Mohammadian Farough, Sadeghi Mehran, Hanifi Saber Moradi, Noorizadeh Najaf, Abedi Kamaladdin, Fazli Zohreh
Department of Occupational Health Engineering, Environmental Health Research Center, Research Institute for Health Development, Kurdistan University of Medical Sciences, Sanandaj, Iran.
Computer Department, Institute of Higher Education of Bahmanyar, Kerman, Iran.
Work. 2022;73(1):189-202. doi: 10.3233/WOR-205271.
Many occupational accidents annually occur worldwide. The construction industry injury is greater than the average injury to other industries. The severity of occupational accidents and the resulting injuries in these industries is very high and severe and several factors are involved in their occurrence.
Modeling important factors on occupational accident severity factor in the construction industry using a combination of artificial neural network and genetic algorithm.
In this study, occupational accidents were analyzed and modeled during five years at construction sites of 5 major projects affiliated with a gas turbine manufacturing company based on census sampling. 712 accidents with all the studied variables were selected for the study. The process was implemented in MATLAB software version 2018a using combined artificial neural network and genetic algorithm. Additional information was also collected through checklists and interviews.
Mean and standard deviation of accident severity rate (ASR) were obtained 283.08±102.55 days. The structure of the model is 21, 42, 42, 2, indicating that the model consists of 21 inputs (selected feature), 42 neurons in the first hidden layer, 42 neurons in the second hidden layer, and 2 output neurons. The two methods of genetic algorithm and artificial neural network showed that the severity rate of accidents and occupational injuries in this industry follows a systemic flow and has different causes.
The model created based on the selected parameters is able to predict the accident occurrence based on working conditions, which can help decision makers in developing preventive strategies.
全球每年都会发生许多职业事故。建筑业的伤害高于其他行业的平均伤害水平。这些行业中职业事故的严重程度以及由此导致的伤害非常高且严重，其发生涉及多个因素。
结合人工神经网络和遗传算法，对建筑业职业事故严重程度的重要因素进行建模。
在本研究中，基于普查抽样，对一家燃气轮机制造公司下属的5个重大项目施工现场5年内的职业事故进行了分析和建模。选取了712起包含所有研究变量的事故进行研究。该过程在MATLAB 2018a软件中使用人工神经网络和遗传算法相结合的方式实现。还通过清单和访谈收集了其他信息。
事故严重率（ASR）的均值和标准差分别为283.08±102.55天。模型结构为21、42、42、2，表明该模型由21个输入（选定特征）、第一个隐藏层的42个神经元、第二个隐藏层的42个神经元和2个输出神经元组成。遗传算法和人工神经网络这两种方法表明，该行业事故和职业伤害的严重率遵循系统流程且有不同原因。
基于选定参数创建的模型能够根据工作条件预测事故发生情况，这有助于决策者制定预防策略。
