Construction Safety Accident Prediction Based On Ontology

Although the research on Construction safety is fruitful,Construction safety accidents still happen frequently.At present,the study of Construction safety accident case analysis mainly relies on expert evaluation,which has the disadvantages of low research efficiency,low-level intelligence and easy to be influenced by subjectivity.Up to now,a large number of construction safety accident investigation reports contain cases information of construction safety accidents need to be mined urgently,and a large number of domain expert knowledge and experience information contained in them cannot be fully reused.In this context,aiming at the shortcomings of the current construction safety accident case study method,a text-mining method is proposed to analyze and excavate the unsafe factors of a construction safety accident.Based on the mutual information and adjacent entropy domain word discovery algorithm,538 key words of unsafe factors in Construction safety accidents were obtained by mining and analyzing the reasons in the investigation report of Construction safety accidents.To solve the problem that traditional Construction safety accident prediction methods do not fully integrate domain knowledge,a Construction safety accident prediction method based on ontology knowledge of Construction safety accident combined with TextCNN(Convolutional Neural Networks for Sentence Classification)text convolutional neural network model is proposed.Based on the analysis of the correlation between unsafe factors and the types of Construction safety accidents,We construct the ontology database of unsafe factors of Construction safety accidents by combining the knowledge of Construction safety domain specification criteria and literature,then use TransH(Translating on Hyperplanes)translation model represented the conceptual knowledge of unsafe factors in the ontology library the conceptual knowledge vectors for unsafe factors.The report data set is respectively converted into the three data sets: one-hot code,word vector and vector that integrates the concept of Construction safety ontology.Naive Bayes,support vector machine,logistic regression,random forest,multi-layer perceptron and TextCNN model.The comparison shows that the TextCNN model with the ontology knowledge proposed in this paper improves the accuracy,precision,F1 value and AUC value.This paper first introduces the research status and significance of Construction safety accidents.Combined with the existing problems such as the lack of effective reuse of knowledge in a large number of relevant fields and the extensive and effective application of ontology in various industries,this paper puts forward the application of ontology technology in the field of Construction safety accident prediction.Then it introduces the main contents and research ideas of this paper.Then it introduces the theory of accident cause and related technologies involved in the experiment in this paper,such as text mining,crawler,ontology construction,TransH model,classification model,etc.,and shows the technical research route of this paper,which lays a theoretical and technical foundation for the research in this paper.Based on the statistical analysis of the construction safety accident information in the past eight years from 2012 to 2019,the main accident types and regions of the investigation report data collection of construction safety accidents are defined.Using the domain word discovery algorithm based on mutual information and adjacency entropy,this paper analyzes the reasons of Construction safety accidents.It provides essential information for the prediction of Construction safety accidents.Based on the relevant literature and national standards in the field of construction safety and the analysis of the relationship between unsafe factors and types of Construction safety accidents,the scope of ontology library construction is clarified.Then the concept classes and the relation attributes of the unsafe factors and accident types of Construction safety accidents are defined.Finally,ontology development software Protégé is used to construct an ontology library.In the prediction of the types of Construction safety accidents,the report data set of Construction safety accidents are represented as the three data sets: one-hot code,word vector and vectorization based on ontology knowledge.Two groups of models are used to predict the three data sets.One group is typical of traditional machine learning methods: Naive Bayes,support vector machine,logistic regression,random forest and multi-layer perceptron.The other group is the TextCNN model.Through the comparative analysis of experimental results,it is shown that the TextCNN model with the ontology proposed in this paper can improve the prediction accuracy,accuracy,F1 value and AUC value.At the end of the paper,the main simulation results and conclusions,as well as future research expectations,are described.Figure[35] Table[16] Reference[60]...