Abstract: The construction site of diversion tunnels are often accompanied by toxic and harmful gases generated from blasting operations, which would affect the safe operation of construction. To address the above problem, a prediction model for toxic and harmful gas concentration based on GA-BP neural net⁃ work is proposed, which can improve the shortcomings of BP neural network, such as difficulty in obtain⁃ ing the initial threshold and weights, and the proneness of falling into local optimal value. The model is applied to the blasting construction site of a diversion tunnel in Chongqing for analysis and verification. The result shows that in the concentration prediction of carbon monoxide, methane and hydrogen sulfide, the convergence error and iteration steps of GA-BP neural network are smaller than BP neural network, and it fits better with the trend of measured toxic and harmful gas concentration, showing better general⁃ ization ability, which can provide effective protection to on-site safe operations after blasting

Key words: toxic and harmful gas, GA-BP, prediction model, tunnel

摘要： 引水隧洞施工现场常伴随着爆破作业产生的有毒有害气体，会对人体造成危害，影响现场人员的安全作 业。针对以上问题，提出一种基于GA-BP神经网络的有毒有害气体浓度预测模型，能改善BP神经网络对初始阈 值、权值获取困难及容易陷入局部最优值的缺点。将模型应用到重庆某引水隧洞的爆破施工现场进行分析验证， 结果表明：在对一氧化碳、甲烷、硫化氢三种气体浓度预测中，该模型收敛误差和迭代步数均小于BP神经网络，与 实测有毒有害气体浓度的趋势贴合更好，表现出较好的泛化能力，为现场爆破后的安全作业提供了有效保障。

关键词: 有毒有害气体, GA-BP, 预测模型, 隧洞