials, and p be the key operating parameters for power materials. Let p be the current value of the key operating parameter p for power equipment, and n be the number of key operating parameters. This represents the historical average value of p, a key operating parameter for power equipment. This represents the lower limit of the rated range of the key operating parameter p for power equipment. Where p is the upper limit of the rated range of the key operating parameter p for power materials, and k is the deviation adjustment coefficient. t represents the rated value of the key operating parameter p for power equipment, and t represents the current time.

[0103] S1023: Time-series factor for acquiring power material data based on lifetime-aware decay mechanism.

[0104] It is understandable that the value (i.e., timeliness) of power material data decreases over time, and the rate of decay varies among different equipment (different types of materials). The lifetime-aware decay mechanism adds the operating time of the power material to the base time decay, dynamically adjusting the decay rate.

[0105] Specifically, based on the lifetime-aware decay mechanism, the time-series factors for acquiring power material data include:

[0106] Determine the basic time-series attenuation coefficient based on the type of power equipment;

[0107] Based on the basic timing decay coefficient and combined with the running time, a lifetime-perceived decay coefficient is given;

[0108] Based on the lifetime-aware decay coefficient, and combined with the current time and the key timestamps of power material data business, a time series factor is given.

[0109] Different types of power equipment (such as transformers, cables, and circuit breakers) have different base time-series attenuation coefficients. Transformers require real-time monitoring of their operating status and have high requirements for data timeliness, resulting in a relatively large base time-series attenuation coefficient. Cables experience slower changes in status and have lower requirements for data timeliness, resulting in a relatively small base time-series attenuation coefficient. In this example, the base time-series attenuation coefficient for transformers is 0.1, and for cables it is 0.08. Based on the base time-series attenuation coefficient, a lifetime weight is determined according to empirical values, and then combined with the operating time, the lifetime-perceived attenuation coefficient can be calculated. The key timestamp for power equipment data business refers to the key time nodes for data generation, such as equipment commissioning time and fault occurrence time. The difference between the current time and the key timestamp is the time difference. The product of the lifetime-perceived attenuation coefficient and the time difference is used as the exponent, resulting in the time-series factor. The time-series factor can quantify the time-sensitivity attenuation of power equipment, allowing knowledge units to prioritize the aggregation of new data while also considering the lifetime of power equipment.

[0110] Furthermore, the time series factor is specifically expressed as:

[0111]

[0112] In the formula, Let i be the time series factor of the i-th type of data. Here, l is the base decay coefficient, l is the lifetime weight, controlling the impact of runtime on decay, and Age is the runtime. T represents the perceived lifespan degradation coefficient. i Let t be the business-critical timestamp for the i-th type of data, and t be the current time.

[0113] S1024: Integrate the first credibility factor, the second credibility factor, and the time series factor to obtain a comprehensive factor, which is then weighted and aggregated with the original data units to give the basic knowledge units.

[0114] The first credibility factor represents the credibility of the data within the business scenario, the second credibility factor represents the credibility of the data quality, and the time series factor represents the timeliness of the data. A comprehensive factor is obtained by multiplying and fusing the first credibility factor, the second credibility factor, and the time series factor, which rigorously filters out unreliable data. If any factor is low (e.g., the time series factor L...),... i (t) = 0.2, indicating that the data is outdated), and the comprehensive factor will decrease significantly (e.g. =0.9、 =0.8, L=0.2, comprehensive factor=0.144), thereby reducing the impact of low-value data on basic knowledge units. By weighting and aggregating the original data units with the corresponding comprehensive factor, the contribution of high-value data is highlighted, ultimately resulting in basic knowledge units that reflect the true status of power materials. The fusion method can also take other forms according to actual needs, and is not limited here.

[0115] Furthermore, the basic knowledge unit is specifically represented as follows:

[0116]

[0117] In the formula, K t Let i be the basic knowledge unit at time t, i be the data category, N be the number of data categories, and D be the data type. i (t) represents the original data unit of the i-th type of data. The first confidence factor for the i-th type of data. The second confidence factor for the i-th type of data. Let i be the time series factor of the i-th type of data. This represents the comprehensive factor, and t is the current time.

[0118] S103: Based on basic knowledge units, obtain spatial, topological, and temporal relationships among power materials, and combine them with fault propagation mechanisms to provide comprehensive relationships.

[0119] Understandably, basic knowledge units represent the attributes and states of power materials themselves, but lack interrelationships. The relationships between power materials are multi-attribute, multi-scenario, and dynamically changing, a complex interplay of physical, functional, temporal, and dynamic attributes. A single dimension cannot fully depict this complexity. Comprehensive relationships supplement the spatial, topological, temporal, and fault propagation relationships between power materials, deeply adapting to business scenarios, making knowledge units more comprehensive, and supporting more complex question-and-answer structures. Comprehensive relationships integrate spatial, topological, temporal, and fault propagation relationships between power materials, comprehensively reflecting the deep relationships between them, and adapting to the needs of different business scenarios through weight adjustments. Omitting any dimension will result in a one-sided description of relationships, failing to meet the complex management needs of power materials.

[0120] In a specific embodiment, such as Figure 3 As shown, based on basic knowledge units, spatial, topological, and temporal relationships among power materials are obtained, and combined with fault propagation mechanisms, comprehensive relationships are presented, including:

[0121] S1031: Obtain the spatial range of each power material, calculate the overlapping area and total area of ​​the spatial range, obtain the spatial intersection-union ratio, set spatial weights according to environmental factors, and combine the maximum value of the spatial range to give the spatial correlation.

[0122] Based on the spatial location of power materials, their spatial influence range, i.e., spatial extent, such as warehouse area or latitude and longitude range, can be determined. The proximity of the physical location of power materials is quantified by the ratio of spatial intersection ratio to the maximum value of the spatial extent. Changes in environmental factors (such as typhoons, high temperatures, etc.) can alter the business importance of physical location associations. Therefore, the weights of spatial associations are dynamically adjusted based on environmental factors to adapt physical location associations to business needs under the current environment, avoiding decision-making lags caused by static weights and meeting the dynamic needs of power material management. The spatial distribution of power materials is their fundamental attribute. Spatial association quantifies the physical spatial relationships of power materials, effectively supporting business scenarios such as warehousing management and distribution planning.

[0123] S1032: Based on the position of each power material in the topology, match the topology correlation between power materials according to the topology relation library and predefined topology hierarchical rules, and set topology weights according to the business stage to give the topology correlation.

[0124] As we can understand, topology refers to the connection relationships (physical or logical connections) of electrical materials (such as transformers, circuit breakers, and cables) within an electrical installation, reflecting the functional coordination between devices. A topology database is a structured database storing the topology of electrical materials, including device ID, bay, busbar, substation, and connected upper / lower-level devices. Topology hierarchical rules classify the tightness and level of topological associations based on the hierarchy of electrical installations; the higher the level, the tighter the association. In this example, the association degree of devices within the same bay is set to 1.0, the association degree of devices on the same busbar but in different bays is set to 0.8, the association degree of devices on the same substation but on different buses is set to 0.5, and the association degree of devices in different substations is set to 0.1.

[0125] Different business phases (such as emergency response, operation and maintenance, procurement, and decommissioning) have different requirements for topology association. In the emergency phase (such as fault repair, where it's necessary to quickly locate the associated equipment of the faulty device), the topology weight should be higher. In the procurement phase, the focus is more on the basic attributes of power materials (such as model and price), and the weight of topology association is lower. Adjusting the topology weight according to the business phase improves the adaptability to business scenarios. Topology association quantifies the functional connections of power materials within power installations, supporting scenarios such as fault handling and operation and maintenance decisions.

[0126] S1033: Obtain the key business timestamps of each power material, calculate the time correlation, set time weights, and provide the time correlation.

[0127] Key business timestamps refer to critical points in the entire lifecycle of power equipment, such as commissioning time, maintenance time, failure time, and decommissioning time. By using these key business timestamps, the time difference between different power equipment can be calculated. Time correlation can be calculated using empirical formulas, and time weights can be set according to business needs to obtain the time relationships. The time correlation of power equipment quantifies the association between equipment in a time series (such as equipment from the same procurement batch or those recently maintained), supporting scenarios such as traceability management and operation and maintenance planning.

[0128] S1034: Obtain the fault type, fault severity, and fault propagation distance from the faulty power material to the target power material. Combine this with the distance scale to calculate the fault distance attenuation, set the fault propagation gain according to the material status, and provide the fault propagation correlation.

[0129] It is understandable that in power installations, a fault in one material can be propagated to other materials through electrical connections or physical proximity, and fault propagation is dynamic and cascading. Without considering fault propagation correlations, the scope of the fault's impact cannot be predicted, leading to decision-making errors and repair delays. Fault type refers to the specific category of fault occurring in the power material (e.g., "short circuit fault" or "insulation fault" in transformers, "partial discharge fault" or "overheating fault" in cables). Fault severity refers to the degree of harm the fault poses to the power installation; in this example, a numerical value of 0 to 1 is used to quantify fault severity. Fault propagation distance refers to the physical distance from the faulty material to the target material; the distance scale is an empirical value used to adjust the rate of fault decay with distance, measuring the effective range of the fault's impact. Fault distance decay refers to the decrease in the degree of fault impact as the propagation distance increases; in this example, an exponential form is used to characterize this. Fault propagation gain is used to adjust the impact of fault propagation correlations, and the probability of fault propagation is closely related to the material's state (e.g., materials in a defective state are more susceptible to fault propagation than materials in a normal state). Therefore, the fault propagation gain needs to be adjusted in conjunction with the material's state to highlight the correlation of high-risk materials. Fault propagation correlation quantifies the dynamic impact of material failure on other materials, supporting scenarios such as fault prediction and emergency repair decision-making.

[0130] S1035: Integrate spatial correlation, topological correlation, temporal correlation and fault propagation correlation to provide a comprehensive correlation.

[0131] The relationships between power equipment are a comprehensive reflection of multiple dimensions, including space, topology, time, and fault propagation; a single dimension cannot fully reflect the true relationships. Integrating multi-dimensional relationships can avoid biased judgments and, through the complementarity of each dimension, provide more accurate basis for fault early warning, operation and maintenance decisions, and risk prevention and control. The integration method can be selected according to business needs, such as weighted summation, vectorized concatenation, or attention-based integration.

[0132] Furthermore, a fusion method is adopted by vectorizing and concatenating the related items to synthesize the relationships, specifically represented as follows:

[0133]

[0134] In the formula, E represents electrical materials a and power materials E b The comprehensive correlation, For power materials E a and power materials E b Spatial association, For power materials E a and power materials E b Topological association, For power materials E a and power materials E bTime correlation, For power materials E a and power materials E b Fault propagation correlation, , , These are spatial weights, topological weights, and temporal weights, respectively. For fault conduction gain, For power materials E a and power materials E b Spatial intersection-union ratio, S a For power materials E a Spatial range, S b For power materials E b spatial range, E represents electrical materials a and power materials E b The maximum value of the spatial range, For power materials E a and power materials E b The topological correlation degree, T a T b E, respectively, power materials a Power materials E b The business-critical timestamps The time correlation coefficient controls the nonlinear decay rate of the time difference on the time correlation. It can be flexibly set according to the business scenario requirements and the time characteristics of the materials, or it can be determined by fitting historical data. Sev is the severity of the fault, Dis is the fault propagation distance, Scl is the distance scale, t is the current time, Env(t) is the environmental factor, Phs(t) is the business stage, and Stat(t) is the material status.

[0135] The intersection-union ratio of spaces is specifically expressed as:

[0136]

[0137] In the formula, E represents electrical materials a and power materials E b Spatial intersection and union ratio, E represents electrical materials a and power materials E b The intersection of spatial ranges, E represents electrical materials a and power materials E b The union of spatial ranges.

[0138] S104: Integrate basic knowledge units and comprehensive relationships to generate enhanced knowledge units and construct a knowledge unit library for power materials.

[0139] Understandably, basic knowledge units represent the attributes and states of power materials themselves, while comprehensive associations represent the spatial, topological, temporal, and fault propagation relationships between power materials. By integrating the two, a complete knowledge representation of power materials is formed, comprehensively describing the current status and potential impact of power materials.

[0140] In a specific embodiment, basic knowledge units and comprehensive relationships are integrated to generate enhanced knowledge units, and a power material knowledge unit library is constructed, specifically including:

[0141] An attention mechanism is used to integrate basic knowledge units with comprehensive connections, and combined with power rule verification, to generate enhanced knowledge units;

[0142] Metadata is associated with each enhanced knowledge unit, an index is built, and a dynamic update mechanism is set up to form a knowledge unit library for power materials.

[0143] In a specific embodiment, basic knowledge units and comprehensive associations are used as input vectors, and basic knowledge units are mapped to the query space, while comprehensive associations are mapped to the key-value space, specifically as follows:

[0144]

[0145]

[0146]

[0147] In the formula, Q, K, and V are the query vector, key vector, and value vector, respectively; BKU and CA are the vectorized representations of basic knowledge units and comprehensive associations, respectively; and W... q W K W v This is the weight matrix.

[0148] Calculate the dot product of the query vector and the transpose of the key vector, then perform dot product scaling and Softmax normalization to obtain the attention weight matrix. The above process can be formally expressed as:

[0149]

[0150] In the formula, Score represents the dot product; the larger the dot product, the higher the relevance. Q represents the query vector. This is the transpose of the key vector K;

[0151]

[0152] In the formula, ScaledScore is the scaled dot product, and Score is the dot product value. d is the scaling factor. k The dimension of the key vector;

[0153]

[0154] In the formula, A is the attention weight matrix, ScaledScore is the scaled dot product, and Softmax represents normalization.

[0155] Multiplying the value vector by the attention weight matrix yields a weighted association vector, which is then residual-joined with the basic knowledge unit. This completes the fusion of the basic knowledge unit and the comprehensive association, as shown below:

[0156]

[0157] In the formula, Fused represents the fusion result vector, A is the attention weight matrix, V is the value vector, and BKU is the vectorized representation of the basic knowledge unit.

[0158] The fusion result vector must undergo power rule validation to avoid generating knowledge units that violate regulations or do not conform to operational logic. Power rules (such as safety procedures and operational standards) are converted into IF-THEN format using rule engines such as Drools and Easy Rules. The fusion result vector is then input into the rule engine to check for compliance with all rules. Violating fusion result vectors are adjusted by adding missing tag information or correcting errors. The fusion result vector validated by power rules is the Enhanced Knowledge Unit (EKU), containing information such as attributes, real-time status, historical data, relationships, and rule tags.

[0159] To facilitate retrieval and management, metadata is associated with each enhanced knowledge unit (EKU). The metadata should include basic information about the power materials (material ID, name, rated parameters, etc.), EKU generation information (generation time, update time, etc.), association information (associated material ID, association type, association strength, etc.), rule validation information (validation time, validation result, etc.), and status information (valid, abnormal). All EKUs are stored in a vector database, and the metadata is stored in a relational database. A similarity index is created for the EKUs, and a B+ tree index is created for key fields in the metadata (such as material ID, association type, and rule tags), thus completing the construction of the power material knowledge unit database.

[0160] The dynamic update mechanism refers to detecting update trigger conditions through sensors, database triggers, or scheduled tasks. These trigger conditions include changes in the status of the power materials themselves, changes in their relationships with other materials, adjustments to power business rules, and the arrival of the update time period. Based on the new status data, the reliability factor and time series factor are updated, the basic knowledge units are recalculated, the comprehensive associations are updated, new enhanced knowledge units are generated to replace the original enhanced knowledge units, and the index is updated synchronously, thus completing the dynamic update of the power material knowledge unit library.

[0161] S105: Using a pre-trained language model, based on the interaction unit and combined with the constraints of the power rule base, it generates answers, responds to the user's answer feedback, updates the power material knowledge unit base, and iteratively optimizes the pre-trained language model until convergence, thereby realizing the construction of interactive power material knowledge units.

[0162] The power materials knowledge unit base uses vectorized knowledge representation, which cannot directly handle natural language interactions. A pre-trained language model (PLM) can transform users' natural language questions into structured queries, connecting to the power materials knowledge unit base's retrieval interface. It can also understand users' implicit needs, extract relevant information from the power materials knowledge unit base, and, combined with the power rule base, transform the structured information in the knowledge unit base into a complete natural language answer.

[0163] The initial construction of the power material knowledge unit base may have issues such as missing information or insufficient timeliness. User feedback on the answers generated by the language model can drive the updating of the knowledge unit base. The initial fine-tuning of the pre-trained language model may also have issues such as comprehension bias or rule violations. User feedback can help the language model learn the correct answer logic.

[0164] In a specific embodiment, such as Figure 4 As shown, a pre-trained language model is used to generate answers based on the interaction units and constraints from the power rule base, including:

[0165] Input the interactive unit into the search engine to retrieve relevant enhanced knowledge units from the power materials knowledge unit database;

[0166] By combining relevant enhanced knowledge units, interaction units, and constraints in the power rule base, the context of the pre-trained language model is constructed;

[0167] The constructed context is input into the pre-trained language model, and the RLAIF mechanism is used to generate candidate answers. Combined with the compliance verification of the power rule base, the final answer is given.

[0168] Pre-trained language models can be chosen from general-purpose language models such as BERT, GPT, T5, and LLaMA. The pre-training process is as follows: general-purpose text corpora and power industry-specific corpora are used as training data. The power industry-specific corpora include power material technical manuals, industry standard documents, operation and maintenance procedures, fault handling records, and procurement contract texts. A masked language model task and a next-sentence prediction task are used. Initial training is first performed on the general-purpose text corpora, and then domain-adaptive training is performed on the power industry-specific corpora.

[0169] Understandably, the interaction unit refers to the user's query or instruction (such as "Which cables will be affected by a fault in transformer T1?"), which is the input question to the pre-trained language model. Figure 5 As shown, keywords such as "power materials" and "demand" are parsed and extracted from the interaction unit and converted into a query vector. Using a retrieval engine, the query vector and the enhanced knowledge unit vector are calculated using cosine similarity. Several enhanced knowledge units with the highest similarity are found in the power materials knowledge unit database. These enhanced knowledge units are then filtered using keywords such as "power materials" and "demand" from the interaction unit to obtain relevant enhanced knowledge units.

[0170] The original wording of the interactive units is retained as the question section; key information from relevant enhanced knowledge units is extracted and described in natural language as the knowledge section; the power rule base contains clauses from power safety work procedures, material management methods, and operation and maintenance specifications. Rules related to the user's question are extracted from the power rule base and embedded in natural language as the rules section. By integrating the question section, knowledge section, and rules section, the construction of the pre-trained language model context is completed.

[0171] The constructed context is input into a pre-trained language model, and the RLAIF mechanism constrains the response generation process. Specifically, the RLAIF mechanism, combined with constraints from the power rule base, increases the retention weight of output content that meets the constraints and decreases the retention weight of output content that violates the constraints during the response generation process. This weight adjustment constrains the output content. After candidate responses are generated, they are converted into computable logical expressions and input into the rule engine for matching against rules in the power rule base. If a candidate response meets all rules, the output is the final response; otherwise, it is regenerated. The final response includes information such as material ID, associated materials, and decision suggestions, conforming to the reading habits of operations and maintenance personnel.

[0172] In a specific embodiment, such as Figure 6 As shown, in response to user feedback, the power material knowledge unit base is updated, and the pre-trained language model is iteratively optimized until convergence, including:

[0173] Analyze user responses, extract feedback values, and locate relevant augmented knowledge units;

[0174] The validity of the feedback value is verified using pre-defined verification rules. Based on the verified feedback value, the knowledge unit is modified and enhanced, and the power material knowledge unit database is updated synchronously.

[0175] A training set is built based on user feedback, and the pre-trained language model is trained on the training set to fine-tune the model parameters.

[0176] Use the validation set to verify the model performance, and obtain the model's answer accuracy on the validation set, the proportion of the model's generated answers that conform to the power rule base, and the error rate reported by users.

[0177] Repeat the training and validation process of the pre-trained language model until the termination condition is met.

[0178] User feedback refers to the feedback users provide to the language model's generated answers. This can include corrections to errors, supplementary information for incomplete answers, and suggestions for new rules. The user feedback is parsed to extract the relevant power materials and corresponding feedback values ​​(i.e., the specific corrections made by the user to errors or missing information in the model's answer). Relevant enhanced knowledge units are then retrieved from the power material knowledge unit base. The relationships in the feedback are checked to ensure they conform to power regulations or historical maintenance records. The existence of the materials mentioned in the feedback in the power material knowledge unit base is verified, as is the consistency between the feedback values ​​and other data in the base. Based on the verified feedback values, the corresponding enhanced knowledge units are corrected (including corrections to basic attributes, relationships, etc.). The corrected enhanced knowledge units replace the original enhanced knowledge units, and the index is updated, thus synchronously updating the power material knowledge unit base.

[0179] Based on user feedback, user questions and correct answers are converted into question-answer pairs. These pairs are added to the existing training set as a new training set. The model is trained on this training set, with fine-tuning parameters such as learning rate, batch size, and training steps. For the fine-tuned model, the accuracy of the generated answers is statistically analyzed on the validation set. A rule engine is used to check whether the generated answers comply with power regulations and calculate the compliance rate. The proportion of errors pointed out by users in the model's generated answers is calculated as a model evaluation metric. When all metrics meet expectations, the model is considered converged, and iteration stops. If any metric fails to meet expectations, the iteration process is repeated until all metrics meet expectations or the number of iterations reaches a preset limit. By converting user feedback into correction signals for the power material knowledge unit base and training data for the language model, the model learns from user feedback, providing more accurate and compliant answers, thus achieving intelligent power material management.

[0180] Secondly, the present invention provides an apparatus for constructing interactive power material knowledge units, employing the aforementioned method for constructing interactive power material knowledge units, specifically including:

[0181] The data acquisition module is used to acquire power material data and quantify it to obtain raw data units;

[0182] The knowledge construction module is used to perform dynamic weighted aggregation based on the original data units, taking into account the reliability factor and the time series factor, to give the basic knowledge units; based on the basic knowledge units, it obtains the spatial, topological and temporal correlations between power materials, and gives the comprehensive correlations in combination with the fault propagation mechanism; it integrates the basic knowledge units and the comprehensive correlations to generate enhanced knowledge units, and builds a power material knowledge unit library.

[0183] The interactive question-and-answer module uses a pre-trained language model to generate answers based on interactive units and constraints from the power rule base. In response to user feedback, it updates the power material knowledge unit base and iteratively optimizes the pre-trained language model until convergence, thereby realizing the construction of interactive power material knowledge units.

[0184] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for constructing an interactive power asset knowledge unit, characterized in that, Specifically comprising the following steps: Obtain power material data and vectorize to obtain original data units; Based on the original data units, combine the trust factors and time sequence factors to perform dynamic weighted aggregation to give basic knowledge units; Based on the basic knowledge units, obtain the spatial correlation, topological correlation and time correlation between the power materials, and combine the fault conduction mechanism to give the comprehensive correlation; Fuse the basic knowledge units and the comprehensive correlation to generate enhanced knowledge units and build a power material knowledge unit library; Use a pre-trained language model to generate answers according to the interaction units, combine the power rule library constraints, update the power material knowledge unit library in response to user feedback, and iteratively optimize the pre-trained language model to convergence to realize the construction of interactive power material knowledge units; Wherein, the trust factors include first trust factors and second trust factors, based on the original data units, combining the trust factors and time sequence factors, performing dynamic weighted aggregation to give basic knowledge units, specifically including: obtaining the business stage, material state and environmental factors of power materials, weighting the business stage, material state and environmental factors with pre-set weight coefficients to give the first trust factors of power material data; Obtain the rated parameters and historical baseline of power materials, and perform verification to give the second trust factors of power material data; Based on the life perception attenuation mechanism, obtain the time sequence factors of power material data; Fuse the first trust factors, the second trust factors and the time sequence factors to obtain the comprehensive factors, and weight aggregate with the original data units to give the basic knowledge units; Obtain the rated parameters and historical baseline of power materials, and perform verification to give the second trust factors of power material data, specifically including: selecting key operating parameters according to the type of power materials, and determining the parameter rated value and parameter rated range; For each key operating parameter, check whether the current value meets the parameter rated range, if not, filter the data, if yes, perform historical deviation verification, obtain the deviation of the current value of the key operating parameter and the historical mean value, combine the deviation adjustment coefficient and the parameter rated value to give the second trust factors; Based on the life perception attenuation mechanism, obtain the time sequence factors of power material data, specifically including: determining the basic time sequence attenuation coefficient according to the type of power materials; Based on the basic time sequence attenuation coefficient, combine the running time to give the life perception attenuation coefficient; Based on the life perception attenuation coefficient, combine the current time and the business key timestamp of the power material data to give the time sequence factors.

2. The method of claim 1, wherein the interactive power asset knowledge unit is constructed by: The basic knowledge units are specifically represented as: ； In the formula, K t is the basic knowledge unit at time t, i is the data category, N is the number of data categories, D i (t) is the original data unit of the i-th data category, is the first trust factor of the i-th data category, is the second trust factor of the i-th data category, is the time sequence factor of the i-th data category, represents the comprehensive factor, and t is the current time. The first trust factors are specifically represented as: ； In the formula, is a first trusted factor of the i-th type of data, is a service stage, is a material state, is an environmental factor, is a weight coefficient; The second trust factors are specifically represented as: ； In the formula, is the second trusted factor of the i-th type of data, P is a set of key operation parameters of power materials, p is a key operation parameter of power materials, is the current value of the key operation parameter p of power materials, is the historical operation average value of the key operation parameter p of power materials, is the lower limit of the rated range of the key operation parameter p of power materials, is the upper limit of the rated range of the key operation parameter p of power materials, k is a deviation adjustment coefficient, is the rated value of the key operation parameter p of power materials, t is the current time; The time sequence factors are specifically represented as: ； wherein is a timing factor for the i-th class of data, is a base decay coefficient, l is a lifetime weight, and Age is a running time, denotes a lifetime-aware decay coefficient, T i is a service critical timestamp for the i-th class of data, and t is a current time.

3. The method of claim 1, wherein the interactive power asset knowledge unit is constructed by: Based on the basic knowledge units, obtain the spatial correlation, topological correlation and time correlation between the power materials, and combine the fault conduction mechanism to give the comprehensive correlation, specifically including: Obtain the spatial range of each power material, calculate the overlapping area and total area of the spatial range to obtain the spatial intersection ratio, and set the spatial weight according to the environmental factors, combine the maximum spatial range to give the spatial correlation; Based on the position of each power material in the topology structure, the topological correlation degree between the power materials is matched according to the topological relationship library and the predefined topological classification rules, and the topological weight is set according to the business stage to give the topological correlation; Obtain the business key timestamp of each power material, calculate the time correlation degree, and set the time weight to give the time correlation; Obtain the fault type, fault severity, and fault conduction distance of the fault power material to the target power material, combine the distance scale, calculate the fault distance attenuation, and set the fault conduction gain according to the material state to give the fault conduction correlation; Fuse spatial correlation, topological correlation, time correlation and fault conduction correlation to give comprehensive correlation.

4. The method of claim 3, wherein the interactive power asset knowledge unit is constructed by: The comprehensive correlation is specifically represented as: ； In the formula, represents the power material E a and the power material E b comprehensive association, respectively, spatial weight, topological weight, time weight, is the fault conduction gain, is the power material E a and the power material E b spatial intersection ratio, S a is the spatial range of the power material E a , S b is the spatial range of the power material E b , represents the maximum value of the spatial range of the power material E a and the power material E b , is the topological correlation degree of the power material E a and the power material E b , is the time correlation coefficient, T a , T b respectively, the business critical time stamp of the power material E a , the power material E b , Sev is the fault severity, Dis is the fault conduction distance, Scl is the distance scale, t is the current time, Env(t) is the environmental factor, Phs(t) is the business stage, Stat(t) is the material state. The spatial intersection ratio is specifically represented as: ； wherein represents the spatial intersection of the electric power asset E a and the electric power asset E b , represents the spatial intersection of the electric power asset E a and the electric power asset E b , represents the spatial union of the electric power asset E a and the electric power asset E b .

5. The method of claim 1, wherein the interactive power asset knowledge unit is constructed by: Fuse basic knowledge units and comprehensive correlation to generate enhanced knowledge units, and build a power material knowledge unit library, specifically including: Using attention mechanism, fuse basic knowledge units and comprehensive correlation, and combine power rules verification to generate enhanced knowledge units; Associate metadata to each enhanced knowledge unit, build an index, set a dynamic update mechanism, and form a power material knowledge unit library.

6. The method of claim 1, wherein the interactive power asset knowledge unit is constructed by: Using a pre-trained language model, generate an answer according to the interaction unit combined with the power rule library constraints, specifically including: Input the interaction unit into the retrieval engine to retrieve relevant enhanced knowledge units from the power material knowledge unit library; Combine the relevant enhanced knowledge units, interaction units and constraint conditions in the power rule library to construct the context of the pre-trained language model; Input the constructed context into the pre-trained language model, use the RLAIF mechanism to generate candidate answers, and combine the power rule library compliance verification to give the final answer.

7. The method of claim 6, wherein the interactive power asset knowledge unit is constructed by: In response to the user's answer feedback, update the power material knowledge unit library and iteratively optimize the pre-trained language model to convergence, specifically including: Parse the user's answer feedback, extract the feedback value, and locate the relevant enhanced knowledge unit; Use the pre-set verification rules to verify the legality of the feedback value, modify the enhanced knowledge unit according to the feedback value that passes the verification, and update the power material knowledge unit library synchronously; According to the user's answer feedback, build a training set, and train the pre-trained language model on the training set to fine-tune the model parameters; Use the validation set to evaluate the model performance, get the answer accuracy of the model on the validation set, the proportion of the answers generated by the model that comply with the power rule library, and the error rate of the user feedback; Repeat the training and verification process of the pre-trained language model until the termination condition is reached.

8. A device for constructing interactive power material knowledge units, characterized in that, The construction method of the interactive power material knowledge unit according to any one of claims 1-7, specifically including: A data acquisition module for acquiring power material data and vectorizing to obtain original data units; The knowledge construction module is configured to perform dynamic weighted aggregation based on the original data unit, combine the trusted factor and the time sequence factor, and give a basic knowledge unit; based on the basic knowledge unit, obtain spatial correlation, topological correlation and time correlation between the power materials, and combine a fault conduction mechanism to give a comprehensive correlation; fuse the basic knowledge unit and the comprehensive correlation to generate an enhanced knowledge unit, and construct a power material knowledge unit library; wherein the trusted factor includes a first trusted factor and a second trusted factor, based on the original data unit, combining the trusted factor and the time sequence factor, performing dynamic weighted aggregation to give the basic knowledge unit, specifically including: obtaining the business stage, the material state and the environmental factor of the power material, weighting the business stage, the material state and the environmental factor with a pre-set weight coefficient to give the first trusted factor of the power material data; obtaining the rated parameter and the historical baseline of the power material, performing verification to give the second trusted factor of the power material data; based on a life perception attenuation mechanism, obtaining the time sequence factor of the power material data; fusing the first trusted factor, the second trusted factor and the time sequence factor to obtain a comprehensive factor, and weighting and aggregating with the original data unit to give the basic knowledge unit; obtaining the rated parameter and the historical baseline of the power material, performing verification to give the second trusted factor of the power material data, specifically including: selecting key operating parameters according to the type of the power material, and determining the parameter rated value and the parameter rated range; for each key operating parameter, checking whether the current value meets the parameter rated range, if not, filtering the data, if yes, performing historical deviation verification, obtaining the deviation of the current value of the key operating parameter and the historical mean value, combining the deviation adjustment coefficient and the parameter rated value to give the second trusted factor; based on the life perception attenuation mechanism, obtaining the time sequence factor of the power material data, specifically including: determining the basic time sequence attenuation coefficient according to the type of the power material; based on the basic time sequence attenuation coefficient, combining the operating time to give the life perception attenuation coefficient; based on the life perception attenuation coefficient, combining the current time and the business key timestamp of the power material data to give the time sequence factor; The interactive question and answer module is configured to use a pre-trained language model to generate an answer according to an interaction unit and combining a power rule library constraint, update the power material knowledge unit library in response to user feedback on the answer, and iteratively optimize the pre-trained language model to convergence to realize interactive construction of the power material knowledge unit.

Images