| China is a major coal production country in the world.As far as the current situation of China’s coal mines is concerned,it mainly adopts the way of underground mining.Due to the low level of industrial production equipment and relatively backward industrial technology,the prevention and treatment of coal miners’ occupational diseases has become extremely difficult.Therefore,early judgment of coal miners’ health status is an important prerequisite for occupational disease of prevention and treatment.In the traditional application scenario,the occupational health of coal miners mainly depends on experienced doctors to analyze and evaluate through various physical signs of the physical examination report.However,there are some problems,such as the complexity of medical data,more redundant information,and it is difficult to judge the potential relationship between data attributes.In view of the above shortcomings,the data reduction method combined with classifiers are proposed to identify the health status of miners based on physical sign parameters in this study.The paper focuses on the identification model,feature construction and recombination,single-level attribute selection and multi-level feature simplification.(1)The better identification model of coal miners’ health status is studied.The qualitative statistical inference analysis is carried out on the original miner data of 21 attributes.The ABC,CS,FA,grey wolf optimization(GWO)and GSA are taken as the intelligent optimization algorithm for the optimal parameters of support vector classification(SVC).The identification model of abnormal coal miner based on intelligent optimization SVC is established by using the original miner data.At the same time,the optimal data preprocessing method and kernel function are determined,and the optimal intelligent optimal SVC identification model is obtained.The experimental results show that under the conditions of normalized preprocessing method with [0,1]and RBF kernel function,the intelligent optimized SVC can achieve the better identification performance.Among them,the GWO-SVC identification model can ensure that on the basis of high average identification accuracy(91.75%),it has a lower average time cost(2.5235 seconds)and the highest identification accuracy of 92.5%.The optimal parameters cost is 39.3192 and gamma is 0.763.The samples of overall error identification are mainly concentrated in the drivage and the coal preparation work area.(2)The identification based on feature construction and recombination is studied.Considering that there are a certain number of potential features in the original physical sign data that directly affect the identification accuracy of the model.Based on this,the physiological index features of BMI,PP,MAP and RPP,the information features of coal coal miners’ basic data and their combination features are added to the original attribute set to expand and transform the coal miner data with 21 attributes.The intelligent optimized SVC identification model is constructed,and the final new feature data set is obtained according to different evaluation indexes.The experiment shows that compared with the original attribute parameter data,the average identification accuracy of optimized SVC algorithm(92.43%)is improved by the combined data of continuous features.The performance of models of common machine learning algorithm has also been improved.In the SVC model of continuous feature combination,the average identification accuracy of many experiments of GWO-SVC is 92.5%,which is very stable.Among them,the time cost of the sixth experiment is the lowest,which is 2.329 seconds.The optimal parameters corresponding to the best result are cost is 39.3192,gammais0.2229.(3)The identification based on method of single-level attribute selection is studied.In view of some redundant attribute information that cannot be judged manually in the newly constructed and supplemented data,the single-level data reduction method is used for the feasibility analysis of important attribute selection.The correlation analysis based on PCC and SCC and the model-free attribute reduction method between LPS and m RMR are compared.The training of RF,EN,SVM-RFE,and NCA model learning methods are used to remove useless and redundant attribute information.The average identification evaluation indexes of GWO-SVC model with different model learning reduction methods,model learning reduction methods and unfiltered data are compared.The NCA model is determined to obtain better identification performance and low time cost.The average identification accuracy is 96%,and it only retains four important attributes.The reduced attribute data of NCA is also applicable to other classification algorithms of machine learning,and improves the identification ability of their models.(4)The identification based on strategy of multi-level feature simplification is studied.The model-free learning and model learning algorithms are combined through the decreasing order of average identification accuracy.The simplification strategies based on model-free learning,model learning,between model-free learning and model learning are constructed to simplify the data of coal miner’s physical signs.The performance of the model under different types of multi-level simplification strategies is compared in detail.Meanwhile,it is compared with the performance of single-level reduction method.The simplified attribute data is used to establish a conventional machine learning model to verify the universality of the optimal simplification strategy.Six features are obtained by reduction strategy based on EN-m RMR are combined with GWO-SVC classifier to obtain the highest average recognition accuracy(97.38%),which is 4.88% higher than that of the unfiltered data,the time cost(1.4906 seconds)is reduced by 1.113 seconds,2.75% higher than that of the single-level selection,and the time cost is reduced by 0.4297 seconds.The simplified optimal features are {DBP,ALT,CHOL,GLU,TG,RPP},which are only 24% of the unfiltered attributes and 50% of the singlelevel selection results respectively.This paper provides a new idea and method for the early judgment of occupational health.The experimental result show that the multi-level data reduction combined with GWO-SVC is feasible for the identification of coal miners’ health status based on physical sign parameters.Compared with the traditional identification methods,this study can realize the accurate and efficient identification of coal miners’ health status by fewer features.The research conclusion can provide reference basis for early intelligent screening,intelligent health management and disease prevention of coal miners’ occupational health,which can realize the intellectualization of miners’ health intervention in advance and reduce the risk of disease.Figure [64] Table [42] Reference [188]... |
PDF Full Text Request