Intelligent analysis and dynamic pricing method for engineering cost data

Abstract

The application relates to the technical field of intelligent analysis of engineering cost data, and discloses a method for intelligent analysis of engineering cost data and dynamic pricing, which comprises the following steps: obtaining a bill of quantities table, a quota sub-item table, a regional pricing rule text, an engineering feature text and material price data, performing field unification, unit mapping and source marking processing on the data to form a standardized data set; performing entity recognition and term standardization on the text data to construct a cross-version list knowledge graph; generating a candidate mapping set for a to-be-mapped list item and constructing a candidate mapping subgraph, and determining a target mapping relationship through constraint propagation reasoning; on the basis of the foregoing, constructing a pricing dependency subgraph, performing local recalculation on affected nodes when a trigger event occurs, and outputting a dynamic pricing result.

Description

Technical Field

[0001] This application relates to the field of intelligent analysis technology for engineering cost data, specifically a method for intelligent analysis and dynamic pricing of engineering cost data. Background Technology

[0002] With the continuous expansion of engineering construction projects and the improvement of the informatization level of engineering cost management, bill of quantities pricing has become the main pricing method in engineering cost management. During project implementation, it is usually necessary to process data from multiple sources simultaneously, including bill of quantities, quota items, regional pricing rules, engineering feature descriptions, and material prices. This data exhibits inconsistencies in expression and significant differences in field structure across different versions of standards, regional pricing systems, and project documents. Because information such as bill of quantities item names, engineering feature descriptions, and measurement standards are often presented in natural language, and the same component or process may be expressed using different terminology under different versions of standards or regional rules, there is a high degree of parsing difficulty in the process of version migration, regional switching, and project reuse of engineering cost data.

[0003] In existing technologies, engineering cost data processing typically relies on manual matching of bill of quantities items with quota sub-items and calculation of costs according to pricing rules. When the standard version of the bill of quantities is updated or regional pricing rules change, the bill of quantities mapping and pricing calculation need to be re-performed. This approach is not only labor-intensive, but also prone to omissions or inconsistencies in complex situations such as bill of quantities item splitting and merging, differences in measurement caliber, and costs triggered by engineering features. Furthermore, when material prices are updated or engineering features change, traditional pricing methods often require recalculation of the entire project, lacking fine-grained management of dependencies, resulting in low calculation efficiency and difficulty in meeting the needs of frequent updates and dynamic adjustments to engineering cost data.

[0004] Therefore, how to achieve unified parsing of bill of quantities data, reliable mapping between bill of quantities items across versions, and effective recalculation of affected parts when data changes in a multi-source engineering cost data environment has become an urgent technical problem to be solved in the information processing of engineering cost. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent analysis and dynamic pricing method for engineering cost data to solve the problems mentioned in the background art.

[0006] According to the first aspect of this application, a method for intelligent analysis and dynamic pricing of engineering cost data is provided, comprising the following steps: Obtain the bill of quantities, quota item table, regional pricing rule text, project feature text, and material price data; perform field unification, unit mapping, and source marking on the data to form a standardized data set. Entity recognition and terminology normalization are performed on the text data in the standardized dataset to obtain list entities, engineering feature entities, measurement caliber entities, and rule entities; a cross-version list knowledge graph is constructed based on the standardized dataset and the entities. For each item in the list to be mapped, a candidate mapping set is generated in the cross-version list knowledge graph; a candidate mapping subgraph is constructed based on the candidate mapping set, and constraint propagation reasoning is performed on the candidate mapping subgraph to obtain the target mapping relationship; Construct a pricing dependency subgraph based on the target mapping relationship; respond to the triggering event, determine the affected nodes based on the pricing dependency subgraph and perform local recalculation, and output dynamic pricing results.

[0007] Preferably, the data undergoes field unification, unit mapping, and source marking to form a standardized data set, including: The coding fields in the bill of quantities and quota item tables are uniformly formatted as strings, the quantity fields are uniformly formatted as numbers, and the unit fields are mapped to standard unit codes according to the standard unit table. For convertible units, the quantity of work is converted based on the conversion factor in the standard unit table; for units with the same dimensions but whose measurement caliber is not consistent, only the unit compatibility status is recorded. Write the source type, source location, version identifier, region identifier, and project identifier for each structured record and each text fragment.

[0008] Preferably, entity recognition and terminology normalization are performed on the text data in the standardized dataset, including: Input a text segment into a sequence labeling model consisting of a RoBERTa whole-word mask pre-trained encoding layer, a bidirectional long short-term memory network layer, and a conditional random field layer connected in sequence, and output an entity label sequence. Based on the entity tag sequence, extract list terms, component locations, material specifications, construction techniques, additional conditions, measurement boundary terms, and version indicator terms; Then, based on the terminology list, thesaurus, and contextual semantic similarity, the extraction results are mapped to standard term identifiers, and the entity type, source fragment, and entity confidence are recorded.

[0009] Preferably, the cross-version list knowledge graph includes list item nodes, quota sub-item nodes, engineering feature nodes, unit of measurement nodes, measurement caliber nodes, cost composition nodes, version nodes, regional rule nodes, and material nodes; The relationships in the cross-version list knowledge graph include synonym relationships, version mapping relationships, parent-child splitting relationships, unit compatibility relationships, unit conversion relationships, caliber dependency relationships, feature triggering relationships, cost composition relationships, and recalculation transmission relationships. Each relation is bound to a relation source and a relation confidence level. The relation source includes at least structured comparison data, rule text parsing results, and verified model extraction results.

[0010] Preferably, generating a candidate mapping set in the cross-version list knowledge graph for the list item to be mapped includes: The cross-version list knowledge graph is filtered by version and region based on the target version and target region; the name, location, engineering features and unit of the list item to be mapped are combined and input into the sentence vector model to obtain the list vector to be mapped; candidate list vectors are generated for the filtered candidate list items; Based on the semantic similarity between the list vector to be mapped and the candidate list vector, and combined with coding proximity, unit compatibility, engineering feature consistency and regional rule compatibility, candidate list items are selected to form a candidate mapping set.

[0011] Preferably, the step of constructing a candidate mapping subgraph based on the candidate mapping set and performing constraint propagation reasoning on the candidate mapping subgraph to obtain the target mapping relationship includes: Construct a candidate mapping subgraph using items to be mapped, candidate items, engineering feature entities, unit of measurement nodes, measurement caliber nodes, version nodes, and regional rule nodes; Name semantic similarity, encoding proximity, unit compatibility, engineering feature consistency, and regional rule compatibility are used as the initial features for candidate edges; The node features and initial candidate edge features are input into the graph attention network to obtain the updated candidate edge scores; the target mapping relationship is determined based on the updated candidate edge scores.

[0012] Preferably, determining the target mapping relationship based on the updated candidate edge scores includes: For each candidate edge, a global consistency score is calculated. The global consistency score consists of at least a split closure consistency item, a measurement caliber consistency item, a feature trigger integrity item, and a region rule adaptation item. For candidate combinations that have a correspondence between the old version parent item and multiple new version sub-items, calculate the combination quantity coverage, feature coverage, and cost composition coverage. Based on the global consistency score and the coverage of the candidate combinations, one of the following is determined as the target mapping relationship: direct correspondence mapping, split mapping, merge mapping, or conditional mapping.

[0013] Preferably, constructing the pricing dependency subgraph based on the target mapping relationship includes: Using the mapped project list item as the core node, connect its associated quota sub-item node, labor node, material node, machinery node, measure cost composition node and fee calculation node to the same graph structure; Establish the following relationships in the graph structure: quantity dependency, consumption dependency, price reference, measure triggering, and cost transmission. The target mapping relationship is then written into the project-level mapping relationship edge to ensure that the list item nodes in the pricing dependency subgraph are consistent with the target mapping relationship in the cross-version list knowledge graph.

[0014] Preferably, the triggering events include list version change events, regional rule switching events, engineering feature change events, and material price update events; the step of determining the affected nodes and performing local recalculation based on the pricing dependency subgraph includes: The initial node set is determined according to the triggering event type; dirty tags are propagated from the initial node set to subsequent nodes based on recalculation propagation relationships, price reference relationships, measure triggering relationships, and fee collection propagation relationships. Nodes marked with "dirty" are recalculated in dependency order, and the node status version number is updated after the node recalculation is completed. The recalculation of nodes marked with "dirty" in dependency order includes: first, calculating the basic direct cost of the list item based on the mapped quota sub-items and resource details; then, updating the measure cost composition node based on the engineering feature change results; and finally, updating the management fee node, regulatory fee node, and tax node based on the fee collection caliber corresponding to the regional rule node. Before recalculation, for list version change events and regional rule switching events, a candidate mapping set is regenerated for the affected list items and the target mapping relationship is redefined. Then, the corresponding pricing dependency subgraph is reconstructed based on the updated target mapping relationship.

[0015] In a second aspect, this application also provides an intelligent analysis and dynamic pricing system for engineering cost data, comprising: The data acquisition and standardization module is used to acquire bill of quantities, quota item tables, regional pricing rule texts, engineering feature texts, and material price data, and to perform field unification, unit mapping, and source marking processing on the data to form a standardized data set. The entity recognition and knowledge graph construction module is used to perform entity recognition and terminology normalization processing on the text data in the standardized data set to obtain list entities, engineering feature entities, measurement caliber entities and rule entities, and to construct a cross-version list knowledge graph based on the standardized data set and the entities. The list mapping reasoning module is used to generate a candidate mapping set in the cross-version list knowledge graph for the list items to be mapped; construct a candidate mapping subgraph based on the candidate mapping set; and perform constraint propagation reasoning on the candidate mapping subgraph to obtain the target mapping relationship. The dynamic pricing calculation module is used to construct a pricing dependency subgraph based on the target mapping relationship; when responding to a triggering event, it determines the affected nodes based on the pricing dependency subgraph and performs local recalculation to output the dynamic pricing result.

[0016] Compared with existing engineering cost data processing methods, this invention performs unified and standardized processing on the bill of quantities, quota item tables, regional pricing rule texts, engineering feature texts, and material price data. It also constructs a cross-version bill of quantities knowledge graph by combining entity recognition and terminology standardization, enabling engineering cost data from different sources, versions, and regional rules to be organized and associated in a unified data structure.

[0017] Furthermore, by introducing a constraint propagation reasoning mechanism on the candidate mapping subgraph, a comprehensive analysis is conducted on the semantic similarity, unit compatibility, measurement caliber, and engineering feature triggering relationships among list items. This determines the target mapping relationships between list items, improving the accuracy and consistency of cross-version list mapping. Based on this, by constructing a pricing dependency subgraph and combining it with a trigger event-driven local recalculation mechanism, in cases of material price updates, engineering feature changes, regional rule switching, or list version changes, calculation updates are performed only on the affected nodes, avoiding duplicate calculations for the entire project. This ensures the consistency of pricing results while improving the efficiency and traceability of engineering cost data processing. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of an intelligent analysis and dynamic pricing method for engineering cost data provided in an embodiment of this application.

[0020] Figure 2 A schematic diagram illustrating the construction of a cross-version inventory knowledge graph provided in this embodiment of the disclosure.

[0021] Figure 3 This is a schematic diagram of the target mapping relationship acquisition process provided in the embodiments of this disclosure.

[0022] Figure 4 This is a schematic diagram of a trigger event-driven partial recalculation process provided in an embodiment of this disclosure.

[0023] Figure 5 This is a schematic diagram of the structure of an intelligent analysis and dynamic pricing system for engineering cost data provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] In practical implementation, this method is applicable to engineering cost processing scenarios deployed on engineering cost data processing servers, cost data platforms, or data processing clusters composed of graph databases and relational databases. The objects to be processed can be a single project or multiple projects within the same region, period, or version migration task. Initially, the system must have the capability to access at least the bill of quantities, quota sub-item tables, regional pricing rule texts, engineering characteristic texts, and material price data, and possess standard unit tables, terminology tables, synonym tables, version identifier tables, and regional identifier tables. The relational database stores standardized structured records, while the graph database stores bill of quantities item nodes, engineering characteristic nodes, unit of measurement nodes, measurement caliber nodes, cost composition nodes, version nodes, regional rule nodes, material nodes, and their relationships. Therefore, upon arrival of input data, standardization, terminology parsing, cross-version mapping, graph constraint reasoning, and dynamic pricing recalculation can be completed within the same processing flow.

[0026] The following detailed description, in conjunction with specific embodiments, illustrates the implementation process of the intelligent analysis and dynamic pricing method for engineering cost data described in this application. It should be noted that these embodiments are merely for explaining this application and not for limiting its scope of protection. Any conventional adjustments or substitutions made by those skilled in the art to the steps without departing from the concept of this application should be included within the scope of protection of this application.

[0027] like Figure 1 As shown in the figure, this application discloses a schematic diagram of an intelligent analysis and dynamic pricing method for engineering cost data, including the following method steps: S1. Obtain the bill of quantities, quota item table, regional pricing rule text, engineering feature text and material price data, and perform field unification, unit mapping and source marking on the data to form a standardized data set; S2, perform entity recognition and terminology normalization on the text data in the standardized data set to obtain list entities, engineering feature entities, measurement caliber entities and rule entities; construct a cross-version list knowledge graph based on the standardized data set and the entities; S3, for the list items to be mapped, generate a candidate mapping set in the cross-version list knowledge graph; construct a candidate mapping subgraph based on the candidate mapping set, and perform constraint propagation reasoning on the candidate mapping subgraph to obtain the target mapping relationship; S4. Construct a pricing dependency subgraph based on the target mapping relationship; respond to the triggering event, determine the affected nodes based on the pricing dependency subgraph and perform local recalculation, and output the dynamic pricing result.

[0028] In some embodiments, for step S1, the bill of quantities, quota item table, regional pricing rule text, project feature text and material price data are obtained, and field unification, unit mapping and source marking are performed on the data to form a standardized data set.

[0029] For structured inputs such as bill of quantities, quota item tables, and material price data, the system first establishes a project task context. This context includes at least the project identifier, target version, target region, target pricing date, target price caliber, and task trigger type. The task trigger type distinguishes between initial parsing tasks, version migration tasks, region rule switching tasks, engineering feature change tasks, and material price update tasks. This task context remains unchanged throughout the processing; all subsequent parsing results, candidate mapping results, graph relationships, and recalculation results are bound to this task context to ensure the data processing chain for the same project at different task points is traceable.

[0030] After receiving the input data, standardization processing is performed first. Specifically, the bill of quantities code field in the bill of quantities table and the quota code field in the quota sub-item table are standardized to string format; the quantity field, unit price field, and consumption field are standardized to numeric format; and the unit field is mapped to the standard unit code according to the standard unit table. The standard unit table pre-records the unit code, basic unit, target standard unit, and conversion factor. For units that can be directly converted, the system performs quantity conversion. For example, if the original quantity is recorded as... The unit conversion factor is denoted as The converted standard engineering quantity is denoted as:

[0031] in, Indicates standard engineering quantity. It can be preset from the standard unit table or entered according to national or industry metrological standards. The value range is determined by the specific unit conversion relationship. If the corresponding units are of the same dimension and can be converted, the coefficient is a positive real number; if there is no direct conversion relationship, no numerical conversion is performed. For units with the same dimension but inconsistent measurement standards, the system does not perform quantity conversion, but only records the unit compatibility status. The unit compatibility status here is a status field used to identify whether the two units are of the same dimension, whether they can be directly converted, and whether further determination is needed based on the measurement standards. This status field will directly participate in candidate mapping scoring and constraint propagation reasoning, so it is written during the standardization phase.

[0032] During the standardization process, the system also writes source tags for each structured record and each text fragment. Source tags include at least source type, source location, version identifier, region identifier, and project identifier. The source type distinguishes different data sources such as bill of quantities, quota data, material price data, drawing instructions, and pricing rule clauses. The source location records the row / column position in a table or the position of a text paragraph. The version identifier corresponds to the bill of quantities version or quota version. The region identifier corresponds to the province, city, or district pricing rule range. The project identifier corresponds to the project object in the current task context. After completing the above processing, a standardized data set is formed. This set consists of a subset of structured records, a subset of text fragments, a subset of price time series, and a subset of rule metadata.

[0033] In some embodiments, for step S2, entity recognition and terminology normalization are performed on the text data in the standardized data set to obtain list entities, engineering feature entities, measurement caliber entities, and rule entities; a cross-version list knowledge graph is constructed based on the standardized data set and the entities. The technical problem addressed by this process is that the same list item may be expressed with different names in different regions, versions, and documents, and engineering features, material specifications, and measurement boundaries are often embedded in free text. If literal string retrieval is used directly, it is easy to encounter situations such as duplicate items with the same name, duplicate items with different names, and loss of features.

[0034] Optionally, before textual data enters terminology parsing, the system further performs formatting normalization and segmentation. Specifically, for drawing descriptions, project feature descriptions, and regional pricing rules, character normalization, whitespace cleanup, paragraph boundary recognition, and table cell splitting are performed first. For cases where a single cell in a table contains a list name, location description, and material specifications simultaneously, the system splits it into a main name field, location field, and additional feature field according to a preset delimiter and domain dictionary. For text paragraphs, they are segmented into atomic fragments based on periods, semicolons, numbering characters, clause numbers, and line breaks, with each fragment bound to a fragment identifier and source location. This processing method facilitates subsequent entity recognition and rule extraction at the fragment level.

[0035] Please see Figure 2 , Figure 2 This is a schematic diagram illustrating the construction of a cross-version inventory knowledge graph provided in an embodiment of this disclosure. For example... Figure 2 As shown, in S201, extract the list entity, engineering feature entity, measurement caliber entity, and rule entity.

[0036] In one implementation, the system employs a sequence labeling model consisting of a RoBERTa full-word mask pre-trained encoding layer, a Bi-Short Memory (BiLSTM) network layer, and a Conditional Random Field (CRF) layer connected sequentially. Text segments first enter a word-level encoding layer, which outputs a word sequence representation of the segment. This word sequence representation then enters the RoBERTa encoding layer to obtain the contextual semantic vector for each position. The contextual semantic vectors are then sequentially input into the Bi-Short Memory (BiLSTM) network layer, forming a sequence state containing forward and backward dependencies. Finally, the sequence state is input into the CRF layer, which outputs a sequence of entity labels that satisfy transition constraints.

[0037] For an input text segment, let its word sequence be denoted as . The context representation is obtained after passing through the RoBERTa encoding layer. After passing through a bidirectional long short-term memory network layer, the sequence state is obtained. The conditional random field layer outputs an entity tag sequence based on the tag transition matrix and the emission scores at each position. The tag set includes at least the following: list terminology tags, component location tags, material specification tags, construction process tags, additional condition tags, measurement boundary tags, and version indication tags. After entity identification, the system extracts list entities, engineering feature entities, measurement caliber entities, and rule entities accordingly.

[0038] In this implementation, the "measuring scope entity" refers to a terminology that describes the measurement boundary, measurement range, content attribution, or inclusion relationship, such as inclusion in the comprehensive unit price, separate measurement, calculation by unfolded area, or calculation by entity volume. The "rule entity" refers to a rule object extracted from the regional pricing rule text and explanatory documents, which may include the scope of application, triggering conditions, and cost attribution. These two types of entities correspond to the measuring scope node and the regional rule node, respectively, in the subsequent cross-version list knowledge graph.

[0039] In S202, the entity extraction results are normalized to form an entity set. The entity extraction results are not directly used for subsequent graph construction but instead enter the terminology normalization stage. Terminology normalization is based on a terminology lexicon, a thesaurus, and contextual semantic similarity. The system first performs exact matching based on the terminology lexicon; for unmatched items, it then performs synonym merging based on the thesaurus; for term objects that are still undetermined, it calls the contextual semantic matcher to calculate their semantic similarity with each candidate term in the standard terminology lexicon. If the similarity reaches a preset terminology normalization threshold, it is mapped to the corresponding standard term identifier. The terminology normalization threshold is a value within a closed interval. The parameter is defined as the minimum acceptable semantic similarity during term merging. This threshold can be optionally determined by validation results on the annotated corpus, or it can be updated in batches based on manually checked samples after deployment. The update method involves statistically analyzing the correctly merged and incorrectly merged samples after manual correction, and then re-searching on the validation set for the threshold that minimizes the combined index of mismerging rate and missed merging rate.

[0040] After terminology normalization is completed, an entity set is formed. Each entity record in the entity set includes an entity identifier, entity type, standard term identifier, original text, source fragment, source location, and entity confidence score. The entity confidence score is defined as the combined value of the model's reliability in recognizing and normalizing the entity. This value is obtained by weighting the output probability of the sequence labeling model with the terminology normalization score. Let the confidence score of the sequence labeling model's output for a certain entity boundary and category be denoted as... The terminology normalization similarity is The overall confidence level can then be expressed as:

[0041] in, Indicates the confidence level of an entity. and The values ​​of all are located within the closed interval. , The fusion weighting coefficients take values ​​within the range . The meaning is to control the proportion of the identification result and the normalization result in the overall confidence score. This parameter can be configured to be obtained by fitting a historical annotation set, or it can be determined by cross-validation of the validation set.

[0042] Optionally, for training this sequence labeling model, in one example, a training set can be constructed using project feature text from historical engineering projects, drawing description text, regional rule text, and manually corrected terminology samples. The RoBERTa encoding layer is initialized with pre-trained parameters, while the bidirectional long short-term memory network layer and conditional random field layer are randomly initialized. During the training phase, training is supervised based on manually labeled entity boundaries and categories, and the loss function consists of the negative log-likelihood of the conditional random field. After training, the model parameters that meet the preset validation metrics are fixed into the online inference version.

[0043] In S203, after obtaining the standardized data set and entity set, the system constructs a cross-version inventory knowledge graph. The technical problem addressed in this part is that engineering cost data not only has name differences, but also version migration, splitting and merging, unit compatibility, differences in measurement standards, and regional rule differences. If only relational tables are used for connection, it is difficult to express multi-type, multi-path, and conditional relationships.

[0044] The cross-version bill of quantities knowledge graph comprises a set of nodes and a set of relationships. The node set includes at least the following nodes: bill of quantities item nodes, quota sub-item nodes, engineering feature nodes, unit of measurement nodes, measurement caliber nodes, cost composition nodes, version nodes, regional rule nodes, and material nodes. Bill of quantities item nodes are generated from the bill of quantities table and historical mapping table, with attributes including bill of quantities code, standard bill of quantities name, unit code, version, region, and source task. Quota sub-item nodes are generated from the quota sub-item table, with attributes including quota code, quota name, labor consumption, material consumption, machinery consumption, and applicable conditions. Engineering feature nodes originate from engineering feature entities in the entity set. Unit of measurement nodes originate from the standard unit table. Measurement caliber nodes originate from measurement caliber entities. Cost composition nodes are generated from the cost composition table or rule metadata. Version nodes and regional rule nodes are generated from the version identifier table and rule metadata, respectively. Material nodes are generated from material price data and resource detail data.

[0045] The relation set includes at least the following: synonym relationships, version mapping relationships, parent-child splitting relationships, unit compatibility relationships, unit conversion relationships, caliber dependency relationships, feature triggering relationships, cost composition relationships, and recalculation propagation relationships. Synonym relationships connect semantically equivalent or mergeable list item nodes. Version mapping relationships describe the substitution or migration relationships between old and new version list items. Parent-child splitting relationships describe the structural mapping of one old version parent item to multiple new version child items. Unit compatibility and unit conversion relationships describe dimensional compatibility and numerical conversion between units, respectively. Calibration dependency relationships describe scenarios where unit compatibility still depends on measurement caliber determination. Feature triggering relationships describe the triggering connections between engineering features and cost components or list items. Cost composition relationships connect list items to basic cost components. Recalculation propagation relationships describe the propagation boundary in subsequent dynamic pricing stages.

[0046] When adding a relation to the knowledge graph, each relation is bound to its source and confidence level. The source refers to the data or computational process from which the relation was generated, including at least structured reference data, rule text parsing results, and verified model extraction results. The confidence level indicates the degree of credibility of the relation when it enters the knowledge graph. If the relation comes from a standard version reference table or unit conversion table, the confidence level is directly recorded as a high-confidence source; if the relation comes from model extraction, it needs to be verified by rules before being written. Therefore, the system sets a relation entry threshold. The relation entry threshold is defined as the minimum confidence level allowed for candidate relations to enter the knowledge graph. This threshold can optionally be determined based on historical samples, and its value falls within a closed interval. When the confidence level of a candidate relation is lower than this threshold, the relation is only written to the candidate relation cache and is not directly entered into the main graph database.

[0047] In some embodiments, for step S3, a candidate mapping set is generated in the cross-version list knowledge graph for the list item to be mapped; a candidate mapping subgraph is constructed based on the candidate mapping set, and constraint propagation reasoning is performed on the candidate mapping subgraph to obtain the target mapping relationship.

[0048] Please see Figure 3 , Figure 3 This is a schematic diagram illustrating the target mapping relationship acquisition process provided in an embodiment of this disclosure. Figure 3 As shown, in S301, the name, location, engineering features, and unit of the list item to be mapped are combined into the text to be mapped, and input into the sentence vector model to obtain the list vector to be mapped and the candidate list vector respectively.

[0049] After the knowledge graph is constructed, the system begins processing the items in the pending mapping list. These items refer to those in the current project for which no corresponding object has yet been determined under the target version and target region. To generate a candidate mapping set, the system first filters the knowledge graph according to the target version and target region. Specifically, it removes version nodes and their adjacent edges that are unreachable from the target version, and removes rule nodes and their conflicting edges whose applicable region does not include the target region. The filtered graph subspace serves as the candidate search space.

[0050] Subsequently, the system combines the name, location, engineering features, and units of the items to be mapped into a text to be mapped, and inputs it into the sentence vector model. The sentence vector model adopts Sentence-BERT, which consists of a dual-tower encoder with shared parameters. The left tower inputs the text to be mapped, and the right tower inputs the candidate list text, outputting two fixed-dimensional vectors. The text to be mapped is composed of the standard list name, location description, normalized engineering features, and standard units concatenated sequentially; the candidate list text consists of the name, additional features, and unit information of the candidate list item nodes. After dual-tower encoding, the vectors of the list to be mapped and the candidate list are obtained respectively.

[0051] In S302, for each item in the list to be mapped, the system calculates semantic similarity in the filtered list item node set and combines it with encoding proximity, unit compatibility, engineering feature consistency, and regional rule compatibility to form candidate scoring features. Let the item to be mapped be... The candidate list items are Then the local score can be expressed as:

[0052] in, The semantic similarity of names is derived from the cosine similarity between the list vector to be mapped and the candidate list vectors; The encoding proximity is derived from the list encoding hierarchy and prefix similarity. The degree of unit compatibility is derived from the unit compatibility relationship and the unit conversion relationship. The consistency of engineering features is determined by the overlap of engineering feature nodes. This indicates the compatibility of regional rules, derived from the adaptation relationship between regional rule nodes and candidate list items. to These are the local scoring weighting coefficients, and the values ​​of each coefficient lie within a closed interval. And the sum is 1. The weighting coefficients are defined to control the proportion of contribution of different scoring factors to local scores. They can be determined by statistical fitting based on historical manually confirmed mapping samples, or by selecting a set of parameters through a validation set grid search to achieve the target value for the combined index of candidate recall and false positive recall.

[0053] In step S303, the system retains a few candidate items with the highest local scores for each item in the list to be mapped, thus forming a candidate mapping set. To avoid introducing obviously irrelevant candidates, a minimum candidate score threshold can be optionally set. Objects with scores below this threshold are removed from the candidate set. At this point, each item in the list to be mapped obtains a candidate mapping set containing several candidate items, while retaining its local score characteristics.

[0054] After the candidate mapping set is generated, the system does not directly determine the target mapping relationship based on local scores. Instead, it constructs a candidate mapping subgraph and performs constraint propagation reasoning. The technical problem addressed by this process is that local scores can only reflect the similarity between a single item in the list to be mapped and a single candidate object, and cannot fully utilize the global constraints between parent-child splitting, measurement caliber, engineering feature linkage, and regional rules. Therefore, it is necessary to perform global consistency determination on the graph structure.

[0055] In S304, for each item in the list to be mapped, the system constructs a candidate mapping subgraph using the item, its candidate list items, the corresponding engineering feature entity, the unit of measurement node, the caliber of measurement node, the version node, and the regional rule node. The edges in the candidate mapping subgraph include candidate mapping edges, unit compatibility edges, unit conversion edges, caliber dependency edges, feature triggering edges, version mapping edges, and regional applicability edges. Candidate mapping edges connect the item in the list to be mapped and the candidate list items, carrying the aforementioned local scoring features. Other edges are inherited from the knowledge graph, retaining the relationship source and relationship confidence.

[0056] In step S305, constraint propagation inference is performed on the candidate mapping subgraph to obtain the target mapping relationship. In some embodiments, the system uses a graph attention network for constraint propagation inference to obtain the target mapping relationship. The network structure sequentially includes a node feature embedding layer, a candidate edge feature mapping layer, a multi-head attention propagation layer, an edge score update layer, and a global consistency calculation layer. The node feature embedding layer receives node attributes from the candidate mapping subgraph and outputs a node representation with a unified dimension; the candidate edge feature mapping layer receives local score features and relationship type features of candidate mapping edges and outputs edge representations; the multi-head attention propagation layer simultaneously receives node representations and edge representations and performs neighborhood information aggregation on multiple attention heads; the edge score update layer calculates the updated candidate edge score based on the propagated node representations and edge representations; the global consistency calculation layer further integrates split closure, measurement caliber, feature triggering, and regional rule constraints to output the final mapping score.

[0057] Specifically, the node feature embedding layer uses a vector embedding algorithm to convert the textual and structural attributes of list item nodes, engineering feature nodes, unit of measurement nodes, version nodes, and regional rule nodes into node vectors of a unified dimension. The node textual attributes are semantically encoded using a sentence vector encoding model, while discrete attributes such as node type, unit code, and version identifier are mapped to vectors through an embedding table, concatenated, and then linearly transformed to obtain the node representation. The candidate edge feature mapping layer inputs the feature vectors composed of name semantic similarity, encoding proximity, unit compatibility, engineering feature consistency, and regional rule compatibility from the candidate mapping relationships into the fully connected mapping network, and converts them into edge vector representations of a unified dimension through a nonlinear activation function.

[0058] The multi-head attention propagation layer employs a graph attention network algorithm to propagate neighborhood information to the candidate mapping subgraph. It calculates the attention weights between nodes and their neighbors through an attention mechanism, allowing nodes to aggregate different neighborhood information across multiple attention heads. This results in an updated node representation that includes structural information such as unit relationships, rule constraints, and feature relationships. The edge score update layer, based on the propagated node representations and corresponding edge representations, recalculates the mapping probabilities of candidate mapping edges through vector concatenation and a linear scoring function, thereby obtaining the candidate mapping scores that reflect the graph's structural constraints. .

[0059] After the candidate edges are updated, the system calculates a global consistency score for each candidate edge. The global consistency score consists of at least four components: a split-closure consistency item, a measurement caliber consistency item, a feature-triggered completeness item, and a regional rule adaptation item. The split-closure consistency item measures whether the combination of the old parent item and multiple new child items fully explains the content of the parent item. The measurement caliber consistency item measures the degree of matching in measurement boundaries, unit conversions, and scope before and after candidate mapping. The feature-triggered completeness item measures whether the key engineering features in the list of items to be mapped are fully accepted by the target candidate object. The regional rule adaptation item measures whether there are conflicts or omissions in the candidate object under the target regional rules.

[0060] The global consistency score can optionally be represented as:

[0061] in, This represents the global consistency score of the candidate edges. For the updated local score, To split the closure consistency term, For consistency of measurement standards, For feature-triggered integrity items, For regional rule adaptation. Parameters to These are the global constraint weight coefficients, representing the proportion of influence of different global constraints on the final score. Each coefficient can be configured to be calibrated based on historical version change samples, and its value range can be set to a non-negative real number interval. If manually reviewed samples are collected after the system goes live, they can be recalibrated in batches based on the target mapping accuracy. To maintain consistency in processing within the same task, this set of coefficients remains fixed within the same task context.

[0062] For candidate combinations with correspondences between old parent items and multiple new sub-items, the system also calculates the combined quantity coverage, feature coverage, and cost composition coverage. Combined quantity coverage characterizes the degree to which the sub-item combination explains the parent item's quantity after unit conversion and caliber verification. Feature coverage characterizes the extent to which the engineering features extracted from the parent item are incorporated into the candidate sub-item combination. Cost composition coverage characterizes whether the quota consumption and cost composition corresponding to the parent item can be explained by the union of the candidate sub-item combinations. If all three coverage indicators meet the preset requirements, the candidate combination is retained; otherwise, it is discarded as an incomplete combination. The coverage requirement can be determined using either a threshold method or a ranking method to select the optimal combination. If a threshold method is used, the threshold can be determined based on historical split mapping samples. If a ranking method is used, the combination with the highest comprehensive coverage score is selected as the candidate mapping.

[0063] After obtaining the final global consistency score for each item in the to-be-mapped list, the system determines the target mapping relationship accordingly. Target mapping relationships include direct correspondence mapping, split mapping, merge mapping, and conditional mapping. Direct correspondence mapping means one item in the to-be-mapped list corresponds to one target item in the list. Split mapping means one item in the to-be-mapped list corresponds to multiple target sub-items. Merge mapping means multiple source items are combined to correspond to one target item. Conditional mapping means the mapping is valid only when specified project features or regional rules are met. The system writes the target mapping relationships into the mapping result table and project-level mapping relationship edges into the knowledge graph. Each project-level mapping relationship edge records the source list item identifier, target mapping object identifier, mapping type, consistency score, generation time, and task identifier.

[0064] In practical implementation, to improve traceability, the system also stores key evidence for the final mapping determination in the mapping interpretation record. The mapping interpretation record includes at least the consistency item scores involved in the calculation, key constraint node identifiers, the compatibility status of the involved units, and the regional rule identifiers. This interpretation record can be directly accessed when the dynamic pricing results need to be audited and displayed later.

[0065] Optionally, after determining the target mapping relationship, the system does not delete the unaccepted candidate mapping relationships, but writes them into the candidate history table for reuse during subsequent rule updates or model retraining. This preserves the historical judgments of the model and graph constraint layer on boundary samples without affecting the determinism of the current task result.

[0066] In some embodiments, for step S4, a pricing dependency subgraph is constructed based on the target mapping relationship; in response to a triggering event, the affected nodes are determined based on the pricing dependency subgraph and a local recalculation is performed, and a dynamic pricing result is output.

[0067] After determining the target mapping relationship, the system enters the pricing dependency subgraph construction phase. This phase uses the mapped project list item as the core node, connecting its associated quota sub-item nodes, labor nodes, material nodes, machinery nodes, measure cost composition nodes, and fee calculation nodes into the same graph structure. The labor, material, and machinery nodes correspond to different resource types in the resource details. The measure cost composition nodes correspond to measure-related cost objects such as high-support formwork, vertical transportation, and finished product protection. The fee calculation nodes correspond to the calculation levels such as management fees, regulatory fees, and taxes.

[0068] The edge relationships in the pricing dependency subgraph include quantity dependency, consumption dependency, price reference, measure triggering relationship, and fee transmission relationship. Quantity dependency describes the association between a bill of quantities item and its quantity. Consumption dependency describes the reference of a bill of quantities item or quota item to resource consumption. Price reference describes the price binding between resource nodes and material price data, labor price data, and machinery price data. Measure triggering relationship describes the conditional connection of engineering feature nodes triggering measure cost constituent nodes. Fee transmission relationship describes the transmission link from basic direct costs and measure costs to management fees, regulatory fees, and tax nodes. When constructing the pricing dependency subgraph, the system simultaneously writes the aforementioned project-level mapping relationship edges into the graph structure, ensuring a one-to-one correspondence between project bill of quantities item nodes and target mapping objects.

[0069] For the routine pricing process, the system calculates the basic direct costs of the bill of quantities items based on the mapped quota sub-items and resource details. Resource details are derived from the labor consumption, material consumption, and machinery consumption in the quota sub-item nodes, as well as the resource details of measures added based on trigger relationships. Material unit prices are derived from the price record in the material price data that best matches the target pricing point. Labor unit prices and machinery unit prices are derived from the regional price database or resource price table. If the resource set associated with a bill of quantities item is... ,resource The consumption is The unit price of the resource is The basic direct cost is recorded as:

[0070] in, Indicates list item The basic direct costs are then calculated. Next, the cost component nodes are updated based on project characteristics and the triggering relationships of the measures. Then, the management fee, regulatory fee, and tax nodes are updated according to the fee calculation criteria in the regional rule nodes. This calculation chain is consistent with the edge relationships in the aforementioned graph structure, thus ensuring that subsequent dynamic recalculation can be directly executed on the same dependency graph.

[0071] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating the event-driven partial recalculation and dynamic pricing result output process provided in an embodiment of this disclosure. Figure 4 As shown, in S401, after the initial pricing result is generated and written to the result table, the system continuously listens for trigger events.

[0072] Triggering events include list version change events, regional rule switch events, engineering feature change events, and material price update events. Each type of event is written to the event queue as an event object. An event object must at least include the event type, trigger time, target effective time, scope, and target project identifier. Material price update events generally also include a price object identifier and a price version identifier. Engineering feature change events generally include the feature value before and after the change. Regional rule switch events generally include a source rule identifier and a target rule identifier. List version change events generally include a source version identifier and a target version identifier.

[0073] In S402, after the system reads the event object from the event queue, it first determines the initial node set based on the event type. If it's a material price update event, the initial node set is the material node referenced by the updated price and its price reference edge. If it's an engineering feature change event, the initial node set is the changed engineering feature node, its connected item list node, and the cost component node. If it's a regional rule switch event, the initial node set is the target regional rule node, the cost component node with constraints on that rule, and the fee collection node. If it's a list version change event, the initial node set is the affected item list node and its corresponding item-level mapping edge.

[0074] In S403, after determining the initial node set, the system propagates dirty tags from the initial node set to subsequent nodes based on recalculation propagation relationships, price reference relationships, measure triggering relationships, and fee collection propagation relationships. Here, a dirty tag is a state identifier used to indicate that the current node has a state that requires recalculation under the current triggering event. Another state field that works in conjunction with it is the node state version number. The node state version number records the event version corresponding to the node's most recent valid calculation. During propagation, if a subsequent node already has the same state version number as the current event, the propagation is not repeated; if it has not been updated, it is marked as dirty and written to the recalculation queue.

[0075] The propagation boundaries differ for different event types. Material price update events propagate along price reference relationships to resource reference nodes, and then along consumption dependency relationships and fee transmission relationships to upstream list item nodes and fee nodes. Engineering feature change events propagate along measure trigger relationships and engineering quantity dependency relationships. Regional rule switch events propagate along the association edges between rule nodes and cost composition nodes and fee nodes. List version change events are handled slightly differently because version changes alter the mapping relationships themselves. Therefore, the system first pauses recalculation on the original dependency graph, returns to the candidate mapping generation stage, regenerates the candidate mapping set for the affected list items, and redetermines the target mapping relationship. Then, it reconstructs the pricing dependency subgraph of the affected part using the updated mapping relationship, and recalculates on the new graph.

[0076] In S404, for nodes with a dirty flag, the system performs partial recalculation according to dependency order. Dependency order means recalculating resource price and resource consumption-related nodes first, then basic direct cost nodes for bill of quantities items, then nodes constituting measure costs, and finally management fee nodes, regulatory fee nodes, and tax nodes. The dependency order is determined by the edge direction in the pricing dependency subgraph, and the system can implement this through topological ordering or hierarchical queues. After node recalculation, its node state version number is updated, and the dirty flag is cleared. If the node value does not change after recalculation, and its successor nodes depend only on this node, propagation terminates on that path; if it changes, propagation continues to successor nodes.

[0077] For engineering feature change events, in one implementation, the system first retrieves the associated measure triggering relationships based on the changed engineering feature node to determine the cost constituent nodes of the newly added or invalidated measures. If a measure changes from invalid to effective, the resource details corresponding to the measure are incorporated into the resource set of the bill of quantities item, and then the basic direct cost and measure cost of the bill of quantities item are recalculated. If a measure changes from effective to invalid, the resource corresponding to the measure is deleted from the resource set, and then the relevant nodes are recalculated. This allows engineering feature changes to have a structured impact at the pricing layer, rather than simply rewriting a single numerical field.

[0078] For regional rule switching events, the system first checks whether the existing cost composition nodes and fee collection nodes are still compatible based on the applicable scope and cost attribution description in the target regional rule nodes. If they are not compatible, the cost composition relationship and fee collection propagation relationship are re-bound. Then, the relevant nodes are recalculated. For list version change events, after re-executing the aforementioned candidate mapping and constraint propagation process, the system deletes the project-level mapping relationship edges of the affected list items under the old version and replaces them with the target mapping relationship edges of the new version, and then reconstructs the pricing dependency subgraph structure within the affected scope.

[0079] In S405, after recalculation, the system generates dynamic pricing results. These results include at least the updated bill of quantities unit price, quantity, total price, details of measures costs, details of charges, and total cost. Simultaneously, for each bill of quantities item, the system also outputs the source bill of quantities item identifier, target mapping object identifier, mapping type, consistency score, trigger event type, and recalculation path. The consistency score is directly derived from the global consistency score obtained during the target mapping relationship determination phase. The recalculation path originates from the propagation path record in the pricing dependency subgraph, recording the propagation sequence of this event from the initial node to the target bill of quantities item node. This maintains both the integrity of the result data and preserves the traceable path of mapping decisions and dynamic recalculation.

[0080] In one example, if an old comprehensive bill of quantities item is split into a main item and an additional measure item in the new version, the system first extracts the name, location, material grade, and engineering characteristics of the old item from the standardized dataset, and then filters the knowledge graph to a set of candidate sub-items under the target version and target region. Next, the system uses a sentence vector model and local scoring to generate a set of candidate mappings, and then uses a graph attention network to propagate unit compatibility, caliber dependency, and feature triggering constraints to obtain a global consistency score for multiple candidate combinations. If one of the combinations meets the requirements in terms of quantity coverage, feature coverage, and cost composition coverage, then that combination is determined as the target mapping relationship. Subsequently, the system generates a new pricing dependency subgraph based on this mapping relationship. If the material price changes later, dirty markings are propagated from the material node, and the resource costs, measure costs, and fee collection nodes of the main item are recalculated, finally outputting the updated bill of quantities total price and total cost. If the engineering characteristics change again later, such as adding high formwork conditions, the system rebinds the relevant measure cost composition nodes from the engineering characteristic nodes and performs a local recalculation again.

[0081] Through the above processing flow, in scenarios such as bill of quantities version migration, regional rule changes, engineering feature adjustments, and material price updates, the system can complete standardized parsing, terminology normalization, knowledge graph modeling, candidate mapping screening, constraint propagation reasoning, and local recalculation within a unified data organization structure. It maintains the consistency of bill of quantities mapping relationships, measurement caliber relationships, engineering feature triggering relationships, and pricing transmission relationships, and outputs dynamic pricing results with mapping basis and recalculation links.

[0082] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0083] Please see Figure 5 , Figure 5 This application provides a structural block diagram of an intelligent analysis and dynamic pricing system for engineering cost data. The system specifically includes: The data acquisition and standardization module 501 is used to acquire the bill of quantities, quota item table, regional pricing rule text, engineering feature text and material price data, and to perform field unification, unit mapping and source marking processing on the data to form a standardized data set; The entity recognition and knowledge graph construction module 502 is used to perform entity recognition and terminology normalization processing on the text data in the standardized data set to obtain list entities, engineering feature entities, measurement caliber entities and rule entities, and to construct a cross-version list knowledge graph based on the standardized data set and the entities. The list mapping reasoning module 503 is used to generate a candidate mapping set in the cross-version list knowledge graph for the list items to be mapped; construct a candidate mapping subgraph based on the candidate mapping set; and perform constraint propagation reasoning on the candidate mapping subgraph to obtain the target mapping relationship. The dynamic pricing calculation module 504 is used to construct a pricing dependency subgraph based on the target mapping relationship; when responding to a triggering event, it determines the affected nodes based on the pricing dependency subgraph and performs local recalculation to output the dynamic pricing result.

[0084] It should be noted that the working process of each module in the intelligent analysis and dynamic pricing system for engineering cost data described in this embodiment can refer to the working process of the intelligent analysis and dynamic pricing method for engineering cost data described in the above embodiments, and the technical effects achieved are the same as those of the intelligent analysis and dynamic pricing method for engineering cost data described in the above embodiments, so they will not be repeated here.

[0085] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for intelligent analysis and dynamic pricing of engineering cost data, characterized in that, include: Obtain the bill of quantities, quota item table, regional pricing rule text, project feature text, and material price data; perform field unification, unit mapping, and source marking on the data to form a standardized data set. Entity recognition and terminology normalization are performed on the text data in the standardized dataset to obtain list entities, engineering feature entities, measurement caliber entities, and rule entities; a cross-version list knowledge graph is constructed based on the standardized dataset and the entities. For each item in the list to be mapped, a candidate mapping set is generated in the cross-version list knowledge graph; a candidate mapping subgraph is constructed based on the candidate mapping set, and constraint propagation reasoning is performed on the candidate mapping subgraph to obtain the target mapping relationship; Construct a pricing dependency subgraph based on the target mapping relationship; In response to the triggered event, the affected nodes are determined based on the pricing dependency subgraph, and a local recalculation is performed to output the dynamic pricing result.

2. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 1, characterized in that, The data is subjected to field unification, unit mapping, and source marking to form a standardized data set, including: The coding fields in the bill of quantities and quota item tables are uniformly formatted as strings, the quantity fields are uniformly formatted as numbers, and the unit fields are mapped to standard unit codes according to the standard unit table. For convertible units, the quantity of work is converted based on the conversion factor in the standard unit table; for units with the same dimensions but whose measurement caliber is not consistent, only the unit compatibility status is recorded. Write the source type, source location, version identifier, region identifier, and project identifier for each structured record and each text fragment.

3. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 2, characterized in that, Perform entity recognition and terminology normalization on the text data in the standardized dataset, including: Input a text segment into a sequence labeling model consisting of a RoBERTa whole-word mask pre-trained encoding layer, a bidirectional long short-term memory network layer, and a conditional random field layer connected in sequence, and output an entity label sequence. Based on the entity tag sequence, extract list terms, component locations, material specifications, construction techniques, additional conditions, measurement boundary terms, and version indicator terms; Then, based on the terminology list, thesaurus, and contextual semantic similarity, the extraction results are mapped to standard term identifiers, and the entity type, source fragment, and entity confidence are recorded.

4. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 3, characterized in that, The cross-version list knowledge graph includes list item nodes, quota sub-item nodes, engineering feature nodes, unit of measurement nodes, measurement caliber nodes, cost composition nodes, version nodes, regional rule nodes, and material nodes. The relationships in the cross-version list knowledge graph include synonym relationships, version mapping relationships, parent-child splitting relationships, unit compatibility relationships, unit conversion relationships, caliber dependency relationships, feature triggering relationships, cost composition relationships, and recalculation transmission relationships. Each relation is bound to a relation source and a relation confidence level. The relation source includes at least structured comparison data, rule text parsing results, and verified model extraction results.

5. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 4, characterized in that, The step of generating a candidate mapping set in the cross-version list knowledge graph for the list items to be mapped includes: The cross-version list knowledge graph is filtered by version and region based on the target version and target region; the name, location, engineering features and unit of the list item to be mapped are combined and input into the sentence vector model to obtain the list vector to be mapped; candidate list vectors are generated for the filtered candidate list items; Based on the semantic similarity between the list vector to be mapped and the candidate list vector, and combined with coding proximity, unit compatibility, engineering feature consistency and regional rule compatibility, candidate list items are selected to form a candidate mapping set.

6. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 5, characterized in that, The step of constructing a candidate mapping subgraph based on the candidate mapping set and performing constraint propagation reasoning on the candidate mapping subgraph to obtain the target mapping relationship includes: Construct a candidate mapping subgraph using items to be mapped, candidate items, engineering feature entities, unit of measurement nodes, measurement caliber nodes, version nodes, and regional rule nodes; Name semantic similarity, encoding proximity, unit compatibility, engineering feature consistency, and regional rule compatibility are used as the initial features for candidate edges; The node features and initial candidate edge features are input into the graph attention network to obtain the updated candidate edge scores; the target mapping relationship is determined based on the updated candidate edge scores.

7. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 6, characterized in that, The step of determining the target mapping relationship based on the updated candidate edge scores includes: For each candidate edge, a global consistency score is calculated. The global consistency score consists of at least a split closure consistency item, a measurement caliber consistency item, a feature trigger integrity item, and a region rule adaptation item. For candidate combinations that have a correspondence between the old version parent item and multiple new version sub-items, calculate the combination quantity coverage, feature coverage, and cost composition coverage. Based on the global consistency score and the coverage of the candidate combinations, one of the following is determined as the target mapping relationship: direct correspondence mapping, split mapping, merge mapping, or conditional mapping.

8. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 7, characterized in that, The step of constructing a pricing dependency subgraph based on the target mapping relationship includes: Using the mapped project list item as the core node, connect its associated quota sub-item node, labor node, material node, machinery node, measure cost composition node and fee calculation node to the same graph structure; Establish the following relationships in the graph structure: quantity dependency, consumption dependency, price reference, measure triggering, and cost transmission. The target mapping relationship is then written into the project-level mapping relationship edge to ensure that the list item nodes in the pricing dependency subgraph are consistent with the target mapping relationship in the cross-version list knowledge graph.

9. The intelligent analysis and dynamic pricing method for engineering cost data according to claim 8, characterized in that, The triggering events include list version change events, regional rule switching events, engineering feature change events, and material price update events; The step of determining the affected nodes and performing local recalculation based on the pricing dependency subgraph includes: The initial node set is determined according to the triggering event type; dirty tags are propagated from the initial node set to subsequent nodes based on recalculation propagation relationships, price reference relationships, measure triggering relationships, and fee collection propagation relationships. Nodes marked with "dirty" are recalculated in dependency order, and the node status version number is updated after the node recalculation is completed. The recalculation of nodes marked with "dirty" in dependency order includes: first, calculating the basic direct cost of the list item based on the mapped quota sub-items and resource details; then, updating the measure cost composition node based on the engineering feature change results; and finally, updating the management fee node, regulatory fee node, and tax node based on the fee collection caliber corresponding to the regional rule node. Before recalculation, for list version change events and regional rule switching events, a candidate mapping set is regenerated for the affected list items and the target mapping relationship is redefined. Then, the corresponding pricing dependency subgraph is reconstructed based on the updated target mapping relationship.

10. A system for intelligent analysis and dynamic pricing of engineering cost data, characterized in that, include: The data acquisition and standardization module is used to acquire bill of quantities, quota item tables, regional pricing rule texts, engineering feature texts, and material price data, and to perform field unification, unit mapping, and source marking processing on the data to form a standardized data set. The entity recognition and knowledge graph construction module is used to perform entity recognition and terminology normalization processing on the text data in the standardized data set to obtain list entities, engineering feature entities, measurement caliber entities and rule entities, and to construct a cross-version list knowledge graph based on the standardized data set and the entities. The list mapping reasoning module is used to generate a candidate mapping set in the cross-version list knowledge graph for the list items to be mapped; construct a candidate mapping subgraph based on the candidate mapping set; and perform constraint propagation reasoning on the candidate mapping subgraph to obtain the target mapping relationship. The dynamic pricing calculation module is used to construct a pricing dependency subgraph based on the target mapping relationship; When responding to a triggering event, the affected nodes are determined based on the pricing dependency subgraph and a local recalculation is performed to output dynamic pricing results.

Images