Method for semantic recognition and automatic compliance review of design parameters of seismic mitigation structure

Abstract

The present invention belongs to the technical field of civil engineering, and relates to a method for semantic recognition and automatic compliance review of design parameters of a seismic mitigation structure. The method comprises: according to the review requirements, performing language segmentation and preprocessing on the seismic mitigation structure design parameter requirements in the design specification, and screening out convertible specification parameter sentences; using a deep neural network model based on a Transformer algorithm to perform semantic labeling and syntax analysis on the specification parameter sentences, and converting the specification parameter sentences into corresponding rule syntax trees; according to the mapping relationship between the seismic mitigation structure design parameters of the specification sentences in the rule syntax trees and the computer language, further converting the specification parameter sentences into automatic compliance review instructions; preprocessing the seismic mitigation structure design file, and using a pre-trained automatic compliance review model to perform automatic review. The present invention can trace the relevant specifications and automatically review, thereby improving the review efficiency and reducing the labor cost.

Description

TECHNICAL FIELD

[0001] The present invention belongs to the technical field of civil engineering, and relates to a method for semantic recognition and automatic compliance review of design parameters of a seismic mitigation and isolation structure. BACKGROUND

[0002] With the continuous expansion and complexity of construction projects, the traditional design mode based on text calculation book and manual review has been difficult to meet the needs of modern engineering construction. This design mode not only depends on the experience and skills of professional technicians, but also has problems such as low efficiency, difficulty in ensuring accuracy, and incomplete review. Especially under the time limit, through the way of spot check, often leads to the incompleteness and uncertainty of the review results.

[0003] In order to solve these problems, compliance intelligent review has become an inevitable trend of industry development. At present, many enterprises have begun to try to realize compliance intelligent review by hard coding. The advantage of hard coding is that the logic is strong and the execution efficiency is high, but at the same time there are some obvious shortcomings. First, hard coding needs to code one by one for different specification texts and index contents, which leads to a lot of repeated and redundant calculation items. Second, the maintenance and update of hard coding is time-consuming, has poor flexibility and maintainability, and has high modification cost. Finally, the logic calculation and judgment method of hard coding is invisible to users, so that users cannot judge whether the calculation process and formula meet the requirements of relevant specification standards.

[0004] In order to overcome the shortcomings of hard coding, some researchers have begun to try to use structured natural language to define and describe review rules. However, the existing structured natural language definition and description of review rules are very limited in depth, and there are certain technical barriers in cross-disciplinary research. Especially for rules with complex semantics and many restrictions, there are problems such as difficulty in combination, difficulty in compound judgment, difficulty in multiple calculation, and difficulty in accurate translation.

[0005] Therefore, the automatic compliance review of calculation book needs a more intelligent, efficient and flexible method. This method should be able to automatically recognize and analyze the key information in the text calculation book, and then make intelligent judgment and calculation according to the relevant specification standards, and finally generate a compliance review report. This method should be able to overcome the shortcomings of hard coding, improve the accuracy and efficiency of review, reduce manual intervention and dependence, and reduce the cost and time cost of review. At the same time, this method should have high maintainability and flexibility, and be able to adapt to the changing specification standards and engineering construction needs. SUMMARY

[0006] To address the aforementioned issues, the purpose of this invention is to provide a method for semantic recognition and automatic compliance review of seismic isolation and damping structure design parameters, enabling users to quickly trace and automatically review relevant standards according to their own review needs.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for semantic recognition and automatic compliance review of seismic isolation and vibration reduction structural design parameters, comprising the following steps:

[0008] S1. Based on the review requirements, perform language segmentation and preprocessing on the design parameters of seismic isolation and damping structures in the design specifications, and select convertible specification parameter statements.

[0009] S2. A deep neural network model based on the Transformer algorithm is used to perform semantic annotation and syntax parsing on the canonical parameter statements, and the canonical parameter statements are converted into corresponding rule syntax trees.

[0010] S3. Based on the mapping relationship between the seismic isolation and vibration reduction structural design parameters in the rule syntax tree and the computer language, the standard parameter statements are further converted into automatic compliance review instructions.

[0011] S4. Preprocess the seismic isolation and vibration reduction structure design documents, use a pre-trained automatic compliance review model to automatically review them, and output the review results.

[0012] Furthermore, step S1 specifically involves: decomposing and structuring the standard according to the chapter, usage conditions, value range, and clause number of the relevant design parameters of the seismic isolation structure, and converting the standard clauses into natural language expressions; further dividing the natural language into convertible individual standard statements using semicolons and periods.

[0013] Furthermore, the deep neural network model based on the Transformer algorithm utilizes the Transformer's self-attention mechanism to perform semantic annotation and syntactic parsing on the relevant specification parameter statements of seismic isolation and vibration reduction structures. The model combines labels according to five dimensions: specification source of seismic isolation and vibration reduction structure design parameters, parameter name, usage conditions, comparison relationship between parameters and target values, and target values, to construct a rule syntax tree corresponding to the seismic isolation and vibration reduction design specifications.

[0014] Furthermore, the specific steps of step S3 are as follows:

[0015] (1) Use natural language processing tools to perform part-of-speech tagging and syntactic analysis on natural language elements in the rule syntax tree.

[0016] (2) The processed word units are converted into high-dimensional semantic vector form by a word embedding model, and then a review-oriented seismic design specification expression method is used to store design parameters and their semantic labels in a unified data format, integrate model information layer by layer, and form an information model containing specification sources, usage conditions, parameter names, conjunction words and target values.

[0017] (3) The semantic similarity between the elements in the rule syntax tree in the information model and the elements in the predefined calculation book language term set is evaluated using the self-attention mechanism of the deep learning model based on Transformer. The rule syntax tree is traversed and mapped using a recursive algorithm, and the specification articles are converted into SPARQL query instructions to realize automatic compliance review.

[0018] Further, the specific steps of the step S4 are:

[0019] (1) Convert the seismic mitigation structure design file into processable text data; perform text cleaning and normalization operations on the text data to obtain a pure text representation of the structure parameters;

[0020] (2) Extract the key information and semantic content of the seismic mitigation structure design parameters contained in the text data.

[0021] Compared with the prior art, the present application has the following advantages:

[0022] 1. The present application realizes the rapid compliance review of calculation books and other seismic mitigation structure design files through an automated process based on computer language, without the need for manual comparison and inspection of seismic mitigation structure design parameters in relevant specifications. Especially for large-scale or high-complexity construction projects, this feature not only greatly improves the review efficiency, but also reduces the labor and time costs.

[0023] 2. The deep learning model is used for semantic recognition and syntax analysis of text data such as calculation books, and through repeated training of the deep model, it can successfully capture the key information points of each seismic mitigation structure design parameter and strictly compare with the specifications, greatly improving the review accuracy and reducing the risk of human error or omission.

[0024] 3. Through the automatic and intelligent review system, the dependence on the professional knowledge of the reviewer is effectively reduced. By recording the review data and related key steps in the review process, the review structure has a high degree of traceability, which provides strong support for subsequent review process optimization and review quality improvement. At the same time, through in-depth analysis of the review data, potential problems and improvement points can be found, which provides a strong guarantee for the quality of construction projects. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flowchart of a method of semantic recognition and automatic compliance review of design parameters of a seismic mitigation structure in embodiments of the present application;

[0026] Figure 2 is a construction flow of a rule syntax tree in embodiments of the present application;

[0027] Figure 3 is a schematic diagram of a parsing process of an automatic compliance review model provided in embodiments of the present application. DETAILED DESCRIPTION

[0028] The present application will be further described below in conjunction with the accompanying drawings and embodiments, but the embodiments of the present application are not limited thereto.

[0029] Embodiment One The present embodiment provides a method of semantic recognition and automatic compliance review of design parameters of a seismic mitigation structure, which can provide technical support for application scenarios such as building planning reporting, text calculation book review, and building specification tracing, etc. Figure 1 As shown in the figure, the method specifically comprises the following steps:

[0030] S1. According to the review requirements, language segmentation and preprocessing are performed on the seismic mitigation structure design parameter requirements in the design specification, and convertible specification parameter sentences are screened out.

[0031] The language segmentation and preprocessing of the seismic mitigation structure design parameters in the design specification comprises: identifying the boundaries of the provisions according to the special rules of the specification punctuation marks and provision numbers, etc., splitting the continuous specification provisions into single independent units or sentences, and finally based on natural language processing technology, further extracting the key information in the provisions, such as parameter name, value, use condition, etc., thereby providing structured data for subsequent interpretation and code generation.

[0032] S2. A deep learning model is used to perform semantic annotation and syntax analysis on the specification parameter sentences, and the specification parameter sentences are converted into corresponding rule syntax trees, as shown in the flowchart. Figure 2

[0033] ​This step aims to achieve the deep semantic annotation and accurate syntax parsing of the input convertible specification parameter sentence, and construct a structured rule syntax tree. In order to overcome the various restrictions brought by the current regular matching for semantic annotation and syntax parsing, this embodiment adopts a deep learning method for semantic annotation. Through a large amount of data training and complex neural network structure, the deep learning method can capture the deep semantic information in the specification parameter sentence and accurately annotate it. At the same time, in order to realize accurate syntax parsing, a context-free grammar (CFG) is also used. By defining a complete set of grammar rules, CFG can accurately describe the grammar structure of the specification parameter sentence and convert it into a structured syntax tree. This method not only avoids the complex problem of regular matching system, increases the flexibility and scalability of the grammar system, but also improves the accuracy and efficiency of semantic annotation and syntax parsing.

[0034] This step uses deep learning to perform semantic recognition and syntax parsing on the specification parameter sentence, and then decomposes it into word segmentation, part-of-speech tagging, named entity recognition, and named entity relationship. Then it is combined into a rule syntax tree that uniformly describes five dimensions of specification source, parameter name, usage condition, comparison word, and target value of the seismic design parameter. For example, "Article 5.5.1 of the Building Seismic Design Code GB50011-2010 stipulates that the elastic inter-story drift angle of multi-story or high-rise steel structures under seismic action shall not be greater than 1 / 250", which can be structured as a rule syntax tree as shown in Figure 2

[0035] This step defines five semantic tags to represent the role played by each word in the specification parameter sentence in the rule syntax tree, as shown in Table 1. The label src represents the specific specification source of the specification parameter sentence; the label cond represents the usage condition required by some specific specification parameter sentence; the label param refers to the specific seismic design parameter name; the label cmp is used to represent the comparison relationship between the selected seismic design parameter and the target value, and contains its obligation type (such as should, should, should not, etc.); the label val represents the target value of the seismic design parameter. Directly applied to the requirements of the selected seismic design parameter:

[0036] Table 1 defines the semantic tags

[0037]

[0038] S3. According to the mapping relationship between the key terms of the specification sentence in the rule syntax tree and the computer language, the specification parameter sentence is further converted into an automatic compliance review instruction based on SPARQL.​

[0039] A seismic isolation and vibration reduction structural design code based on review requirements is adopted. A unified data format is established to fully store various design parameters and their semantic tag fields in the data format. Various types of model information are integrated layer by layer to form an information model. This information model consists of an information matrix composed of five semantic tags: source of the code, conditions of use, relevant parameter names, conjunctions, and target values, together with the total number of code provisions.

[0040] In the processing of rule-based syntax trees (RDBML), the Transformer algorithm from deep learning is used to capture sequence information and represent semantics for linguistic elements. Specifically, natural language text is input into a deep learning model, which generates corresponding vector representations through learning. These vectors not only contain information about the words themselves but also reflect the meaning and association of words in context. These vectors are compared with the word vectors of the computer language, and the most similar mapping is determined by calculating their similarity. Finally, they are stored in a structured form to form a multi-dimensional, unified rule-based syntax tree.

[0041] When constructing mapping rules, the language characteristics of the review instructions to be transformed are considered. After ensuring the correspondence between the rule syntax tree and the review instructions is correct, a recursive algorithm is used to traverse the rule syntax tree and perform mapping transformation according to preset rules. This transforms the annotated specification clauses into SPARQL query instructions to achieve automated compliance review, as shown in Table 2. Combining rule syntax trees with expertise in structural engineering ensures that each node of the rule syntax tree is linked to a computer language.

[0042] Table 2 Automated Compliance Review Template

[0043]

[0044] S4. Preprocess the calculation sheet content, extract it into a processable text format, and use a pre-trained automatic compliance review model to automatically review the calculation sheet and other seismic isolation structure design documents according to the automatic compliance review instructions, and finally output the review results.

[0045] In the data preprocessing stage, the text calculation report needs to undergo deep text cleaning to remove irrelevant information and standardize the format, facilitating subsequent data extraction and review. Optical Character Recognition (OCR) technology is used to convert the preprocessed text calculation report into computer-processable text data. The standardized data is then integrated into a comprehensive and complete text calculation report model for automatic review. This model presents all key information in a structured and standardized format (such as XML or JSON).

[0046] This step uses an automatic compliance review model (a deep learning neural network) to automatically review and label the output of the text calculation book model to be automatically reviewed. The underlying algorithm of the deep learning neural network is an automatic compliance review instruction, as shown in Figure 3 The specific process is as follows:

[0047] (1) Collect a large number of structural design parameters in the text calculation book of the seismic mitigation structure design that conforms to the specification as a data set, normalize the data set, and divide the training set and the test set according to 8:2.

[0048] (2) Set the input variable as the seismic mitigation structure design parameter (such as the inter-story drift angle, additional damping ratio, base shear, structure height-width ratio, bearing tensile stress, isolation bearing stiffness, etc.) in the calculation book to be reviewed; the output variable is the labeled structure parameter statement.

[0049] (3) Adjust the neural network learning rate, momentum gradient, number of hidden layers, and other hyperparameters, train in the training set until convergence. The hidden layer processes the data:

[0050] a1 = σ(z1) = σ(W1a 1-1 +b1)

[0051] Where a1 is the current layer, a 1-1 is the input matrix in the current layer, which is also the output matrix of the previous layer, W1 is the weight parameter matrix of the current layer, b1 is the bias parameter matrix, σ is the activation function, and the RELU function is used to map each layer output: let x be the horizontal coordinate value, and RELU(x) be the function result, then:

[0052]

[0053] (4) According to the automatic compliance review instruction, calculate the compliance of the seismic mitigation structure design parameter, and use the back propagation algorithm for training, and the loss function is:

[0054]

[0055] Where, is the probability that the prediction sample of the text calculation book model to be automatically reviewed is a positive example; y is the sample label, and if the sample is a positive example, it takes the value 1, otherwise it takes the value 0.

[0056] (5) Use the Sigmoid function as the output layer neuron to limit the output of the neural network between 0 and 1 to adapt to the binary classification task.

[0057] The automatic compliance review model reads the seismic reduction structure design parameter information, outputs the multi-label design parameter statement, and obtains the seismic reduction structure design parameter result that does not conform to the specification. At the same time, the source and reason of not conforming to the relevant specification are labeled.

[0058] The above embodiments are only preferred embodiments of the present application, but the protection scope of the present application is not limited to these embodiments. Those skilled in the art can make various modifications and improvements to the above embodiments without departing from the principles of the present application, and these modifications and improvements shall be included in the protection scope of the present application. The core of the present application is to provide an automatic compliance review method based on text calculation book, which has the advantages of efficient automation, accurate identification, flexible expansion, standardized processing, comprehensive tracing, reduced professional threshold, and seamless integration. Therefore, any improvement, equivalent replacement or modification of the present application, as long as it does not deviate from the spirit and scope of the present application, shall be considered as included in the protection scope of the present application. The protection scope of the present application shall be subject to the scope defined in the claims.

Claims

1. A method for semantic recognition and automatic compliance review of seismic isolation and vibration reduction structural design parameters, characterized in that, Includes the following steps: S1. Based on the review requirements, perform language segmentation and preprocessing on the design parameters of seismic isolation and damping structures in the design specifications, and select convertible specification parameter statements. S2. A deep neural network model based on the Transformer algorithm is used to perform semantic annotation and syntax parsing on the normative parameter statements, converting the normative parameter statements into corresponding rule syntax trees. This step defines 5 semantic tags to represent the role of a single word in each normative parameter statement in the rule syntax tree: the tag src represents the specific normative source of the normative parameter statement; the tag cond represents the usage conditions required for certain normative parameter statements; the tag param refers to the specific design parameter name of the seismic isolation and vibration reduction structure; and the tag cmp is used to represent the comparison relationship between the selected seismic isolation and vibration reduction structure design parameter and the target value, and includes its obligation type. The label val represents the target value of the design parameters for the seismic isolation and vibration reduction structure; S3. Based on the mapping relationship between the seismic isolation and vibration reduction structural design parameters in the rule syntax tree and the computer language, the standard parameter statements are further transformed into SPARQL-based automatic compliance review instructions; the SPARQL review rule is defined as: SELECT ?a WHERE{?a ifc:category inst:(cond). ?a ifc:(param) ?b. filter(?b(cmb)(val)}; S4. Preprocess the seismic isolation and vibration reduction structure design documents, use a pre-trained automatic compliance review model to automatically review them, and output the review results.

2. The method for semantic recognition and automatic compliance review of seismic isolation and vibration reduction structural design parameters according to claim 1, characterized in that, Step S1 specifically involves: decomposing and structuring the standard according to the chapters, usage conditions, value ranges, and clause numbers of the relevant design parameters of the seismic isolation and damping structure; at the same time, converting the standard clauses into natural language expressions; and further dividing the natural language into convertible individual standard statements using semicolons and periods.

3. The method for semantic recognition and automatic compliance review of seismic isolation and vibration reduction structural design parameters according to claim 1, characterized in that, In step S2, the deep neural network model based on the Transformer algorithm uses the self-attention mechanism of Transformer to perform semantic annotation and syntactic parsing on the relevant specification parameter statements of seismic isolation and vibration reduction structures. The model combines labels according to five dimensions: specification source of seismic isolation and vibration reduction structure design parameters, parameter name, usage conditions, comparison relationship between parameters and target values, and target values, to construct a rule syntax tree corresponding to the seismic isolation and vibration reduction design specifications.

4. The method for semantic recognition and automatic compliance review of seismic isolation and vibration reduction structural design parameters according to claim 1, characterized in that, The specific steps of step S3 are as follows: (1) Use natural language processing tools to perform part-of-speech tagging and syntactic analysis on the natural language elements in the rule syntax tree; (2) The processed vocabulary units are converted into high-dimensional semantic vectors using the word embedding model, and the semantic similarity between elements in the rule syntax tree and elements in the predefined computational book language term set is evaluated using the self-attention mechanism of the Transformer model. The most similar mapping relationship is determined based on similarity.

5. The method for semantic recognition and automatic compliance review of seismic isolation and vibration reduction structural design parameters according to claim 1, characterized in that, In step S4, the specific steps of preprocessing are as follows: (1) Convert the seismic isolation and vibration reduction structure design documents into processable text data; perform text cleaning and normalization operations on the text data to obtain a plain text representation of the structural parameters; (2) Extract key information and semantic content of seismic isolation and vibration reduction structural design parameters contained in text data.

Images