An artificial intelligence-based construction plan review method and system for the construction industry

Abstract

The present invention discloses an artificial intelligence-based construction plan review method and system thereof. First, the system is trained and learned based on the sorted sample materials of historical manual review of construction plans and the historical manual review rules. Then, the system that has completed training and learning is used to conduct an intelligent review of the construction plan to be reviewed, and the intelligent review results are manually rechecked. The artificial intelligence-based construction plan review method and system provided by the present invention innovatively use AI intelligence to accurately identify and review data, saving the labor cost and time cost of plan review, improving the review efficiency, review quality and intelligence level of the plan. On this basis, an organic integration with manual secondary review is also realized, enabling the targeted conduct of re-review work according to the results of machine review, with extremely high degrees of freedom and flexibility. The combination of man and machine accurately breaks through various content risks and improves the operation efficiency.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to an auditing scheme for construction project construction plans. Background Art

[0002] With the increasingly fierce market competition in the construction industry, construction enterprises are facing huge pressure for survival and development. Strengthening the foundation, making up for weaknesses, and focusing on new infrastructure and major projects are important development directions for construction enterprises. Utilizing information technology means to pursue refined management will be the key development direction for construction enterprises.

[0003] Project construction plans are document materials that construction enterprises need to write a large amount of daily, which contain a large amount of proofreading and auditing work.

[0004] In the current scheme, the proofreading and auditing of project construction plans are still based on manual methods. For staff, the work labor intensity is high, and for project construction, the efficiency is low and the accuracy is insufficient.

[0005] Therefore, it can be seen that how to improve the efficiency and accuracy of proofreading and auditing of project construction plans is an urgent problem to be solved in this field. Summary of the Invention

[0006] In view of the problems of low efficiency and insufficient accuracy in the manual auditing of current construction plans, the present invention provides an auditing method for construction project construction plans based on artificial intelligence. This scheme accurately identifies and audits data based on AI intelligence, and on this basis, integrates manual secondary auditing, greatly improving the efficiency and accuracy of proofreading and auditing of project construction plans; on this basis, the present invention also provides an auditing system scheme that can implement this auditing method.

[0007] In order to achieve the above object, the auditing method for construction project construction plans based on artificial intelligence provided by the present invention includes:

[0008] S1: Sort out the sample materials of the construction plans audited manually in history, and sort them according to different types of construction plans; and form an auditing content semantic table according to all the contents and specific extraction formats of the historical manual auditing, and obtain the rules of the historical manual auditing;

[0009] S2: Parse and identify different formats of documents included in the construction plan and convert the document formats;

[0010] S3: Mark and extract the feature contents after parsing and identifying in step S2. Here, only the relevant contents in the auditing content semantic table formed in step S1 are marked, and the document format document data is extracted;

[0011] S4: Conduct manual auditing on the data marked in step S3;

[0012] S5: Train a model using the labeled data after manual review in step S4 to train the paragraph positions of the labeled content in the document, and form paragraph position coordinate data;

[0013] S6: Establish a system rule library based on various review rules determined in step S1;

[0014] S7: Parse and identify the construction plan document to be reviewed and convert the document format, and combine it with the rule library for intelligent review; in this step, compare the review rules in the rule library established in step S6 with the parsed and identified content to check whether they are consistent, and form a review result accordingly: if they are inconsistent, display that the review fails; if they are consistent, display that the review passes;

[0015] S8: Classify the status of documents with different reviews;

[0016] S9: Generate a review report for the construction plans that have been reviewed by the system and manually reviewed, which is convenient for review experts to consult and subsequent archiving.

[0017] In some examples of the present invention, the review method includes a step of optimizing the training result of step S5 through rules.

[0018] In some examples of the present invention, the review method further includes a step of manually revising the review result of step S7.

[0019] To achieve the above object, the construction plan review system based on artificial intelligence provided by the present invention includes

[0020] A document parsing and recognition module, which is used to parse, recognize and convert the document format of the files included in the construction plan;

[0021] A document annotation module, which is used to annotate the files processed by the document parsing and recognition module, and annotate the content segments related to extraction and semantic table;

[0022] A model training module, which trains a model based on the sample files marked and processed by the document annotation module to train the paragraph positions of the marked content in the document, and form paragraph position coordinate data;

[0023] An intelligent review library module, which constructs an intelligent review rule corpus based on the sorted manual review rules;

[0024] An intelligent recognition and review module, which retrieves the document parsing and recognition module, and based on the model trained by the model training module and the intelligent review rule corpus constructed by the intelligent review library module, performs intelligent extraction and review of the review content of the construction plan to be reviewed;

[0025] The review task management module identifies the status of extraction review tasks at different stages;

[0026] The review report generation module generates a review report for the reviewed construction plan.

[0027] In some examples of the present invention, the document parsing and recognition module sequentially performs image correction processing, text detection processing, text recognition processing, and semantic correction processing on the file.

[0028] In some examples of the present invention, the model training module performs structured extraction according to the paragraph position of the training annotation content in the document to form a structured triple <head entity, relationship, tail entity>, and then performs sequence annotation to extract keywords in the semantic table, and finally forms business extraction data for comparison with the rules in the rule library.

[0029] In some examples of the present invention, the system further includes an effect optimization module. The effect optimization module performs data interaction with the model training module to optimize the effect of some content after model training, and performs text classification and feature extraction on the data that needs to improve the integrity or accuracy of the business extraction content.

[0030] In some examples of the present invention, when the effect optimization module performs text classification, according to the set of training documents that have been annotated in model training, through feature extraction, it finds the relationship model between the document features and the document categories, and then uses the obtained relationship model to judge the category of new documents to obtain the most similar semantic table data.

[0031] In some examples of the present invention, the intelligent review library module imports the rules of manual review into the system database, then corresponds to the extracted content for relationship, performs consistency comparison, and displays the comparison results.

[0032] In some examples of the present invention, the intelligent recognition and review module uses the document parsing and recognition module to parse the text and tables in the relevant photos and scanned copies in the construction plan to be reviewed, uses the model training module to perform model training on the text recognition and table parsing content to form semantic table extraction content, and performs consistency comparison on the semantic table extraction content.

[0033] The artificial intelligence-based construction plan review method and system provided by the present invention innovatively uses AI intelligence to accurately identify and review data, saving the labor cost and time cost of plan review, improving the plan review efficiency, review quality, and intelligence level; on this basis, it also realizes an organic integration with the secondary manual review, realizes the targeted development of the review work according to the machine review results, with extremely high freedom and flexibility, combines human and machine, accurately breaks through various content risks, and improves the operation efficiency.

[0034] The artificial intelligence-based construction plan review method and system provided by the present invention can, in actual application, provide a unified plan review platform for construction plan review, convert the offline plan review method into an online intelligent centralized plan review method, and realize intelligent plan review by solidifying the plan review standard and providing technical knowledge big data that can quickly extract the target content. At the same time, according to the plan content, a plan review report is intelligently generated for reference and execution by on-site production personnel. After the system is applied, the integrity, compliance, and accuracy requirements of the plan can be guaranteed, while standardized management of plan review can be realized, the review efficiency and quality of the plan can be improved, and the protection of the core technical knowledge of the enterprise can be strengthened. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0036] Figure 1 This is an example diagram of the process of reviewing construction plans in the construction industry based on artificial intelligence in an example of the present invention;

[0037] Figure 2 An example diagram of the audit calculation diagram in the example of the present invention;

[0038] Figure 3 This is an example diagram of a composite node computation graph formed in an example of the present invention. DETAILED DESCRIPTION

[0039] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below with reference to specific diagrams.

[0040] In response to the problems of low efficiency and insufficient accuracy in the current manual review of construction plans, the present invention provides an artificial intelligence-based review solution for construction plans in the construction industry. This solution trains on massive samples and integrates a large amount of manual review experience to connect the machine review and human review processes. It uses AI intelligence to accurately identify and review data, and then integrates manual secondary review with the review results of intelligent machine review. In this way, re-examination work can be carried out in a targeted manner according to the results of machine review, with extremely high freedom and flexibility. The combination of man and machine can accurately break down various content risks and improve operational efficiency.

[0041] Based on the in-depth research and application of natural language processing technology (NLP), the inventor of this invention application uses deep learning machine learning technology to conduct multi-model training on massive sample libraries, custom configurations of violation vocabulary, and tens of thousands of variant types, and combines internal and external data to conduct accurate review of business materials.

[0042] Specifically, the present invention provides a review scheme for construction plans in the construction industry based on artificial intelligence to complete the intelligent pre-review of typical construction plans for building projects, saving the labor cost and time cost of plan review, and improving the review efficiency, review quality and intelligence level.

[0043] Combined with Figure 1 As shown, the review scheme for construction plans in the construction industry based on artificial intelligence provided by the present invention specifically includes the following process.

[0044] S1. Preparation for organizing model training materials.

[0045] In this step, historical construction plan materials are organized according to different types and used as sample data for subsequent model training.

[0046] As an example, here the historical construction plan materials can be the construction plan for cantilever scaffolding, the construction plan for construction organization design, the construction plan for high formwork, etc.

[0047] Furthermore, about 200 pieces are organized for each type. The more complete the types organized in this step, the more accurate the data extracted by artificial intelligence in the later stage.

[0048] S2. Preparation for organizing model annotation content.

[0049] In this step, classification and organization are carried out according to different format types: paragraphs, tables, and pictures.

[0050] As an example, here it can be according to the construction personnel list, the construction progress schedule, the calculation book for the formwork support of fastener steel pipes for floors, etc.

[0051] In this step, all the contents of the historical manual review and the specific extraction format are specifically confirmed and organized into a semantic table of review content. For example, the basis for compilation (paragraph), project overview (paragraph), general situation of building design (table), calculation book (paragraph), etc. And the content of the formed document will be used as the elements for later artificial intelligence extraction, directly affecting the integrity of the subsequent extracted content.

[0052] S3. Preparation for the content of model review rules.

[0053] In this step, compliance, integrity, accuracy, consistency, etc. are determined according to different review rule types. For example, the compliance of the basis for compilation, the integrity of the project overview, the standardization of building design parameters, the accuracy of the results of the calculation book, etc.

[0054] As an example, here whether the basis for compilation complies with the national administrative document "Work Safety Law of the People's Republic of China", the local administrative document "Regulations on the Management of Construction Safety in Shandong Province", the national industry standard / national standard / industry standard "Safety Technical Code for Construction of Fastener Steel Pipe Scaffolds", the contract, the design document, the enterprise technical standard, etc.

[0055] The content determined in this step will serve as the source of the rule database for later artificial intelligence review. The clearer the rules, the more efficient the later review process will be.

[0056] S4. Establish the OCR recognition engine.

[0057] The OCR recognition engine constructed in this step can support the text parsing and recognition of different document formats, such as.pdf,.doc,.docx,.txt,.jpg,.jpeg,.png, etc.

[0058] In addition, for the construction plan documents which are usually in word format, to ensure that the recognition accuracy rate of the calculation formula reaches 100% during parsing and recognition, the OCR recognition engine constructed in this step can accurately recognize the calculation symbols of the calculation formula. Based on this, this solution will be able to accurately parse and recognize WORD documents.

[0059] For example, the stability calculation of the vertical pole in the construction plan:

[0060] (1) When not considering the wind load, the stability calculation formula of the vertical pole

[0061]

[0062] σ = 1.00×11896 / (0.387×450) = 68.250N / mm2;

[0063] (2) When considering the wind load, the stability calculation formula of the vertical pole is:

[0064]

[0065] MTk = 0.461×6.1×1.20×(0.5×6.1 + 0.60) = 12.127kN.m;

[0066] Nwk = 6×8 / (8 + 1) / (8 + 2)×(12.127 / 8.00) = 0.808kN;

[0067] Nw = 1.200×6.553 + 1.400×2.880 + 1.40×0.6×0.808 = 12.575kN;

[0068] σ = 1.00×(12575 / (0.387×450) + 105000 / 4730) = 94.241N / mm2;

[0069] For this calculation formula, the OCR recognition engine constructed in this step can accurately recognize it.

[0070] Specifically, in order to support the text parsing and recognition of different document formats, the OCR recognition engine constructed in this step is mainly composed of an image correction module, a text detection module, a text recognition module, and a semantic correction module, which cooperate with each other in sequence.

[0071] Among them, the image correction module is used to restore pictures with rotation, tilt, perspective, wrinkles, etc. to flat and regular pictures, which is convenient for the subsequent model to play, thereby improving the recognition effect.

[0072] As an example, the image correction module preferably uses algorithms such as Fast RCNN and YOLO to implement.

[0073] The text detection module interacts with the image correction module and is used to find the text area as accurately as possible from the pictures corrected by the image correction module.

[0074] As an example, the text detection module preferably uses algorithms such as CTPN and FPN to implement.

[0075] The text recognition module interacts with the text detection module and is used to accurately recognize the specific characters in the detected text pictures, so as to convert them into character sequences that can be understood by a computer.

[0076] As an example, the text recognition module preferably uses algorithms such as CRNN, Attention OCR, and Bi-LSTM to implement.

[0077] The semantic correction module interacts with the text recognition module and is used to perform semantic-level correction on the output results of the text recognition module based on natural language processing methods, thereby further improving the accuracy of OCR recognition.

[0078] As an example, the semantic correction module preferably uses algorithms such as BERT, Bi-LSTM, and CNN to implement.

[0079] In this step, the document parsing and recognition function is realized through the established OCR recognition engine, and different formats of documents are parsed, recognized, and document format conversion is performed. The processed files will be used as the basis for subsequent construction plan material parsing and annotation, and intelligent extraction and review of construction plans.

[0080] S5. Establish an artificial annotation platform.

[0081] The artificial annotation platform established in this step can cooperate with the constructed OCR recognition engine and is used to perform artificial annotation on the content of the construction plan after OCR parsing and recognition by the OCR recognition engine for different purposes.

[0082] The manual annotation platform provides a corresponding human-computer interaction module to perform manual annotation processing on the content of the construction plan after OCR parsing and recognition by the OCR recognition engine.

[0083] As an example, the human-computer interaction module here supports basic operation modes such as [select by drawing], [select by bounding box], [editing], etc.

[0084] Based on this manual annotation platform, the corresponding file content formed after OCR parsing and recognition of the annotation content prepared in step S2 can be annotated. The annotated file content will be used as the basis for subsequent model training. The more accurate the annotation data content in this step, the higher the accuracy of the extraction and recognition results in subsequent model training.

[0085] S6. Check and review the annotation quality.

[0086] This step is based on the manual annotation platform built in step S5 to review and confirm the manually annotated data content through human-computer interaction to improve the annotation quality.

[0087] Preferably, here, the review and confirmation are carried out by sampling and checking the manually annotated data content. If there is any deviation in the annotation content, all tasks within the scope involved in the sample are repeatedly checked.

[0088] S7. Establish a machine learning model.

[0089] This step performs model training on the materials annotated, trained, and reviewed in steps S5 and S6; here, it mainly trains the paragraph positions of the annotation content in the document to form paragraph position coordinate data.

[0090] Specifically, the implementation of this step's solution is mainly achieved through the mutual cooperation of a structured extraction module and a sequence annotation module.

[0091] 1) Structured extraction module

[0092] This structured extraction module is used to perform structured extraction according to the paragraph positions of the training annotation content in the document to form a structured triple <head entity, relationship, tail entity>.

[0093] Regarding the specific composition and implementation solution of this structured extraction module, no limitations are imposed here and it can be determined according to actual needs.

[0094] 2) Sequence annotation module

[0095] This sequence annotation module includes a series of basic sequence annotation methods, including the following:

[0096] BiLSTM-CRF

[0097] CRF conditional random field

[0098] Transformer

[0099] The sequence annotation module extracts characteristic training sequence annotation models from the annotated text data for extracting text fragments with specified meanings from continuous text data.

[0100] Accordingly, this step is based on the machine learning model composed of the structured extraction module and the sequence annotation module. First, according to the paragraph positions of the training annotations in the document, structured extraction is performed to form structured triples <head entity, relationship, tail entity>, and then sequence annotation is performed to extract keywords in the semantic table. Finally, the business extraction data is compared with the rules in the rule library.

[0101] The machine learning model established in this step can be used to perform intelligent identification and review processing on the processed construction plan documents after corresponding training and learning. The obtained results will be used as the basis for the review extraction content.

[0102] Establish the S8 model training effect optimization mechanism.

[0103] In this step, a corresponding effect optimization mechanism is constructed to optimize the file content processed by the model training in step S7.

[0104] The optimization processing of the file content processed by the model training in step S7 in this step mainly includes text classification and feature extraction for the data that needs to improve the integrity or accuracy of the business extraction content.

[0105] Furthermore, when optimizing the effects of some content after model training in this step, it is preferable to regularly (such as every 3 months) collect the semantic table extraction content that may need to improve the integrity and accuracy, and after reaching a certain threshold, for example, the same type of material appears more than 10 times, perform centralized processing, so as to continuously improve the accuracy of the extraction content. In this way, text classification and feature extraction for the data that needs to improve the integrity or accuracy of the business extraction content can make it tend to be consistent with the semantic table of the review content.

[0106] The solution of this step is mainly composed of the cooperation between the feature extraction module and the text classification module when implemented specifically.

[0107] 1) Feature extraction module

[0108] The feature extraction module is used to perform feature extraction and feature weight calculation to find the semantic keywords of the set of annotated training documents.

[0109] Feature extraction here can reduce the dimensionality of the vector space without damaging the core information, simplify the calculation, and improve the speed and efficiency of text processing. Feature extraction uses methods based on information gain and PCA to find the most classification-informative features.

[0110] Feature weights are used to measure the importance or discriminative ability of a certain feature term in document representation. The feature weight calculation function uses multiple dimensions such as TF-IDF, part of speech, position, syntactic structure, and professional word library to calculate the weights of word features in the text, so as to improve the classification effect of text classification.

[0111] 2) Text classification module

[0112] This text classification module calls the IDPS intelligent classification module. For different extraction scenarios such as chapter location, paragraph extraction, table extraction, etc., it finds the relationship model between document features (annotated documents) and document categories (semantic table documents), and then uses the obtained relationship model to judge the category of new documents and obtain the most similar semantic table data.

[0113] Specifically, this text classification module uses, but is not limited to, the following basic classifiers:

[0114] Support Vector Machine (SVM)

[0115] xgboost

[0116] TextCNN

[0117] The basic classifier accepts the feature vectors generated by the feature processing module and trains the classification model.

[0118] Accordingly, in this step, when optimizing the file content processed by the model training in step S7 based on the established model training effect optimization mechanism, according to the set of training documents that have been annotated in model training, through feature extraction, the relationship model between document features (annotated documents) and document categories (semantic table documents) is found, and then the obtained relationship model is used to judge the category of new documents and obtain the most similar semantic table data.

[0119] S9 Establishment of intelligent review library.

[0120] This step is based on the manual review rules sorted out in step S3 to establish an intelligent review library. Through configuration and model algorithm training, it can quickly replace manual work for construction plan review.

[0121] In this step, the rules for manual review are imported into the system database to form an intelligent review library. This intelligent review library corresponds to the extracted content based on the imported manual review rules for consistency comparison and displays the comparison results.

[0122] In specific implementation, the intelligent review library constructed in this step serves as the basis for implementing intelligent review of construction plan documents. In this step, based on the manual review rules sorted out in step S3, summarization and data processing are carried out, and the corresponding business rules are expressed in computer language and imported into the system database to form the corresponding intelligent review library. On this basis, combined with underlying information extraction, the manual review rules imported into the review library are matched and compared with the content extracted by the model to check consistency, so that the manual rules configured in the system can correspond to the content extracted by the model, and the results of inconsistent and consistent comparisons are displayed to achieve a complete end-to-end review business.

[0123] Specifically, in this step, the review business is first sorted out and regularized:

[0124] (1) The object is clear, and the review subject can be clearly determined.

[0125] (2) Most of the sentence patterns are in the subject-predicate-object structure. For example, "The amount is greater than 1 million", and complex rules can be split into several simple subject-predicate-object structures.

[0126] (3) The expression is flexible and changeable, and it is difficult to directly learn through the model.

[0127] (4) There are dependencies among the review rules.

[0128] Based on the above sorted out and regularized content, the solution of the present invention innovatively uses the way of computational graph to express the review business, and at the same time makes the review rules correspond to the nodes on the computational graph. Each node is expressed by the SPO (Subject, Predicate, Object) triple. The document content and the extracted results flow through the computational graph in the form of data stream. When the data flows through the entire computational graph, the entire review business is completed.

[0129] SPO (Subject, Predicate, Object) triple:

[0130] Adopt the SPO expression method commonly used in knowledge graphs. These three attributes can be understood as (entity one, predicate, entity two). This predicate defines the relationship between entity one and entity two.

[0131] As an example, in the above "The amount is greater than 1 million", in the SPO expression system, entity one is "amount", the predicate is "greater than", and entity two is "1 million".

[0132] Accordingly, a triple is defined as a node, which serves as the audit logic at the smallest granularity. The output of each node is whether the business rules of this node are satisfied, and this result is passed in the graph.

[0133] In addition to the business logic nodes, the solution of the present invention further adds two special nodes: the starting point and the ending point, which respectively represent the entrance and the exit of the graph.

[0134] It should be noted here that for each sub-graph, such as Figure 2 Rule 1 (amount < 1 million, interest rate < 5%) and Rule 2 (amount > 1 million, interest rate < 5%) in it can also be combined into a composite node to form a combined rule (interest rate < 5%, amount < 1 million or amount > 1 million), reducing logical judgments.

[0135] The audit calculation graph is thus defined as: starting from the starting point, if there exists a path that satisfies the audit conditions from the entrance to the exit, the audit passes; otherwise, it fails.

[0136] As an example:

[0137] Rule 1: The amount is less than 1 million and the interest rate is less than 5%.

[0138] Rule 2: The amount is greater than 1 million and the interest rate is less than 4%

[0139] Rule 3: Satisfy Rule 1 or Rule 2, and the loan period does not exceed 2 years.

[0140] The corresponding calculation graph is as shown in Figure 2 If there are two paths in the graph and one of them passes, it means the audit passes.

[0141] See Figure 3 , when there are rule nestings or the sub-graphs are extremely complex, the sub-graphs can be further abstracted into a composite node through the extensibility of the graph, thereby simply expressing various complex business logics, realizing the conversion from business to computer language, and further realizing the automated audit business.

[0142] S10. Intelligent recognition and audit.

[0143] This step is based on the foundation constructed in the previous steps 1 - 9. Based on the OCR recognition engine and the machine learning model, the content to be audited in the construction plan is intelligently extracted and audited, and the audit result is output.

[0144] In this step, the corresponding OCR recognition engine is used to parse the text and tables in the photos and scanned copies of the construction plan; on this basis, the text recognition and table parsing content are model-trained based on the machine learning model to form the semantic table extraction content, and further the consistency comparison is carried out on the semantic table extraction content.

[0145] S11. Human-machine review.

[0146] In this step, a manual review is conducted on the comparison results obtained after the intelligent recognition review in step 10, thereby realizing the combination of the system's intelligent review results and manual review, and improving the accuracy of the review results.

[0147] When implementing this step, for the extracted review results, the original text files with annotations corresponding to the review results are synchronously extracted; then the original text files with annotations and the review results are displayed side by side, and at that time, manual editing and processing of the review results are carried out to complete the manual review.

[0148] As an example, after extracting the original text files with annotations corresponding to the review results, the original text with annotations can be displayed on one side of the display screen of the human-machine interaction device, and the review results on the other side; at the same time, the human-machine interaction device provides corresponding editing function modules, so as to realize manual intervention in the review results, edit the review status as passed, not passed, passed (manually), and the original text on the left can be selected by box for content annotation to add review comments.

[0149] S12. Establish the review task status.

[0150] In this step, status recognition is carried out for the extraction review tasks in different stages in step S11.

[0151] As an example, the task status recognized in this step can be: recognition failure, recognition in progress, recognition result, not reviewed, review in progress, system has reviewed, manual has reviewed, etc., so as to efficiently screen and control the document processing flow and improve the efficiency of document human-machine combined processing.

[0152] S13. Create an extraction review report.

[0153] Based on steps S11 and S12, in this step, corresponding review reports are constructed for the construction plans that have been reviewed by the system and manually.

[0154] In this step, the specific composition of the review report is not limited and can be determined according to actual needs.

[0155] When implementing this step specifically, it supports exporting the review report. The exported review report can display the system review opinions and expert review opinions, and highlight the key review contents. Thus, it realizes strict control of the construction plan risks, does a good job in the pre-construction work, and reduces the subsequent construction risks.

[0156] For the construction plan review solution based on artificial intelligence given in this example solution, during specific application, it can form a corresponding software program to form a corresponding intelligent construction plan review system. When this software program runs, it will execute the above-mentioned construction plan review method solution based on artificial intelligence and be stored in a corresponding storage medium for the processor to retrieve and execute.

[0157] The intelligent construction plan review system formed hereby mainly includes, in terms of functions: a database 100, an artificial annotation platform 200, an intelligent review platform 300, and a user platform 400.

[0158] The database 100 here is used to store the basic data required by the review system and various data generated during the operation process.

[0159] Specifically, this database 100 conducts data interaction with the intelligent review platform 300 and the artificial annotation platform 200, and stores the model training materials processed by OCR-recognized documents and the construction plan files to be reviewed processed by OCR-recognized documents. The specific content is as described above and will not be elaborated here.

[0160] The artificial annotation platform 200 in this system conducts data interaction with the intelligent review platform 300, and provides a human-computer interaction method to conduct artificial annotation and annotation content review on the construction plan content processed by OCR-recognized documents.

[0161] Specifically, a corresponding document annotation module 210 and an annotation content review module 220 are running in this artificial annotation platform 200.

[0162] Among them, this document annotation module 210 is based on manual operation to annotate the files obtained by processing OCR-recognized documents, and annotate the semantic table content to be extracted.

[0163] This annotation content review module 220 conducts data interaction with the document annotation module 210, and conducts review and confirmation operations on the manually annotated data content in a human-computer interaction manner.

[0164] The specific solutions of the aforementioned steps S5 and S6 can be completed through this artificial annotation platform 200.

[0165] The intelligent review platform 300 in this system conducts data interaction with the artificial annotation platform 200 and the database 100, completes the review model training based on the prepared model training materials, constructs an intelligent review library based on the sorted artificial review rules, and then conducts intelligent extraction and review of the content to be reviewed of the construction plan based on the trained review model and the constructed intelligent review library, and outputs the review results.

[0166] Specifically, the intelligent audit platform 300 mainly includes a document parsing and recognition module 310, a model training module 320, an effect optimization module 330, an intelligent audit library module 340, and an intelligent recognition and audit module 350.

[0167] Among them, the document parsing and recognition module 310 is used to perform OCR parsing and recognition on the documents included in the construction plan and convert the document format.

[0168] As an example, the document parsing and recognition module 310 can sequentially perform image correction processing, text detection processing, text recognition processing, and semantic correction processing on the file to be recognized and processed.

[0169] The specific solution of the foregoing step S4 can be completed through the document parsing and recognition module 310.

[0170] The model training module 320 in this intelligent audit platform can perform model training based on the sample files that have been marked and audited by the manual marking platform 200, and train the paragraph positions of the marked content in the document to form paragraph position coordinate data.

[0171] Furthermore, the model training module 320 performs structured extraction according to the paragraph positions of the training marked content in the document to form a structured triple <head entity, relationship, tail entity>, and then performs sequence marking to extract keywords in the semantic table, and finally forms business extraction data for comparison with the rules in the rule library.

[0172] The specific solution of the foregoing step S7 can be completed through the model training module 320.

[0173] The effect optimization module 330 in this intelligent audit platform, the effect optimization module 330 performs data interaction with the model training module 320 to optimize and process the effects of some contents after model training.

[0174] The effect optimization module 330 specifically performs text classification and feature extraction on the data that needs to improve the integrity or accuracy of the business extraction content.

[0175] In some examples of the present invention, when the effect optimization module performs text classification processing, according to the set of training documents that have been marked in model training, through feature extraction, the relationship model between the document features (marked documents) and the document categories (semantic table documents) is found, and then the obtained relationship model is used to judge the category of the new document to obtain the most similar semantic table data. The effect optimization module processes the image through feature extraction.

[0176] The specific solution of the foregoing step S8 can be completed through the effect optimization module 330.

[0177] The intelligent review library module 340 in this intelligent review platform constructs an intelligent review rule corpus based on the sorted manual review rules.

[0178] In some examples of the present invention, the intelligent review library module 340 imports the manual review rules into the system database, then corresponds with the extracted content, conducts consistency comparison, and displays the comparison results.

[0179] The specific solution of the foregoing step S9 can be completed through the intelligent review library module 340.

[0180] The intelligent recognition and review module 350 in this intelligent review platform conducts data interaction with the document parsing and recognition module 310, the model training module 320, and the intelligent review library module 340. By invoking the document parsing and recognition module, and based on the model trained by the model training module and the intelligent review rule corpus constructed by the intelligent review library module, it conducts intelligent extraction and review of the review content for the construction plan to be reviewed.

[0181] In some examples of the present invention, the intelligent recognition and review module invokes the OCR recognition engine through the document parsing and recognition module 310 to parse the text and tables in the relevant photos and scanned documents in the construction plan to be reviewed. Through the model training module 320, it conducts model training on the text recognition and table parsing content based on machine learning to form semantic table extraction content, and conducts consistency comparison on the semantic table extraction content based on the core algorithm engine.

[0182] The specific solution of the foregoing step S10 can be completed through the intelligent recognition and review module 350.

[0183] The user platform 400 in this system conducts data interaction with the intelligent review platform 300, and is used to cooperate with the intelligent review platform 300 to conduct secondary manual review on the results of intelligent review.

[0184] Specifically, this user platform 400 includes a human-computer interaction module 410, a document upload module 420, an information extraction module 430, a human review module 440, a review task management module 450, and a review report generation module 460.

[0185] The human-computer interaction module 410 in this platform provides human-computer functions and can invoke all other modules in this platform.

[0186] The document upload module 420 in this platform runs in the background. It conducts data interaction with the intelligent review platform 300, obtains the results of the review by the intelligent review platform 300, and uploads the result files to the user platform 400 for subsequent extraction.

[0187] The specific composition of the document upload module 420 will not be elaborated here and can be determined according to actual needs.

[0188] The information extraction module 430 in this platform interacts with the document upload module 420, processes documents in formats such as pdf and.txt into WORD format document data, retains its data format, and obtains relevant content of the semantic table through the model.

[0189] The human-machine review module 440 in this platform can be invoked by the human-machine interaction module 410 to perform manual verification operations on the intelligent review results of the system.

[0190] The specific solution of the aforementioned step S11 can be completed through the human-machine review module 440.

[0191] The review task management module 450 in this platform can be invoked by the human-machine interaction module 410 and interacts with the human-machine review module 440 to identify the status of extraction review tasks at different stages.

[0192] The specific solution of the aforementioned step S12 can be completed through the review task management module 450.

[0193] The review report generation module 460 in this platform can be invoked by the human-machine interaction module 410 and interacts with the review task management module 450 to generate a review report for the reviewed construction plan.

[0194] The specific solution of the aforementioned step S13 can be completed through the review report generation module 460.

[0195] When the intelligent review system for construction plans in the construction industry based on artificial intelligence formed accordingly is running, first, the system is trained and learned based on the sorted sample materials of historical manually reviewed construction plans and historical manual review rules; then, the system after completing the training and learning is used to perform intelligent review on the construction plan to be reviewed, and the intelligent review results are manually re-verified.

[0196] As an example, its specific implementation process is as follows:

[0197] (1) Sort out the sample materials of historical manually reviewed construction plans and sort them according to different construction plan types, such as cantilever scaffold construction plans, construction organization design construction plans, high formwork construction plans, etc.

[0198] (2) Confirm all the contents and specific extraction formats of historical manual reviews and organize them into a review content semantic table, such as compilation basis (paragraph), project overview (paragraph), building design overview (table), calculation book (paragraph), etc.

[0199] (3) Confirm the rules for historical manual review, such as the compliance of the compilation basis, the integrity of the project overview, the standardization of building design parameters, the accuracy of the calculation results, etc.

[0200] (4) Through the document parsing and recognition module in the intelligent recognition platform of the system based on optical character recognition technology (OCR), parse and recognize documents in different formats and perform document format conversion, and process documents in formats such as pdf and.txt into WORD format document data, retaining their data formats.

[0201] (5) Through the manual annotation platform in the system, based on computer vision technology (CV), label and extract the feature content after parsing and recognition.

[0202] (6) Through the manual annotation platform in the system, business experts conduct manual review on the labeled data to ensure the accuracy of the labels.

[0203] (7) Input the obtained labeled data into the intelligent recognition platform in the system to train the model training module in the platform.

[0204] (8) Through the effect optimization module in the intelligent recognition platform, optimize the content to be improved in the effect after model training through rules.

[0205] (9) Through the intelligent review library module in the intelligent recognition platform, establish a system rule library based on various review rules sorted out manually.

[0206] (10) For the construction plan documents to be reviewed, after being recognized and processed by the document parsing and recognition module in the intelligent recognition platform, then through the intelligent recognition and review module, perform intelligent extraction and review of the review content for the construction plan to be reviewed after recognition and processing in combination with the rule library.

[0207] (11) Through the human-machine review module in the user platform of the system, conduct manual review on the intelligent review results formed by the intelligent recognition platform, and support manual modification of the content with errors in the system review.

[0208] (12) Through the review task management module in the user platform, classify the status of different reviewed documents, such as recognition failure, recognition in progress, recognition result, not reviewed, review in progress, system reviewed, manually reviewed, etc.

[0209] (13) Through the review report generation module in the user platform, support the system to export the review report for the construction plans that have been reviewed by the system and manually reviewed, which is convenient for review experts to consult and subsequent archiving.

[0210] In summary, the construction plan review solution based on artificial intelligence of the present invention can provide a unified plan review platform for the review of construction plans, convert the offline plan compilation method into an online intelligent centralized review method for plans, realize intelligent plan review by solidifying plan review criteria and providing a big data of technical knowledge that can quickly extract target content, and at the same time, intelligently generate a review report according to the plan content for on-site production personnel to refer to and execute. After the system is applied, it can ensure the integrity, compliance, accuracy, etc. of the plan, and at the same time realize the standardized management of the plan, improve the review efficiency and quality of the plan, and strengthen the protection of the enterprise's core technical knowledge.

[0211] The method of the present invention, or a specific system unit, or a part of its unit, is a pure software architecture and can be deployed on a physical medium through program code, such as a hard disk, a CD-ROM, or any electronic device (such as a smart phone, a computer-readable storage medium). When the machine loads the program code and executes (such as a smart phone loads and executes), the machine becomes a device for implementing the present invention. The method and device of the present invention can also be transmitted in the form of program code through some transmission media, such as cables, optical fibers, or any transmission type. When the program code is received, loaded, and executed by a machine (such as a smart phone), the machine becomes a device for implementing the present invention.

[0212] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based construction plan review method for the construction industry, characterized in that, Including: S1: Sort out the sample materials of the construction plan for historical manual review, and sort them according to different types of construction plans; and form a semantic table of review content and obtain the rules of historical manual review according to all the content and specific extraction formats of historical manual review; S2: Parse and identify the documents in different formats included in the construction plan and convert the document formats; S3: Mark and extract the characteristic content after parsing and identifying in step S2; S4: Conduct manual review on the marked data in step S3; S5: Train the marked data after manual review in step S4 to train the paragraph positions of the marked content in the document and form paragraph position coordinate data; S6: Establish a system rule library based on various review rules determined in step S1; S7: Parse and identify the construction plan file to be reviewed and convert the document format, and conduct intelligent review in combination with the rule library; this step verifies whether it is consistent with the rules by comparing the review rules in the rule library established in step S6 with the parsed and identified content; S8: Classify the status of different reviewed documents; S9: Generate a review report for the construction plans that have been reviewed by the system and manually, which is convenient for review experts to consult and subsequent archiving.

2. The method for reviewing construction plans in the construction industry based on artificial intelligence according to claim 1, wherein The review method includes a step of optimizing the training result of step S5 through rules.

3. The method for auditing construction project plans based on artificial intelligence according to claim 1, characterized in that, The review method further includes a step of manually revising the review result of step S7.

4. An artificial intelligence-based construction plan review system for the construction industry, characterized in that, Including A document parsing and identifying module, which is used to parse and identify the files included in the construction plan and convert the document formats; A document marking module, which is used to mark the files obtained after being processed by the document parsing and identifying module, and mark and extract the content segments related to the semantic table; A model training module, which conducts model training based on the sample files marked and processed by the document marking module to train the paragraph positions of the marked content in the document and form paragraph position coordinate data; An intelligent review library module, which constructs an intelligent review rule corpus based on the sorted manual review rules; An intelligent identification and review module, which retrieves the document parsing and identifying module, and based on the model trained by the model training module and the intelligent review rule corpus constructed by the intelligent review library module, conducts intelligent extraction and review of the review content of the construction plan to be reviewed; A review task management module, which identifies the status of extraction and review tasks at different stages; A review report generation module, which generates a review report for the reviewed construction plan.

5. The artificial intelligence-based construction plan review system according to claim 4, wherein, The document parsing and identifying module sequentially performs image correction processing, text detection processing, text recognition processing, and semantic correction processing on the files.

6. The artificial intelligence-based construction plan review system according to claim 4, wherein, The model training module conducts structured extraction according to the training of the paragraph positions of the marked content in the document to form a structured triple <head entity, relationship, tail entity>, and then conducts sequence annotation to extract the keywords in the semantic table, and finally forms business extraction data to compare with the rules in the rule library.

7. The artificial intelligence-based construction plan review system according to claim 4, wherein The system further includes an effect optimization module, and the effect optimization module conducts data interaction with the model training module to optimize the effect of some content after model training.

8. The artificial intelligence-based construction plan review system according to claim 7, characterized in that The effect optimization module performs text classification and feature extraction on data that requires improvement in integrity or accuracy in the content extracted from the business.

9. The artificial intelligence-based construction plan review system according to claim 4, wherein The intelligent audit library module imports the rules of manual audit into the system database, then corresponds with the extracted content, conducts consistency comparison, and displays the comparison results.

10. The artificial intelligence-based construction plan review system according to claim 4, wherein The intelligent recognition and audit module uses the document parsing and recognition module to parse the text and tables in the relevant photos and scanned documents in the construction plan to be audited. The model training module trains the text recognition and table parsing content to form semantic table extraction content, and conducts consistency comparison on the semantic table extraction content.

Images