Research On The Self-classification Model And Risk Pre-control Of Safety Production Accidents Based On Text Data

Workplace safety production research is an important field concerning the safety of people’s lives and property.However,safety accidents often occur in practical production,causing not only significant personal and property losses but also adverse impacts on society and the economy.Therefore,it is particularly important to classify,analyze,and prevent safety production accidents.With the development of big data and natural language processing technology,natural language processing and machine learning methods based on unstructured text data have been widely applied in various fields.They also provide new ideas and methods for integrating intelligent and informational technologies into safety production accident automatic classification and risk pre-control,which can provide reference for safety production management decision-making.However,current research on processing text data for knowledge extraction and decision systems using digital technologies is still relatively limited,especially in China where relevant studies are lacking.There is an urgent need to conduct in-depth research on safety production accident classification and risk pre-control based on digital science and information management technologies,and to further explore effective ways to improve the informationization process of workplace safety production.This dissertation proposes a text-based safety production risk management framework through a data-driven research paradigm and artificial intelligence techniques.The framework utilizes data processing techniques such as natural language processing,convolutional neural networks,association rule mining,fuzzy Bayesian networks,and random forests to construct a self-classification model for safety production text accident reports and a risk pre-control system.The framework has been validated using a case study in the construction industry,preliminarily demonstrating its effectiveness in self-classifying accident types,identifying risk factors,extracting critical risk factors and their chain propagation relationships,conducting safety production risk assessments,and formulating risk control plans.The combination of these tools and techniques can help safety production managers better identify safety risks,further improve safety production management efficiency,and provide support for reducing or eliminating occupational injuries and losses,as well as providing theoretical guidance and practical guidance for accident prevention and hazard investigation in the field of safety production.The main research content,research methods,and conclusions of this dissertation are as follows.(1)Identification of safety risk factors using an improved text preprocessing method.Based on the core techniques and task execution process of natural language processing,an improved approach to text preprocessing is proposed for safety production accident text reports.Four main processing steps are introduced:hierarchical data cleaning,Chinese text segmentation,feature representation and extraction,and contextual processing of feature words.These steps aim to comprehensively identify safety production risk factors.Through the validation with actual construction accident cases,unstructured implicit information contained in construction accident reports is identified,and a set of construction risk factors consisting of 52 subcategories is successfully constructed.These risk factors are classified into four categories: human factors(HF),facility factors(FF),environmental factors(EF),and management factors(MF).Through the structured data transformation process,a dataset that can be automatically recognized by computers is obtained,providing data foundation for learning accident causation mechanisms and risk control strategies from historical experiences in subsequent language modeling tasks.The research results demonstrate that the improved text processing approach can effectively identify safety production risk factors and provide core data support for safety production management.(2)Construction of a data-driven self-classification model for safety production accident texts.Due to the presence of a large amount of redundant information in accident text reports,it is difficult for managers to manually extract valuable information.Therefore,the use of artificial intelligence methods is needed to extract valuable text features and automatically classify text accident reports.Based on the convolutional neural network(CNN)algorithm,this dissertation proposes a data-driven research paradigm for a self-classification model for safety production accident texts.The model utilizes a multi-layer learning structure to automatically capture key feature information and has been successfully applied to the task of classifying accident types in construction text reports.Experimental results show that the proposed CNN-based deep learning model achieves an overall accuracy of 76% in the text classification task,while the overall accuracy indicators of three classical shallow learning models(Support Vector Machine,Logistic Regression,and Naive Bayes)are 73%,66%,and 54% respectively.The research results show that the proposed self-classification model for accident texts exhibits higher accuracy and stability,possesses practical and promotional value,and helps managers to effectively obtain valuable information from safety production accident text reports,thus promoting the process of safety production informatization.(3)Importance and correlation analysis of safety production risk factors.This dissertation applies the Apriori algorithm to conduct association analysis on collected safety production accident cases from the perspectives of support,confidence,and lift.It systematically uncovers the underlying implicit patterns of safety production accidents.The experimental results of association rule mining using construction accident data as an application example show that there is a strong chain transmission relationship between inadequate safety hazard handling,insufficient performance of safety management personnel,and incomplete safety hazard investigation and between weak safety awareness,inadequate performance of safety management personnel,and lack of safety training and education.Additionally,this dissertation also adopted a mixed method of HFACS,Bayesian networks,and fuzzy set theory to investigate the information related to human and organizational factors in accident reports,infer the key human and organizational factors that cause safety production accidents and their interdependent relationships.The results of Bayesian network deduction and sensitivity analysis using construction accident data show that on-site safety management loopholes,lack of safety culture,inadequate safety supervision,inspection and acceptance,non-compliance in operation,and non-standard equipment operation are the core risk factors affecting safety production.The research results show that studying the internal mechanism of safety production risk factors through the above combined methods can help to explain the internal connections between risk factors and propose targeted safety risk prevention measures,thereby providing theoretical and practical guidance for comprehensive and efficient risk prevention and control.(4)Construction and optimization of safety production risk assessment model.This dissertation aims to construct a machine learning-based safety production risk assessment model and set up three experimental scenarios to optimize the model for effectively predicting the severity of occupational accidents.Additionally,focusing on the interpretability of the model,the underlying predictive mechanism of the model is investigated to enhance the understanding of machine learning algorithms and provide better explanations for the prediction results.By validating the model performance using data on construction collapse incidents,the evaluation of the model shows that the accuracy of the safety assessment model using partial key attribute features,all attribute features,and hyperparameter optimization is 77%,80%,and 82%respectively.Furthermore,the results of the interpretability analysis indicate that the emphasis of project managers on safety,government regulation,safety design,and emergency response are critical attribute factors affecting the severity of collapse accidents.These factors can reasonably explain the underlying logical relationships that lead to varying degrees of injury severity.The research findings demonstrate that the safety production risk assessment model constructed using the random forest algorithm,based on the effective identification of key attribute factors from unstructured accident text,exhibits good predictive accuracy.The proposed model can meet the comprehensive accident prevention requirements of hazard investigation and risk classification management,providing technical support for occupational safety and health issues.(5)Development of safety production risk control plan based on risk assessment results.This dissertation constructs a closed-loop control system for safety production risks based on the principles of closed-loop control,dynamic control,and hierarchical multi-level control in accordance with the principles of safety production risk control and the risk assessment results provided in the previous sections.Meanwhile,using the bow-tie model,specific risk control schemes are proposed from the aspects of risk assessment prevention measures and emergency response measures.And as a practical application case,focusing on construction safety production,targeted risk prevention measures for safety production have been formulated from the perspectives of government regulatory departments,organizational level,on-site management level,and workers.In order to avoid the failure of prevention measures that may lead to the further loss of control over hazardous factors,emergency response measures are formulated,emphasizing timely actions such as rescue operations,implementing crowd evacuation,and immediately contacting emergency rescue units to quickly control the impact of risk occurrence.The research results show that the safety production risk control system based on the risk assessment results can comprehensively guide the scientific practice of safety production risk management,and the provided risk control plans can effectively control safety production risks from multiple aspects,providing practical guidance for safety production management departments.