Electric power unstructured data feature and framework extraction method and system

Abstract

The application belongs to the technical field of electric power data processing and artificial intelligence, and provides an electric power unstructured data feature and framework extraction method and system. According to power supply and consumption contract text data, a power supply and consumption contract framework constraint graph is constructed. According to the power supply and consumption contract framework constraint graph, a framework constraint path is generated. Relationship extraction is carried out according to the framework constraint path. The framework constraint path set after backfilling is obtained through conversion. According to the framework constraint path set after backfilling, feature extraction is carried out to obtain power supply and consumption contract value structured data objects. Through the power supply and consumption contract framework constraint graph and the framework constraint path control large model extraction process, the large model can be changed from open field extraction to relationship extraction under the relationship constraint of the contract field, can adapt to different power supply and consumption contract templates and field expression differences, can express the business relationship in the contract, and solves the problems of field omission, field ownership error and cross-clause relationship loss.

Description

Technical Field

[0001] This invention belongs to the field of power data processing and artificial intelligence technology, specifically relating to a method and system for extracting features and frameworks from unstructured power data. Background Technology

[0002] Currently, power supply contracts are mostly stored in text, Portable Document Format (PDF), or scanned text formats. Key fields in these contracts are scattered across different clauses, and issues such as inconsistent field names, inconsistent clause wording, and failure to synchronize system ledgers in a timely manner after contract changes exist. Due to the lack of calculable structured expression in contract content, tasks such as verifying consistency between marketing system ledgers and contract texts, identifying contract expiration dates, and managing renewals still mainly rely on manual review and judgment, which is inefficient and prone to omissions.

[0003] Existing methods based on fixed templates or ordinary field extraction are difficult to adapt to the differences in different power supply contract templates and field descriptions. They are also difficult to express the business relationships between fields such as contract capacity, metering method, contract term, and property rights responsibility, which can easily lead to field omissions, incorrect field attribution, and missing cross-clause relationships. Summary of the Invention

[0004] This invention proposes a method and system for extracting features and frames from unstructured power data. By controlling the extraction process through the framework constraint graph and framework constraint path of power supply and consumption contracts, the extraction process can be transformed from open field extraction to relational extraction constrained by contract field relationships. This method can adapt to the differences in different power supply and consumption contract templates and field descriptions, and can express the business relationships in the contract. It solves the problems of field omission, incorrect field attribution, and missing cross-clause relationships.

[0005] In a first aspect, the present invention provides a method for extracting features and frames from unstructured power data, comprising the following steps: Based on the obtained power supply contract text data, a power supply contract framework constraint graph is constructed using the relationship set of nodes, which includes field alias relationships, field coupling relationships, cross-clause association relationships, responsibility grouping relationships, and field validation relationships. Based on the power supply contract framework constraint graph, extraction control generates a framework constraint path to represent the organizational relationship and extraction order between fields in the contract business information; Based on the mapping relationship between the framework constraint path and the semantic unit of the clause, the framework constraint path is converted into a field group extraction task; relational extraction is then performed within the corresponding clause text based on the field group extraction task. The relational extraction results are backfilled into the corresponding frame constraint paths, resulting in a backfilled set of frame constraint paths with field values, field relationships, original evidence, and evidence locations. Based on the set of constraint paths in the backfilled framework, feature extraction is performed to obtain a structured data object of power supply contract value.

[0006] Furthermore, the set of nodes in the constraint graph of the power supply contract framework includes contract type nodes, clause semantic nodes, standard field nodes, field alias nodes, and rule nodes. The set of node attributes includes field name, field type, field unit, field enumeration value, whether it is required, the type of contract it belongs to, the semantics of the clause it belongs to, and the evidence requirements. The set of constraint rules includes field integrity rules, field format rules, unit rules, date rules, enumeration value rules, field relationship consistency rules, and evidence traceability rules.

[0007] Furthermore, the generation of the framework constraint path includes: matching the applicable contract framework scope, clause modules, standard field sets, field alias relationships, field coupling relationships, cross-clause association relationships, responsibility grouping relationships, and field validation rules in the power supply contract framework constraint graph to obtain field nodes and relationship edges, thereby generating the framework constraint path set corresponding to the current contract.

[0008] Furthermore, the relational extraction includes: generating a field group extraction task based on the field nodes, field aliases, field relationship edges, and validation constraints in the framework constraint path, combined with the clause text range corresponding to the clause semantic unit set; and performing group extraction according to the field group extraction task within the clause semantic unit range defined by the clause semantic unit set, based on the field relationships defined by the framework constraint path, to obtain the relational extraction result. : ; in, Indicates the field name; Indicates the field value; Indicates the relationship between fields; Indicate the original text as evidence; Indicate the location of the original evidence; Indicates the first The number of fields in each field group extraction task.

[0009] Furthermore, the generation of the structured data object of the power supply contract value includes: traversing the backfilled set of framework constraint paths and grouping multiple framework constraint paths involving the same business theme into the same business theme; determining the degree of semantic mutual exclusion between evidence texts for the set of evidence texts within the same business theme, in order to quantify the logical conflict characteristics across clauses within the text; and constructing dynamic temporal difference operators and ladder-style state mapping operators for static fields involving the contract validity period in the structured framework of the power supply contract, transforming them into lifecycle timeliness characteristics that evolve dynamically over time.

[0010] Furthermore, the build size is Association semantic matrix : ; Among them, the main diagonal elements This indicates that the evidence itself does not have semantic conflicts; off-diagonal elements Indicates the probability of clause conflicts between the original evidence: ; in, and The evidence texts are respectively Original evidence With the Original evidence The semantic feature vector generated after processing; This represents the Euclidean distance between the semantic feature vectors; This is a mutual exclusion indicator function for electrical engineering terminology; This is the preset conflict penalty coefficient; For the association semantic matrix By aggregating the maximum risk of all upper triangular elements, the probability feature value of conflict of terms in the power supply contract is extracted. : .

[0011] Furthermore, the static cutoff time nodes are extracted and standardized; the static cutoff time nodes are determined using dynamic time-series difference operators. Compared with the current system standard execution time Dynamic time-series difference days : ; in, (.) represents a day-level time-series difference function, used to quantify the time span from the current time to the contract expiration boundary; a ladder-style state mapping operator is constructed to map the dynamic time-series difference days into a standard format time-derived feature status code; the time-derived feature status code undergoes dynamic state transitions as the system standard execution time advances, serving as a dynamic high-order feature index characterizing the contract time dimension.

[0012] Secondly, the present invention provides a system for extracting features and frames from unstructured power data, comprising: The constraint graph construction module is configured to: construct a constraint graph of the power supply contract framework based on the obtained power supply contract text data, using the relationship set of nodes as field alias relationship, field coupling relationship, cross-clause association relationship, responsibility grouping relationship and field verification relationship; The framework constraint path generation module is configured to: extract and generate framework constraint paths that represent the organizational relationships and extraction order between fields in the contract business information based on the power supply contract framework constraint graph; The extraction module is configured to: convert the framework constraint path into a field group extraction task based on the mapping relationship between the framework constraint path and the semantic unit of the clause; and perform relational extraction within the corresponding clause text based on the field group extraction task. The backfilling module is configured to: backfill the relational extraction results into the corresponding frame constraint paths, and transform them into a backfilled set of frame constraint paths with field values, field relationships, original evidence, and evidence locations; The feature extraction module is configured to extract features from the backfilled set of framework constraint paths to obtain a structured data object of power supply contract value.

[0013] Thirdly, the present invention provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the above-mentioned method for extracting features and frames of unstructured power data.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described method for extracting features and frames of unstructured power data.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention first constructs a framework constraint graph of the power supply contract based on the acquired power supply contract text data, using a set of relationships among nodes, including field alias relationships, field coupling relationships, cross-clause association relationships, responsibility grouping relationships, and field verification relationships. Then, based on the framework constraint graph, it extracts and controls the generation of framework constraint paths to represent the organizational relationships and extraction order between fields in the contract's business information. According to the mapping relationship between the framework constraint paths and the semantic units of the clauses, the framework constraint paths are converted into field group extraction tasks. Relational extraction is performed within the corresponding clause text based on the field group extraction tasks. The relational extraction results are backfilled into the corresponding framework constraint paths, resulting in a backfilled set of framework constraint paths containing field values, field relationships, original text evidence, and evidence locations. Finally, feature extraction is performed based on the backfilled set of framework constraint paths to obtain a structured data object representing the value of the power supply contract. By using the power supply contract framework constraint graph and framework constraint path control extraction process, the extraction process can be transformed from open field extraction to relational extraction constrained by contract field relationships. This can adapt to different power supply contract templates and differences in field descriptions, express the business relationships in the contract, and solve the problems of field omission, incorrect field attribution, and missing cross-clause relationships. Attached Figure Description

[0016] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0017] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the power supply contract framework constraint diagram of Embodiment 1 of the present invention; Figure 3 This is a flowchart of the structured framework extraction of power supply contracts based on frame constraint paths in Embodiment 1 of the present invention; Figure 4 This is a flowchart of the feature extraction process for the structured framework in Embodiment 1 of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] This embodiment provides a method and system for extracting features and frameworks from unstructured power data. Power supply contracts may include high-voltage power supply contracts, low-voltage residential power supply contracts, low-voltage non-residential power supply contracts, temporary power supply contracts, and other power customer power supply contracts. Power supply contracts can be electronic text, scanned documents recognized by OCR, text exported from a contract management system, or other data capable of forming parsable text content.

[0021] Example 1: like Figure 1 As shown, this embodiment provides a method for feature and framework extraction of unstructured power data. Targeting power supply contract text data, it performs semantic segmentation of clauses and contract type identification on the power supply contract. It then introduces a power supply contract framework constraint graph to generate framework constraint paths, converting these paths into field group extraction tasks. This drives a large model to complete the relational extraction of field values, field relationships, original evidence, and confidence levels. Furthermore, based on the extracted structured contract framework, it determines value features such as contract term, capacity management, metering and billing, property rights and responsibilities, framework integrity, and evidence credibility. The extraction results are optimized through rule verification and supplementary extraction mechanisms, ultimately generating computable, comparable, predictable, and reusable structured data of power supply contract value. The specific steps include: S1. Constructing the constraint diagram of the power supply contract framework: like Figure 2 As shown, this step transforms the contract type, clause semantics, standard fields, field aliases, field relationships, and validation rules in the power supply contract into framework constraint information that can be called by the large model, providing a basis for subsequent framework constraint path generation and structured framework extraction. Specifically, the framework constraint graph of the power supply contract is constructed as follows: ; Wherein, V represents the graph node set, which includes contract type nodes, clause semantic nodes, standard field nodes, field alias nodes, and rule nodes; E represents the set of relationships between nodes, which includes field alias relationships, field coupling relationships, cross-clause association relationships, responsibility grouping relationships, and field validation relationships; A represents the set of node attributes, which includes field name, field type, field unit, field enumeration value, whether it is required, the contract type to which it belongs, the clause semantics to which it belongs, and evidence requirements; C represents the set of constraint rules, which includes field integrity rules, field format rules, unit rules, date rules, enumeration value rules, field relationship consistency rules, and evidence traceability rules.

[0022] The relations in the relation set E can be represented in the form of triples: ; Among them, v i and v j 'r' represents two graph nodes; 'r' represents the type of relationship between them.

[0023] The set of relationships E between the nodes is used to characterize the normalization, coupling, cross-clause association, responsibility grouping, and validation constraints among the fields of the power supply contract. Specifically, field aliasing relationships address the issue of multiple representations of the same field in different contract texts; field coupling relationships constrain the joint extraction of fields such as electricity usage nature, electricity price category, metering method, and electricity billing method; cross-clause association relationships constrain the associated extraction of scattered fields such as contract term, renewal clauses, and modification / termination clauses; responsibility grouping relationships constrain the grouped extraction of responsibility boundary fields such as property demarcation points, maintenance responsibilities, and safety responsibilities; and field validation relationships constrain field value formats, units, date logic, mandatory status, and original evidence location.

[0024] In this embodiment, the large model can be selected from Qianwen, Wenxin Yiyan, or other large models.

[0025] S2. Extracting the structured framework of power supply contracts based on frame constraint paths: like Figure 3 As shown, the power supply contract framework constraint graph serves as an executable constraint source in the large model extraction control. It is used to generate framework constraint paths and control the field group extraction, result backfilling, missing data identification, conflict identification, and supplementary extraction processes through the framework constraint paths. This transforms the large model from open field extraction to phased relational extraction constrained by the field relationships of the power supply contract.

[0026] S21. Semantic Analysis of Contract Clauses: The power supply contract text is semantically segmented. Based on clause number, chapter title, paragraph level, and contract format characteristics, the power supply contract text is initially segmented to obtain candidate clause fragments. A large-scale model is then used to perform semantic recognition on these candidate clause fragments to determine their corresponding semantic tags. These semantic tags characterize the contract business category to which the candidate clause fragment belongs, including one or more of the following: contract parties, electricity usage information, power supply method and capacity, metering and billing, contract term, property rights and liabilities, modification / termination, and liability for breach of contract.

[0027] Adjacent candidate clause fragments are merged. During merging, it is determined whether adjacent fragments revolve around the same contractual business theme and whether they have a hierarchical relationship. Adjacent fragments belonging to the same business theme or having a subordinate relationship are merged into a single clause semantic unit. Each clause semantic unit includes the clause text, clause number, clause title, text location, clause semantic tag, and clause hierarchy information.

[0028] Based on the contract title, user type, voltage level, electricity usage nature, power supply method, capacity expression, and contract keywords, determine the contract type for electricity supply and consumption. Contract types include one of the following: high-voltage electricity supply and consumption contract, low-voltage residential electricity supply and consumption contract, low-voltage non-residential electricity supply and consumption contract, and temporary electricity supply contract.

[0029] This step converts the power supply contract text into a contract type. and the set of semantic units of clauses : ; Each clause semantic unit u i Represented as: ; in, Indicates the type of power supply contract; This represents a set of semantic units for terms; Indicates the first Each clause is a semantic unit; This indicates the text of the terms and conditions; Indicates the clause number; Indicates the title of the clause; Indicates text position; Indicates the semantic tags of the terms; This indicates the hierarchical information of the terms and conditions.

[0030] S22. Frame constraint path generation: Based on the contract type and the set of semantic units of the clauses, the applicable contract framework scope, clause modules, standard field sets, field alias relationships, field coupling relationships, cross-clause association relationships, responsibility grouping relationships, and field validation rules are matched in the power supply contract framework constraint graph. Based on the matched field nodes and relationship edges, a set of framework constraint paths corresponding to the current contract is generated.

[0031] The framework constraint path is used to represent the organizational relationship and extraction order between fields in a certain type of contract business information. Each framework constraint path is represented as follows: ; in, Indicates the first Path constraints within a framework; to This represents clause nodes, field nodes, field alias nodes, or rule nodes in the path; to This indicates the relationship type between adjacent nodes, including fields, field coupling, cross-clause association, responsibility grouping, alias normalization, and validation constraints.

[0032] Establish a correspondence between the generated framework constraint paths and the semantic units of the clauses: ; in, Represents the set of framework constraint paths; It represents the mapping relationship between the frame constraint path and the semantic unit of the clause, and is used to characterize the scope of the clause text corresponding to each frame constraint path.

[0033] This step transforms the static field relationships in the power supply contract framework constraint graph into framework constraint paths applicable to the current contract text, providing a path basis, field scope, and clause scope for subsequent relational extraction by field group in the large model.

[0034] S23. Relational Extraction from Large Models: Based on the framework constraint path set And the mapping relationship between framework constraint paths and clause semantic units Each framework constraint path is converted into a large model field group extraction task, and the large model is called to perform relational extraction within the corresponding clause text range.

[0035] For any frame constraint path According to the mapping relationship Determine the set of semantic units corresponding to the clauses: ; in, Indicates the relationship with the first Path of Constraints The corresponding set of semantic units of the clauses, This represents a set of semantic units for terms.

[0036] According to the framework constraint path The field nodes, field aliases, field relationship edges, and validation constraints in the clause semantic unit set are combined with the clause semantic unit set. For the corresponding clause text range, generate a field group extraction task: ; in, Indicates the first Extraction task for each field group; This represents the set of fields to be extracted. This represents a collection of field aliases; Indicates field relationship constraints; Indicates field validation constraints; This indicates the output format and evidence return requirements.

[0037] Field group extraction task Input a large model, and make the large model in the set of semantic units of terms. Within the defined semantic unit of the clause, according to the framework constraint path The defined field relationships are grouped for extraction, resulting in relational extraction results: ; in, Indicates the field name; Indicates the field value; Indicates the relationship between fields; Indicate the original text as evidence; Indicate the location of the original evidence; Indicates the first The number of fields in each field group extraction task.

[0038] S24. Path backfilling and dynamic replenishment: The relational extraction results are backfilled into the corresponding frame constraint paths, so that each frame constraint path is transformed from a state to be extracted into a path state with field values, field relationships, original evidence, and evidence location.

[0039] For any framework-constrained path, based on the field nodes and relation edges in that path, the field names, field values, field relationships, original evidence, and evidence locations extracted from the large model are filled into the corresponding path nodes. After backfilling, the path is status-determined, including complete state, missing field state, conflicting relation state, and insufficient evidence state.

[0040] Among them, the complete status indicates that all the necessary fields in the path have been filled in, and the relationship between the fields satisfies the field coupling relationship, cross-clause association relationship or responsibility grouping relationship in the power supply contract framework constraint diagram; the missing field status indicates that there are necessary fields in the path that have not been filled in; the relationship conflict status indicates that the relationship between the filled fields does not satisfy the path constraints; the insufficient evidence status indicates that the field value lacks corresponding original text evidence, or the evidence location cannot be located to the specific clause semantic unit.

[0041] When a path contains missing fields, conflicting relationships, or insufficient evidence, a supplementary extraction task, a relationship verification task, or an evidence supplementation task is generated based on the semantic unit range of the corresponding clause and the upstream and downstream relationships of the missing fields. The large model is then invoked again to perform targeted supplementary extraction or verification within the corresponding semantic unit range. The supplementary extraction results are then backfilled to the corresponding path nodes, and the path status is reassessed.

[0042] In one implementation, taking the "power supply method - voltage level - contract capacity" framework constraint path as an example, when the contract capacity field has been filled in but the voltage level field is missing, a voltage level supplement extraction task is generated based on the power supply condition coupling relationship between power supply method, voltage level and contract capacity, so that the large model can perform targeted supplement extraction within the power supply method and capacity clauses corresponding to this path.

[0043] By assembling the various framework constraint paths that meet the requirements of field integrity, field relationship consistency, and evidence traceability, a structured framework for power supply contracts is formed: ; ; Where F represents the structured framework of the power supply contract; F a Represents the a-th subframe; field b Indicates the field name; value b Represents the field value; rel b Indicates the relationship between fields; evi b Indicates original evidence; pos b Indicates the location of the evidence; m a This indicates the number of field items in the a-th sub-frame. The sub-frames include one or more of the following: contract subject frame, electricity information frame, power supply information frame, metering and billing frame, contract term frame, property rights and liability frame, and breach of contract liability frame.

[0044] S3. Feature extraction based on a structured framework: like Figure 4 As shown, based on the transformation of the state to be extracted into a set of frame constraint paths with field values, field relationships, original evidence, and evidence location, feature extraction is performed on the discrete frame basic data, which is then encapsulated to generate a standard-formatted structured data object for the value of power supply contracts. Specifically, the steps include: S31. Business Theme Evidence Aggregation and Alignment: After the operator traverses and backfills the set of frame constraint paths, multiple frame constraint paths involving the same core business theme are grouped into the same theme set according to the preset power core business classification rules.

[0045] The core power business classification rules include at least clustering the framework constraint paths in power supply contracts involving property boundary points, maintenance responsibilities, and safety responsibilities into a property responsibility business theme set; and extracting the relational extraction results corresponding to each framework constraint path within the theme set, wherein the relational extraction results include field values. Original evidence and the corresponding original text evidence location ; Extract original text evidence from all framework constraint paths within the aforementioned topic set By performing deduplication and spatiotemporal alignment, a global collection of evidence texts under this specific topic is generated. ,in This represents the total number of original textual evidence gathered under this topic. This indicates the first [topic] in the set of topics. A unique fragment of original evidence.

[0046] S32. Extraction of clause conflict probability features based on the association semantic matrix: For aggregated collections of evidence texts within the same business theme The semantic mutual exclusion between the evidence texts is calculated to quantify the logical conflict characteristics across clauses within the texts. Build size is Association semantic matrix The associated semantic matrix The representation of is as follows: ; Among them, matrix row index With column index Corresponding to the aforementioned evidence text sets The first in Original evidence With the Original evidence The main diagonal elements of the matrix (that is, when) The value of 0 (when) always indicates that the evidence itself does not have semantic conflicts; Any off-diagonal element in the matrix ( () indicates original evidence and The probability value of conflict between clauses is calculated using the following formula: ; in, and The evidence texts are respectively and The semantic feature vector is generated after processing by a preset lightweight text vectorization model; This represents the Euclidean distance between the semantic feature vectors; This is a mutual exclusion indicator function for electrical engineering terminology. If two texts simultaneously trigger a pre-defined pair of opposing terms (e.g., one statement contains "power supplier asset maintenance," and the other contains "power consumer asset maintenance"), then... ,otherwise ; This is the preset conflict penalty coefficient; For the association semantic matrix Aggregate the maximum risk of all upper triangular elements, extract and output the feature value of the probability of conflict of terms in the power supply contract. : ; Furthermore, when multiple systemic logical conflicts exist within the contract text, resulting in several high values ​​close to 1 (e.g., 0.98, 0.97, 0.96) appearing simultaneously in the upper triangular region of the association semantic matrix, the maximum risk aggregation operator employed in this invention can accurately pinpoint the core contradiction with the highest conflict intensity (i.e., 0.98) as a global representation indicator, ensuring the absolute triggering of the risk red light signal. Simultaneously, the association semantic matrix, as a bottom-level feature map, is fully preserved to support subsequent decision-making processes in fine-grained tracing and highlighting of multi-point chain conflicts, thus balancing the robustness of risk interception with the completeness of conflict verification.

[0047] S33. Determination of time-derived features based on temporal boundary difference: For the static fields involving the contract validity period in the structured framework of power supply contracts, a dynamic temporal difference operator and a ladder-style state mapping operator are constructed to transform them into lifecycle time-dependent features that evolve dynamically over time. The specific steps include the following: Extracting static deadline nodes and performing standard alignment, the operator reads the field value related to the contract validity period from the backfilled set of framework constraint paths corresponding to the power supply contract structured framework, and records it as the static deadline node. Obtain the current system standard execution time through the standard clock interface. .

[0048] To calculate the dynamic time-series difference days, the dynamic time-series difference operator is invoked to calculate the static cutoff time node. Compared with the current system standard execution time Dynamic time-series difference days Its mathematical formula is: ; in, (.) represents a day-level time-series difference function, used to quantify the time span from the current time to the contract expiration boundary. Later hour, It automatically converges to a negative value.

[0049] Perform a ladder-like state feature mapping and construct a ladder-like state mapping operator. The calculated dynamic time series difference days Mapped to time-derived feature status codes in standard format The mapping rule formula is as follows: ; in, , , and These represent the contract's expired status, urgent reminder status, near-expiration warning status, and normal performance status characteristics under the current clock slice, respectively. The time-derived characteristic status codes... With system standard execution time The dynamic state transition occurs as the contract progresses, serving as a dynamic high-order feature indicator characterizing the contract's time dimension.

[0050] S34. Generation and accumulation of valuable data objects; The operator performs the fusion and standardization of the multi-dimensional feature space, and outputs a global high-value feature entity to characterize the power supply contract. Specifically, this includes the following steps: Multidimensional feature space integrated alignment and stitching, the operator as a feature fusion module, extracts clause conflict probability feature values. and time-determined feature status codes And establish the characteristic value of the probability of conflict of the terms. Time-derived feature status codes It establishes topological relationships between the basic structured fields in the power supply contract structured framework and performs homogeneous alignment and cross-dimensional splicing of multidimensional heterogeneous features.

[0051] The high-level value data object is standardized, encapsulated, and deposited. The spliced ​​and aligned global feature fields are mapped to a preset standard data exchange structure, and encapsulated to generate a power supply contract value structured data object. The power supply contract value structured data object is persistently deposited into a preset contract feature database, which is used as a standardized feature input source for subsequent cross-system compliance verification and early warning decision flow, and completes the dynamic extraction of unstructured contract text into high-dimensional composite high-value data assets.

[0052] This embodiment utilizes a power supply contract framework constraint graph and framework constraint path control to control the large-scale model extraction process. This transforms the large-scale model from open-ended field extraction to relational extraction constrained by contract field relationships, reducing field omissions, incorrect field attribution, and missing cross-clause relationships. Through path status backfilling and dynamic supplementary extraction mechanisms, targeted supplementation is provided for extraction results with missing fields, conflicting relationships, and insufficient evidence, improving the completeness, consistency, and traceability of the structured framework of the power supply contract. Furthermore, based on the structured framework, clause conflict probability features and time-derived features are extracted, enabling the power supply contract text to be transformed into a computable, predictable, verifiable, and reusable value structured data object.

[0053] In a specific embodiment, the method, taking a power supply contract for a high-voltage industrial and commercial user as an example, includes: The contract text includes the following clauses: Article 3: Power Supply Method and Capacity: The power supplier will supply power to the user at 10kV, with a contracted capacity of 315kVA. Article 4: Metering and Billing: The electricity usage is for industrial and commercial purposes, subject to general industrial and commercial electricity pricing, using a high-voltage metering method, with electricity bills settled monthly. Article 6: Contract Term: This contract is valid from January 1, 2024 to December 31, 2026, and may be renewed by mutual agreement 30 days prior to expiration. Article 8: Property Rights and Responsibilities: The property rights demarcation point is the outgoing side of the metering device. The user is responsible for the maintenance of equipment below the demarcation point and bears corresponding safety responsibilities.

[0054] Constructing a constraint diagram for the power supply contract framework: Based on the business rules of power supply and consumption contracts, a framework constraint graph G is constructed. In this graph, a contract type node "High Voltage Power Supply and Consumption Contract" is set, and clause nodes "Power Supply Method and Capacity", "Metering and Billing", "Contract Term" and "Property Rights" are set. Field nodes "Voltage Level", "Contract Capacity", "Electricity Use Nature", "Electricity Price Category", "Metering Method", "Electricity Billing Method", "Contract Termination Date", "Renewal Clause", "Property Boundary Point", "Maintenance Responsibility", and "Safety Responsibility" are set.

[0055] At the same time, set field relationship edges in the graph, such as: from power supply method to voltage level to contract capacity; from electricity usage nature to electricity price category to metering method to electricity billing method; from contract termination date to renewal clause; from property boundary point to maintenance responsibility to safety responsibility.

[0056] Structured frame extraction based on frame constraint paths: First, the contract text is semantically analyzed to obtain the contract type τ = high-voltage power supply contract, and the set of semantic units U of the clauses is obtained. For example, "Article 3" is marked as the power supply method and capacity clause, "Article 4" is marked as the metering and billing clause, "Article 6" is marked as the contract term clause, and "Article 8" is marked as the property rights liability clause.

[0057] Based on the contract type and clause semantic units, generate a set P of frame constraint paths in the frame constraint graph, and establish the correspondence Ω between the paths and clause semantic units. For example: P1: From power supply method to voltage level to contract capacity, corresponding to item 3; P2: From electricity usage type to electricity price category to metering method to electricity billing method, corresponding to item 4; P3: Contract termination date and renewal clause, corresponding to Article 6; P4: From the property rights demarcation point to maintenance responsibility and then to safety responsibility, corresponding to Article 8.

[0058] Subsequently, the large model was invoked to extract relationships along each path, resulting in the following structured framework: Power Supply Information Framework: Voltage Level = 10kV, Contract Capacity = 315kVA, Evidence Location = Article 3; Metering and Billing Framework: Electricity Use Type = Industrial and Commercial Electricity Use, Electricity Price Category = General Industrial and Commercial Electricity Price, Metering Method = High-Level Supply and Metering, Electricity Billing Settlement Method = Monthly Settlement, Evidence Location = Article 4; Contract Term Framework: Contract Start Date = January 1, 2024, Contract Termination Date = December 31, 2026, Renewal Clause = Renewal negotiated by both parties 30 days before expiration, Evidence Location = Article 6; Property Rights and Responsibilities Framework: Property Boundary Point = Outgoing Line Side of Metering Device, Maintenance Responsibility = The electricity user is responsible for maintaining equipment below the boundary point, Safety Responsibility = The electricity user assumes corresponding safety responsibilities, Evidence Location = Article 8.

[0059] If a field is missing in a certain path, such as "safety responsibility" not being extracted, a supplementary extraction task will be generated based on the responsibility grouping relationship from "ownership demarcation point to maintenance responsibility to safety responsibility", and targeted supplementary extraction will be performed again within the scope of Article 8.

[0060] Feature extraction based on a structured framework: First, based on the structured framework F of the power supply contract, the field values, original evidence, and evidence locations in each framework constraint path are extracted and aggregated according to business themes. For example, the path from "ownership demarcation point to maintenance responsibility to safety responsibility" is categorized under the property rights responsibility theme, and the path from "contract termination date to renewal clause" is categorized under the contract term theme.

[0061] For the topic of property rights and liability, the system extracts original text evidence under that topic. For example, the extracted original text evidence includes: evi1: the property rights boundary point is the outlet side of the metering device; evi2: the electricity user is responsible for maintaining the equipment below the boundary point; and evi3: the electricity user bears the corresponding safety responsibility. After deduplication and alignment of the above evidence, a set of evidence texts under the topic of property rights and liability is formed, E={evi1, evi2, evi3}.

[0062] Subsequently, an association semantic matrix M is constructed based on the evidence text set E, and the probability of clause conflict between the evidence texts is calculated. Since evi1, evi2, and evi3 all revolve around the same property rights and responsibility boundary, and there are no mutually exclusive expressions such as "the power supplier is responsible for maintenance" and "the power consumer is responsible for maintenance", the corresponding clause conflict probability is low, and the output clause conflict probability feature Pconflict=0.08.

[0063] Regarding the contract term, the system reads the contract termination date Tend = December 31, 2026, and obtains the current system standard execution time Tcurrent. When Tcurrent = July 1, 2026, the system calls the dynamic time-series difference operator to calculate the dynamic time-series difference in days between the contract termination date and the current system time, obtaining Δt = 183, and maps it to the normal performance state Stime = CODE_NORMAL according to the preset step-by-step state mapping rules.

[0064] If the current system standard execution time changes to Tcurrent = December 10, 2026, the dynamic time series difference days will be Δt = 21. Since the contract includes a renewal clause stipulating that the contract can be renewed 30 days before expiration, this time-derived feature is mapped to an impending expiration warning status Stime = CODE_WARN.

[0065] Finally, the clause conflict probability feature Pconflict and the statute of limitations derived feature status code Stime are aligned and concatenated with the basic fields, field relationships, original evidence, and evidence location in the power supply contract structured framework F to generate a structured data object representing the value of the power supply contract. This object includes the contract's basic fields, clause conflict probability feature, remaining valid days of the contract, statute of limitations status code, source field, original evidence, and evidence location.

[0066] Through this step, the structured framework of the power supply contract is further transformed into a value structured data object containing characteristics of the probability of clause conflicts and the characteristics derived from the time limit.

[0067] Example 2: This embodiment provides a system for extracting features and frames from unstructured power data, including: The constraint graph construction module is configured to: construct a constraint graph of the power supply contract framework based on the obtained power supply contract text data, using the relationship set of nodes as field alias relationship, field coupling relationship, cross-clause association relationship, responsibility grouping relationship and field verification relationship; The framework constraint path generation module is configured to: extract and generate framework constraint paths that represent the organizational relationships and extraction order between fields in the contract business information based on the power supply contract framework constraint graph; The extraction module is configured to: convert the framework constraint path into a field group extraction task based on the mapping relationship between the framework constraint path and the semantic unit of the clause; and perform relational extraction within the corresponding clause text based on the field group extraction task. The backfilling module is configured to: backfill the relational extraction results into the corresponding frame constraint paths, and transform them into a backfilled set of frame constraint paths with field values, field relationships, original evidence, and evidence locations; The feature extraction module is configured to extract features from the backfilled set of framework constraint paths to obtain a structured data object of power supply contract value.

[0068] The working method of the system is the same as that of the power unstructured data feature and framework extraction method in Example 1, and will not be repeated here.

[0069] Example 3: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power unstructured data feature and framework extraction method described in Embodiment 1.

[0070] Example 4: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the steps of the power unstructured data feature and framework extraction method described in Embodiment 1.

[0071] Example 5: This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the power unstructured data feature and framework extraction method described in Embodiment 1.

[0072] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for feature and frame extraction of unstructured power data, characterized in that, include: Based on the obtained power supply contract text data, a power supply contract framework constraint graph is constructed using the relationship set of nodes, which includes field alias relationships, field coupling relationships, cross-clause association relationships, responsibility grouping relationships, and field validation relationships. Based on the power supply contract framework constraint graph, extraction control generates a framework constraint path to represent the organizational relationship and extraction order between fields in the contract business information; Based on the mapping relationship between the framework constraint path and the semantic unit of the clause, the framework constraint path is converted into a field group extraction task; relational extraction is then performed within the corresponding clause text based on the field group extraction task. The relational extraction results are backfilled into the corresponding frame constraint paths, resulting in a backfilled set of frame constraint paths with field values, field relationships, original evidence, and evidence locations. Based on the set of constraint paths in the backfilled framework, feature extraction is performed to obtain a structured data object of power supply contract value.

2. The method for extracting features and frames from unstructured power data as described in claim 1, characterized in that, The set of nodes in the constraint graph of the power supply contract framework includes contract type nodes, clause semantic nodes, standard field nodes, field alias nodes, and rule nodes. The set of node attributes includes field name, field type, field unit, field enumeration value, whether it is required, the type of contract it belongs to, the semantics of the clause it belongs to, and the evidence requirements. The set of constraint rules includes field integrity rules, field format rules, unit rules, date rules, enumeration value rules, field relationship consistency rules, and evidence traceability rules.

3. The method for extracting features and frames from unstructured power data as described in claim 1, characterized in that, The generation of the framework constraint path includes: matching the applicable contract framework scope, clause modules, standard field sets, field alias relationships, field coupling relationships, cross-clause association relationships, responsibility grouping relationships, and field validation rules in the power supply contract framework constraint graph to obtain field nodes and relationship edges, thereby generating the framework constraint path set corresponding to the current contract.

4. The method for extracting features and frames from unstructured power data as described in claim 1, characterized in that, The relational extraction includes: generating a field group extraction task based on the field nodes, field aliases, field relationship edges, and validation constraints in the framework constraint path, combined with the clause text range corresponding to the clause semantic unit set; and performing group extraction according to the field relationships defined by the framework constraint path within the clause semantic unit range defined by the field group extraction task, based on the field group extraction task, to obtain the relational extraction result. : ； in, Indicates the field name; Indicates the field value; Indicates the relationship between fields; Indicate the original text as evidence; Indicate the location of the original evidence; Indicates the first The number of fields in each field group extraction task.

5. The method for extracting features and frames from unstructured power data as described in claim 1, characterized in that, The generation of the structured data object of the power supply contract value includes: traversing the backfilled set of framework constraint paths and grouping multiple framework constraint paths involving the same business theme into the same business theme; determining the degree of semantic mutual exclusion between evidence texts for the set of evidence texts within the same business theme, in order to quantify the logical conflict characteristics across clauses within the text; and constructing dynamic temporal difference operators and ladder-style state mapping operators for static fields involving the contract validity period in the structured framework of the power supply contract, transforming them into lifecycle timeliness characteristics that evolve dynamically over time.

6. The method for extracting features and frames from unstructured power data as described in claim 5, characterized in that, Build size is Association semantic matrix : ； Among them, the main diagonal elements This indicates that the evidence itself does not have semantic conflicts; off-diagonal elements Indicates the probability of clause conflicts between the original evidence: ； in, and The evidence texts are respectively Original evidence With the Original evidence The semantic feature vector generated after processing; This represents the Euclidean distance between the semantic feature vectors; This is a mutual exclusion indicator function for electrical engineering terminology; This is the preset conflict penalty coefficient; For the association semantic matrix By aggregating the maximum risk of all upper triangular elements, the probability feature value of conflict of terms in the power supply contract is extracted. : 。 7. The method for extracting features and frames from unstructured power data as described in claim 5, characterized in that, Extract the static cutoff time nodes and perform standard alignment; determine the static cutoff time nodes using dynamic time-series difference operators. Compared with the current system standard execution time Dynamic time-series difference days : ； in, (.) represents a day-level time-series difference function, used to quantify the time span from the current time to the contract expiration boundary; a ladder-style state mapping operator is constructed to map the dynamic time-series difference days into a standard format time-derived feature status code; the time-derived feature status code undergoes dynamic state transitions as the system standard execution time advances, serving as a dynamic high-order feature index characterizing the contract time dimension.

8. A system for extracting features and frames from unstructured power data, characterized in that: include: The constraint graph construction module is configured to: construct a constraint graph of the power supply contract framework based on the obtained power supply contract text data, using the relationship set of nodes as field alias relationship, field coupling relationship, cross-clause association relationship, responsibility grouping relationship and field verification relationship; The framework constraint path generation module is configured to: extract and generate framework constraint paths that represent the organizational relationships and extraction order between fields in the contract business information based on the power supply contract framework constraint graph; The extraction module is configured to: convert the framework constraint path into a field group extraction task based on the mapping relationship between the framework constraint path and the semantic unit of the clause; and perform relational extraction within the corresponding clause text based on the field group extraction task. The backfilling module is configured to: backfill the relational extraction results into the corresponding frame constraint paths, and transform them into a backfilled set of frame constraint paths with field values, field relationships, original evidence, and evidence locations; The feature extraction module is configured to extract features from the backfilled set of framework constraint paths to obtain a structured data object of power supply contract value.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the program, it implements the steps of the power unstructured data feature and framework extraction method as described in any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the method for extracting features and frames of unstructured power data as described in any one of claims 1-7.

Images