Power transmission and transformation project design document intelligent analysis system based on multistage preprocessing

Abstract

The invention relates to a power transmission and transformation project design document intelligent analysis system based on multistage preprocessing, and belongs to the technical field of power transmission and transformation project document processing. Wherein the engineering feature enhancement preprocessing unit outputs a structured image set with attribute marks through pixel-level layered deconstruction, cross-modal feature alignment and symbol-oriented repair; the double-domain collaborative intelligent recognition unit depends on a professional semantic enhancement framework, and outputs a recognition text with professional confidence in combination with standard corpus constraint and dynamic confusion set correction; the specification-driven deep analysis unit generates a structured data set with a specification reference mark through text-specification bidirectional mapping and parameter association reasoning based on a design specification knowledge tree; and the engineering scene adaptive storage unit constructs an engineering analysis library supporting standard traceability and time sequence query by means of an engineering node time sequence data model, version evolution increment fusion and document fragment time sequence anchoring. And intelligent analysis of the power transmission and transformation project design document from feature extraction to storage application is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power transmission and transformation engineering document processing, and particularly relates to a power transmission and transformation engineering design document intelligent analysis system based on multi-stage preprocessing. BACKGROUND

[0002] The power transmission and transformation engineering design document is a core technical basis throughout the whole life cycle of engineering planning, equipment selection, construction, and completion acceptance, and its content covers multi-modal information such as electrical parameter text, wiring topology chart, and special equipment symbol, and needs to strictly match industry standards such as overhead transmission line design specification and substation design technical regulation, and the analysis quality directly determines the safety, compliance and efficiency of engineering construction.

[0003] However, the current power transmission and transformation engineering design document analysis technology still stays at the traditional data extraction level, and has not formed a whole-process intelligent scheme of multi-modal fusion-specialty recognition-specification verification-time sequence management, and in actual engineering application, many technical bottlenecks are exposed, and it is difficult to meet the fine engineering management demand.

[0004] Firstly, the multi-modal feature processing capability is weak, which leads to the fracture of parameter-equipment-scene correlation. The traditional analysis technology takes basic OCR as the core, and can only extract the text information in the document, ignoring the spatial correlation logic among text, chart and symbol - for example, the scatteredly labeled equipment parameters and corresponding equipment symbols in the design drawing are often mismatched due to lack of spatial mapping, and the rework problem of parameter and equipment model mismatch often occurs in subsequent construction; for the documents with mixed layout of vector graphics and raster images, the traditional method does not perform cross-modal spatial calibration, which directly affects the symbol recognition accuracy; for the worn and incomplete special symbols, the existing technology relies on manual page-by-page repair, and the repair result is affected by the professional level of personnel, and has poor consistency, which cannot meet the large-scale engineering document processing demand.

[0005] Secondly, the specialty recognition accuracy and specification adaptability are insufficient, and the error transmission risk is high. There are a large number of equipment symbols with similar appearance and professional terms with similar semantics in the power transmission and transformation field, the traditional recognition adopts simple template matching algorithm, and does not combine the topological connection rules and specification constraints of power transmission and transformation engineering; and the recognition process lacks dynamic correction mechanism, once the symbol or term is misjudged, the error will be directly transmitted to the subsequent parameter verification and equipment selection link - for example, misjudging "load switch" as "isolating switch" will cause protection parameter setting deviation and bury the line overload hidden danger; at the same time, the traditional recognition result does not carry professional confidence score, and the engineering personnel cannot quickly judge the data reliability, which seriously reduces the work efficiency.

[0006] Furthermore, the specification association and parameter reasoning capability are missing, and it is difficult to meet the engineering compliance requirements. The design of power transmission and transformation projects needs to strictly follow hundreds of industry specification clauses. The traditional analysis technology only performs literal extraction on the text of the document, does not establish a bidirectional association mechanism between the document text and the design specification, and cannot automatically check whether the text content meets the specification constraints. In the parameter checking link, the existing technology only checks the value range of the isolated parameters of a single device, ignores the hierarchical association logic and cross-loop association logic of the power transmission and transformation project, and causes parameter combination contradictions, for example, the problem that the rated current of the main device is less than the total of the branch load current cannot be identified, which easily causes device overheating failure.

[0007] Finally, the engineering timing and version management mechanism are backward, the data traceability and iteration efficiency are low. The construction period of the power transmission and transformation project is as long as several months to several years, and the design document needs to be iterated for dozens of times along with the project progress. The traditional method adopts full replacement type version update, does not distinguish between structural increment and parameter increment, and needs to process the complete document again each time. The time consumption of large-scale document update can be up to several hours. At the same time, the document fragments are not anchored in time with the engineering nodes, and manual screening is required in multiple version documents when inquiring. Moreover, the version change lacks complete trajectory record, the source of the incremental data and the correction basis are unknown, and the corresponding design clauses cannot be located when tracing the specification, so it is difficult to trace back the engineering quality problems.

[0008] In summary, the existing power transmission and transformation engineering design document analysis technology has significant limitations in the core links such as multi-modal feature processing, professional specification adaptation, parameter association reasoning, and timing version management, resulting in low analysis efficiency, high error rate, and poor engineering adaptability. It is difficult to support the fine and intelligent management needs of the power transmission and transformation project, and an intelligent analysis scheme that integrates multi-level preprocessing, professional specification driving, and timing dynamic management is urgently needed to break through the existing technical bottlenecks SUMMARY

[0009] To solve the above problems in the prior art, the present application provides an intelligent analysis system for power transmission and transformation engineering design documents based on multi-level preprocessing; The object of the present application can be achieved by the following technical solutions: An intelligent analysis system for power transmission and transformation engineering design documents based on multi-level preprocessing, comprising: an engineering feature enhancement preprocessing unit, a dual-domain collaborative intelligent identification unit, a specification-driven deep analysis unit, and an engineering scene adaptation storage unit; The engineering feature enhancement preprocessing unit: performs pixel-level hierarchical decomposition through a power transmission and transformation document feature perception network, identifies the spatial association of text, charts, and device symbols; simultaneously performs cross-modal feature alignment operation, calibrates the features of multiple types of elements in the spatial dimension; combines the symbol-oriented repair module to complete the engineering features in the fuzzy area, and outputs a structured image set with attribute labels; Dual-domain collaborative intelligent recognition unit: based on power transmission and transformation professional semantic enhancement recognition framework, through engineering symbol topology extraction subnetwork to obtain device identification and wiring symbol spatial features, call standard corpus constraint subnetwork for context reasoning; add dynamic confusion set correction operation, generate real-time correction rules for symbols prone to confusion, output recognition text with professional confidence; Standard-driven deep analysis unit: based on power transmission and transformation design specification knowledge tree, cross-chapter semantic completion is performed through text-standard bidirectional mapping mechanism; parameter association reasoning operation is executed, and the internal association of power transmission and transformation design is used to verify the extraction result, and a structured data set with specification reference marks is generated; Engineering scene adaptation storage unit: based on engineering node time sequence data model, the analysis result is mapped to each professional scene dimension; through the version evolution incremental fusion module, the database is updated, the parameter position change in version iteration is recorded by adding document segment time sequence anchoring operation, and the engineering analysis library supporting specification tracing and time sequence query is output.

[0010] As a preferred technical scheme of the present application, the power transmission and transformation document feature perception network has the following specific process: The multi-modal input adaptation layer performs pixel-level normalization on the vector graphics, raster images and mixed layout formats of the design document, and marks the high information density area through the symbol density mapping mechanism of the power transmission and transformation document; The engineering feature extraction subnetwork adopts a three-layer progressive convolution structure, the first layer extracts basic character edge features, the middle layer identifies the spatial connection relationship of special symbols such as circuit breakers and transformers through symbol topology perception convolution kernel, and the top layer generates a feature atlas in combination with the symbol combination rules in the power transmission and transformation specification; The cross-region association reasoning module establishes a spatial association weight matrix of text blocks and chart regions based on the inherent layout rules of devices, lines and parameters in design drawings, and performs feature binding on scattered parameter annotations and corresponding device symbols; By comparing the typical layout templates of power transmission and transformation documents, the feature coordinates of inclined and deformed areas are corrected, and a structured image feature set with modal labels and spatial correlation degrees is output.

[0011] Specifically, the cross-modal feature alignment operation has the following specific process: Modal feature initialization: for the text modal, extract parameter keyword-position coordinate feature pairs; for the chart modal, extract device symbol outline-topology coordinate feature pairs; for the symbol modal, extract special identification coding-pixel coordinate feature pairs, and perform feature deconstruction of the core modal of the power transmission and transformation document; Construct an alignment benchmark, based on the layout rules of device symbols, parameter annotations and chart regions in power transmission and transformation engineering documents, and based on the geometric center of the device symbol as the origin, an polar coordinate alignment coordinate system is established, and the typical distribution angle range of parameter text in the coordinate system is set. The association calculation is bound to the correction, through a modal association score algorithm, combining the matching rules of parameters and equipment in the power transmission and transformation specification, calculating the distance deviation and semantic matching degree of different modal features in polar coordinates, generating alignment confidence; for the feature pairs with confidence lower than the threshold, calling the power transmission and transformation document layout template library for offset compensation, and binding the aligned multi-modal features as the association unit of equipment-parameter-position.

[0012] Specifically, the symbol-oriented repair module comprises: Based on the special symbol feature library of power transmission and transformation, the topological structure parameters, typical edge features and standard drawing standards of core equipment symbols are stored, and the mapping relationship between symbol damage mode and repair scheme is established by analyzing the wear and tear and incomplete samples of symbols in historical design documents; A pre-set hierarchical repair engine is used for engineering symbol damage mode identification, edge feature residue extraction of damaged parts, and contour completion of geometric features of damaged symbols through feature library matching; at the same time, key details such as contacts and terminal connections are supplemented based on the power transmission and transformation symbol drawing specification; The first recheck is performed through the matching of symbol type and adjacent electrical parameter engineering attributes; the second recheck is performed in combination with the power transmission and transformation drawing layout specification; the secondary repair is performed based on the abnormal items calling the typical forms of similar symbols in the feature library, and the complete symbol conforming to the power transmission and transformation engineering design specification is output.

[0013] Specifically, the spatial features of equipment identification and connection symbols are obtained through the engineering symbol topology extraction subnetwork, and the specific process is as follows: The symbol area is accurately positioned, the typical size range and line features of the power transmission and transformation symbol are used, the scale sliding window traversal structure is used, and the feature filtering mechanism is used to exclude the pixel area of equipment identification and connection symbols, and the symbol mask with a bounding box is generated; The key connection points of the symbol are located through edge gradient change detection, including the center position of the device terminal and the intersection point of the line diameter; the contour features of the switch symbol are extracted through contour direction analysis, and the series and parallel topological association between symbols are identified based on the device layout and connection specification in the power transmission and transformation connection rule; The geometric parameters, connection point coordinates and relative position relationship with adjacent symbols of the symbol are converted into standardized quantitative feature values, and the topological feature vectors containing electrical connection attributes are generated according to the dimensions of equipment type, connection relationship and spatial parameters.

[0014] Specifically, the specification corpus constraint subnetwork comprises: A pre-set hierarchical corpus library of power transmission and transformation specification is used for structured storage through design stages and professional categories, and contains professional corpus of equipment parameter verification rules, symbol labeling specification and text expression paradigm, and a mapping index of corpus and engineering scene is established; The semantic constraint reasoning module parses the device type and parameter information in the recognized text, calls the specification corpus of the corresponding professional category to generate constraint conditions, and performs context consistency verification on the recognition result; Effective cases that pass the recognition in actual engineering are collected, expression forms that conform to the specification and are not included in the corpus are extracted, new expressions are classified and supplemented to the corresponding corpus through similarity calculation with the structure of the existing corpus; the mapping weight of the corpus and the scene is optimized based on the new cases to improve the adaptation accuracy of corpus calling in different design scenarios.

[0015] Specifically, the new dynamic confusion set correction operation includes: First, a confusion set specific to the power transmission field is constructed. By analyzing historical recognition errors and field expert annotated confusing objects, device symbols with similar shapes and professional terms with similar semantics are classified and stored, and the core distinguishing features are recorded simultaneously. The visual features and semantic attributes of the current recognition object are extracted, and the feature similarity with each item in the confusion set is calculated. If the similarity exceeds a certain threshold, the correction mechanism is triggered, the core distinguishing standards of the device in the power transmission specification are called, and targeted correction rules are generated. The correction rules are applied to the current recognition result for accurate differentiation of symbols or terms. At the same time, the correction case is supplemented to the confusion set, and the distinguishing feature weight is updated through incremental learning for dynamic iteration of the confusion set.

[0016] Specifically, the text-specification bidirectional mapping mechanism includes: A mapping index library is constructed to extract the core features in the text segments of the power transmission documents and analyze the key elements in the power transmission design specification items. The association index between the text segments and the specification items is established through feature hashing algorithm to form the basis of bidirectional mapping. The semantic splitting is performed on the recognized text to extract the device attributes and parameter information in the text. Based on the mapping index library, the corresponding specification items are matched, the consistency of professional terms is compared, the parameter values are checked against the specification thresholds, and the degree of fit between the text content and the specification requirements is determined. The text segments deviating from the specification are marked. Based on the specified specification items, the relevant text segments in the mapping index library are retrieved to check the complete coverage of the mandatory clauses and recommended requirements in the text. Supplementary prompts are generated for missing parameter descriptions and logical explanations. The consistency of the bidirectional mapping result is verified through the power transmission engineering design logic, and the matching relationship between the text and the specification and the deviation annotation are output.

[0017] Specifically, the execution parameter association reasoning operation includes: Perform structured extraction on the identified acquired device parameters and wiring parameters, organize exclusive parameter values through device type classification, sort out the adaptation requirements of parameters in loop connection based on wiring relationship, and form the corresponding relationship matrix of parameters and devices, parameters and wiring; Construct a parameter dynamic correlation reasoning model based on power transmission and transformation engineering principles: Through the hierarchical sequence of the power supply layer, busbar layer, main device layer, and branch layer, sort out the parameter transmission path level by level, and clarify the constraint rules and influence weights between levels; Based on cross-loop scenarios, define the quantitative calculation method and correlation strength threshold of parameter interaction influence; Through voltage level and wiring form classification, construct a scene rule set including basic parameter correlation logic and boundary conditions; Call the model to perform hierarchical verification on the structured parameter matrix: Through parameter attribute tag matching degree analysis, verify the internal consistency of single device parameters and functions; Through correlation dimension identification mapping, calculate the cooperative correlation degree of devices in the same loop and generate a quantitative score; Through hierarchical transmission path tracing, check the conduction adaptability of device parameters between upper and lower levels, and accurately locate the abnormal nodes of correlation.

[0018] Specifically, the engineering node time sequence data model: Based on the division of the whole life cycle of power transmission and transformation engineering into design, installation, debugging, and acceptance core nodes and corresponding process sub-nodes, a hierarchical structure is formed; Define the time sequence dimension of the node, establish the mapping relationship between the time axis and the node attribute; Connect the device parameters, wiring state, and detection results of each node based on time series, mark the mutation threshold and gradual change range of parameters when the node is converted, and set the adjacent node data connection, the same node parameter time sequence consistency, and the cross-node parameter trend correlation verification rules; At the same time, extract the time sequence data mapping to the time axis, identify abnormalities through rule verification, generate a time sequence correlation diagram, and show the parameter change trajectory and data transmission relationship between nodes as the project progresses.

[0019] Specifically, the version evolution increment fusion module: Construct an engineering version pedigree management architecture, establish a version sequence through timestamp and change identifier, record the generation time, triggering scenario, and associated version number of each version; Store the data of each version in layers according to the basic architecture layer-parameter layer-topology layer, extract the incremental data between versions, and distinguish between structural increments and parameter increments to provide structured input for incremental fusion; Establish a multi-dimensional incremental fusion mechanism, locate the difference areas between versions through feature hashing comparison: For structural increments, use topology alignment algorithm to match the device connection relationship of new and old versions, retain the common topology structure and embed new nodes; For parameter increments, perform weighted fusion of incremental values and historical values based on engineering logic weight, and simultaneously mark the influence range of parameter changes; Set incremental conflict detection rules, call the change initiation time and approval level to determine the priority; The integrated dynamic evolution adaptation function is based on the historical version fusion case to construct a change type-fusion strategy mapping model, based on the detected new change scene, to generate a candidate fusion scheme and select the optimal solution through engineering compliance verification; during the fusion process, the incremental source, processing rule and fusion basis are recorded in real time to form a version evolution track archive.

[0020] Specifically, the new document segment timing anchoring operation includes: Based on the power transmission and transformation engineering document, the segment structure is split, the core segments of equipment parameter description, wiring logic description and detection result record are extracted, engineering node tags and key time identifiers are added to each segment to form a segment library with timing attributes; An anchoring mapping mechanism of segments and engineering node timing is established, the corresponding engineering node is matched through segment tag matching, and the segment is accurately associated to the specific time point of the node timing axis combined with the key time identifier; anchoring verification is performed synchronously to check the time logic consistency and content correlation of the segment content and the corresponding node, and the anchoring abnormal segments are removed; A timing anchoring dynamic updating function is set, when the engineering node timing is adjusted or a new document segment is added, the association relationship between the segment and the node is automatically re-matched, the segment anchoring position on the timing axis is updated, and the anchoring change track is recorded.

[0021] The beneficial effects of the present application are: (1) Through the engineering feature enhanced preprocessing unit, the pixel-level hierarchical decomposition, cross-modal feature alignment and symbol-oriented repair are realized, the spatial correlation and complete repair of text, chart and device symbols are realized, the parameter and device mismatch is avoided, the repair consistency of incomplete symbols is greatly improved, and the distortion rate of feature extraction in inclined and deformed areas is reduced; combined with the dynamic confusion set correction of the double-domain collaborative intelligent recognition unit, the recognition error rate of similar symbol and similar semantic term is reduced, the recognition text with professional confidence can reduce the manual review time, and the analysis efficiency and reliability are improved.

[0022] (2) The specification compliance and engineering timing adaptability of the analysis result are guaranteed to support the fine management of power transmission and transformation engineering. The specification-driven deep analysis unit can automatically mark the text segments deviating from the design specification through the text-specification bidirectional mapping, accurately locate the parameter association abnormal node combined with the parameter association reasoning, avoid compliance omissions and device parameter combination contradictions; the version evolution incremental fusion of the engineering scene adaptation storage unit reduces the version update redundancy, the document segment timing anchoring shortens the parameter tracing time to minutes, and cooperates with the version evolution track archive to realize the controllable full process from specification verification to timing query of the analysis result, and provides reliable data support for engineering quality control and progress tracing. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to facilitate the understanding of those skilled in the art, the present application will be further described below with reference to the drawings.

[0024] Fig. 1 A framework diagram of a power transmission and transformation engineering design document intelligent analysis system based on multi-level preprocessing of the present application; Fig. 2 A timing diagram of a power transmission and transformation engineering design document intelligent analysis system based on multi-level preprocessing of the present application. DETAILED DESCRIPTION

[0025] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purposes, the specific embodiments, structures, features and effects according to the present application are described in detail below in combination with the drawings and preferred embodiments.

[0026] Please refer to Figs. 1-2 A power transmission and transformation engineering design document intelligent analysis system based on multi-level preprocessing includes: an engineering feature enhanced preprocessing unit, a dual-domain collaborative intelligent recognition unit, a specification driven deep analysis unit and an engineering scene adaptive storage unit. The engineering feature enhanced preprocessing unit: performs pixel-level hierarchical deconstruction through the power transmission and transformation document feature perception network, identifies the spatial correlation of text, charts and device symbols; synchronously performs cross-modal feature alignment operation, calibrates the spatial dimension of multiple types of element features; combines the symbol oriented repair module to complete the fuzzy area engineering feature, and outputs the structured image set with attribute labels; The dual-domain collaborative intelligent recognition unit: based on the power transmission and transformation professional semantic enhancement recognition framework, obtains the spatial features of device identification and wiring symbols through the engineering symbol topology extraction subnetwork, calls the specification corpus constraint subnetwork for context reasoning; adds a dynamic confusion set correction operation to generate real-time correction rules for symbols prone to confusion, and outputs the recognized text with professional confidence; The specification driven deep analysis unit: based on the power transmission and transformation design specification knowledge tree, performs cross-chapter semantic completion through the text-specification bidirectional mapping mechanism; performs parameter association reasoning operation, verifies the extraction result by using the internal association of power transmission and transformation design, and generates a structured data set with specification reference labels; The engineering scene adaptive storage unit: based on the engineering node time sequence data model, maps the analysis result to each professional scene dimension; updates the database through the version evolution incremental fusion module, adds the document segment time sequence anchoring operation to record the position change of parameters in version iteration, and outputs the engineering analysis library supporting specification tracing and time sequence query.

[0027] Specifically, the power transmission and transformation document feature perception network has the following specific process: The multi-modal input adaptation layer performs pixel-level normalization on the vector graphics, raster images and mixed layout formats of the design document, and marks the high information density area through the symbol density mapping mechanism of the power transmission and transformation document; The engineering feature extraction sub-network adopts a three-layer progressive convolution structure. The first layer extracts basic character edge features. The middle layer identifies the spatial connection relationship of special symbols such as circuit breakers and transformers through symbol topology perception convolution kernels. The top layer generates a feature atlas in combination with the symbol combination rules in the power transmission and transformation specification. The cross-region association reasoning module establishes a spatial association weight matrix of text blocks and chart regions based on the inherent layout rules of equipment-line-parameters in design drawings, and performs feature binding on scattered parameter annotations and corresponding equipment symbols. By comparing the typical layout templates of power transmission and transformation documents, the feature coordinates of inclined and deformed regions are corrected, and a structured image feature set with modal labels and spatial association degrees is output.

[0028] In this embodiment, the design documents of a new substation project are taken as the application scenario. The documents include a vector wiring diagram (containing equipment symbols such as main transformers and circuit breakers) exported by CAD software, a scanned raster version of the equipment parameter table (containing handwritten annotations), and a protection configuration description with mixed text and chart layout. The power transmission and transformation document feature perception network performs the following specific processing: In the multi-modal input adaptation layer, first, pixel-level normalization is performed on different formats of documents: the vector wiring diagram is converted to a 512x512 pixel matrix at a resolution of 150DPI, preserving the vector outline accuracy of equipment symbols; Gaussian filtering is used on the raster version of the parameter table to remove scanning noise, and binary processing is used to enhance the contrast between parameter text and background; for the mixed layout protection configuration description, a region segmentation algorithm (based on RGB color channel differences) is used to separate text blocks (black font) and chart areas (blue equipment symbols). Subsequently, the symbol density mapping algorithm is started, and the whole image is traversed with a 32x32 pixel sliding window to count the number of power transmission and transformation special symbols (such as circuit breaker "□" symbols and mutual inductor "○" symbols) in each window. Regions with a symbol density greater than or equal to 5 windows are marked as high information density regions - ultimately identifying the main transformer wiring area and the bus protection configuration area as two core areas, focusing on the target for subsequent key feature extraction.

[0029] The input image is normalized and then input into the engineering feature extraction subnetwork. The normalized image is processed by a three-layer progressive convolution structure: the first layer uses a 3x3 convolution kernel to extract basic character edge features, such as the character outline of "120MVA" in the parameter table and the straight edge of the conductor in the wiring diagram, and outputs an edge feature map; the middle layer loads a symbol topology perception convolution kernel designed specifically for power transmission and transformation equipment. For the topology feature of the circuit breaker symbol "circular contact + horizontal connection end", a 5x5 special convolution kernel is designed to identify the series connection relationship between the circuit breaker and the bus by calculating the convolution response value. Similarly, a 7x7 convolution kernel is used to identify the spatial topology of the transformer "three-phase winding + neutral point grounding". Finally, the connection logic of the key equipment is captured. The top layer combines the symbol combination rules in the "DL / T5445-2010 Design Technical Regulations for Electric Power System Communication Station" that "current transformers need to be arranged in series with circuit breakers, and voltage transformers need to be connected in parallel with buses". The topology features extracted by the middle layer are screened for compliance, and false connections that do not comply with the specifications (such as the misidentified series relationship between the voltage transformer and the circuit breaker) are removed to generate a three-dimensional feature map containing "device symbol-topology connection-specification label".

[0030] In the cross-region association reasoning module, based on the inherent layout rules of 220kV substation design drawings (device symbols are mostly distributed on the left side of the drawing, corresponding parameter labels are mostly within 5-10mm range on the right side of the symbol, and line parameters are mostly labeled on the middle section of the conductor), a 3x3 spatial association weight matrix is constructed: the pixel point where the device symbol is located is set as the center, 1-2 pixel columns on the right side are set as high weight area, the middle section of the conductor is set as medium weight area, and the remaining area is set as low weight area. For the parameters scattered around the main transformer symbol, "main transformer type S11-120" labeled on the left symbol area, "rated voltage 220 / 110kV" labeled on the right text block, and "short circuit impedance 10.5%" labeled on the middle section of the conductor, the spatial association degrees of the three with the main transformer symbol are calculated by the weight matrix, all of which are higher than the set threshold, feature binding is performed, and an associated unit of "main transformer symbol-multiple parameter set" is generated, solving the problem of disconnection between scattered parameters and device symbols in traditional analysis.

[0031] Finally, a coordinate correction operation is performed, a typical layout template of the main wiring diagram of the substation is called, a standard coordinate range containing the main transformer, bus, and circuit breaker is compared with the feature coordinates of the current processing image: it is found that the bus symbol is tilted away from the horizontal direction of the template, and the circuit breaker symbol coordinate is offset by 8 pixels. The offset matrix is calculated by the affine transformation algorithm, and the feature coordinates of the tilted and offset areas are dynamically corrected. Finally, a structured image feature set with modal labels (such as "text: main transformer rated voltage", "symbol: circuit breaker", and "chart: bus topology") and spatial association degrees is output. The feature set can be directly input into the dual-domain collaborative intelligent recognition unit later.

[0032] Specifically, the cross-modal feature alignment operation is performed, and the specific process is as follows: Modal feature initialization: extract parameter keyword-location coordinate feature pairs for text modalities, extract equipment symbol outline-topological coordinate feature pairs for chart modalities, extract special identifier code-pixel coordinate feature pairs for symbol modalities, and perform feature deconstruction of the core modalities of power transmission and transformation documents; To establish an alignment benchmark, based on the layout rules of equipment symbols, parameter annotations, and chart areas in power transmission and transformation engineering documents, a polar coordinate alignment coordinate system is established with the geometric center of the equipment symbol as the origin, and the typical distribution angle range of parameter text in the coordinate system is set. The correlation calculation and correction are linked. By using the modal correlation degree scoring algorithm and combining the matching rules of parameters and equipment in the power transmission and transformation specifications, the distance deviation and semantic matching degree of different modal features in polar coordinates are calculated to generate alignment confidence. For feature pairs with confidence scores below the threshold, the power transmission and transformation document layout template library is called to perform offset compensation, and the aligned multimodal features are bound as the correlation unit of equipment-parameter-location.

[0033] In this embodiment, the 110kV substation design document is used as the processing object. This document contains CAD vector wiring diagrams, scanned equipment parameter text, and special identifiers such as "CT" and "grounding ⊕". When performing cross-modal feature alignment operations, the process follows a coherent flow of "modal feature initialization - building alignment benchmark - association calculation and correction binding". The specific process is as follows: First, modal feature initialization is performed, and feature deconstruction is carried out on the three core modalities of the document: text, charts, and symbols. At the text modal level, a pre-trained BERT model for power transmission and transformation is used to identify parameter keywords, denoted as K_txt (e.g., "circuit breaker rated current" and "disconnector insulation level"). Then, combined with OCR text line segmentation algorithm and pixel coordinate localization technology, the top-left pixel coordinates P_txt(x,y) of each keyword are recorded, forming a set of text modal feature pairs: F_txt={(K_txt1,P_txt1(x1,y1)),...,(K_txtn,P_txtn(x n ,y n ))}, where n is the total number of text feature pairs, ensuring that each parameter keyword is bound to a spatial location. At the chart modality level, for equipment symbols such as circuit breakers and disconnectors in CAD wiring diagrams, the Canny edge detection algorithm (the threshold is adaptively determined by the maximum inter-class variance method) is used to extract the symbol contour features C_gra. At the same time, a topological coordinate system is established with the lower left corner of the drawing as the origin, and the topological coordinates T_gra(u,v) of the geometric center of each symbol are recorded, forming a set of chart modality feature pairs: F_gra={(C_gra1,T_gra1(u1,v1)),...,(C_gram,T_gram(u m ,vm ))}, wherein m is the total number of chart features, achieving spatial anchoring of the device symbols. At the symbol modality level, a pre-stored library of unique symbol encodings Lib_sym (containing 32 types of symbol unique encodings, such as "CT" corresponding to E_sym=CT-01, and ground ⊕ corresponding to E_sym=G-03) is called, matching the special identification in the document and generating the encoding E_sym, synchronously recording the pixel coordinates Pix_sym(p, q) of the identification, forming the symbol modality feature pair set F_sym={(E_sym1, Pix_sym1(p1, q1)),..., (E_symk, Pix_symk(pk, qk)}, wherein k is the total number of symbol features, completing the digital identification of the symbol. k k ))}, wherein k is the total number of symbol features, completing the digital identification of the symbol.

[0034] After completing the initialization of the modal features, an alignment reference is constructed based on the inherent layout rules of the power transmission and transformation engineering document. According to the requirement in "DL / T5210.6-2019 Electrical Construction Quality Acceptance Regulations" that "device parameter annotations should be close to the device and easy to read", combined with the layout features statistically analyzed from historical 110kV distribution station design documents, the geometric center of each device symbol is taken as an independent origin O i (i is the device number, such as O1 for circuit breaker and O2 for disconnector), a multi-origin polar coordinate alignment coordinate system is established to avoid alignment deviation of remote modal features caused by a single origin. At the same time, according to the conventional association relationship between parameter text, special identification and device symbol, the typical distribution range of each modality feature is set: the polar angle θ∈[θ1, θ2] (corresponding to the right side to the lower area of the symbol) and the polar radius r∈[r1, r2] of the text parameter, ensuring that the parameter and the symbol are not overlapped and readable; the polar angle θ∈[θ3, θ4] (corresponding to the area close to the device symbol) and the polar radius r≤r3 of the special identification, ensuring the direct association between the identification and the device body.

[0035] Subsequently, the association calculation and correction binding phase is entered, and a modality association degree scoring algorithm is used to evaluate the feature pair matching degree from two dimensions. In the spatial distance dimension, the polar angle deviation Δθ and the polar radius deviation Δr of the text and symbol modality features in the corresponding polar coordinates are calculated, and the distance score ω 距 (maximum score ω 总 ) is generated by matching the deviation value with the typical distribution range; in the semantic matching dimension, the power transmission and transformation specification semantic library such as "GB10963.1-2020 Household and Similar Use Circuit Breaker" is called to verify the functional matching of the parameters and the device (such as the matching of "rated current" with the circuit breaker and the matching of "insulation level" with the disconnector), and the semantic score ω 语 is generated. The two scores are combined according to the formula "confidence=ω 距 ×α+ω 语 ​Xb" calculation, wherein a, b are preset weight coefficients. Set a confidence threshold T, for feature pairs below the threshold (such as the "circuit breaker - rated current" feature pair), call the 110kV power distribution station document layout template library, match the parameter labeling templates of similar devices, calculate the coordinate offset compensation amount Ad, and correct the feature position (such as adjusting the pixel coordinates by horizontal offset Ad), so that the corrected feature polar coordinates return to the typical distribution range, at which time the recalculated confidence needs to satisfy "confidence >= T". Finally, the aligned device symbol features, parameter text features, and special identification features are bound to form a "device - parameter - position" association unit (such as "circuit breaker O1 - rated current K_txt1 - corrected coordinates P_txt1'(x1', y1')"), completing the precise association of cross-modal features.

[0036] Specifically, the symbol-oriented repair module comprises: Based on the power transmission and transformation special symbol feature library, the topological structure parameters, typical edge features and standard drawing standards of core device symbols are stored, and the mapping relationship between symbol damage patterns and repair schemes is established by analyzing the wear and tear samples of symbols in historical design documents; A preset hierarchical repair engine is used to identify the damaged shape of engineering symbols, extract the edge feature residues of damaged parts, and complete the contour of damaged symbols by matching the geometric features in the feature library; At the same time, the key details of the contacts and terminals are supplemented based on the power transmission and transformation symbol drawing standards; The first recheck is performed by matching the engineering properties of the symbol type and the adjacent electrical parameters; the second recheck is performed in combination with the power transmission and transformation paper layout specifications; the secondary repair is performed based on the abnormal items calling the typical forms of similar symbols in the feature library, and the complete symbol conforming to the power transmission and transformation engineering design specifications is output.

[0037] Specifically, the spatial features of the device identification and connection symbols are obtained by the engineering symbol topology extraction subnetwork, and the specific process is as follows: The symbol area is accurately positioned, the typical size range and line features of the power transmission and transformation symbols are used, the structured image is traversed using a scale sliding window, and the pixel area of the device identification and connection symbols is excluded by combining the feature filtering mechanism, and a symbol mask with a bounding box is generated; The key connection points of the symbol are located by edge gradient change detection, including the center position of the device terminal and the intersection of the line diameter; the contour features of the switch symbol are extracted by contour direction analysis, and the series and parallel topological associations between symbols are identified based on the device layout and connection specifications in the power transmission and transformation connection rules; The geometric parameters, connection point coordinates and relative position relationship with adjacent symbols of the symbol are converted into standardized quantitative feature values, and are integrated according to the dimensions of device type, connection relationship and space parameter to generate a topological feature vector containing electrical connection attributes.

[0038] Specifically, the specification corpus constraint subnetwork comprises: A preset power transmission and transformation specification hierarchical corpus is stored in a structured manner through a design stage and a professional category, and contains professional corpus of device parameter verification rules, symbol annotation specifications, and text expression paradigms, and a mapping index of the corpus and an engineering scene is established; A semantic constraint reasoning module is based on, device types and parameter information in the recognized text are parsed and recognized, constraint conditions are generated by calling the specification corpus of the corresponding professional category, and context consistency verification is performed on the recognized results; Effective cases that pass the recognition in actual engineering are collected, expression forms that conform to the specifications and are not included in the corpus are extracted, new expressions are classified and supplemented to the corresponding corpus through structure similarity calculation with existing corpus, and the mapping weight of the new cases is optimized to improve the adaptation accuracy of corpus calling in different design scenes.

[0039] Specifically, the new dynamic confusion set correction operation comprises: First, a confusion set exclusive to the power transmission and transformation field is constructed, similar device symbols and similar professional terms are classified and stored by analyzing historical recognition error cases and field expert annotated confusion objects, and core distinguishing features are recorded synchronously; The visual features and semantic attributes of the current recognition object are extracted, and the feature similarity of each item in the confusion set is calculated; if the similarity exceeds a set threshold, the correction mechanism is triggered, the core distinguishing standard of the device in the power transmission and transformation specification is called, and a targeted correction rule is generated; The correction rule is applied to the current recognition result to accurately distinguish the symbols or terms; meanwhile, the correction case is supplemented to the confusion set, the distinguishing feature weight is updated through incremental learning, and the confusion set is dynamically iterated.

[0040] Specifically, the text-specification bidirectional mapping mechanism comprises: A mapping index library is constructed, core features in the text segments of the power transmission and transformation documents are extracted, key elements in the specification items of the power transmission and transformation design are parsed, an association index of the text segments and the specification items is established through a feature hashing algorithm, and a bidirectional mapping basis is formed; The recognized text is subjected to semantic splitting, device attributes and parameter information in the text are extracted, corresponding specification items are matched based on the mapping index library, the consistency of professional terms is compared, parameter values are checked against specification thresholds, the degree of fit between the text content and the specification requirements is determined, and text segments deviating from the specification are marked; Based on the specified specification items, the relevant text fragments in the mapping index library are retrieved in reverse, the mandatory clauses and recommended requirements in the text complete coverage specification are checked, the missing parameter description and logical explanation are generated, the consistency of the two-way mapping results is checked through the power transmission and transformation engineering design logic, and the matching relationship and deviation annotation of the text and the specification are output.

[0041] In this embodiment, the main transformer design documents (including parameter table and selection explanation) of 220 kV substation and “GB / T6451-2015 Oil-immersed Power Transformer” and “DL / T5747-2010 220 kV Substation Design Technical Specification” are taken as processing objects, and two-way mapping is realized through the process of “mapping index library construction-semantic split matching-reverse retrieval verification”, and the specific process is as follows: When constructing the mapping index library, the core features are first extracted from the document text fragments: for the content of “main transformer type S11-120, rated capacity VA 额 , rated voltage V 额 , short-circuit impedance Z 阻 , cooling mode ONAN”, the structured features such as “device type: oil-immersed main transformer, voltage grade: V 额 , parameter set {VA 额 , Z 阻 , ONAN}” are disassembled; at the same time, the key elements of the specification items are analyzed, such as “220 kV level capacity range 50-240 MVA” “short-circuit impedance should be 10%-12%” “ONAN is suitable for capacity ≤100 MVA” in “GB / T6451-2015” and “top oil temperature rise ≤55K (mandatory)” in “DL / T5747-2010”. The feature hashing algorithm is adopted to generate hash value H1 for the document feature combination and hash value H2 for the specification element combination, to establish the association index of H1 and H2, to record the item number and clause type (mandatory / recommended) synchronously, and to form the mapping index library.

[0042] The semantic split is performed on the recognized text: through the power transmission professional NLP model, the main transformer selection “type S11-120, VA 额 , V 额 , Z 额 , cooling mode ONAN, no temperature rise requirement” is disassembled, and the device attributes (oil-immersed main transformer, V 额 ) and parameter information (VA 额 , Z 额 , ONAN) are extracted. After matching the corresponding specification items based on the index library, double verification is carried out: on the professional term level, “rated capacity” and “short-circuit impedance” are consistent with the specification description; on the parameter verification level, VA 额In the range of 50-240 MVA, Z exceeds the interval of 10%-12%, and ONAN is not suitable for the capacity main transformer, which does not meet the requirement of "≤100 MVA". The label "Z 额 The "cooling mode ONAN" is a deviation from the specification section, and the deviation type is marked respectively.

[0043] Based on the specified specification items, perform reverse search: select the mandatory clause "220kV oil-immersed main transformer top layer oil temperature rise ≤55K" in GB / T6451-2015 4.5.1, generate feature hash value H3 to search the index library, and no corresponding document fragment is matched. It is determined that the description is missing, and a supplementary prompt "Need to supplement the parameter description of the main transformer top layer oil temperature rise ≤55K (according to GB / T6451-2015 4.5.1)". Check the consistency of the bidirectional mapping results combined with the power transmission and transformation engineering design logic (such as capacity and cooling mode matching, impedance and system short-circuit current correlation), and finally output the matching relationship table and deviation list of the text and the specification (including Z 阻 Exceeding threshold, cooling mode mismatch, temperature rise description missing.

[0044] Specifically, the execution parameter association reasoning operation includes: Perform structured extraction on the identified and obtained device parameters and wiring parameters, sort the exclusive parameter values by device type classification, and clarify the adaptation requirements of parameters in the loop connection based on the wiring relationship, to form the corresponding relationship matrix of parameters and devices, and parameters and wiring; Construct a parameter dynamic association reasoning model based on the principle of power transmission and transformation: through the hierarchical sequence of power supply layer, busbar layer, main device layer, and branch layer, clarify the constraint rules and influence weight between levels; based on cross-loop scenarios, define the quantitative calculation method and correlation strength threshold of parameter interaction influence; through voltage level and wiring form classification, construct a scene rule set including basic parameter association logic and boundary conditions; Call the model to perform hierarchical verification on the structured parameter matrix: through parameter attribute label matching degree analysis, verify the internal consistency of single device parameters and functions; through association dimension identification mapping, calculate the cooperative correlation degree of devices in the same loop and generate a quantitative score; through hierarchical transmission path tracing, check the conduction adaptability of upper and lower level device parameters, and locate the associated abnormal nodes.

[0045] Specifically, the engineering node timing data model includes: Based on the whole life cycle of power transmission and transformation engineering, the design, installation, debugging, acceptance core nodes and corresponding process sub-nodes are divided to form a hierarchical structure; the time sequence dimension of the nodes is defined, the mapping relationship between the time axis and the node attributes is established; the device parameters, connection state and detection results of each node are connected based on time sequence, the mutation threshold and gradual change range of the parameters when the node is converted are marked, and the adjacent node data connection, the same node parameter time sequence consistency and the cross-node parameter trend correlation checking rules are set; at the same time, the time sequence data is extracted and mapped to the time axis, the abnormality is identified through rule checking, the time sequence correlation diagram is generated, and the change trajectory of the parameters with the progress of the project and the data transmission relationship between the nodes are displayed; Specifically, the version evolution increment fusion module comprises: An engineering version pedigree management architecture is constructed, a version sequence is established through a time stamp and a change identifier, the generation time, triggering scene and associated version number of each version are recorded, version data is stored in layers according to the basic architecture layer-parameter layer-topology layer, the incremental data between versions is extracted, and structural increments and parameter increments are distinguished to provide structured input for increment fusion; A multi-dimensional increment fusion mechanism is established, and the difference area between versions is located through feature hash comparison: for structural increments, a topology alignment algorithm is used to match the device connection relationship of new and old versions, the common topology structure is retained and the new nodes are embedded; for parameter increments, the weighted fusion of incremental values and historical values is performed based on engineering logic weight, and the influence range of parameter change is marked; the increment conflict detection rule is set, and the change initiation time and approval level are called to determine the priority; A dynamic evolution adaptation function is integrated, a change type-fusion strategy mapping model is constructed based on historical version fusion cases, based on the detected new change scene, a candidate fusion scheme is generated and the optimal solution is selected through engineering compliance verification; the increment source, processing rule and fusion basis are recorded in real time during the fusion process, and a version evolution track file is formed.

[0046] In this embodiment, the three key version iterations (V1.0→V2.0→V3.0) of the design document of a certain substation are taken as the actual application scene, the module solves the core pain points of high data redundancy rate, low fusion efficiency and poor adaptability to complex scenes in traditional "full replacement" version management through the innovative architecture of "pedigree hierarchical management-differential increment fusion-dynamic strategy adaptation", and the specific implementation process deeply lands the technical innovation points.

[0047] Firstly, the engineering version pedigree management architecture is constructed to establish a traceable and structured basis for version iteration. For three versions, a unique version sequence is generated through timestamp and change identification: V1.0 (initial design final draft, triggering scenario "project preliminary design review passed"), V2.0 (parameter optimization version, triggering scenario "main transformer capacity review and adjustment"), V3.0 (topology expansion version, triggering scenario "new reactive power compensation device special requirement"), and a tree-like pedigree is formed by associating version numbers (V2.0 is associated with parent version V1.0, V3.0 is associated with parent version V2.0) to ensure that the inheritance relationship between versions is clear. The core innovation lies in the hierarchical storage strategy of "basic architecture layer-parameter layer-topology layer": the basic architecture layer stores fixed core data such as substation voltage level and outgoing line loop number that do not change with version changes to avoid repeated storage; the parameter layer stores variable parameters by device type; and the topology layer records device connection relationships in the electrical adjacency matrix. By comparing the hash value differences of data in each layer through the feature hash algorithm, incremental data between versions is extracted: only parameter increment is generated from V1.0 to V2.0; and structural increment and parameter increment are contained from V2.0 to V3.0.

[0048] Then, a multi-dimensional incremental fusion mechanism is established, and different types of increments are processed differently, which is the core innovative link of the module. When processing the structural increment from V2.0 to V3.0, the topology alignment algorithm is adopted to break through the traditional inefficient way of redrawing topology: first, match the common topology structure in the new and old versions through the unique identification of the device to keep the core connection logic of the main transformer busbar-outgoing line switch unchanged; then, according to the specification requirement in "DL / T5747-2010 220kV Substation Design Technical Regulations" that "reactive power compensation devices should be connected in parallel to the busbar", calculate the optimal connection point coordinates of the new capacitor and the busbar based on the principle of uniform busbar load distribution, and embed the connection relationship of the new capacitor node and the supporting disconnecting switch into the original topology adjacency matrix to ensure that the fused topology not only retains the historical logic but also meets the electrical design specifications. When processing parameter increment, the engineering logic weight system is innovatively introduced to assign weights according to the impact of parameters on engineering safety and operation: safety-related parameters (such as main transformer insulation level and circuit breaker breaking current) are set to ω1, and ordinary operation parameters (such as the number of cooling fans and busbar color identification) are set to ω2; taking the main transformer capacity adjustment from V1.0 to V2.0 as an example, instead of directly replacing the historical value, the fusion value is calculated based on: incremental value × weight ω1 + historical value × weight ω2, and the impact range of the parameter change is automatically marked, including main transformer protection setting value re-tuning, busbar current-carrying capacity re-checking, cooling system pipe reconstruction, etc., providing clear guidance for subsequent engineering adjustment. For incremental conflict scenarios.

[0049] Finally, the dynamic evolution adaptation function is integrated, breaking through the limitation of traditional modules that can only handle regular scenarios, and realizing intelligent response to new change scenarios. The module first fuses cases based on historical power transmission and transformation engineering document versions, constructs a "change type-fusion strategy" mapping model, covers 12 types of regular scenarios such as "single device parameter adjustment-weighted fusion" and "single loop topology addition-node embedding", and ensures efficient processing of regular iterations. When a new change scenario is detected, such as V3.0 needing to be merged with the design document of another substation across voltage levels, the model automatically analyzes the scene characteristics (across voltage levels, topology independent but needs to be associated), generates candidate fusion schemes. Then call the mandatory clause in "Power System Design Technical Regulations" that "cross-voltage level system connection should be set up with tie-in transformers to ensure voltage matching and fault isolation", perform engineering compliance check on the candidate scheme, eliminate schemes that do not comply with the specification, and further optimize the parameter matching rules of the tie-in transformer. During the entire fusion process, the module records key information in real time: incremental source, processing rules (topology alignment algorithm version V2.1, engineering logic weight system V1.0), fusion basis (specific specification clause number and review opinion number), forming a complete version evolution track archive. The archive supports retrospective query of any version (such as tracing the approval process and calculation basis of V2.0 main transformer capacity adjustment), and can also perform incremental contribution analysis to provide data support for engineering iteration optimization.

[0050] Specifically, the new document segment time sequence anchoring operation: Based on the structured splitting of power transmission and transformation engineering documents, the core segments of device parameter description, wiring logic description, and test result record are extracted, engineering node labels and key time identifiers are added to each segment, and a segment library with time sequence attributes is formed; An anchoring mapping mechanism is established between segments and engineering node time sequences, which matches corresponding engineering nodes through segment labels and associates segments to specific time points on the node time sequence axis through key time identifiers; Anchoring verification is performed synchronously to check the consistency of segment content and corresponding node time logic, content association, and eliminate anchoring abnormal segments; Set up dynamic update function of time sequence anchoring, when the engineering node time sequence is adjusted or new document segments are added, automatically re-match the association between segments and nodes, update the anchoring position of segments on the time sequence axis, and record the anchoring change track.

[0051] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed with the preferred embodiments as above, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, as long as the changes or modifications do not deviate from the technical solution of the present application. Any modification, change, equivalent change and modification of the above embodiments made according to the technical essence of the present application still belong to the scope of the technical solution of the present application.

Claims

1. A multi-stage preprocessing-based power transmission project design document intelligent analysis system, characterized in that, Comprise: Engineering feature enhancement preprocessing unit, dual domain collaborative intelligent recognition unit, specification driven deep analysis unit and engineering scene adaptive storage unit; Engineering feature enhancement preprocessing unit: execute pixel-level hierarchical deconstruction through power transmission and transformation document feature perception network, identify the spatial correlation of text, chart and device symbols; simultaneously perform cross-modal feature alignment operation, calibrate multi-type element features in spatial dimension; combine symbol-oriented repair module to complete fuzzy area engineering features, output structured image set with attribute labels; Dual domain collaborative intelligent recognition unit: based on power transmission and transformation professional semantic enhancement recognition framework, obtain spatial features of device identification and wiring symbols through engineering symbol topology extraction subnetwork, call specification corpus constraint subnetwork for context reasoning; Add dynamic confusion set correction operation, generate real-time correction rules for symbols prone to confusion, output recognized text with professional confidence; Specification driven deep analysis unit: based on power transmission and transformation design specification knowledge tree, perform cross-chapter semantic completion through text-specification bidirectional mapping mechanism; Perform parameter association reasoning operation, verify extraction results using the inherent association of power transmission and transformation design, generate structured data set with specification reference labels; Engineering scene adaptive storage unit: based on engineering node time sequence data model, map analysis results to each professional scene dimension; Through version evolution incremental fusion module, update database, add document segment time sequence anchoring operation to record parameter position changes in version iteration, output engineering analysis library supporting specification tracing and time sequence query.

2. The system according to claim 1, characterized in that The power transmission and transformation document feature perception network specifically comprises: Multi-modal input adaptation layer, performing pixel-level normalization on vector graphics, raster images and mixed layout formats of design documents, marking high information density areas through power transmission and transformation document symbol density mapping mechanism; Engineering feature extraction subnetwork, adopting three-layer progressive convolution structure, the first layer extracts basic character edge features, the middle layer identifies the spatial connection relationship of special symbols such as circuit breakers and transformers through symbol topology perception convolution kernel, and the top layer generates feature atlas combining symbol combination rules in power transmission and transformation specifications; Cross-region association reasoning module, based on the inherent layout rules of devices, lines and parameters in design drawings, establishes a spatial correlation weight matrix between text blocks and chart areas, and performs feature binding on scattered parameter annotations and corresponding device symbols; By comparing typical layout templates of power transmission and transformation documents, correct the feature coordinates of inclined and deformed areas, and output structured image feature set with modal labels and spatial correlation degrees.

3. The system according to claim 1, characterized in that, The specific process of performing cross-modal feature alignment operation is as follows: Modal feature initialization, extracting parameter keyword-position coordinate feature pairs for text modal, device symbol outline-topology coordinate feature pairs for chart modal, and special identification code-pixel coordinate feature pairs for symbol modal, to deconstruct the features of core modal of power transmission and transformation documents; Construct alignment benchmark, based on the layout rules of device symbols, parameter annotations and chart areas in power transmission and transformation engineering documents, establish a polar coordinate alignment coordinate system with the geometric center of device symbols as the origin, and set the typical distribution angle range of parameter text in the coordinate system; The association calculation is bound to the correction, the distance deviation and the semantic matching degree of different modal features in the polar coordinate are calculated through the modal association degree scoring algorithm combined with the matching rules of parameters and equipment in the power transmission and transformation specification, and the alignment confidence is generated; for the feature pairs with confidence lower than the threshold, offset compensation is called through the power transmission and transformation document layout template library, and the aligned multi-modal features are bound as the association unit of equipment-parameter-position.

4. The system according to claim 1, characterized in that The symbol-oriented repair module includes: Based on the special symbol feature library of power transmission and transformation, the topological structure parameters, typical edge features and standard drawing standards of core equipment symbols are stored, the mapping relationship between symbol damage mode and repair scheme is established by analyzing the wear and tear samples of symbols in historical design documents; A pre-set hierarchical repair engine is used for engineering symbol damage mode identification, edge feature residual of the damaged part is extracted, and the contour of the damaged symbol is completed through geometric feature library matching; at the same time, the key details of the contact and terminal are supplemented based on the power transmission and transformation symbol drawing specification; The first re-verification is performed through the matching of symbol type and adjacent electrical parameter engineering properties; the second re-verification is performed in combination with the power transmission and transformation drawing layout specification; the secondary repair is performed based on the abnormal items calling the typical forms of the same type of symbols in the feature library, and the complete symbol conforming to the power transmission and transformation engineering design specification is output.

5. The system according to claim 1, characterized in that, The spatial features of equipment identification and wiring symbols are obtained through the engineering symbol topology extraction sub-network, and the specific process is as follows: Precise positioning of the symbol area is performed, based on the typical size range and line features of the power transmission and transformation symbol, a scale sliding window traversal structure image is used combined with a feature filtering mechanism to exclude the pixel area of equipment identification and wiring symbols, and a symbol mask with a bounding box is generated; The key connection points of the symbol are located through edge gradient change detection, including the center position of the device terminal and the intersection of the line diameter; the contour features of the switch symbol are extracted through contour analysis, and the series and parallel topological association between symbols are identified based on the device layout and connection specification in the power transmission and transformation wiring rules; The geometric parameters, connection point coordinates and relative position relationship with adjacent symbols of the symbol are converted into standardized quantitative feature values, which are integrated according to the dimensions of equipment type, connection relationship and spatial parameters, and the topological feature vector containing electrical connection attributes is generated.

6. The system of claim 1, wherein, The specification corpus constraint sub-network includes: A pre-set hierarchical corpus library of power transmission and transformation specification is used for structured storage through design stages and professional categories, which contains professional corpus of equipment parameter verification rules, symbol labeling specification and text expression paradigm, and mapping index of corpus and engineering scene is established; Based on the semantic constraint reasoning module, the device type and parameter information in the text are analyzed and identified, the constraint conditions are generated by calling the specification corpus of the corresponding professional category, and the context consistency verification is performed on the identification result; Effective cases that pass the identification in actual engineering are collected, expression forms that conform to the specification and are not included in the corpus library are extracted, new expressions are classified and supplemented to the corresponding corpus library through structure similarity calculation with existing corpus; the mapping weight of new cases is optimized, and the adaptation accuracy of corpus calling in different design scenarios is improved.

7. The system according to claim 1, characterized in that, The new dynamic confusion set correction operation includes: Firstly, a confusion set specific to the power transmission and transformation field is constructed. By combing historical errors and objects easily confused as marked by field experts, similar device symbols and similar professional terms are classified and stored, and the core distinguishing features are recorded simultaneously; The visual features and semantic attributes of the current recognition object are extracted, and the feature similarity with each item in the confusion set is calculated; If the similarity exceeds the set threshold, the correction mechanism is triggered, the core distinguishing standards of the device in the power transmission and transformation specification are called, and the targeted correction rules are generated; The correction rules are applied to the current recognition result to accurately distinguish the symbols or terms. At the same time, the correction case is supplemented to the confusion set, the distinguishing feature weight is updated through incremental learning, and the confusion set is dynamically iterated.

8. The system of claim 1, wherein, The text-specification bidirectional mapping mechanism includes: A mapping index library is constructed to extract the core features in the text segments of the power transmission and transformation documents and analyze the key elements in the power transmission and transformation design specification items. The association index between the text segments and the specification items is established through feature hashing algorithm to form the basis of bidirectional mapping; The semantic split of the recognized text is performed to extract the device attributes and parameter information in the text. Based on the mapping index library, the corresponding specification items are matched. Through consistency comparison of professional terms, parameter value and specification threshold verification, the degree of fit between the text content and the specification requirements is determined, and the text segments deviating from the specification are marked; Based on the specified specification items, the related text segments in the mapping index library are retrieved to check the complete coverage of the mandatory clauses and recommended requirements in the specification. The missing parameter description and logical explanation are generated to check the consistency of the bidirectional mapping results through the power transmission and transformation engineering design logic. The matching relationship between the text and the specification and the deviation annotation are output.

9. The system according to claim 1, characterized in that, The execution parameter association reasoning operation includes: Structured extraction is performed on the device parameters and connection parameters obtained by recognition to sort the exclusive parameter values based on device type classification and analyze the adaptation requirements of parameters in the loop connection based on connection relationship to form the corresponding relationship matrix of parameters and devices and parameters and connections; A parameter dynamic association reasoning model is constructed based on the principle of power transmission and transformation engineering: the parameter transmission path is analyzed through the hierarchical sequence of power supply layer, bus layer, main device layer and branch layer, and the constraint rules and influence weights between levels are determined; Based on the cross-loop scenario, the quantitative calculation method and correlation strength threshold of parameter interaction influence are defined. The scenario rule set including basic parameter association logic and boundary conditions is constructed through voltage level and connection form classification; The model is called to perform hierarchical verification on the structured parameter matrix: the internal consistency of single device parameters and functions is verified through parameter attribute label matching degree analysis; the correlation degree of device parameters in the same loop is calculated and a quantitative score is generated through correlation dimension identification mapping; the transmission adaptability of device parameters between levels is checked through hierarchical transmission path tracing to accurately locate the abnormal nodes.

10. The system of claim 1, wherein, The engineering node timing data model includes: Based on the whole life cycle of power transmission and transformation project, the design, installation, debugging, acceptance core node and corresponding process sub-node are divided to form a hierarchical structure; the time sequence dimension of the node is defined, the mapping relationship between the time axis and the node attribute is established; the device parameters, connection state and test results of each node are connected based on time sequence, the mutation threshold and gradual change range of parameters are marked when the node is converted, the data connection of adjacent nodes, the time sequence consistency of parameters in the same node and the trend correlation of cross-node parameters are set; At the same time, the time sequence data is mapped to the time axis, the abnormality is identified through rule checking, the time sequence correlation diagram is generated, and the change trajectory of parameters with the progress of the project and the data transmission relationship between nodes are displayed.

11. The system of claim 1, wherein, The version evolution increment fusion module: An engineering version pedigree management architecture is constructed, a version sequence is established through a timestamp and a change identifier, the generation time, triggering scenario and associated version number of each version are recorded; the incremental data between versions is extracted by storing the data of each version according to the infrastructure layer-parameter layer-topology layer, and the structural increment and the parameter increment are distinguished to provide structured input for increment fusion; A multi-dimensional increment fusion mechanism is established, and the difference area between versions is located through feature hash comparison: for structural increment, the topology alignment algorithm is used to match the device connection relationship of new and old versions, and the common topology structure is retained and the new node is embedded; For parameter increment, the weighted fusion of incremental value and historical value is performed based on engineering logic weight, and the influence range of parameter change is marked synchronously; the increment conflict detection rule is set, and the change initiation time and approval level are called to determine the priority; The dynamic evolution adaptation function is integrated, a change type-fusion strategy mapping model is constructed based on historical version fusion cases, a candidate fusion scheme is generated based on the detected new change scenario, and the optimal solution is selected through engineering compliance verification; the increment source, processing rule and fusion basis are recorded in real time during the fusion process, and the version evolution trajectory file is formed.

12. The system of claim 1, wherein, The new document segment time sequence anchoring operation: Based on the structured splitting of the power transmission and transformation engineering document, the core segments of device parameter description, connection logic description and test result record are extracted, engineering node labels and key time identifiers are added to each segment to form a segment library with time sequence attributes; An anchoring mapping mechanism of segment and engineering node time sequence is established, the corresponding engineering node is matched through segment label matching, and the segment is accurately associated to the specific time point of the node time sequence axis combined with the key time identifier; Anchoring verification is performed synchronously to check the time logic consistency and content correlation of the segment content and the corresponding node, and the anchoring abnormal segment is excluded; Set the dynamic update function of time sequence anchoring, when the engineering node time sequence is adjusted or the new document segment is added, the association relationship between the segment and the node is automatically re-matched, the anchoring position of the segment on the time sequence axis is updated, and the anchoring change trajectory is recorded.

Images