Long And Short Term Prediction Of Miners’ Breathing Flow Based On Random Forest And LSTM Neural Network

Coal mines are an industry with a high incidence of pneumoconiosis in China,and wearing individual particulate respirators is an important means of preventing pneumoconiosis.However,the existing anti-particulate respirators have defects such as excessive breathing resistance and insufficient protection,while the new autonomous air supply and on-demand air supply respirators have problems such as lagging air volume and insufficient air supply.In view of this,it is of great practical significance to investigate the breathing characteristics of coal miners and to realize the advanced diagnosis of breathing demand and on-demand air supply in order to solve the problem of lagging air supply.In this thesis,the key parameters such as instantaneous heart rate,relative heart rate index,metabolic rate and respiratory resistance are selected for the representative dust catching jobs in coal mines,and a respiratory flow prediction model based on RF algorithm and LSTM neural network is established,which can analyze the change of respiratory flow of coal miners and the importance of influencing factors,and then realize the long and shortterm prediction of respiratory flow of coal miners,and the results can be used for the development of intelligent The results can provide important support for the research and development of intelligent air supply respirator.The main work contents and conclusions are as follows:1)Monitoring of the respiratory characteristics of 164 dust collection miners in a mine showed that the ventilation volume per minute(MV,55～59 L/min),average inhalation flow rate(MIF,123～130 L/min),peak inhalation flow rate(PIF,187～203 L/min),and maximum peak inhalation flow rate(PIFmax,257～280 L/ min)are much larger than those of the coal processing plant.It shows that the work intensity of the underground work types in coal mines is much higher than that of the coal processing plant work types;with the increase of working time,the MV,PIF,MIF and PIFmax of each work type show an obvious rising trend.The analysis of the influence of the characteristic variables of the measured dust catcher types revealed that the miners’ continuous work in deep shafts would cause a sharp increase in body metabolic rate and oxygen consumption,which would lead to an increase in respiratory rate,respiratory flow rate and MV,MIF and PIF data,and then the clean airflow through the filter elements of individual respiratory protective equipment could not meet the rapidly increasing respiratory demand.2)In order to reduce the input layer dimensionality at the pre-training stage of the prediction model and avoid over-referencing of the prediction model constructed at a later stage,this study assessed the importance of features based on the RF algorithm for out-ofsample laboratory data,in which the feature variable with the greatest influence on respiratory flow was respiratory resistance,with the highest contribution of 36.8% in the resting state,followed by light labor(34.4%),heavy labor(33.0%)and moderate labor(31.8%);human relative heart rate index had the highest contribution of 30.8% in the moderate and heavy states.The two variables of metabolic rate and instantaneous heart rate were the next highest,with the highest contribution rates of 18.1% and 14.7% in the light and resting states,respectively.Finally,according to the contribution rate of different characteristic variables(contribution rate: respiratory resistance > relative heart rate index >instantaneous heart rate > metabolic rate),respiratory resistance,relative heart rate index,instantaneous heart rate,and metabolic rate were selected as the main characteristic indexes,and the subsequent prediction model was also limited to these four characteristic variables as the input layer of the model,which further enhanced the accuracy and robustness of the miner’s respiratory flow prediction model.3)The cross-sectional comparison of the prediction results between different models revealed that the hybrid RF-LSTM neural network processed the prediction of the respiratory flow of the dust catching miners with higher accuracy.Among them,the results of shortterm prediction of miners’ respiratory flow showed that the MAPE values of hybrid RFLSTM neural network were reduced by about 11.36%,8.68%,8.95% and 1.04%,respectively,compared with other models;the results of long-term prediction of miners’ respiratory flow showed that the MAPE values of hybrid RF-LSTM neural network were reduced by about 4.37%,0.21%,9.18% and 2.59%,respectively,compared with other models,0.21%,9.18% and about 2.59%,respectively,compared to the other models.Therefore,the hybrid RF-LSTM model proposed in this paper has higher generalizability and is more suitable for predicting the respiration of miners in dust catching miners’ operating conditions.There are 47 figures,16 tables and 81 references in this thesis.