Building industry accident report evaluation method and equipment
By constructing an accident tree and automatically extracting the causal relationship in the accident report using the BERT-Bi-LSTM-CRF model, the problem of automatic evaluation of the quality of accident report in the construction industry is solved, the accuracy and depth of the report are improved, and safety risks are reduced.
Patent Information
- Application Number
- CN202410984522.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-07-22
AI Technical Summary
It is difficult for the existing technology to achieve automatic assessment of the quality of accident reporting in the construction industry, resulting in uneven reporting quality, which may cause incorrect risk assessment and preventive measures, increasing the risk of recurring similar accidents.
By building an accident tree, the logical relationship between the root node, the accident-induced intermediate node and the leaf node is used to evaluate the accuracy and analysis depth of the accident report, and the BERT-Bi-LSTM-CRF model is combined with the BERT-Bi-LSTM-CRF model to automatically extract the entity in the accident report and the causal relationship between entities, perform content mapping and structure matching, and optimize report quality.
It improves the quality of construction accident reports, reduces long-term safety risks, ensures the causal logic accuracy and analysis depth of reports, and provides optimization suggestions to improve the quality of reports.
Smart Images

Figure CN119066196A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the field of construction industry safety accident analysis, and in particular to a construction industry accident report evaluation method and device. Background Art
[0002] Construction safety issues are important issues that affect life and property, and accident reports are important tools for accident analysis and prevention. The quality of accident reports is directly related to the accuracy of accident cause analysis and the effectiveness of subsequent preventive measures, and will also have an impact on subsequent training and education, risk management and improvement, and safety culture construction based on accident reports.
[0003] However, since report writers often face challenges such as large amounts of information, numerous professional terms, and lack of experience, the difficulty of report writing and the possibility of errors are increased, and it is difficult to objectively restore the facts. As a result, the quality of existing accident reports is uneven, which may lead to erroneous risk assessments and preventive measures, thereby increasing the risk of similar accidents occurring again.
[0004] Therefore, how to realize the automatic evaluation and optimization of the quality of accident reports in the construction industry is an important aspect of improving construction safety. In the prior art, patent CN114580978A provides an environmental impact assessment report quality inspection system and method, and patent CN116050872A provides a scientific and technological report quality evaluation method based on the hierarchical analysis method, neither of which can realize the automatic evaluation of the quality of accident reports in the construction industry. Summary of the invention
[0005] The embodiment of the present invention provides a construction industry accident report evaluation method and device, which utilizes the logical relationship in the accident tree to evaluate the accuracy and analysis depth of the accident report to evaluate the report quality.
[0006] In a first aspect, an embodiment of the present invention provides a construction industry accident report evaluation method, comprising: Obtain construction industry accident reports for evaluation; Extracting multiple entities and relationships between entities in the accident report, wherein the entity categories include accident types and accident causes, and the relationships between entities include causal relationships; Evaluate the accuracy and analysis depth of the accident report based on the levels and connection paths of two entities with causal relationship in the accident tree; The accident tree has the accident type entity as the root node and the accident cause entity as the intermediate node and leaf node.
[0007] In a second aspect, an embodiment of the present invention provides an electronic device, the electronic device comprising: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the construction industry accident report evaluation method described in any embodiment.
[0008] An embodiment of the present invention provides an evaluation method for accident reports in the construction industry. An accident tree with the accident type as the root node and the accident cause as the intermediate node and leaf node is used to reflect the causal elements and causal logic of most construction accidents, and provide a knowledge reference for the causal logic of construction industry accidents. The report to be evaluated is then content-mapped and structure-matched with the accident tree. By analyzing the hierarchy and connection path of the report entity in the accident tree, the accuracy and analysis depth of the report are judged, providing a basis for report optimization, effectively improving the quality of accident reports in the construction industry and reducing long-term safety risks. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0010] Figure 1 is a flow chart of a construction industry accident report evaluation method provided by an embodiment of the present invention; Figure 2 is a flow chart of another construction industry accident report evaluation method provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0011] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.
[0012] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.
[0013] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0014] Figure 1 is a flow chart of a construction accident report evaluation method provided by an embodiment of the present invention. The method is applicable to the situation of quality evaluation of construction accident reports, especially the evaluation of the accuracy and analysis depth of the reports. The method is executed by an electronic device, such as Figure 1 As shown, the specific steps include: S110. Obtain construction accident reports to be evaluated.
[0015] The report may be provided by a user, such as uploaded by a user through a web interface, as an object of evaluation.
[0016] S120. Extract multiple entities and relationships between entities in the accident report, wherein entity categories include accident types and accident causes, and relationships between entities include causal relationships.
[0017] The entities here refer to the main concepts and elements in the accident report, including the type of accident, the danger of the accident, the location of the accident, the operation link, the responsible party, the accident prevention, as well as the rectification measures and the causes of the accident. Among them, the cause of the accident can be further divided into unsafe operation behavior, unsafe management behavior, unsafe state of things, personal state, and organizational safety culture.
[0018] It should be noted that there is a difference between entities and entity categories, while the relationship between entities is consistent with the relationship between the categories to which the entities belong. For example, suppose there are two entities in the report, "installer" and "weak safety awareness", and their categories are "responsible subject" and "personal status", respectively, and the entities and entity categories are different; while the relationship between the entity "installer" and "weak safety awareness" is "yes", and the relationship between the entity categories "responsible subject" and "personal status" is also "yes", and the two are the same. Among all the above entity categories, accident type and accident cause are the necessary entity categories for accident reports, and the relationship between these two entity categories, as well as the relationship between these two types of entities are causal relationships.
[0019] In a specific implementation, a set of entity categories and a set of relationships between entity categories in the field of construction accidents can be predefined, with the purpose of establishing a conceptual system in the field of construction accident report analysis and clarifying the main concepts / elements and the structural relationships between them. In this embodiment, these two sets are referred to as the knowledge ontology model of the field of construction accidents, which serves as a basis for the automatic extraction of entities and entity relationships. Optionally, the predefined entity category set is shown in Table 1: Table 1 Entity category set
[0020] Correspondingly, the possible relationship set between the entity categories in Table 1 is shown in Table 2. This set can be obtained based on the analysis of the text features of the accident report and expert opinions, covering one or more relationships for each possible combination of entity categories, and Table 2 only shows a part of them. Among them, relationship names such as "attributed to", "affect", "affected by", "caused by", "responsible for", "implemented by", "implemented by", and "caused by" can all be classified as "causal relationship".
[0021] Table 2 Relationship set between entity categories
[0022]
[0023]
[0024]
[0025] Optionally, you can use Protégé to define the above ontology model. Protégé is an open source ontology editing software that provides a powerful mold mechanism for advanced construction of ontology modules, including parent-child classification through methods such as "nested classes" and "synonyms". It also has a built-in Reasoner for automated reasoning. In addition, it uses an easy-to-use graphical user interface, making the construction, editing and management of ontologies simple and clear. In summary, Protégé is a powerful and user-friendly visual ontology editing tool and an efficient tool for knowledge management.
[0026] Based on the above knowledge ontology, this embodiment can pre-label the entities, entity categories and relationships between entities in the existing construction accident reports and / or regulatory documents, and then use the labeled data set to train the BERT-Bi-LSTM-CRF model, so that the trained model can be applied to S120 to automatically extract the entities and relationships between entities in the report to be evaluated. In order to illustrate the process, the BERT-Bi-LSTM-CRF model is first introduced. This model is a deep learning structure for sequence labeling tasks, which can be used to perform complex text processing tasks. It combines BERT (Bidirectional Encoder Representations from Transformers), BiLSTM (Bidirectional Long Short-Term Memory) and CRF (Conditional Random Field). Among them, BERT is a Transformer-based model used for pre-training in natural language processing. It can understand the meaning of words in different contexts by reading the entire sentence (rather than just one-way reading), which makes BERT particularly effective in processing the diversity and complexity of word meanings. BiLSTM is a special type of recurrent neural network used to process sequence data. Unlike the traditional unidirectional LSTM (Long Short-Term Memory), BiLSTM processes the past (forward) and future (reverse) information of the sequence at the same time, using forward and backward LSTM to capture the left and right context in the word sequence. CRF is a statistical modeling method for structured prediction. It considers the dependencies between the annotation results and predicts the sequence of output labels. In the named entity recognition task, the CRF layer is used to decode the output of the BiLSTM layer and determine the optimal label sequence. Combining the above three technologies, the BERT-BiLSTM-CRF model can effectively handle complex natural language processing tasks, especially in scenarios that require context understanding and sequence label prediction.
[0027] In a specific implementation, in order to reduce the burden of labeling large-scale accident report data, the BERT-BiLSTM-CRF model can be trained using transfer learning and semi-supervised learning methods to enable it to automatically extract entities from accident reports. Specifically, the training process may include the following steps: Step 1: Annotate the existing construction accident reports and regulatory documents according to the predefined entity category set and relationship set between entity categories in the field of construction accidents. The evaluation of the quality of accident reports not only depends on common sense knowledge, but also needs to be based on reasoning knowledge. Therefore, this embodiment uses large-scale construction accident reports and regulatory documents as training data for the entity extraction model to comprehensively learn the entities and relationships in the accident reports. Among them, the accident report data contains rich information, including key information such as accident type and danger, operation link and location, accident cause and treatment measures. These data can be used to discover hidden patterns and associations to improve the safety management of the construction site; the system document data contains detailed information such as specific requirements for each operation link, the job responsibilities of each responsible entity, and the construction of corporate safety culture. This information effectively supplements the accident report data and can improve the model's ability to learn construction industry knowledge.
[0028] Optionally, select 20% of the accident report data and 20% of the rules and regulations documents for annotation, respectively, to ensure that the selected accident report data covers various accident scenarios and entity categories and relationships, and the selected rules and regulations documents cover different types, entity categories and relationships. The annotated contents are shown in Table 1 and Table 2. During the annotation process, you can first annotate the category of each entity according to the entity category set in Table 1. Assume that the text A series of words Composition, the entities in the annotation text are ,in, Represents an entity, Indicates the starting position of the entity in the text, Indicates the end position. Indicates its entity category. For example, if the accident report states that "due to inadequate safety training by managers, workers have weak safety awareness of working at heights", the entity category labeling result is: E1 = (6, 9, “SafetyCulture”) E2 = (22, 25, “PersonalState”) Then, the relationship between the entities is marked according to the category of each entity and the relationship set between the entity categories shown in Table 2. For example, the relationship between "SafetyCulture" and "PersonalState" in Table 2 is "influences", then the relationship between the above entities E1 and E2 is: R1 = (E1, E2, \text{“influences”}) Step 2: Use the labeled data to train the BERT-BiLSTM-CRF model. During the training process, each paragraph of text in the construction accident report or regulations document is input into the BERT-BiLSTM-CRF model in turn, the loss function is calculated according to the model state, and the model parameters are continuously updated so that the model output keeps approaching the labeled entities and the relationships between entities in each paragraph of text.
[0029] The following first introduces the data processing process inside the BERT-BiLSTM-CRF model after a piece of text is input into the model. Optionally, the model first uses BERT's tokenizer to split the text into tokens, and adds special tags (B-Entity, I-Entity) based on the tokens to indicate the boundaries of the entities. BERT is used for embedding to generate word vector representations for the text (including entities and context). That is, for each Token Xi, BERT outputs an embedding vector Di. The embedded representation of the text obtained by the BERT layer provides a basis for subsequent steps to capture local and global relationships between entities.
[0030] Then, the output of BERT (i.e., the embedding vector Di of each token) is used as the input of BiLSTM. BiLSTM captures the contextual information from left to right and from right to left through its forward and backward LSTM layers, respectively. This helps to understand the role and relationship of entities in the entire document or longer text fragment, and further capture the sequential dependencies between tokens. For the i-th token of the input sequence, the output of BiLSTM can be expressed as ,in and are the hidden states of the forward LSTM and backward LSTM at position i, respectively.
[0031] Optionally, in BiLSTM, in order to enable the model to focus on the most important part of the current task, an attention layer is added to help the model focus on keywords or phrases of key entities and relationships. This weighted information focus helps improve the performance of the model, especially when processing long texts with a lot of irrelevant information. The self-attention mechanism dynamically allocates the "attention" of each token by analyzing and evaluating the interaction of each token in the input sequence and its contribution to the current task. Suppose the output of BiLSTM is ,in is the hidden state vector of the ith token in the sequence. The self-attention mechanism first evaluates the relationship between each token in the sequence by calculating the interaction between each token and all other tokens in the sequence. , the model learns to generate three vectors: query vector (Query, Q), key vector (Key, K), and value vector (Value, V). Next, the model obtains the attention score by calculating the dot product of each query vector (Q) and key vector (K) with all key vectors to evaluate the influence of each token on other tokens in the sequence. If the score is regarded as a kind of "relationship strength", the higher the score means the more important the relationship between the two tokens. The score calculation formula is:
[0032] Secondly, use the Softmax function to normalize the scores to get the attention weight of each token to other tokens. The normalization formula is:
[0033] in, is the dimension of the key vector. This scaling factor helps prevent the vanishing gradient problem caused by dot products being too large.
[0034] Finally, the model uses the value vector (Value, V) for weighted summation to obtain the final representation of each token, which is the output of BiLSTM. This representation not only contains the information of the token itself, but also incorporates the information of other tokens in the sequence, especially those that are closely related to it. The formula for the weighted sum is:
[0035] The CRF layer receives the BiLSTM output with attention weights and predicts the entity category label of each token through the global optimal decoding strategy to ensure that the sequence label is optimal overall.
[0036] In addition, a relation classifier is added to predict the type of relationship between entities using the extracted entity and text features. First, the representation of each entity pair (i.e., the attention-weighted BiLSTM output) is concatenated with the representation of their common context (the average value of the BiLSTM output of the tokens between the entities) into a feature vector. Second, a fully connected layer is used as a classifier, and the feature vector is used as the input of the relation classifier. The relation classifier is trained using the training data (feature vectors and relation labels) prepared in the previous steps to output the possible relationship types between entity pairs.
[0037] Next, we will jump out of the model and explain the overall training process of the model. Optionally, first, based on the large-scale pre-trained language model, the pre-trained model is fine-tuned using the annotated construction industry accident report data. Large-scale pre-trained language models have strong generalization learning capabilities, but they may not work well when directly applied to professional fields. This is because the general corpus used in pre-training differs from the professional field data in distribution and pragmatic features, and it is difficult for the model to take into account the subtle semantics and usage of the professional field. The fine-tuning process is a method of retraining model parameters with professional field annotated corpus based on the pre-trained model, which can include the following steps: First, load the BERT model weights pre-trained on the general language task and fix the BERT layer parameters. Second, unlock and train the task-related layers (BiLSTM, attention layer, CRF, and relation classifier), define the loss functions for entity recognition and relation classification, and train on the annotated data. Subsequently, gradually unfreeze the BERT parameters for training updates. The CRF loss for entity recognition and the cross entropy loss for relation classification are used as optimization targets. The model minimizes the total loss function through the gradient descent method to adjust its parameters to meet the specific needs of construction industry accident reports. The loss function L is defined as:
[0038] in, is the input of the model, is the corresponding annotated entity or relation, is the probability predicted by the model.
[0039] Through the above fine-tuning, the model parameters can gradually adapt to and learn the specific pragmatic features of professional field report texts, such as industry terminology, event description methods, etc., so that the model can better process and understand construction industry report texts while maintaining general understanding capabilities, and generate more accurate analysis results. This helps to improve the effectiveness of the model in the construction field.
[0040] The fine-tuned BERT-BiLSTM-CRF model is further applied to unlabeled construction accident report data to predict entity categories and entity relationships. The pseudo-labels generated by this process are used to expand the initial labeled dataset to form a richer and more diverse training dataset. The pseudo-labels are combined with the initial labeled data to form the expanded labeled data. A semi-supervised learning strategy is used to iteratively use the expanded labeled data to further train the model. Each round of training includes the use and fine-tuning of pseudo-labels (the specific process is the same as the above process). The core of this strategy is to make the model more robust by gradually introducing more information from the labeled data, while improving its accuracy in identifying various entities and relationships in construction accident reports. In each round of iteration, the model not only learns to recognize new entity and relationship types, but also adjusts its own parameters to better adapt to the characteristics and complexity of the dataset. After the iteration, the model performance is evaluated on an independent test set to ensure that it has good generalization performance on unseen data.
[0041] In order to distinguish the training stages, the BERT-BiLSTM-CRF model trained at this time is called the first model in this embodiment. After a piece of text is input into the first model, the model can output the key entities, entity categories and relationships between entities in the text according to the logical relationships in the accident report. A specific input and output example is given below: 「Example Input 1 IN1」
[0042] On June 30, 2023, an electric shock accident occurred at No. 356, Yingrenshi Village, Yingrenshi Community, Shiyan Street, Bao'an District, resulting in one death and direct economic losses of 1.5 million yuan.
[0043] 1. How the accident happened On June 30, 2023, the gas pipeline installation team of Runyi Construction Company worker Yan Liyang (deceased) was installing natural gas pipelines on the scaffolding on the outer wall of No. 356 Yingrenshi Old Village, Yingrenshi Community, Shiyan Street. There are natural gas pipes to be installed on the west and north sides of the house, and there are low-voltage overhead lines at the junction of the west and north sides. Before the incident, Yan Liyang and his co-worker Cui Ziliang prefabricated natural gas pipelines on top of the scaffolding, and his co-worker Zhou Guoan assisted in delivering the pipelines on the ground. After delivering the pipelines, Zhou Guoan left the work site and went to other locations to continue working. Cui Ziliang then climbed down the scaffolding to collect the welding machine, leaving Yan Liyang alone working on the scaffolding. At around 10:30, when Zhou Guoan returned to the scene to get tools, he found that Yan Liyang had fallen to the ground below the scaffolding in the northwest corner of the involved building No. 356, his safety helmet rolled to one side, and there was a ladder next to him leaning against a scaffolding on one floor.
[0044] 2. Analysis of the Cause of the Accident Direct cause: The scaffolding working surface did not maintain a safe distance from the low-voltage overhead lines and four connectors involved, and no safety protection measures were set up; the natural gas pipeline installation workers had a weak sense of safety and illegally climbed across the low-voltage overhead lines. Their calves touched the exposed hexagonal bolts with dangerous voltage on the connectors, and their other limbs touched the scaffolding, causing the current to form a circuit through their bodies, resulting in this electric shock accident.
[0045] Indirect causes: 1. Runyi Construction Company did not provide adequate safety education and training... 2. Runyi Construction Company project manager Zhou Li did not conscientiously perform his safety duties. 3. Runyi Construction Company safety manager Zhang Wen did not conscientiously perform his safety management duties and did not conduct adequate daily inspections... III. Accident rectification and prevention measures (I) Runyi Construction Company should seriously learn the profound lessons of this accident, further implement the main responsibility of enterprise safety production, prevent similar accidents from happening again, and meet the following requirements... (II) Qijun Consulting Company should seriously learn the profound lessons of this accident, strictly perform supervision duties, urge and inspect construction units to implement construction safety protection measures, and promptly discover and eliminate hidden dangers of production safety accidents at construction sites. (III) Shiyan Sub-district Office should take this as a warning and carry out site safety hazard investigation and rectification work in gas-to-home construction projects throughout the street... (IV) The District Housing and Construction Bureau should strengthen accident warning education for gas-to-home construction projects throughout the district, and coordinate the district power supply department to carry out special inspections on projects close to low-voltage overhead lines.
[0046] 「Example output 1 OUT1」 List of entity recognitions and their categories:
[0047] Accident type: [Electric shock accident] Danger of accident: [1 person died, direct economic loss of 1.5 million yuan] Responsible parties: [construction company, consulting company, subdistrict office, district housing construction bureau, installer, safety officer, project manager] Occurrence location: [Scaffolding work surface, low voltage overhead line] Operation link: [Natural gas pipeline installation] Unsafe work behavior: [Illegal climbing and crossing low-voltage overhead lines] Unsafe state of objects: [The scaffolding work surface is not kept at a safe distance from the low-voltage overhead line and the four connectors, no safety protection measures are set, and live bolts are exposed] Unsafe management behaviors: [failure to carry out safety production education and training, failure to perform safety production duties, and inadequate daily inspections] Personal status: [Low security awareness] Organizational safety culture: [Inadequate safety production education and training] Accident prevention and rectification measures: [comprehensive safety hazard investigation, strengthening the deployment and inspection of safety management personnel, safety education and training, construction site safety production management, strengthening safety supervision, and strengthening accident warning education for gas household construction projects] Relation extraction list:
[0048] (Installation workers have a weak sense of safety, have) (The scaffolding work surface did not maintain a safe distance from the low-voltage overhead line and the four connectors.) (No safety protection measures have been set up on the scaffolding work surface) (Construction company, production safety education and training are not in place, shapes) (Construction company, comprehensive safety hazard investigation and rectification, shouldIimplement) ... Furthermore, considering that the language style, structure and content of the rules and regulations documents are different from those of the accident reports, the model can be fine-tuned using the annotated rules and regulations documents on the basis of the first model to adapt to the specific language environment of the rules and regulations documents. After fine-tuning, pseudo labels are generated for the unannotated rules and regulations documents again, the pseudo labels are merged with the initial annotated data, and the model is further trained iteratively using the expanded annotated data using a semi-supervised learning strategy. The trained model is used as the final BERT-BiLSTM-CRF model, which is referred to as the second model in this embodiment. Rules and regulations documents usually contain detailed information on safe operations, risk management, accident prevention measures, and emergency response. By performing in-depth entity extraction and relationship extraction on these documents, various safety specifications, operating procedures, risk factors, and recommended preventive measures can be identified. This information is essential to supplement and enrich the data obtained from the accident report, especially in covering the diversity of accident causes. After a piece of text is input into the second model, the model can integrate the logical relationships in the construction report and rules and regulations of the construction industry, and output the key entities, entity categories, and relationships between entities in the text. A specific input and output example is given below: 「Example Input 2 IN2」
[0049] The evacuation passages and emergency exits should be kept unobstructed at the construction site. It is strictly prohibited to occupy the evacuation passages, and it is strictly prohibited to install fences or other obstacles that may affect evacuation on the emergency exits or evacuation passages.
[0050] 「Example output 2 OUT2」 List of entity recognitions and their categories:
[0051] Regulation name: [Fire safety management system] Occurrence location: [construction site, emergency exit, evacuation passage] Unsafe work behavior: [Occupying evacuation passages, installing barriers and other obstacles that affect evacuation] Unsafe state of things: [occupying evacuation passage] Unsafe management behaviors: [failure to carry out safety production education and training, failure to perform safety production duties, and inadequate daily inspections] Relation extraction list:
[0052] (Construction site, occupying the evacuation passage, has) (Emergency exit, installation of barriers and other obstacles that affect evacuation, has) (Evacuation passages, barriers and other obstacles that affect evacuation have been installed) It is worth mentioning that in the above example, the text input into the first model is a paragraph of text in the accident report, and the text input into the second model is a paragraph of text in the rules and regulations. In actual application, both models can extract entities from any paragraph of text related to construction safety in the construction industry. Therefore, in S120, the first model or the second model can be used to extract entities and relationships between entities in the report to be evaluated, and this embodiment does not make specific restrictions.
[0053] S130. Evaluate the accuracy and analysis depth of the accident report based on the hierarchy and connection path of two entities with a causal relationship in the accident tree, wherein the accident tree has the accident type entity as the root node and the accident cause entity as the intermediate node and leaf node.
[0054] The accident tree is a model or data body used to analyze the cause of an accident. After the entity extraction is completed, the present embodiment uses the pre-built accident tree as a reference to compare the extracted causal logic with the causal logic in the accident tree to evaluate the accuracy and depth of the causal analysis in the report. In order to illustrate this step, the construction process of the accident tree is introduced first. In a specific implementation, the accident tree can be constructed using knowledge graph technology, which specifically includes the following steps: Step 1. Extract entities and relationships between entities from existing construction accident reports and regulatory documents. Optionally, the first model or the second model in S120 can be used to extract entities and entity relationships from existing construction accident reports, and store them in entity sets and relationship sets between entities. Since there may be a problem in the extraction results that different entity expressions actually refer to the same entity, the cosine similarity can be further used to calculate the similarity between entities, and the K-means algorithm is used to cluster entities, and the entity with the highest frequency in each cluster is selected as the normalized representation of this type of entity. Optionally, the clustering results and normalized entities can be submitted to professionals for manual verification to ensure correctness and consistency, and to make adjustments and optimizations when necessary. Finally, based on the normalized results, the entities in the entity set and relationship set are updated to ensure that each entity is a normalized entity representation. Among them, the cosine similarity calculation formula is:
[0055] in, is the vector representation of entity e.
[0056] At the same time, the second model mentioned in S120 can be used to extract entities and entity relationships from existing rules and regulations documents, and to normalize entities by calculating similarity and clustering, and to add the normalized entities to the entity set and relationship set.
[0057] Step 2: According to the categories of each extracted entity, the relationship between the extracted entities is matched with the set of relationships between entity categories in the predefined construction accident field. Optionally, the entities and entity relationships identified in step 1 are grouped into multiple triples, each triple including two entities and the relationship between the two entities. For each triple, first determine the categories to which the two entities in the group belong, and then execute the standardized query language SPARQL query on the ontology model to obtain the possible association relationship between the categories to which the two entities belong. Then, the possible association relationship is compared with the relationship between the two entities stored in the triple, and if the semantics are consistent, the relationship is considered to match. Optionally, the string matching algorithm Levenshtein distance algorithm is used to calculate the text similarity of the two relationships, and when the similarity exceeds a predetermined threshold, it is confirmed as a semantically consistent matching relationship. Among them, the Levenshtein distance L(a,b) is a function that measures the difference between two strings a and b, and is defined as the minimum number of single-character edits (insertion, deletion or replacement) required to convert from string a to string b. The similarity can be calculated by the following formula:
[0058] Where len(a) and len(b) represent the length of strings a and b respectively.
[0059] Exemplarily, the categories to which the two entities in a triple belong are "personal status" and "organizational safety culture", respectively. As can be seen from Table 2, the possible associations between these two categories include "reflection" and "influence"; the text similarities between the relationship between the entities stored in the triple and "reflection" and "influence" are calculated respectively. If the text similarity between the relationship and "reflection" exceeds the set threshold, it is considered that the relationship matches "reflection", that is, the triple matches the entry {personal status, organizational safety culture, reflection} in the ontology model.
[0060] Step 3: Construct a knowledge graph in the field of building accidents based on matching entities and relationships between entities. Optionally, use the graph database management system Neo4j to convert all triples obtained in step 2 that can match an entry in the ontology model into a graph structure. In this structure, entities are converted into nodes in the graph, and relationships between entities are converted into edges in the graph. The knowledge graph can be quality evaluated, indexed, and optimized to support efficient data retrieval and analysis.
[0061] Step 4: For triples that fail to match any entry in the ontology model, based on the frequency of occurrence of entities and relationships in the triples and expert judgment, triples with an occurrence frequency greater than the set threshold and triples that experts consider important are added to the above knowledge graph to improve the knowledge system.
[0062] Step 5: traverse and search the path from the accident type entity node to the cause entity node in the knowledge graph, output the accident tree through operations such as node and path construction, repeated node / path merging, and combine probability calculation and industry expert scoring to supplement the probability values for the intermediate and bottom event nodes, and finally generate an accident tree with the accident type entity as the root node and the accident cause entity as the intermediate node and leaf node. Specifically, the SPARQL query can be first executed on the knowledge graph to retrieve all entities and relationships of the specific accident type and accident cause entity in the knowledge graph, forming a subgraph associated with the specific accident type and accident cause entity. Since the standard query language of SPARQL is mainly aimed at graph pattern matching, all cause entities directly associated with the accident type can be retrieved first. For each found cause entity, the query is executed again to find other entities associated with it until the end node is reached or a specific condition is met. Then, a depth-first search graph traversal algorithm is applied in the subgraph to traverse and record all path information from the accident type node to the cause entity node. Again, for each path, an accident tree is constructed with the accident type entity as the root node; nodes are added to the tree in the order of the paths, each node represents an entity on the path, and the connection between nodes represents the causal relationship; for multiple paths, based on the string matching algorithm Levenshtein, the nodes of the accident tree are traversed to search for duplicate nodes, and the corresponding subtrees or subsequent paths of the duplicate nodes are merged into the existing nodes; finally, Graphviz is used to draw and display the accident tree. Finally, probability information is added to each node in the accident number. Optionally, for the entity relationship derived from the statistical data in the accident report, the triggering probability between entities is calculated, and the calculation formula is as follows:
[0063] In the formula, is the number of relationships from entity 1 to entity 2, is the number of relations that lead to entity 2; the calculated probability It indicates the proportion of events that occurred in entity 2 that were caused by entity 1.
[0064] For entity relationships that are supplemented by rule and system documents and not mentioned in the accident report, domain experts are invited to score their triggering probabilities based on the probability data obtained from the accident report statistics. The probability information is stored in each tree structure node in the form of a key-value pair, where the key is the relationship identifier between entities and the value is the corresponding probability. The accident tree is constructed accordingly.
[0065] It can be seen that the above-mentioned knowledge graph and accident tree present different entities and relationships in large-scale construction industry accident reports and regulatory documents through graphs or tree structures, revealing the complex causal logic behind accident safety management. Among them, the logical relationship in the knowledge graph is more comprehensive and complex, and the accident tree focuses on the causal logic between accident type entities and accident cause entities. This embodiment will use the accident tree as a knowledge framework to conduct quality evaluation of the accident report to be evaluated.
[0066] In general, the completeness, accuracy and analysis depth of the accident report can be comprehensively evaluated from two aspects: whether the entity category is missing and whether the entity and entity relationship are missing. Among them, to determine whether the entity and entity relationship are missing, it is necessary to compare the causal logic in the report with the causal logic in the accident tree, and analyze the hierarchy and connection path of the entities with causal relationships in the report in the accident tree. S130 corresponds to this process. In a specific embodiment, the report evaluation combined with the accident tree may include the following steps: Step 1. Generate a causal chain for the accident report based on every two entities in the accident report that have a direct causal relationship. As described above, multiple triples are obtained in S120. If the relationship in a triple is a causal relationship, the two entities in the triple are called two entities with a direct causal relationship. Furthermore, if there are two entity pairs with a direct causal relationship in the accident report to be evaluated, and the cause of one is the result of the other, the two entity pairs can be connected in series to form a causal chain. Exemplarily, A causes B, and B causes C, then a causal chain of A—>B—>C can be obtained, with the entity at the front end being the cause and the entity at the back end being the result.
[0067] Step 2: Determine the connection path of each pair of entities in the causal chain in the accident tree. Specifically, for each causal chain, execute SPARQL query and string matching algorithm Levenshtein on the accident tree to locate the corresponding nodes and edges of the causal chain in the tree. For each retrieved path, record the nodes and edges of the path and the relationship between them. Optionally, first locate the node corresponding to the first event in the causal chain in the accident tree, for example, using the string matching algorithm, take the node in the accident tree that is most similar to the event as the node corresponding to the event; then starting from the node, search the accident tree for the node that is most similar to the next event in the causal chain through the depth-first search method; when a matching adjacent node is found, construct a mapping between the causal chain and the accident tree to obtain the path of the entire causal chain or part of the causal chain in the accident tree, thereby obtaining the connection path of any pair of entities in the causal chain in the accident tree. In this process, if any node in the accident tree cannot correspond to the current event, it will go back to the previous node of the accident tree and try other possible matches. Repeat the above process until all causal chain search markers are completed.
[0068] Step 3: Compare the connection relationship of each pair of entities in the causal chain with their link path in the accident tree from different angles to verify the matching degree between the two, so as to evaluate the accuracy and analysis depth of the report. This embodiment provides the following three verification methods: Method 1: Verify whether there are any missing intermediate events in the causal chain to evaluate the analytical completeness and accuracy of the report. Specifically, determine whether there are intermediate nodes not mentioned in the accident report in the connection path of the two entities in the causal chain in the accident tree. If so, record it as a potential inconsistent relationship, which indicates that there may be negligence or deviation in the analysis of the reported accident. Then, the accuracy of the accident report is evaluated based on the network centrality and / or occurrence probability of the intermediate node in the accident tree, where the higher the network centrality and / or occurrence probability, the lower the accuracy.
[0069] Optionally, the network centrality of the node is Z1, and the probability of node occurrence is Z2. The calculation method of network centrality is:
[0070] Among them, L is the sum of the shortest path lengths from the node to other nodes in the network, and n is the number of nodes in the network; degree centrality represents the influence of the node on other nodes, and neighbor centrality represents the influence of the node on neighboring nodes.
[0071] Node comprehensive score
[0072] If the score of node A Greater than the combined score of all nodes If the average score of the node is greater than the set percentage (e.g. 120%), the node is considered to be a node with "high network centrality and / or high accident probability" that may be missed in the accident report. Record the number of missed nodes , two points will be deducted for each missing item, and the calculation formula for the analysis completeness and accuracy of the accident report is:
[0073] Method 2: Verify the level of the bottom event in the causal chain in the accident tree to evaluate the analysis depth of the report, especially focusing on the discussion of factors such as personal status, unsafe management, and organizational safety culture. Specifically, calculate the shortest path from the first cause entity in the causal chain as the top event to the leaf node in the accident tree. The level of this path is called the shortest chain path level. The more shortest chain path levels there are, it means that the bottom event in the causal chain has multiple levels of deeper causal entities, but these entities are not paid attention to in the accident report, and the analysis depth is insufficient. In particular, when the report ignores deep factors such as personal status, unsafe management, and organizational safety culture, it is also marked as insufficient report analysis depth.
[0074] Optionally, record the number of levels of the shortest chain path as , you can assign deductions based on the number of levels :
[0075] At the same time, define weight values for the accident cause categories. If the report of this part is ignored, the corresponding score will be subtracted:
[0076] Therefore, the analysis depth calculation formula for the accident report is:
[0077] in, Indicates the existence status of each cause category, Indicates existence, Indicates missing.
[0078] Method 3: Verify the difference between the causal chain path and the accident tree path to evaluate the accuracy of the report. Specifically, for two entities in the causal chain, compare their connection paths in the causal chain with their connection paths in the accident tree. If the two paths are very different, it indicates that there may be a big problem with the report logic, and it is marked as incorrect potential identification relationship type.
[0079] Optionally, the calculation process of the path difference includes: recording the relationship path of the entity pair AB in the report description as path T1, recording the path with the most repeated nodes with path T1 among all the paths of the entity pair AB matched in the accident tree as T2; calculating the number of common nodes in T1 and T2, recording as ; Calculate the sum of the number of different nodes in T1 and T2, recorded as , then the path difference Correspondingly, the calculation formula for the accuracy of relation type recognition of accident reports is:
[0080] It should be noted that the above three verification methods can be executed simultaneously, separately, or partially as needed, and this embodiment does not impose specific limitations. In addition, if it is detected that there are additional nodes or relationships in the report that do not belong to the accident tree, further analysis by experts can be performed to determine the reasons for the omission or mismatch in the report, and decide whether it is necessary to update the accident tree to include new information, or to revise the report to reflect the known accident pattern.
[0081] S140. Evaluate the overall quality of the accident report based on the accuracy and analysis depth, as well as the completeness of the entity categories in the accident report.
[0082] In addition to accuracy and analysis depth, report completeness is also an important aspect of quality assessment. S130 has explained the impact of personal status, unsafe management and organizational safety culture on the completeness of the report in the analysis depth of the accident report. In addition, the accident report must also include entity categories such as accident type and cause of the accident. If they are missing, the report is unqualified; the remaining entity categories, including accident hazard, location, operation link, responsible party and accident prevention and rectification measures, also affect the quality of the report to varying degrees.
[0083] In a specific implementation, the full score of the accident report content and structure completeness scoremax_c=40 points can be set, and a weight value is defined for each important entity category:
[0084] When a required entity category is detected to be missing, the weight value corresponding to that category is subtracted from the total score. If multiple categories are missing, points are deducted for each missing category. If the accident type or cause category is missing, the completeness score is 0, and the report is directly considered unqualified without further deductions. In summary, the formula for calculating the completeness of the accident report content and structure is:
[0085] in, Indicates the existence status of each entity category, Indicates existence, Indicates missing.
[0086] Finally, the completeness of entity categories, the accuracy of accident cause analysis, and the depth of analysis are integrated to form a comprehensive evaluation of the quality of the accident report. score_f (out of 100 points) can be expressed as:
[0087] At the same time, a detailed feedback report can be generated based on the recorded deduction points and the basis for the deduction, pointing out specific inconsistencies and possible optimization directions, guiding report writers on how to fill in logical gaps and how to enhance the causal logical coherence in the report.
[0088] Figure 2It is a flowchart of another construction industry accident report evaluation method provided by an embodiment of the present invention, including the main links in the above-mentioned optional implementations, and indicating the data flow of each link according to the timeline (as shown by the arrows). In addition, the figure also divides the overall method into modules such as data extraction, knowledge graph construction, accident tree construction, and accident report quality self-assessment according to the logic and close relationship between each operation, wherein each module can be replaced by other equivalent modules or methods, all of which belong to the protection scope of this embodiment.
[0089] In summary, this embodiment provides an evaluation method for accident reports in the construction industry, which comprehensively utilizes text materials such as accident reports and rules and regulations documents to construct a knowledge graph, covers common sense knowledge and reasoning knowledge from different angles and text styles, and provides a complete knowledge framework; then, an accident tree is constructed based on the correlation between accidents and accident causes in the knowledge graph, reflecting the causal elements and causal logic of most construction accidents, and providing a knowledge reference for the causal logic of construction industry accidents; finally, the report to be evaluated is content mapped and structure matched with the accident tree, and accident reports with high completeness and logic are screened out through content completeness test, intermediate event test, bottom event test and path consistency test, and possible improvement directions are provided in terms of content, analysis logic, etc., so as to effectively improve the quality of accident reports in the construction industry and reduce long-term safety risks.
[0090] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown, the device includes a processor 60, a memory 61, an input device 62 and an output device 63; the number of processors 60 in the device can be one or more. Figure 3 A processor 60 is taken as an example; the processor 60, the memory 61, the input device 62 and the output device 63 in the device can be connected by a bus or other means. Figure 3 The example of connecting through bus is taken in the following.
[0091] The memory 61 is a computer-readable storage medium that can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the construction industry accident report evaluation method in the embodiment of the present invention. The processor 60 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 61, that is, realizing the above-mentioned construction industry accident report evaluation method.
[0092] The memory 61 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 61 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 61 may further include a memory remotely arranged relative to the processor 60, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0093] The input device 62 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 63 may include a display device such as a display screen.
[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.
Claims
1. A method for evaluating accident reports in the construction industry, It is characterized in that include: Obtain construction industry accident reports for evaluation; Extracting multiple entities and relationships between entities in the accident report, wherein the entity categories include accident types and accident causes, and the relationships between entities include causal relationships; Evaluate the accuracy and analysis depth of the accident report based on the levels and connection paths of two entities with causal relationship in the accident tree; The accident tree has the accident type entity as the root node and the accident cause entity as the intermediate node and leaf node.
2. The method according to claim 1, It is characterized in that Before extracting the multiple entities and the relationships between the entities in the accident report, the method further includes: The existing construction industry accident reports and regulatory documents are input into the BERT-Bi-LSTM-CRF model for training, so that the trained model can extract multiple entities and the relationships between entities in the input text.
3. The method according to claim 2, It is characterized in that The existing construction industry accident reports and regulatory documents are input into the BERT-Bi-LSTM-CRF model for training, including: According to the predefined entity category set in the field of construction accidents, the categories of entities in existing construction accident reports and regulatory documents are annotated; According to the categories of each entity and the set of relationships between entity categories in the predefined construction accident field, the relationships between entities are labeled for model training.
4. The method according to claim 1, It is characterized in that Before evaluating the accuracy and analysis depth of the accident report according to the levels and connection paths of the two entities with causal relationship in the accident tree, the method further includes: Extract multiple entities and relationships between entities from existing construction industry accident reports and regulatory documents; According to the categories of each entity, the relationship between the entities is matched with the set of relationships between entity categories predefined in the field of building accidents; Construct a knowledge graph in the field of construction accidents based on matching entities and relationships between entities; The path from the accident type entity to the accident cause entity is traversed and searched in the knowledge graph to generate an accident tree with the accident type entity as the root node and the accident cause entity as the intermediate node and leaf node.
5. The method according to claim 1, It is characterized in that The step of evaluating the accuracy and analysis depth of the accident report according to the levels and connection paths of two entities with a causal relationship in the accident tree includes: Identify whether there are intermediate nodes in the accident tree between two entities with a causal relationship that are not mentioned in the accident report; If so, the accuracy of the accident report is evaluated based on the network centrality and / or occurrence probability of the intermediate node in the accident tree, wherein the higher the network centrality and / or occurrence probability, the lower the accuracy.
6. The method according to claim 1, It is characterized in that The step of evaluating the accuracy and analysis depth of the accident report according to the levels and connection paths of two entities with a causal relationship in the accident tree includes: Generating a causal chain of the accident report according to entities with causal relationships in the accident report; The accuracy of the accident report is evaluated based on the difference between the paths of two entities in the causal chain and the paths in the accident tree, wherein the greater the difference, the lower the accuracy.
7. The method according to claim 1, It is characterized in that The step of evaluating the accuracy and analysis depth of the accident report according to the levels and connection paths of two entities with a causal relationship in the accident tree includes: Generating a causal chain of the accident report according to entities with causal relationships in the accident report; The analysis depth of the accident report is evaluated based on the shortest path from the first cause entity in the causal chain as the top event to the leaf node in the accident tree, wherein the more levels the shortest path has, the less sufficient the analysis depth is.
8. The method according to claim 1, It is characterized in that Also includes: The overall quality of the incident report is assessed based on the accuracy and depth of analysis, as well as the completeness of the entity categories in the incident report.
9. The method according to claim 1, It is characterized in that The overall quality of the incident report is evaluated based on the accuracy and depth of analysis, as well as the completeness of the entity categories in the incident report, including at least one of the following: If an accident type entity or an accident cause entity is missing from the multiple entities, the accident report is assessed to be incomplete; If any of the entities of responsibility subject, accident hazard, occurrence location, operation link, accident prevention and corrective measures is missing from the multiple entities, the accident report is assessed to be insufficiently complete; If any of the categories of personal status, unsafe management and organizational safety culture are missing in the accident cause entity, the accident report is assessed to be incomplete.
10. An electronic device, It is characterized in that include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the construction industry accident report evaluation method described in any one of claims 1-9.
Citation Information
Patent Citations
Automatic compiling system for safety evaluation report
CN110008457A
Accident causal reasoning method and device, electronic equipment and readable storage medium
CN113657599A
Chemical engineering abnormal event detection method and system based on affair atlas
CN114153988A
System and method for performing automated system management
US7120559B1
System and method for automated problem diagnosis
US7379846B1
Cited By
Coal mine accident cause analysis method based on knowledge graph
CN120218609A
