A power industry policy information identification method and system

Abstract

The application discloses a kind of electric power industry policy information identification method and system, method includes: according to the policy text to be identified to the electric power industry feature library of multiple levels and gradually feature matching, obtain electric power industry feature matching degree, and the text of matching degree greater than first threshold is marked as initial screening candidate text;According to the parameter extraction of initial screening candidate text according to electric power system physical model knowledge graph, generate projection subgraph, calculate electrical connected domain scale, determine electric power system physical coverage according to the ratio of connected domain scale and total node number of knowledge graph, and the text of coverage greater than second threshold is marked as re-screening candidate text;According to the time sequence correlation analysis of re-screening candidate text according to electric power industry time sequence database, and the text of time sequence consistency coefficient greater than third threshold is marked as electric power industry policy text and output. Through three-stage progressive screening, the accuracy and explainability of electric power industry policy information identification are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of power industry policy information processing technology, and in particular relates to a method and system for identifying power industry policy information. Background Technology

[0002] Policy texts come from a wide range of sources, covering various policy documents at the national, provincial, and local levels, and involving multiple aspects such as power generation, transmission, distribution, and sales, as well as dimensions such as electricity prices, carbon emissions, and renewable energy consumption. Power companies, investment institutions, and researchers need to accurately identify policy texts relevant to the power industry from this vast amount of policy information to support business decision-making and research analysis.

[0003] In existing technologies, policy text identification mainly employs keyword matching or text classification methods. However, these methods suffer from the following problems: First, relying solely on keyword matching easily generates a large number of false positives, incorrectly identifying policy texts containing electricity-related terms but unrelated to the power industry; second, existing text classification models lack an understanding of the physical characteristics of power systems, making it impossible to accurately identify the correlation between policy texts and power system operating parameters; and third, the lack of time-series verification using actual operational data after policy implementation makes it difficult to verify the accuracy of the identification results.

[0004] Therefore, there is an urgent need for a method that can accurately identify power industry policy information to solve the technical problems of low identification accuracy, lack of physical mechanism support, and lack of time-series verification in existing technologies. Summary of the Invention

[0005] This invention provides a method and system for identifying power industry policy information, which addresses the technical problems of low accuracy, lack of physical mechanism support, and lack of time-series verification in the existing technology for identifying power industry policy information.

[0006] In a first aspect, the present invention provides a method for identifying power industry policy information, comprising: At least one policy text to be identified is obtained, and the policy text to be identified is matched step by step according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified. Policy texts whose matching degree with power industry characteristics is greater than the first preset threshold are marked as preliminary candidate texts. Based on the preset knowledge graph of the power system physical model, parameters are extracted from the initial candidate texts to obtain the power system physical parameters; The power system physical parameters are mapped to the power system physical model knowledge graph to generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph. The projection subgraph contains at least one knowledge graph node and a corresponding parameter constraint value. Calculate the electrical connectivity domain size of the projected subgraph in the knowledge graph of the power system physical model, where the electrical connectivity domain size is the total number of nodes that can be connected through electrical topology connections in the projected subgraph; The power system physical coverage of the initial candidate texts is determined based on the ratio of the size of the electrical connectivity domain to the total number of nodes in the knowledge graph of the power system physical model. Initially screened candidate texts with a power system physical coverage greater than the second preset threshold are marked as secondary screened candidate texts. Based on the preset power industry time series database, time series correlation analysis is performed on the candidate texts for re-screening, power quantitative indicators involved in the candidate texts are extracted, and the power quantitative indicators are compared with the actual operating data of the corresponding time period in the power industry time series database to obtain the time series consistency coefficient. Candidate texts with a temporal consistency coefficient greater than the third preset threshold are marked as power industry policy texts, and the marked power industry policy texts are output as power industry policy information recognition results.

[0007] Secondly, the present invention provides a power industry policy information identification system, comprising: The acquisition module is configured to acquire at least one policy text to be identified, and perform step-by-step feature matching on the policy text to be identified according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified; The first marking module is configured to mark the policy text to be identified that has a power industry feature matching degree greater than a first preset threshold as a preliminary candidate text; The extraction module is configured to extract parameters from the initially screened candidate texts based on a preset knowledge graph of the power system physical model to obtain power system physical parameters. The generation module is configured to map the power system physical parameters to the power system physical model knowledge graph, and generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph, wherein the projection subgraph contains at least one knowledge graph node and a corresponding parameter constraint value. The calculation module is configured to calculate the size of the electrical connectivity domain of the projected subgraph in the knowledge graph of the power system physical model, wherein the size of the electrical connectivity domain is the total number of nodes that can be connected by electrical topology connections in the projected subgraph; The determination module is configured to determine the power system physical coverage of the initial candidate texts based on the ratio of the size of the electrical connectivity domain to the total number of nodes in the power system physical model knowledge graph. The second marking module is configured to mark the initial screening candidate texts with a power system physical coverage greater than a second preset threshold as secondary screening candidate texts. The analysis module is configured to perform time-series correlation analysis on the candidate texts for re-screening based on a preset power industry time-series database, extract the power quantitative indicators involved in the candidate texts for re-screening, and compare the power quantitative indicators with the actual operating data of the corresponding time period in the power industry time-series database to obtain the time-series consistency coefficient. The output module is configured to mark the candidate texts with a temporal consistency coefficient greater than a third preset threshold as power industry policy texts, and output the marked power industry policy texts as power industry policy information recognition results.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the power industry policy information identification method according to any embodiment of the present invention.

[0009] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the power industry policy information identification method according to any embodiment of the present invention.

[0010] The power industry policy information identification method and system of this application achieves initial screening by performing step-by-step feature matching on the policy texts to be identified through a multi-level power industry feature database; it then performs parameter extraction, projection subgraph construction, and electrical connectivity domain size calculation on the candidate texts selected through the power system physical model knowledge graph, using the power system physical coverage metric to quantify the correlation between the policy texts and the power system, thus achieving secondary screening; finally, it performs time-series correlation analysis through a power industry time-series database, using the time-series consistency coefficient to verify the consistency between the policy texts and actual operating data, thus achieving final screening. This significantly improves the accuracy and reliability of power industry policy information identification and solves the technical problems of low identification accuracy, lack of physical mechanism support, and lack of time-series verification in existing technologies. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a method for identifying power industry policy information according to an embodiment of the present invention; Figure 2This is a structural block diagram of a power industry policy information identification system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Please see Figure 1 The diagram shows a flowchart of a method for identifying power industry policy information according to this application.

[0015] like Figure 1 As shown, the method for identifying power industry policy information specifically includes the following steps: Step S101: Obtain at least one policy text to be identified, and perform step-by-step feature matching on the policy text to be identified according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified.

[0016] In this step, the multi-level power industry feature library includes a primary feature library, a secondary feature library, and a tertiary feature library. The primary feature library contains core power industry terms, the secondary feature library contains power equipment terms and operating parameter terms, and the tertiary feature library contains power market terms and regulatory terms.

[0017] The policy text to be identified is segmented and part-of-speech tagged to obtain a text word sequence; Each word in the text word sequence is matched with the core terms of the power industry in the first-level feature library. The number of successfully matched words is counted, the first-level matching score is calculated, and the position of the successfully matched word in the text is marked as the first-level matching position. Using the primary matching position as the center, extend the preset window length forward and backward, and extract the text within the extended window as the primary matching context. The words in the first-level matching context are matched with the power equipment terms and operating parameter terms in the second-level feature library. The number of successfully matched words is counted, and the second-level matching score is calculated. Using the location of the word that successfully matches the secondary feature library as the center, expand the second preset window length forward and backward, and extract the text within the expanded window as the secondary matching context. The words in the secondary matching context are matched with the electricity market terms and regulatory terms in the tertiary feature library. The number of successfully matched words is counted, and the tertiary matching score is calculated. The matching scores of Level 1, Level 2, and Level 3 are weighted and summed to obtain the matching degree of the power industry characteristics. The weight coefficient of Level 1 matching score is greater than that of Level 2 matching score, and the weight coefficient of Level 2 matching score is greater than that of Level 3 matching score.

[0018] In one specific embodiment, the multi-level power industry feature database includes a primary feature database, a secondary feature database, and a tertiary feature database. Wherein: The primary feature library contains core terms related to the power industry, used to identify whether policy texts fall under the scope of the power industry. Specifically, it includes: Power generation-related terms include: "power generation enterprise", "generator set", "coal-fired power plant", "gas-fired power plant", "nuclear power plant", "hydropower station", "wind farm", "photovoltaic power station", "biomass power generation", "distributed power generation", and "self-owned power plant".

[0019] Power grid terminology includes: "power grid", "transmission line", "substation", "distribution network", "transmission and distribution price", "power grid dispatch", "power trading center", and "inter-provincial and inter-regional power transmission".

[0020] Market-related terms: "electricity market", "electricity trading", "electricity sales company", "electricity user", "direct transaction", "bilateral negotiation", "centralized bidding", "electricity spot market", etc.

[0021] Regulatory terms include: "electricity regulation", "National Energy Administration", "National Development and Reform Commission", "electricity pricing policy", "electricity system reform", "electricity law", and "fair and open access to the power grid".

[0022] The secondary feature library contains power equipment terminology and operating parameter terminology, used to identify specific power equipment and operating parameters mentioned in policy texts. Specifically, it includes: Equipment terminology includes: generator, steam turbine, boiler, transformer, circuit breaker, disconnector, reactor, capacitor, converter valve, inverter, photovoltaic module, wind turbine generator set, etc.

[0023] Operating parameter terms include: "active power", "reactive power", "voltage", "current", "frequency", "power factor", "load rate", "output", "generation", "electricity consumption", "line loss", "plant power consumption rate", and "load rate".

[0024] The Level 3 feature library contains electricity market terminology and regulatory terminology, used to identify market mechanisms and regulatory measures mentioned in policy texts. Specifically, it includes: Electricity market terminology includes: “grid connection price”, “sales price”, “peak-valley price”, “tiered pricing”, “transmission and distribution price”, “ancillary services”, “capacity market”, “electricity trading”, “green certificate trading”, “carbon emission trading”, and “renewable energy consumption responsibility weight”.

[0025] Regulatory terms: “grid connection and dispatch agreement”, “power purchase and sale contract”, “electricity business license”, “generation rights trading”, “alternative generation”, “emergency dispatch”, “orderly electricity consumption”, “demand response”, etc.

[0026] The specific implementation process of stepwise feature matching is as follows: The policy text to be identified is preprocessed, including punctuation removal and stop word filtering. Then, deep learning-based natural language processing tools (such as BERT or LSTM-CRF models) are used for word segmentation and part-of-speech tagging to obtain a text word sequence. For example, the input text is "A certain province's Development and Reform Commission issued the 'Notice on Further Regulating the Power Generation Dispatch of Coal-fired Units,' requiring that coal-fired units must ensure that their active power output does not exceed 95% of their rated capacity during peak load periods." After word segmentation and part-of-speech tagging, the word sequence obtained is: ["certain province", "development and reform commission", "issued", "about", "further", "regulate", "coal-fired units", "power generation", "dispatch", "of", "notice", "requirement", "coal-fired units", "during", "peak load", "must", "ensure", "active power output", "not exceeding", "rated capacity", "of", "95%"].

[0027] Perform a string-by-string match between each word in the text word sequence and the core terms of the power industry in the primary feature library. Iterate through each word; if the word exists in the primary feature library, a match is successful. Count the number of successfully matched words and calculate the primary matching score. Level 1 matching score = (number of successfully matched words / total number of words in the text) × Level 1 weight base; The primary weight base is preset to 1.0. Simultaneously, the position index of each successfully matched word in the text is recorded as the primary matching position.

[0028] Taking the above text as an example, the core terms that were successfully matched include: "coal-fired unit" (matched twice), "power generation", and "dispatch". The number of successfully matched words is 4, and the total number of words in the text is 22. The first-level matching score is 4 / 22×1.0≈0.182. The first-level matching positions are: the 7th word "coal-fired unit", the 8th word "power generation", the 9th word "dispatch", and the 13th word "coal-fired unit".

[0029] Centered on each primary matching position, expand the window forward and backward by a preset window length (e.g., 5 words), and extract the text within the expanded window as the primary matching context. For multiple primary matching positions, take their union to avoid duplication.

[0030] Taking the above text as an example, the first primary matching position is the 7th word "coal-fired unit". Expanding forward by 5 words, we get the 2nd-6th words ("about", "further", "standard", "coal-fired unit", "power generation"), and expanding backward by 5 words, we get the 8th-12th words ("power generation", "dispatch", "of", "notification", "requirement"). After expanding all primary matching positions, we take the union to obtain the primary matching context.

[0031] The system matches each word in the first-level matching context with power equipment terms and operating parameter terms in the second-level feature library. Fuzzy matching is used, allowing for the matching of synonyms and near-synonyms; for example, "output" and "active power output" are considered a match.

[0032] The formula for calculating the second-level matching score is: Second-level matching score = (number of successfully matched words / total number of words in the first-level matching context) × second-level weight base; The secondary weight base is preset to 0.6. Simultaneously, the location of successfully matched words is recorded.

[0033] Taking the above text as an example, the successful matching of secondary feature words in the first-level matching context includes: "coal-fired unit" (equipment term), "active power output" (operating parameter), and "rated capacity" (operating parameter). The number of successfully matched words is 3, the total number of words in the first-level matching context is approximately 15, and the secondary matching score is 3 / 15 × 0.6 = 0.12.

[0034] Centered on the location of a word that successfully matches the secondary feature library, expand the second preset window length (e.g., set to 8 words) both forward and backward, and extract the text within the expanded window as the secondary matching context. Similarly, take the union of multiple matching locations.

[0035] The words in the secondary matching context are matched with electricity market terms and regulatory terms in the tertiary feature library. Fuzzy matching is also used for the matching method.

[0036] The formula for calculating the score of a level 3 match is: Level 3 matching score = (number of successfully matched words / total number of words in the context of Level 2 matching) × Level 3 weight base; The base weight for the three levels is preset to 0.3.

[0037] Taking the above text as an example, the secondary matching context may contain words such as "scheduling", "peak load", and "ensure". If it matches the tertiary feature library successfully, the corresponding score is calculated.

[0038] The matching scores of Level 1, Level 2, and Level 3 are weighted and summed to obtain the power industry feature matching degree. The weight coefficients satisfy the following order: Level 1 matching score weight coefficient > Level 2 matching score weight coefficient > Level 3 matching score weight coefficient. Example weight settings: The weighting coefficient for the first-level matching score is 0.6. Second-level matching score weighting coefficient: 0.3; The weighting coefficient for the third-level matching score is 0.1. The power industry feature matching degree = 0.182 × 0.6 + 0.12 × 0.3 + third-level matching score × 0.1.

[0039] In summary, by constructing a multi-level power industry feature database and employing a progressively expanding matching strategy, accurate identification of power industry policy texts was achieved. Compared with existing single-keyword matching methods, this approach offers the following technical advantages: The core terms in the first-level feature library can quickly identify whether policy texts involve the main body of the power industry; the equipment and operating parameter terms in the second-level feature library can verify whether the text involves the specific technical content of the power system; and the market and regulatory terms in the third-level feature library can confirm whether the text involves the policy mechanisms of the power industry. The three-level feature libraries are progressive and mutually verifying, effectively avoiding the misjudgment of policy texts containing general terms such as "power" but unrelated to the power industry (such as power equipment procurement contracts, power industry news, etc.) as power industry policies. By expanding the context window centered on the matching position, secondary feature matching can be performed within the context of primary feature words, and tertiary feature matching can be performed within the context of secondary feature words, forming a progressive chain of "core term location → context expansion → related term matching", which ensures the semantic coherence and consistency of the matching results. Even if the core terms in the policy text to be identified appear infrequently, relevant equipment parameter terms or market regulation terms can still be captured through context expansion, thereby obtaining a high degree of comprehensive matching and avoiding missed judgments due to the scarcity of core terms. At the same time, the feature library can be updated and maintained according to the development dynamics of the power industry to keep the identification method timely.

[0040] Step S102: Mark the policy texts to be identified that have a power industry feature matching degree greater than the first preset threshold as preliminary candidate texts.

[0041] In this step, Step S103: Based on the preset knowledge graph of the power system physical model, extract parameters from the initial candidate texts to obtain the power system physical parameters.

[0042] In this step, the knowledge graph of the power system physical model includes generator nodes, transmission line nodes, load nodes and bus nodes, as well as the electrical topology connections and operating parameter constraints between nodes. Each node is associated with a node identifier, node name and constrainable parameter template. Based on the pre-defined knowledge graph of the power system physical model, parameters are extracted from the initially screened candidate texts to obtain the power system physical parameters, including: Dependency parsing is performed on the initial candidate texts to construct a dependency parsing tree. Based on the dependency parsing tree, statements containing constraint semantics in the text are identified to obtain constraint statements. The constraint semantics are identified through the dependency relationship between constraint keywords and predicate verbs. Power system entity identification is performed on the constraint statement to extract the power system entities in the constraint statement, and the entity name, entity type and entity position of each power system entity in the constraint statement are recorded. The power system entities include generators, transmission lines, loads and buses, and the entity position is the start character position and end character position of the power system entity in the constraint statement. Based on the extracted entity names and entity types of the power system entities, node retrieval is performed in the power system physical model knowledge graph to obtain entity identifiers corresponding to the power system entities, specifically including: Using the entity name of a power system entity as the search keyword, the target node matching the entity type is found in the knowledge graph of the power system physical model. When a target node that exactly matches the entity name is found, the node identifier of the target node is used as the entity identifier of the power system entity. When multiple target nodes that partially match the entity name are found, the string similarity between the entity name and each candidate node name is calculated, and the node identifier of the candidate node with the highest similarity is selected as the entity identifier. When no matching target node is found, the entity identifier of the power system entity is marked as empty; Obtain the target constraint parameter template associated with the retrieved target node; Numerical parameters corresponding to the power system entity are extracted from the constraint statement, and the numerical parameters are matched with the parameter types in the constrainable parameter template to determine the parameter type corresponding to the numerical parameters. The numerical parameters include power values, voltage values, current values, emission values, and energy values. The parameter type corresponding to the numerical parameters is the specific type of constraint imposed on the numerical parameters at the target node, including one of the following: active power output limit, reactive power output limit, emission limit, power flow limit, energy consumption limit, power factor limit, voltage limit, and voltage lower limit. The entity identifier, parameter type, and numerical parameter are combined to form the physical parameters of the power system.

[0043] In one specific embodiment, a knowledge graph of the power system physical model is first constructed. This knowledge graph is based on the actual physical structure of the power system and includes generator nodes, transmission line nodes, load nodes, and bus nodes, as well as the electrical topology connections and operating parameter constraints between the nodes. Each node is associated with a node identifier (e.g., "GEN_001", "LINE_012", "LOAD_005", "BUS_003") and a node name (e.g., "..."). "Haimen Power Plant Unit #1", "500kV Sunan Line", The templates for "Industrial Park Load Center", "220kV Mudu Substation Bus", and constrainable parameters define the types of policy constraints that are allowed for the node. For example, the constrainable parameters for generator nodes include upper limits for active power output, upper limits for reactive power output, and emission limits; the constrainable parameters for transmission line nodes include upper limits for power flow; the constrainable parameters for load nodes include upper limits for electricity consumption and lower limits for power factor; and the constrainable parameters for bus nodes include upper limits for voltage and lower limits for voltage.

[0044] The specific process for extracting physical parameters of a power system is as follows: Step 1: Dependency parsing and constraint statement identification Dependency parsing was performed on the initial candidate texts to construct dependency parsing trees. Mature Chinese dependency parsing tools (such as LTP and Stanford CoreNLP) were used to parse the texts and obtain the part of speech and dependency relations (such as subject-verb relation SBV, verb-object relation VOB, attributive-head relation ATT, etc.) for each word.

[0045] Within the dependency syntax tree, statements containing constraint semantics are identified. Constraint semantics are identified through the dependency relationship between constraint keywords and predicate verbs. Preset constraint keywords include: "must not exceed," "must not be lower than," "upper limit is," "lower limit is," "controlled within," "should satisfy," "must not be higher than," "must not be lower than," "must be within," "limited to," etc. The specific identification method is as follows: traverse the dependency syntax tree, search for predicate verbs (such as "exceed," "lower than," "controlled," "satisfy"), and if the predicate verb has a noun-head relationship or adverbial-head relationship with the constraint keyword, then the clause or complete sentence containing the predicate verb is considered a constraint statement.

[0046] For example, the input text is: "The Development and Reform Commission of a certain province issued a notice requiring that the active power output of coal-fired power units shall not exceed 95% of the rated capacity during peak load periods." After dependency parsing, "shall not exceed" is identified as the constraint keyword, and the object it governs is "95%". The clause containing this clause is "active power output shall not exceed 95% of the rated capacity". This clause is then extracted as the constraint statement.

[0047] Step 2: Power System Entity Identification Power system entity identification is performed on the identified constraint statements. Specifically, this includes: Construct a power system entity dictionary, including a generator name dictionary (such as "Unit #1", "Turbine #2"), a transmission line name dictionary (such as "500kV line", "220kV line"), a load name dictionary (such as "industrial load", "residential load"), a busbar name dictionary (such as "busbar A", "220kV busbar section"), and various general power equipment names (such as "generator", "transformer", "line").

[0048] The constraint statement is segmented into words, and the segmentation results are matched against an entity dictionary. Matching methods support full word matching and partial matching (e.g., "fired coal unit" matches "unit").

[0049] When a match is successful, the word segment is marked as a power system entity, and the entity type (generator, transmission line, load, bus) is determined according to the matching dictionary type.

[0050] Record the starting and ending character positions (character offsets relative to the entire policy text) of the power system entity in the constraint statement as the entity position.

[0051] For example, in the constraint statement "Active power output shall not exceed 95% of rated capacity," "active power output" is not a device entity, therefore it does not match; "rated capacity" is an operating parameter, not a device entity, so it also does not match. Therefore, no power system entity is identified in this constraint statement. As another example, in the constraint statement "The transmission power of the Sunan Line shall not exceed 2000MW," "Sunan Line" matches the transmission line dictionary, the entity name is "Sunan Line," the entity type is transmission line, and the entity position is recorded as the start character offset and the end character offset.

[0052] Step 3: Node Retrieval and Entity Identifier Acquisition Based on the extracted entity names and types of power system entities, node retrieval is performed in the power system physical model knowledge graph to obtain the entity identifiers corresponding to the power system entities. Specifically, this includes: Using the entity name of a power system entity as the search keyword, the system searches the knowledge graph for nodes that match the entity type. For example, if the entity type is "transmission line", the system searches the knowledge graph for a set of transmission line nodes.

[0053] Exact match: When a target node that exactly matches the entity name is found (e.g., the entity name is "Sunan Line" and there is a transmission line node with the name "Sunan Line" in the knowledge graph), the node identifier of the target node (e.g., "LINE_012") is directly used as the entity identifier of the power system entity.

[0054] Partial Matching: When multiple target nodes that partially match the entity name are found (e.g., the entity name is "Haimen Power Plant #1 Unit", and there are two candidate nodes in the knowledge graph: "Haimen Power Plant #1 Unit" and "Haimen Power Plant #2 Unit"), a string similarity algorithm (such as edit distance or Jaccard similarity) is used to calculate the similarity between the entity name and each candidate node name. The node identifier of the candidate node with the highest similarity is selected as the entity identifier. Simultaneously, the similarity value is recorded as the matching similarity score for subsequent projection confidence.

[0055] Unmatched: When no matching target node is found, the entity identifier of the power system entity is marked as empty, and the reason for the unmatch is recorded (e.g., "No such device in the knowledge graph").

[0056] For example, consider a constraint statement: "The active power output of Unit #1 of Haimen Power Plant shall not exceed 500MW," with the entity name "Unit #1 of Haimen Power Plant" and the entity type being generator. Searching the knowledge graph for generator nodes, if a node named "Unit #1 of Haimen Power Plant" exists, it's a perfect match, and the entity is identified as "GEN_001". If the knowledge graph only contains "Unit #1 of Haimen Power Plant," then a string similarity calculation is performed. If a high similarity is found (e.g., 0.95), this node is selected as the target node, identified as "GEN_002," and the matching similarity is recorded as 0.95.

[0057] Step 4: Obtain the template for constrainable parameters Retrieve the associated constrainable parameter templates from the retrieved target nodes. Each node has a pre-set constrainable parameter template during knowledge graph construction. For example, the generator node template is "active power output limit", "reactive power output limit", and "emission limit"; the transmission line node template is "power flow limit"; the load node template is "electricity consumption limit" and "power factor limit"; and the bus node template is "voltage limit" and "voltage limit".

[0058] Step 5: Numerical parameter extraction and parameter type matching Extract numerical parameters corresponding to the power system entities from the constraint statements. Numerical parameters are usually represented as combinations of numbers and units, such as "500MW", "220kV", "1000A", "200 tons", and "10 million kWh". Use regular expressions or rule-based methods to extract the numerical values ​​and units, and perform unit normalization (e.g., convert "million kWh" to "kWh").

[0059] The extracted numerical parameters are matched with the parameter types in the constrainable parameter template to determine the corresponding parameter type. Matching methods include: The type of parameter can be inferred from the unit of the numerical parameter. For example, the units "MW" and "kW" usually correspond to power parameters (active power output, reactive power output, power flow limit); the units "kV" and "V" correspond to voltage parameters; the unit "A" corresponds to current parameters; the units "ton" and "kg" correspond to emission parameters; and the units "kWh" and "MWh" correspond to energy parameters.

[0060] The parameter type can be inferred from the keywords in the constraint statement. For example, "active power output" corresponds to the upper limit of active power output; "reactive power output" corresponds to the upper limit of reactive power output; "emissions" corresponds to the emission limit; "power flow" corresponds to the upper limit of power flow; "electricity consumption" corresponds to the upper limit of applied electricity; "power factor" corresponds to the lower limit of power factor; and "voltage" corresponds to the upper or lower limit of voltage.

[0061] Entity identifiers, parameter types, and numerical parameters are combined to form power system physical parameters, creating structured data. For example, for the constraint statement "The active power output of Haimen Power Plant Unit #1 shall not exceed 500MW", the extracted entity identifier is "GEN_001", the parameter type is "active power output limit", and the numerical parameter is "500MW", combined to form the power system physical parameter (GEN_001, active power output limit, 500MW). For the constraint statement "The transmission power of the Sunan Line shall not exceed 2000MW", the extracted entity identifier is "LINE_012", the parameter type is "power flow limit", and the numerical parameter is "2000MW", combined to form the power system physical parameter (LINE_012, power flow limit, 2000MW).

[0062] If a power system entity fails to retrieve the target node (entity identifier is empty), parameter extraction for that entity is skipped, and the reason for not extracting is recorded, without affecting the processing of other entities.

[0063] In summary, by combining knowledge graphs of power system physical models with natural language processing techniques, accurate extraction of power system physical parameters from policy texts was achieved, yielding the following technical benefits: This step transforms the semantic constraints in policy texts into physical parameters that can be recognized by the power system. Existing technologies can only identify keywords in policy texts and cannot understand the specific constraints of policy clauses on power equipment. This step identifies constraint statements through dependency parsing, locates constrained power equipment through entity recognition, and associates equipment names with physical nodes in the knowledge graph through node retrieval. Finally, the natural language "not exceeding 500MW" is transformed into the structured "active power output limit: 500MW", so that policy constraints can be directly applied to power system analysis and simulation. In actual policy texts, the same power equipment may have multiple names (such as "Unit #1", "Unit 1", "Unit 1"). This step, through partial matching and string similarity calculation, can map different expressions to unique nodes in the knowledge graph, thereby improving the recall rate of entity recognition. Different power equipment has different types of constrainable parameters. This step obtains the corresponding constrainable parameter template according to the node type, matches the numerical parameters with the parameter types in the template, and ensures that the extracted parameters have the correct physical meaning. For example, for generator nodes, the "voltage value" will not be misidentified as the upper limit of active power output, thus avoiding parameter type confusion. The extracted power system physical parameters (entity identifiers, parameter types, and numerical parameters) form the basis for constructing the projection subgraph, enabling policy constraints to be visualized as nodes in the knowledge graph, thus laying a data foundation for quantifying policy coverage.

[0064] Step S104: Map the power system physical parameters to the power system physical model knowledge graph, and generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph. The projection subgraph contains at least one knowledge graph node and the corresponding parameter constraint value.

[0065] In this step, target nodes corresponding to entity identifiers in the power system physical parameters are identified in the power system physical model knowledge graph, and the target nodes are added to the projection node set. Using nodes in the projection node set as seed nodes, a preset breadth-first search algorithm is used to expand outward along the electrical topology connection relationship, and all nodes visited during the expansion process are added to the projection subgraph node set. The projection subgraph is formed by the set of nodes in the projection subgraph and the electrical topology connection edges between the nodes.

[0066] In one specific embodiment, the power system physical parameters extracted in step S103 are mapped to a power system physical model knowledge graph, and a projection subgraph is generated by expanding seed nodes. The projection subgraph is used to characterize the scope of influence of policy constraints in the power grid, laying the foundation for subsequent calculation of the electrical connectivity domain size.

[0067] Traverse the list of power system physical parameters generated in step S103, and obtain the entity identifier for each power system physical parameter. The entity identifier is a unique identifier of the power system physical model knowledge graph node obtained through node retrieval in step S103 (such as "GEN_001" or "LINE_012").

[0068] In the knowledge graph of the power system physical model, hash indexes or B-tree indexes are used to quickly locate the target node corresponding to the entity identifier. Each node is stored as structured data in the knowledge graph, containing the following attributes: Node identifier: A unique ID for a node; Node name: The Chinese name of the equipment (e.g., "Huaneng Haimen Power Plant #1 Unit"); Node types: generator node, transmission line node, load node, or bus node; Electrical connection edge list: Identifiers of all adjacent nodes connected to this node and the type of the edge; Constraint Parameter Template: The types of constraint parameters that this node is allowed to accept; Current constraint parameter values: Stores the constraint parameter values ​​for this mapping (initially empty); For each power system physical parameter, perform the following operations: If the entity identifier is not empty (i.e., the target node was successfully matched in step S103), the target node is located in the knowledge graph based on the entity identifier.

[0069] The target node is added to the projection node set. The projection node set is a collection data structure that automatically removes duplicates to ensure that the same node is added only once.

[0070] The parameter types (such as "active power output limit") and numerical parameters (such as "500MW") in the power system physical parameters are used as constraint parameter values ​​for the target node and recorded in the node attributes. If the same node is constrained by multiple parameters (such as a generator having both "active power output limit" and "emission limit"), multiple key-value pairs are added to the constraint parameter value dictionary of that node.

[0071] If the entity identifier is empty (no target node was matched in step S103), the power system physical parameter is skipped, it is not added to the projection node set, and the reason for skipping is recorded (such as "the entity identifier is empty") for subsequent analysis reference.

[0072] For example, the power system physical parameters extracted in step S103 include: Parameter 1: (GEN_001, maximum active power output, 500MW); Parameter 2: (LINE_012, upper limit of current flow, 2000MW); Parameter 3: (Empty, maximum power consumption, 1 million kWh) ← Entity identifier is empty, skip; Based on parameter 1, locate the generator node "GEN_001", add it to the projection node set, and record "Active Power Output Limit: 500MW" in the node attributes. Based on parameter 2, locate the transmission line node "LINE_012", add it to the projection node set, and record "Power Flow Limit: 2000MW" in the node attributes. At this time, the projection node set contains two nodes: {GEN_001, LINE_012}.

[0073] Using nodes in the projection node set as seed nodes, a breadth-first search algorithm is used to expand outward along the electrical topology connections, and all nodes visited during the expansion process are added to the projection subgraph node set.

[0074] The specific implementation of breadth-first search is as follows: Initialization: Create a queue and add all nodes from the projection node set to the queue; create a collection visited to record visited nodes, initially adding all nodes from the projection node set to visited; create a collection projection_subgraph_nodes, initially containing all nodes from the projection node set.

[0075] Iterative expansion: When the queue is not empty, perform the following operations: Take a node current_node from the head of the queue; Obtain the list of electrical connection edges of current_node from the knowledge graph of the power system physical model, that is, all adjacent nodes directly connected to this node; Iterate through each neighbor node: If neighbor_node is not in the visited set, then add it to the visited set, add it to the projection_subgraph_nodes set, and add neighbor_node to the tail of the queue; Continue the loop until the queue is empty; Search termination condition: When the queue is empty, it means that all reachable nodes have been visited, and the breadth-first search ends.

[0076] For example, the above projection node set contains nodes GEN_001 and LINE_012. Assume the electrical topology connections in the knowledge graph are as follows: GEN_001 is connected to bus node BUS_001; BUS_001 is connected to LINE_012 and load node LOAD_005; LINE_012 is connected to bus node BUS_002; BUS_002 is connected to generator node GEN_003 and load node LOAD_008; Start expanding with GEN_001 as the seed node: Access GEN_001, obtain its adjacent node BUS_001, and add BUS_001 to the projection subgraph node set and queue.

[0077] Take BUS_001 from the queue, obtain its adjacent nodes GEN_001 (visited), LINE_012 (visited), and LOAD_005, and add LOAD_005 to the projection subgraph node set and queue.

[0078] LOAD_005 is retrieved from the queue. Its adjacent node BUS_001 has already been visited, and there are no new nodes.

[0079] Take LINE_012 from the queue (which already exists in the projection node set), get its adjacent nodes BUS_001 (which has been visited) and BUS_002, and add BUS_002 to the projection subgraph node set and queue.

[0080] Take BUS_002 from the queue, obtain its adjacent nodes LINE_012 (visited), GEN_003, and LOAD_008, and add GEN_003 and LOAD_008 to the projection subgraph node set and queue.

[0081] Continue expanding until the queue is empty.

[0082] The final projected subgraph node set contains: {GEN_001,LINE_012,BUS_001, LOAD_005,BUS_002,GEN_003,LOAD_008}, a total of 7 nodes.

[0083] The projection subgraph is formed by the set of nodes in the projection subgraph and the electrical topology connections between the nodes. The projection subgraph is a subgraph of the knowledge graph of the power system physical model and contains the following elements: Node set: The node set of the projected subgraph, which contains all nodes connected to the seed node through electrical topology connections.

[0084] Edge set: The electrical topological connection edge between any two nodes in the projected subgraph node set that exists in the knowledge graph. Edges retain their original attributes, including edge type (e.g., "generator-bus connection", "line-bus connection"), connection relationship (e.g., "direct connection", "connection via transformer"), etc.

[0085] The method for constructing the projection subgraph is as follows: traverse each node in the projection subgraph node set, obtain the list of electrical connection edges for that node from the knowledge graph, and if an adjacent node is also in the projection subgraph node set, add that edge to the edge set of the projection subgraph. This ultimately forms the complete subgraph structure.

[0086] The projected subgraph is stored in memory as a graph structure. Each node retains its original attributes (node ​​identifier, node name, node type) and the constraint parameter values ​​recorded in step one; each edge retains its original attributes (edge ​​type, connectivity). The projected subgraph can be exported as JSON format or used directly for subsequent calculations of the electrical connectivity domain size.

[0087] When the projection node set is empty (i.e., the entity identifiers of all power system physical parameters in step S103 are empty), the projection subgraph node set is also empty, and the projection subgraph is an empty graph. This situation indicates that although the initial candidate text contains power industry feature words, no specific power equipment matching the knowledge graph has been identified. The policy text may not have a direct physical connection with the power system, and the subsequent electrical connectivity domain size is 0, and the power system physical coverage is 0.

[0088] When an isolated node (without any electrical connection edges in the knowledge graph) exists in the projection node set, the breadth-first search expansion process only adds the isolated node itself to the projection subgraph node set, without expanding other nodes. This indicates that policy constraints apply to equipment without electrical connection to the power grid (such as decommissioned equipment), and its impact is limited to that node itself.

[0089] In summary, by mapping the physical parameters of the power system to a knowledge graph and performing topology expansion to generate a projected subgraph, the following technical effects are achieved: Existing technologies can only identify keywords in policy texts, failing to understand the actual objects and scope of impact of policy constraints on the power system. This step uses entity identifiers to locate knowledge graph nodes, transforming abstract policy constraints (such as "generator active power output shall not exceed 500MW") into constraint parameter values ​​on specific grid equipment, thus giving the policy content a clear semantic anchor in the physical space of the power system. This mapping mechanism from "text space" to "physical space" provides a physically interpretable basis for subsequent quantitative assessment of policy impact. By employing a breadth-first search algorithm to extend the influence of policy constraints outward along the electrical topology connections, the scope of policy constraints is expanded from individual devices to the entire electrical connectivity domain. This reveals the cascading propagation effect of policy constraints. For example, for a power flow upper limit constraint on a transmission line, topology expansion can identify all generators, loads, and buses affected by the power flow restriction of that line, providing a topological basis for a comprehensive assessment of the policy impact.

[0090] The projected subgraph clearly records the set of power grid nodes affected by policy constraints and the connection relationships between the nodes. It is the basis for subsequent calculation of the electrical connectivity domain size. The size of the node set in the projected subgraph directly reflects the breadth of the policy constraints in the power grid, while the constraint parameter values ​​of the nodes in the projected subgraph reflect the strength of the policy constraints. This structured data form makes the quantitative assessment of the policy impact computable and repeatable. The projected subgraphs are stored in a graph structure and can be easily exported to standard graph formats such as GraphML and GEXF for further analysis or visualization by external systems. For example, the projected subgraphs can be overlaid on the power grid geographic wiring diagram to generate a policy impact heat map, which can intuitively show the coverage area and impact intensity of policy constraints and enhance the interpretability of policy analysis. When processing multiple policy texts, projection subgraphs for each policy text can be generated separately, and then the projection subgraphs can be merged to generate a composite projection subgraph. The composite projection subgraph can intuitively show the comprehensive impact range of multiple policies after they are superimposed, providing data support for the analysis of the synergy of multiple policies.

[0091] Step S105: Calculate the electrical connectivity domain size of the projected subgraph in the knowledge graph of the power system physical model. The electrical connectivity domain size is the total number of nodes that can be connected through electrical topology connections in the projected subgraph.

[0092] In this step, the node identifiers of each node in the projection subgraph node set are obtained in the knowledge graph of the power system physical model; Using nodes in the projection subgraph node set as initial activated nodes, a preset depth-first search algorithm is used to traverse the knowledge graph of the power system physical model and explore all reachable nodes through electrical topology connections. During the traversal, the order in which nodes are visited is recorded, as well as the path length from the initial node to a given node. All visited nodes are assigned to the electrical connectivity node set, and the size of the electrical connectivity node set is determined as the size of the first connectivity domain. Obtain high-confidence nodes in the projection subgraph node set whose projection confidence is greater than a preset confidence threshold, and use the high-confidence nodes as seed nodes to perform a depth-first search again to obtain the size of the second connected component. The weighted average of the sizes of the first and second connected components is determined as the size of the electrical connected component.

[0093] In one specific embodiment, a projection subgraph node set is extracted from the projection subgraph generated in step S104. The projection subgraph node set is a collection of all nodes connected to the seed node via electrical topology connections, and each node has a unique node identifier in the knowledge graph. The projection subgraph node set is traversed to obtain the node identifier for each node, forming an initial list of activated nodes.

[0094] For example, the projection subgraph node set generated in step S104 contains the following node identifiers: {GEN_001,LINE_012,BUS_001,LOAD_005,BUS_002,GEN_003,LOAD_008}, a total of 7 nodes.

[0095] Using all nodes in the projection subgraph node set as initial activated nodes, a depth-first search algorithm is used to traverse the entire power system physical model knowledge graph, exploring all reachable nodes through electrical topology connections.

[0096] The specific implementation of depth-first search is as follows: Initialization: Create a stack and push all nodes from the projection subgraph node set onto the stack; create a set visited to record visited nodes, initially adding all nodes from the projection subgraph node set to visited; create a list visit_order to record the node visit order; create a dictionary path_length to record the shortest path length from the initial node to each node, initially setting the path length of nodes in the projection subgraph node set to 0.

[0097] Iterative search: When the stack is not empty, perform the following operations: Pop a node current_node from the top of the stack; Record the access order of the current node in visit_order; Obtain the list of electrical connection edges of current_node from the knowledge graph of the power system physical model, that is, all adjacent nodes directly connected to this node.

[0098] Iterate through each neighbor node: If neighbor_node is not in the visited set, add it to the visited set, set its path length to path_length[current_node]+1, and push neighbor_node onto the stack.

[0099] Continue the loop until the stack is empty.

[0100] Search termination condition: When the stack is empty, it means that all reachable nodes have been visited, and the depth-first search ends. The visited set is the set of electrically connected nodes.

[0101] For example, using seven nodes such as GEN_001 and LINE_012 from the projection subgraph node set as initial activated nodes, a depth-first search is used to traverse the knowledge graph. Assume there are other nodes connected to these nodes in the knowledge graph, such as GEN_003 and LOAD_008 connected via BUS_002 (already included), LOAD_005 connected via BUS_001 (already included), and BUS_002 connected via LINE_012 (already included). If a transformer node TR_001 connected via LOAD_008 also exists, then TR_001 will also be visited and added to the visited set. The final size of the visited set is the size of the first connected component.

[0102] During the depth-first search, the visiting order of each node (i.e., which node was visited) and the shortest path length from any initial node to that node are recorded. Although this information is not a direct input to the calculation of the electrical connectivity domain size, it provides data support for subsequent visualization and sensitivity analysis.

[0103] Access order: Records the order in which each node is first accessed, which can be used to generate a traversal tree to show the propagation path of the impact of policy constraints.

[0104] Path length: Records the shortest path length (in terms of edges) from the initial node to this node. The path length reflects the degree to which this node is indirectly affected by policy constraints. The smaller the path length, the closer the node is electrically to the directly constrained equipment, and the more directly it is affected.

[0105] All visited nodes are assigned to the electrical connectivity node set, and the size of the electrical connectivity node set is determined as the size of the first connectivity domain.

[0106] Let C be the set of nodes in the electrically connected domain, and let its size be |C|. Then the size of the first connected domain is S1 = |C|.

[0107] For example, if 25 nodes are visited through depth-first search, then the size of the first connected component is S1 = 25.

[0108] Nodes with high confidence (projection confidence greater than a preset confidence threshold) are selected from the node set of the projected subgraph. Projection confidence is a string similarity value recorded when an entity name partially matches a knowledge graph node name during node retrieval in step S103. For nodes with a perfect match, the default projection confidence is 1.0 (maximum value).

[0109] The preset confidence threshold can be set according to the actual application scenario, for example, 0.85. A node is considered a high-confidence node only when it has a projected confidence score greater than 0.85.

[0110] Using the high-confidence node as the seed node, perform a depth-first search again to obtain the size of the second connected component. The specific steps are the same as the second step, except that the initial activated nodes only include the high-confidence node, instead of all nodes in the projected subgraph.

[0111] For example, the projection subgraph node set contains 7 nodes, where GEN_001 has a projection confidence of 1.0 (exact match), LINE_012 has a projection confidence of 0.95 (exact match), BUS_001 has a projection confidence of 0.90 (indirectly associated through GEN_001), LOAD_005 has a projection confidence of 0.85 (indirectly associated through BUS_001), and BUS_002, GEN_003, and LOAD_008 all have a projection confidence of 0.80. If the confidence threshold is set to 0.85, then the high-confidence nodes include 4 nodes: GEN_001, LINE_012, BUS_001, and LOAD_005.

[0112] Using these 4 high-confidence nodes as seed nodes, a depth-first search is performed. Assuming that the total number of nodes searched is 22, the size of the second connected component S2 = 22.

[0113] The weighted average of the sizes of the first and second connected components is determined as the electrical connected component size. The weighting coefficients can be adjusted according to the actual application scenario. Generally, the weighting coefficient of the first connected component size is smaller than that of the second connected component size because high-confidence nodes have higher reliability and should be given higher weights.

[0114] Let the size weight of the first connected component be α, and the size weight of the second connected component be β, and α + β = 1. Then the size of the electrically connected component is S = α × S1 + β × S2.

[0115] For example, if α = 0.3 and β = 0.7, then S = 0.3 × 25 + 0.7 × 22 = 7.5 + 15.4 = 22.9, which is 23 after rounding.

[0116] When the node set of the projected subgraph is empty (i.e., no projected nodes are generated in step S104), the size of the electrical connectivity domain is 0.

[0117] When the projection confidence of all nodes in the projection subgraph node set is lower than the confidence threshold, the high confidence node set is empty. At this time, the size of the second connected component is 0, and the size of the electrically connected component is S = α × S1.

[0118] When there is only one node in the projection subgraph node set and that node has no electrical connection edges in the knowledge graph, the depth-first search only visits that node itself. The size of the first connected component and the size of the second connected component are both 1 (if the node is a high-confidence node) or 1 and 0 (if the node is not a high-confidence node).

[0119] In summary, calculating the size of the electrically connected domain using a depth-first search and confidence-weighted mechanism has the following technical advantages: Using a depth-first search algorithm, starting from the directly constrained power equipment nodes, all reachable nodes are explored along the electrical topology connections. The scope of impact is quantified into the size of the connected domain, providing an objective and calculable indicator for the quantitative assessment of policy impact. For example, for the power flow constraints of a certain transmission line, the size of its electrical connected domain reflects the total number of generators, loads and buses that may be affected by the power flow changes of the line, directly reflecting the scope of the policy constraints. In step S103, during node retrieval, there may be matching uncertainty between some entity names and knowledge graph node names (e.g., "Unit 1" matching "#1 Unit"). Projection confidence is calculated using string similarity. This step uses this confidence to filter high-confidence nodes and performs a second depth-first search based on these high-confidence nodes. The weighted average of the two search results is used as the final electrical connectivity domain size. This weighting mechanism makes the evaluation results more robust and avoids significant deviations due to single matching errors. The node access order recorded during the depth-first search process can generate a propagation path tree of the policy constraint impact, which can intuitively show the trajectory of the constraint spreading from the source device to the surrounding devices. The recorded path length reflects the degree of indirectness of the policy constraint impact on each device. The smaller the path length, the closer the electrical distance between the device and the device directly constrained, and the more directly it is affected. The size of the electrical connectivity region is the numerator of the power system physical coverage calculation formula in step S106. A larger connectivity region indicates a wider range of policy constraints' influence on the power grid, higher power system physical coverage, and a stronger correlation between policy and the power system. This quantitative relationship makes the policy identification results physically interpretable. Step S106: Determine the power system physical coverage of the initial candidate texts based on the ratio of the size of the electrical connectivity domain to the total number of nodes in the power system physical model knowledge graph.

[0120] In this step, the total number of nodes in the knowledge graph of the power system physical model is obtained, and the electrical connectivity domain size is divided by the total number of nodes to obtain the connectivity domain coverage. The ratio of the number of nodes in the projection subgraph node set to the total number of nodes is used to obtain the projection node coverage. The power system physical coverage is obtained by weighted summation of the connected domain coverage and the projected node coverage.

[0121] In one specific embodiment, the electrical connectivity domain size calculated in step S105 is compared with the total number of nodes in the knowledge graph, and combined with the projection node coverage rate, the power system physical coverage of the initially screened candidate texts is comprehensively calculated. This coverage is used to quantify the degree of physical correlation between policy texts and the power system, and is the core basis for determining whether a text belongs to power industry policy.

[0122] Specifically, the power system physical model knowledge graph is pre-built and contains all power equipment nodes of the target power grid region. The total number of nodes in the knowledge graph is a fixed value, determined during the construction of the knowledge graph. The method for obtaining the total number of nodes is as follows: The total number of nodes is retrieved from the metadata of the knowledge graph. After the knowledge graph is built, it records the total number of nodes, including the number of generator nodes, transmission line nodes, load nodes, bus nodes, and other auxiliary equipment nodes.

[0123] If the knowledge graph supports dynamic updates (such as adding a new device), the total number of nodes is recalculated after each update and stored in the metadata for later retrieval.

[0124] For example, suppose the knowledge graph of the power system physical model of a certain regional power grid contains the following nodes: Generator nodes: 15; Transmission line nodes: 30 (each line is considered as one node); Load nodes: 25; Busbar nodes: 20; Transformer nodes: 10; The total number of nodes is 15 + 30 + 25 + 20 + 10 = 100.

[0125] Divide the electrical connectivity domain size calculated in step S105 by the total number of nodes in the knowledge graph to obtain the connectivity domain coverage rate. The connectivity domain coverage rate reflects the breadth of the impact of policy constraints on the power grid, that is, the proportion of the number of devices indirectly affected by policies to the total number of devices in the power grid.

[0126] The calculation formula is: Connectivity coverage = Electrical connectivity size / Total number of nodes in the knowledge graph; For example, if the electrical connectivity domain size calculated in step S105 is 23 and the total number of nodes in the knowledge graph is 100, then the connectivity domain coverage = 23 / 100 = 0.23.

[0127] Obtain the number of nodes in the projection subgraph node set, which is the number of nodes initially added to the projection node set in step S104 and successfully mapped (excluding nodes added in subsequent topology expansions). The number of projection nodes reflects the number of power devices directly mentioned in the policy text, i.e., the number of devices directly constrained by the policy.

[0128] Dividing the number of projected nodes by the total number of nodes in the knowledge graph yields the projected node coverage rate. The projected node coverage rate reflects the strength of the policy text's focus on specific power equipment.

[0129] Projection node coverage = Number of nodes in the projection subgraph node set / Total number of nodes in the knowledge graph; For example, in step S104, the initial projection subgraph node set contains two nodes, GEN_001 and LINE_012 (other nodes were added in subsequent topology expansions, but the initial projection nodes are only these two). The number of projection nodes is 2, and the total number of nodes in the knowledge graph is 100. Therefore, the projection node coverage rate is 2 / 100 = 0.02.

[0130] The physical coverage of the power system is obtained by weighted summation of the coverage of the connected components and the coverage of the projected nodes. The weighting coefficients reflect the difference in importance between the two coverage rates in the assessment.

[0131] Power system physical coverage = α1 × connectivity coverage + β1 × projection node coverage Where α1+β1=1, α1 is the weight coefficient of connected component coverage, and β1 is the weight coefficient of projected node coverage.

[0132] Principles for setting weighting coefficients: The weight α1 for connectivity coverage should be greater than the weight β1 for projected node coverage, because connectivity coverage reflects the breadth of the actual impact of policy constraints on the power grid and is a core indicator for assessing the correlation between policy and the power system. Even if the number of devices directly mentioned in the policy text is small, as long as these devices occupy a critical position in the power grid and can influence a large number of other devices through topology expansion, it indicates that the correlation between policy and the power system is still very high.

[0133] The projection node coverage weight β1 is relatively small and serves as an auxiliary indicator to supplement the assessment of the policy text's directive strength towards specific power equipment.

[0134] Example weight settings: α1 = 0.7 (connected component coverage weight); β1 = 0.3 (projection node coverage weight); Then the physical coverage of the power system = 0.7 × 0.23 + 0.3 × 0.02 = 0.161 + 0.006 = 0.167.

[0135] When the total number of nodes in the knowledge graph is 0 (which is theoretically impossible because the knowledge graph contains at least one node), the physical coverage of the power system is defined as 0.

[0136] When the electrical connectivity domain size is 0 (i.e., no nodes are explored in step S105), the connectivity domain coverage is 0, and the power system physical coverage = β1 × projected node coverage. If the projected node coverage is also 0, then the power system physical coverage is 0.

[0137] When the number of nodes in the projection subgraph node set is 0 (i.e., no valid power system physical parameters are extracted in step S103), the projection node coverage rate is 0, and the power system physical coverage rate = α1 × connected component coverage rate. At this time, the connected component coverage rate is also 0 (because there are no seed nodes for topology expansion), so the power system physical coverage rate is 0.

[0138] In this embodiment, the scale of the electrical connectivity domain reflects the breadth of the policy constraints in the power grid, and the number of projected nodes reflects the strength of the policy text's directivity towards specific devices. The two are weighted and fused to form a quantifiable physical coverage index. This index directly reflects the degree of physical coupling between the policy text and the power system, providing an objective and comparable quantitative basis for policy selection.

[0139] Step S107: Mark the initial screening candidate texts whose physical coverage of the power system is greater than the second preset threshold as secondary screening candidate texts.

[0140] In this step, the second preset threshold is a critical value used to determine whether the initially screened candidate texts have sufficient physical relevance to the power system. This threshold can be determined using the following method: Empirical Value Method: Based on the actual needs of identifying power industry policies, a fixed threshold is set. For example, a second preset threshold is set to 0.05. This means that only text with a physical coverage of the power system greater than 5% is considered to have a strong physical connection with the power system. This value can be adjusted according to the actual application scenario. If you want to improve the recall rate (identify more potential power policies), you can lower the threshold; if you want to improve the accuracy rate (reduce false positives), you can raise the threshold.

[0141] Statistical analysis method: Collect a batch of labeled power industry policy texts and non-power industry policy texts as a training set, calculate the power system physical coverage of each text, and draw a histogram of coverage distribution. Select the coverage value that best distinguishes the two types of texts as the second preset threshold. For example, if the coverage of power industry policy texts is generally higher than 0.03, while the coverage of non-power industry policy texts is generally lower than 0.01, the threshold can be set to 0.02 or 0.03.

[0142] Dynamic thresholding: The threshold is dynamically determined based on the coverage distribution of the initial candidate text set. For example, the average and standard deviation of the coverage of all initial candidate texts are calculated, and the threshold is set as the average minus 0.5 times the standard deviation, or the quantile method (such as the quartiles above) is used as the threshold. The dynamic threshold can adaptively adjust according to the characteristics of different batches of texts, and is suitable for scenarios with large variations in coverage distribution.

[0143] Step S108: Based on the preset power industry time series database, perform time series correlation analysis on the candidate texts for re-screening, extract the power quantitative indicators involved in the candidate texts for re-screening, and compare the power quantitative indicators with the actual operating data of the corresponding time period in the power industry time series database to obtain the time series consistency coefficient.

[0144] In this step, the policy release time is extracted from the candidate texts that have been screened, and the level of the policy-issuing organization is identified. The width of the analysis time window is determined based on the level of the issuing organization. The relevant power quantification indicators are extracted from the candidate texts of the secondary screening, and the indicator type and indicator value corresponding to the power quantification indicators are recorded. The power quantification indicators include power generation indicators, electricity consumption indicators, electricity price indicators, carbon emission quota indicators and renewable energy consumption indicators. Based on the policy release time and analysis time window, the actual operating data sequence of the corresponding indicator type within the analysis time window is extracted from the power industry time series database. The actual operating data sequence includes timestamps and indicator values. The indicator values ​​in the candidate texts are compared with the actual operational data sequence, and the difference between the indicator values ​​and the indicator values ​​at the corresponding time points in the actual operational data sequence is calculated to obtain the single-point deviation value. The relative deviation coefficient is obtained by calculating the ratio of the index value to the average value of all index values ​​within the time window in the actual operation data sequence. The time series consistency coefficient is obtained by weighting and summing the single-point deviation value and the relative deviation coefficient, where the sum of the weighting coefficient of the single-point deviation value and the weighting coefficient of the relative deviation coefficient is 1.

[0145] In one specific embodiment, a time-series database of the power industry is used to perform time-series correlation analysis on the candidate texts for secondary screening. By comparing the consistency between the quantitative indicators stipulated in the policy and the actual operational data, the accuracy and executability of the policy texts are verified. The time-series consistency coefficient is the core indicator of the third-level screening and is used to finally confirm whether the policy text belongs to the power industry policy.

[0146] Specifically, the policy release date is extracted from the candidate texts after secondary screening. The policy release date is usually located at the beginning of the policy text and is presented in the form of "Document Number," "Release Date," or "Printing Date." The extraction method is as follows: Use regular expressions to match date formats such as "January 1, 2023", "2023-01-01", "2023 / 01 / 01", etc.

[0147] If there are multiple dates in the text, determine the policy release date based on the context (usually the date in the document number or the printing date at the end of the document).

[0148] If the release date cannot be extracted, it should be inferred from the release time of other policy documents cited in the text or from the textual context, and the basis for the inference should be recorded.

[0149] Identify the hierarchy of policy-issuing agencies. The hierarchy can be determined by analyzing the issuing unit of the policy text. Issuing units typically appear at the beginning of the text in the form of "Provincial Development and Reform Commission," "Municipal People's Government," or "National Energy Administration," etc. The hierarchy of issuing agencies is divided into: National level: The issuing unit is a national ministry or agency directly under the State Council (such as the National Energy Administration or the National Development and Reform Commission). Provincial level: The issuing authority is the provincial people's government or its constituent departments (such as "Guangdong Provincial Development and Reform Commission"). At the prefecture-level city level: the issuing authority is the prefecture-level city people's government or its constituent departments (such as "Suzhou Municipal Bureau of Industry and Information Technology"). At the district / county level: the issuing authority is the district / county people's government or its constituent departments.

[0150] The width of the analysis time window is determined by the issuing agency's level. The time window reflects the time required from policy announcement to the generation of observable impact. Policies at different levels have different implementation and effect timelines. National-level policies have a wide impact and a long implementation chain, so the analysis time window is set to 12 months after the policy is released. Provincial-level policies have a wide impact; the analysis time window is set to 9 months after the policy is released. Municipal-level policies: The scope of impact is relatively concentrated, and the analysis time window is set to 6 months after the policy is released; District and county-level policies: have a smaller impact and are implemented quickly. The analysis time window is set to 3 months after the policy is released.

[0151] For example, if a policy is issued by the "National Energy Administration," which is a national-level agency, and the analysis time window is 12 months, and the policy was issued on January 1, 2023, then the analysis time window would be from January 1, 2023 to December 31, 2023.

[0152] Extract relevant quantitative indicators of electricity from the candidate texts in the secondary screening. These quantitative indicators are specific numerical requirements for the operating status of the power system as stated in the policy texts, including but not limited to: Electricity generation targets: such as "annual electricity generation growth of 5%" or "coal-fired power generation controlled within XX billion kilowatt-hours"; Electricity consumption indicators: such as "the growth rate of total social electricity consumption shall not exceed 3%" or "industrial electricity consumption shall decrease by XX%"; Electricity price indicators: such as "the on-grid electricity price is reduced by 0.01 yuan / kWh" and "the retail electricity price remains unchanged"; Carbon emission quota indicators: such as "carbon emission quota reduced by 10% compared to the previous year" or "carbon emissions per unit of electricity generated decreased by XX grams"; Renewable energy consumption indicators: such as "renewable energy consumption ratio reaches 20%" or "the proportion of wind and solar power generation increases to XX%"; The extraction method is as follows: Dependency parsing is performed on the candidate texts to identify statements containing numerical values ​​and units.

[0153] The type of indicator is determined based on the unit and contextual semantics of the numerical parameters. For example, units such as "ten thousand kilowatt-hours" and "hundred million kilowatt-hours" correspond to electricity-related indicators; units such as "yuan / kilowatt-hour" and "fen / kilowatt-hour" correspond to electricity price-related indicators; units such as "ton" and "ten thousand tons" correspond to carbon emission-related indicators; and percentages related to "proportion" or "percentage" correspond to consumption-related indicators.

[0154] Record the indicator type (e.g., "power generation indicator"), indicator value (e.g., "5%)" and its position in the text for each power quantification indicator.

[0155] For example, if a policy text contains the statement "the proportion of renewable energy consumption reaches 20%", then the extracted quantitative indicator is: the indicator type is "renewable energy consumption indicator", and the indicator value is "20%".

[0156] Based on the policy release time and analysis time window, the actual operation data sequence of the corresponding indicator type is extracted from the power industry time series database.

[0157] The power industry time-series database is pre-built and stores time-series data for various indicators of the power system. Data sources include: Statistics on the power industry released by the National Bureau of Statistics; The National Energy Administration's report on power operation; Industry data released by the China Electricity Council; Transaction data released by provincial power trading centers; Dispatch and operation data released by the power grid company.

[0158] The actual method for extracting the data sequence is as follows: Based on the indicator type (such as "renewable energy consumption indicator"), locate the corresponding data table or data field in the time series database.

[0159] Extract all data points within the analysis time window (e.g., from January 1, 2023 to December 31, 2023).

[0160] Data points include timestamps (e.g., "January 2023") and metric values ​​(e.g., "18.5%").

[0161] If the data is monthly, extract 12 data points; if it is quarterly, extract 4 data points; if it is annual, extract 1 data point. All data is organized as monthly data, with annual data treated as having the same value for all 12 months, and quarterly data split by month.

[0162] For example, the renewable energy consumption ratio indicator, the actual data sequence from January to December 2023 is: [18.2%, 18.5%, 19.0%, 19.3%, 19.8%, 20.1%, 20.3%, 20.5%, 20.2%, 19.8%, 19.5%, 19.2%].

[0163] The indicator values ​​in the candidate texts are compared with the actual operational data sequence, and the difference between the indicator values ​​and the indicator values ​​at the corresponding time points in the actual operational data sequence is calculated to obtain the single-point deviation value.

[0164] The method for determining the corresponding time point is as follows: If the policy is released at the beginning of the year or the policy implementation begins at the beginning of the year, then the annual data or the average of 12 months within the time window shall be used as the comparison benchmark.

[0165] If the policy is released in the middle of the year, the comparison point is determined based on the start time of policy implementation. For example, if the policy is released in June 2023, then the data from December 2023 is used as the comparison point (the effect six months after the policy is implemented).

[0166] If the policy specifies a particular implementation period (e.g., "to reach 20% by 2025"), then the data at the end of the time window (e.g., December 2025) will be used as the comparison point.

[0167] The formula for calculating the single-point deviation value is: Single-point deviation value = |Indicator value - Actual indicator value at the corresponding time point|; For example, if the policy stipulates that "the proportion of renewable energy consumption should reach 20%", and the actual value at the corresponding time point (December 2023) is 19.2%, then the single-point deviation value = |20% - 19.2%| = 0.8%.

[0168] The relative deviation coefficient is obtained by calculating the ratio of the indicator value to the average value of all indicator values ​​within the time window of the actual operating data series. The relative deviation coefficient reflects the degree of deviation between the policy indicator and the actual operating average level.

[0169] First, calculate the average of all indicator values ​​within the time window: Average value = (Indicator value 1 + Indicator value 2 + ... + Indicator value n) / n; For example, the average value of the actual data series from January to December 2023 is (18.2+18.5+19.0+19.3+19.8+20.1+20.3+20.5+20.2+19.8+19.5+19.2) / 12=19.53%.

[0170] Then calculate the relative deviation coefficient: Relative deviation coefficient = |index value - average value| / average value; For example, the difference between the indicator value of 20% and the average value of 19.53% is 0.47%, and the relative deviation coefficient is approximately 0.47% / 19.53% ≈ 0.0241.

[0171] The temporal consistency coefficient is obtained by weighting and summing the single-point deviation value with the relative deviation coefficient. The weighting coefficient reflects the difference in importance between the two types of deviation in the evaluation.

[0172] The formula for calculating the timing consistency coefficient is: Timing consistency coefficient = ×(1-single-point deviation value)+ ×(1-relative deviation coefficient) in, + =1, The single-point deviation value weighting coefficient, This is the relative deviation coefficient and weighting coefficient.

[0173] Since a smaller deviation value indicates better consistency, (1 - deviation value) is used to convert the deviation into a consistency measure. Both the single-point deviation value and the relative deviation coefficient are normalized, with a value range of [0,1].

[0174] Principles for setting weighting coefficients: Single-point deviation reflects the gap between policy indicators and actual values ​​at specific points in time, demonstrating the precision of policy objectives. (Weighting coefficient) It should be relatively large.

[0175] The relative deviation coefficient reflects the degree of deviation between policy indicators and the overall average level, demonstrating the rationality of policy objectives. The weighting coefficient... It should be relatively small.

[0176] Example weight settings: =0.6 (weight of single-point deviation value) =0.4 (Relative Deviation Coefficient Weight) The single-point deviation value is 0.008 after normalization (0.8%), the relative deviation coefficient is 0.0241 after normalization, and the time series consistency coefficient is 0.6×(1-0.008)+0.4×(1-0.0241)=0.6×0.992+0.4×0.9759=0.5952+0.39036=0.98556.

[0177] When no power quantification indicators are extracted from the candidate texts in the second round of screening, the time series consistency coefficient is defined as 0, and the text directly enters the next round of screening (when the time series consistency coefficient is 0, it cannot pass the third preset threshold).

[0178] When there is no actual operating data of the corresponding indicator type in the power industry time series database, the time series consistency coefficient is defined as 0, and the reason for the missing data is recorded.

[0179] When the actual running data sequence is empty (no data within the time window), the timing consistency coefficient is defined as 0.

[0180] When the indicator value or the actual indicator value is 0, the single-point deviation value and the relative deviation coefficient are calculated according to the formula. If the denominator is 0, it is defined as infinity, and the time series consistency coefficient is defined as 0.

[0181] Step S109: Mark the candidate texts with a temporal consistency coefficient greater than the third preset threshold as power industry policy texts, and output the marked power industry policy texts as power industry policy information recognition results.

[0182] In this step, the third preset threshold is a critical value used to determine whether the candidate texts undergoing secondary screening have sufficient temporal consistency. The temporal consistency coefficient reflects the degree of agreement between the quantitative indicators stipulated in the policy and the actual operational data. This threshold can be determined using the following method: Empirical Value Method: Based on the actual needs of identifying business operations related to power industry policies, a fixed threshold is set. For example, a third preset threshold is set to 0.85. This means that only text with a temporal consistency coefficient greater than 85% is considered to have a high degree of consistency between policy indicators and actual operational data, possessing high accuracy and executability. This value can be adjusted according to the actual application scenario; if a higher recall rate is desired, the threshold can be appropriately lowered; if a higher precision rate is desired, the threshold can be increased.

[0183] Statistical analysis method: Collect a batch of labeled power industry policy texts and non-power industry policy texts as a training set, calculate the temporal consistency coefficient of each text, and draw a distribution histogram. Select the coefficient value that best distinguishes the two types of texts as the third preset threshold. For example, if the temporal consistency coefficient of power industry policy texts is generally higher than 0.80, while the coefficient of non-power industry policy texts is generally lower than 0.60, the threshold can be set to 0.70 or 0.75.

[0184] Dynamic thresholding: The threshold is dynamically determined based on the temporal consistency coefficient distribution of the candidate text set undergoing secondary screening. For example, the mean and standard deviation of the coefficients of all candidate texts undergoing secondary screening are calculated, and the threshold is set as the mean minus 0.5 times the standard deviation, or a quantile method (such as the median) is used as the threshold. The dynamic threshold can be adaptively adjusted according to the characteristics of different batches of texts.

[0185] In summary, the method of this application performs step-by-step feature matching on the policy text to be identified based on a multi-level power industry feature database to obtain the power industry feature matching degree. Texts with a matching degree greater than a first threshold are marked as initial screening candidate texts. Parameters are extracted from the initial screening candidate texts based on the power system physical model knowledge graph to generate a projection subgraph. The size of the electrical connectivity domain is calculated, and the power system physical coverage is determined based on the ratio of the connectivity domain size to the total number of nodes in the knowledge graph. Texts with a coverage greater than a second threshold are marked as secondary screening candidate texts. Temporal association analysis is performed on the secondary screening candidate texts based on the power industry time series database. Texts with a temporal consistency coefficient greater than a third threshold are marked as power industry policy texts and output. Through this three-level progressive screening, the accuracy and interpretability of power industry policy information identification are significantly improved.

[0186] Please seeFigure 2 The diagram shows a structural block diagram of a power industry policy information identification system according to this application.

[0187] like Figure 2 As shown, the power industry policy information identification system 200 includes an acquisition module 210, a first marking module 220, an extraction module 230, a generation module 240, a calculation module 250, a determination module 260, a second marking module 270, an analysis module 280, and an output module 290.

[0188] The acquisition module 210 is configured to acquire at least one policy text to be identified, and perform step-by-step feature matching on the policy text to be identified according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified; the first marking module 220 is configured to mark the policy text to be identified with a power industry feature matching degree greater than a first preset threshold as preliminary candidate texts; the extraction module 230 is configured to extract parameters from the preliminary candidate texts according to a preset power system physical model knowledge graph to obtain power system physical parameters; the generation module 240 is configured to map the power system physical parameters to the power system physical model knowledge graph to generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph, wherein the projection subgraph contains at least one knowledge graph node and a corresponding parameter constraint value; the calculation module 250 is configured to calculate the electrical connectivity domain size of the projection subgraph in the power system physical model knowledge graph, wherein the electrical connectivity domain size of the projection subgraph is calculated as follows: The scale of the electrical connectivity domain is the total number of nodes that can be connected through electrical topology connections in the projected subgraph; the determination module 260 is configured to determine the power system physical coverage of the initial screening candidate texts based on the ratio of the electrical connectivity domain scale to the total number of nodes in the power system physical model knowledge graph; the second marking module 270 is configured to mark the initial screening candidate texts with a power system physical coverage greater than a second preset threshold as secondary screening candidate texts; the analysis module 280 is configured to perform time-series correlation analysis on the secondary screening candidate texts based on a preset power industry time-series database, extract the power quantitative indicators involved in the secondary screening candidate texts, and compare the power quantitative indicators with the actual operating data of the corresponding time period in the power industry time-series database to obtain the time-series consistency coefficient; the output module 290 is configured to mark the secondary screening candidate texts with a time-series consistency coefficient greater than a third preset threshold as power industry policy texts, and output the marked power industry policy texts as power industry policy information recognition results.

[0189] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text.Figure 2 The various modules in the document will not be described in detail here.

[0190] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the power industry policy information identification method in any of the above method embodiments. In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows: At least one policy text to be identified is obtained, and the policy text to be identified is matched step by step according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified. Policy texts whose matching degree with power industry characteristics is greater than the first preset threshold are marked as preliminary candidate texts. Based on the preset knowledge graph of the power system physical model, parameters are extracted from the initial candidate texts to obtain the power system physical parameters; The power system physical parameters are mapped to the power system physical model knowledge graph to generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph. The projection subgraph contains at least one knowledge graph node and a corresponding parameter constraint value. Calculate the electrical connectivity domain size of the projected subgraph in the knowledge graph of the power system physical model, where the electrical connectivity domain size is the total number of nodes that can be connected through electrical topology connections in the projected subgraph; The power system physical coverage of the initial candidate texts is determined based on the ratio of the size of the electrical connectivity domain to the total number of nodes in the knowledge graph of the power system physical model. Initially screened candidate texts with a power system physical coverage greater than the second preset threshold are marked as secondary screened candidate texts. Based on the preset power industry time series database, time series correlation analysis is performed on the candidate texts for re-screening, power quantitative indicators involved in the candidate texts are extracted, and the power quantitative indicators are compared with the actual operating data of the corresponding time period in the power industry time series database to obtain the time series consistency coefficient. Candidate texts with a temporal consistency coefficient greater than the third preset threshold are marked as power industry policy texts, and the marked power industry policy texts are output as power industry policy information recognition results.

[0191] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the power industry policy information identification system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely located relative to a processor, and these remote memories can be connected to the power industry policy information identification system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0192] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the power industry policy information identification method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the power industry policy information identification system. The output device 340 may include a display screen or other display device.

[0193] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0194] In one implementation, the aforementioned electronic device is applied to a power industry policy information identification system for a client, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: At least one policy text to be identified is obtained, and the policy text to be identified is matched step by step according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified. Policy texts whose matching degree with power industry characteristics is greater than the first preset threshold are marked as preliminary candidate texts. Based on the preset knowledge graph of the power system physical model, parameters are extracted from the initial candidate texts to obtain the power system physical parameters; The power system physical parameters are mapped to the power system physical model knowledge graph to generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph. The projection subgraph contains at least one knowledge graph node and a corresponding parameter constraint value. Calculate the electrical connectivity domain size of the projected subgraph in the knowledge graph of the power system physical model, where the electrical connectivity domain size is the total number of nodes that can be connected through electrical topology connections in the projected subgraph; The power system physical coverage of the initial candidate texts is determined based on the ratio of the size of the electrical connectivity domain to the total number of nodes in the knowledge graph of the power system physical model. Initially screened candidate texts with a power system physical coverage greater than the second preset threshold are marked as secondary screened candidate texts. Based on the preset power industry time series database, time series correlation analysis is performed on the candidate texts for re-screening, power quantitative indicators involved in the candidate texts are extracted, and the power quantitative indicators are compared with the actual operating data of the corresponding time period in the power industry time series database to obtain the time series consistency coefficient. Candidate texts with a temporal consistency coefficient greater than the third preset threshold are marked as power industry policy texts, and the marked power industry policy texts are output as power industry policy information recognition results.

[0195] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0196] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying power industry policy information, characterized in that, include: At least one policy text to be identified is obtained, and the policy text to be identified is matched step by step according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified. Policy texts whose matching degree with power industry characteristics is greater than the first preset threshold are marked as preliminary candidate texts. Based on the preset knowledge graph of the power system physical model, parameters are extracted from the initial candidate texts to obtain the power system physical parameters; The power system physical parameters are mapped to the power system physical model knowledge graph to generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph. The projection subgraph contains at least one knowledge graph node and a corresponding parameter constraint value. Calculate the electrical connectivity domain size of the projected subgraph in the knowledge graph of the power system physical model, where the electrical connectivity domain size is the total number of nodes that can be connected through electrical topology connections in the projected subgraph; The power system physical coverage of the initial candidate texts is determined based on the ratio of the size of the electrical connectivity domain to the total number of nodes in the knowledge graph of the power system physical model. Initially screened candidate texts with a power system physical coverage greater than the second preset threshold are marked as secondary screened candidate texts. Based on the preset power industry time series database, time series correlation analysis is performed on the candidate texts for re-screening, power quantitative indicators involved in the candidate texts are extracted, and the power quantitative indicators are compared with the actual operating data of the corresponding time period in the power industry time series database to obtain the time series consistency coefficient. Candidate texts with a temporal consistency coefficient greater than the third preset threshold are marked as power industry policy texts, and the marked power industry policy texts are output as power industry policy information recognition results.

2. The method for identifying power industry policy information according to claim 1, characterized in that, The multi-level power industry feature database includes a primary feature database, a secondary feature database, and a tertiary feature database. The primary feature database contains core power industry terms, the secondary feature database contains power equipment terms and operating parameter terms, and the tertiary feature database contains power market terms and regulatory terms. The step of performing step-by-step feature matching on the policy text to be identified based on a preset multi-level power industry feature database to obtain the power industry feature matching degree of the policy text to be identified includes: The policy text to be identified is segmented and part-of-speech tagged to obtain a text word sequence; Each word in the text word sequence is matched with the core terms of the power industry in the first-level feature library. The number of successfully matched words is counted, the first-level matching score is calculated, and the position of the successfully matched word in the text is marked as the first-level matching position. Using the primary matching position as the center, extend the preset window length forward and backward, and extract the text within the extended window as the primary matching context. The words in the first-level matching context are matched with the power equipment terms and operating parameter terms in the second-level feature library. The number of successfully matched words is counted, and the second-level matching score is calculated. Using the location of the word that successfully matches the secondary feature library as the center, expand the second preset window length forward and backward, and extract the text within the expanded window as the secondary matching context. The words in the secondary matching context are matched with the electricity market terms and regulatory terms in the tertiary feature library. The number of successfully matched words is counted, and the tertiary matching score is calculated. The matching scores of Level 1, Level 2, and Level 3 are weighted and summed to obtain the matching degree of the power industry characteristics. The weight coefficient of Level 1 matching score is greater than that of Level 2 matching score, and the weight coefficient of Level 2 matching score is greater than that of Level 3 matching score.

3. The method for identifying power industry policy information according to claim 1, characterized in that, in, The knowledge graph of the power system physical model includes generator nodes, transmission line nodes, load nodes and bus nodes, as well as the electrical topology connections and operating parameter constraints between nodes. Each node is associated with a node identifier, node name and constrainable parameter template. The step of extracting parameters from the initially screened candidate texts based on a preset power system physical model knowledge graph to obtain power system physical parameters includes: Dependency parsing is performed on the initial candidate texts to construct a dependency parsing tree. Based on the dependency parsing tree, statements containing constraint semantics in the text are identified to obtain constraint statements. The constraint semantics are identified through the dependency relationship between constraint keywords and predicate verbs. Power system entity identification is performed on the constraint statement to extract the power system entities in the constraint statement, and the entity name, entity type and entity position of each power system entity in the constraint statement are recorded. The power system entities include generators, transmission lines, loads and buses, and the entity position is the start character position and end character position of the power system entity in the constraint statement. Based on the extracted entity names and entity types of the power system entities, node retrieval is performed in the power system physical model knowledge graph to obtain entity identifiers corresponding to the power system entities, specifically including: Using the entity name of a power system entity as the search keyword, the target node matching the entity type is found in the knowledge graph of the power system physical model. When a target node that exactly matches the entity name is found, the node identifier of the target node is used as the entity identifier of the power system entity. When multiple target nodes that partially match the entity name are found, the string similarity between the entity name and each candidate node name is calculated, and the node identifier of the candidate node with the highest similarity is selected as the entity identifier. When no matching target node is found, the entity identifier of the power system entity is marked as empty; Obtain the target constraint parameter template associated with the retrieved target node; Numerical parameters corresponding to the power system entity are extracted from the constraint statement, and the numerical parameters are matched with the parameter types in the constrainable parameter template to determine the parameter type corresponding to the numerical parameters. The numerical parameters include power values, voltage values, current values, emission values, and energy values. The parameter type corresponding to the numerical parameters is the specific type of constraint imposed on the numerical parameters at the target node, including one of the following: active power output limit, reactive power output limit, emission limit, power flow limit, energy consumption limit, power factor limit, voltage limit, and voltage lower limit. The entity identifier, parameter type, and numerical parameter are combined to form the physical parameters of the power system.

4. The method for identifying power industry policy information according to claim 3, characterized in that, The step of mapping the power system physical parameters to the power system physical model knowledge graph and generating projection subgraphs of the power system physical parameters in the power system physical model knowledge graph includes: In the knowledge graph of the power system physical model, target nodes corresponding to entity identifiers in the power system physical parameters are identified, and the target nodes are added to the projection node set; Using nodes in the projection node set as seed nodes, a preset breadth-first search algorithm is used to expand outward along the electrical topology connection relationship, and all nodes visited during the expansion process are added to the projection subgraph node set. The projection subgraph is formed by the set of nodes in the projection subgraph and the electrical topology connection edges between the nodes.

5. The method for identifying power industry policy information according to claim 3, characterized in that, The calculation of the electrical connectivity domain size of the projected subgraph in the knowledge graph of the power system physical model includes: Obtain the node identifiers of each node in the knowledge graph of the power system physical model in the node set of the projected subgraph; Using nodes in the projection subgraph node set as initial activated nodes, a preset depth-first search algorithm is used to traverse the knowledge graph of the power system physical model and explore all reachable nodes through electrical topology connections. During the traversal, the order in which nodes are visited is recorded, as well as the path length from the initial node to a given node. All visited nodes are assigned to the electrical connectivity node set, and the size of the electrical connectivity node set is determined as the size of the first connectivity domain. Obtain high-confidence nodes in the projection subgraph node set whose projection confidence is greater than a preset confidence threshold, and use the high-confidence nodes as seed nodes to perform a depth-first search again to obtain the size of the second connected component. The weighted average of the sizes of the first and second connected components is determined as the size of the electrical connected component.

6. The method for identifying power industry policy information according to claim 1, characterized in that, The step of determining the power system physical coverage of the initial candidate texts based on the ratio of the electrical connectivity domain size to the total number of nodes in the power system physical model knowledge graph includes: Obtain the total number of nodes in the knowledge graph of the power system physical model, and divide the size of the electrical connectivity domain by the total number of nodes to obtain the connectivity domain coverage. The ratio of the number of nodes in the projection subgraph node set to the total number of nodes is used to obtain the projection node coverage. The power system physical coverage is obtained by weighted summation of the connected domain coverage and the projected node coverage.

7. The method for identifying power industry policy information according to claim 1, characterized in that, The step involves performing time-series correlation analysis on the candidate texts for secondary screening based on a preset power industry time-series database, extracting the power quantitative indicators involved in the candidate texts, and comparing the power quantitative indicators with the actual operating data of the corresponding time period in the power industry time-series database to obtain the time-series consistency coefficient, including: The policy release time is extracted from the candidate texts that have been screened, and the level of the policy-issuing organization is identified. The width of the analysis time window is determined based on the level of the issuing organization. The relevant power quantification indicators are extracted from the candidate texts of the secondary screening, and the indicator type and indicator value corresponding to the power quantification indicators are recorded. The power quantification indicators include power generation indicators, electricity consumption indicators, electricity price indicators, carbon emission quota indicators and renewable energy consumption indicators. Based on the policy release time and analysis time window, the actual operating data sequence of the corresponding indicator type within the analysis time window is extracted from the power industry time series database. The actual operating data sequence includes timestamps and indicator values. The indicator values ​​in the candidate texts are compared with the actual operational data sequence, and the difference between the indicator values ​​and the indicator values ​​at the corresponding time points in the actual operational data sequence is calculated to obtain the single-point deviation value. The relative deviation coefficient is obtained by calculating the ratio of the index value to the average value of all index values ​​within the time window in the actual operation data sequence. The time series consistency coefficient is obtained by weighting and summing the single-point deviation value and the relative deviation coefficient, where the sum of the weighting coefficient of the single-point deviation value and the weighting coefficient of the relative deviation coefficient is 1.

8. A power industry policy information identification system, characterized in that, include: The acquisition module is configured to acquire at least one policy text to be identified, and perform step-by-step feature matching on the policy text to be identified according to a preset multi-level power industry feature library to obtain the power industry feature matching degree of the policy text to be identified; The first marking module is configured to mark the policy text to be identified that has a power industry feature matching degree greater than a first preset threshold as a preliminary candidate text; The extraction module is configured to extract parameters from the initially screened candidate texts based on a preset knowledge graph of the power system physical model to obtain power system physical parameters. The generation module is configured to map the power system physical parameters to the power system physical model knowledge graph, and generate a projection subgraph of the power system physical parameters in the power system physical model knowledge graph, wherein the projection subgraph contains at least one knowledge graph node and a corresponding parameter constraint value. The calculation module is configured to calculate the size of the electrical connectivity domain of the projected subgraph in the knowledge graph of the power system physical model, wherein the size of the electrical connectivity domain is the total number of nodes that can be connected by electrical topology connections in the projected subgraph; The determination module is configured to determine the power system physical coverage of the initial candidate texts based on the ratio of the size of the electrical connectivity domain to the total number of nodes in the power system physical model knowledge graph. The second marking module is configured to mark the initial screening candidate texts with a power system physical coverage greater than a second preset threshold as secondary screening candidate texts. The analysis module is configured to perform time-series correlation analysis on the candidate texts for re-screening based on a preset power industry time-series database, extract the power quantitative indicators involved in the candidate texts for re-screening, and compare the power quantitative indicators with the actual operating data of the corresponding time period in the power industry time-series database to obtain the time-series consistency coefficient. The output module is configured to mark the candidate texts with a temporal consistency coefficient greater than a third preset threshold as power industry policy texts, and output the marked power industry policy texts as power industry policy information recognition results.

9. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 7.

Images