Prediction of Tunnel Blasting Outcomes Based on SSA-DBN

SHI Long①， CUI Dayong①， LILong^（2） ， CHEN Di③， ZHOU Changchun① ① 1st Engineering Co.，Ltd.，China Railway Construction Bridge Enginering Bureau Group（Liaoning Dalian，16000） （204 ② School of Civil Engineering，Xi'an University of Architecture and Technology（Shaanxi Xi'an，710055） ③ Hubei Jiaotou Yichu Construction Management Co.，Ltd.（Hubei Yichang，443200）

[ABSTRACT]A prediction study on tunnel blasting outcomes was conducted using the Qilinguan Tunnel project as an example.SSA-DBN prediction model based on sparrow search algorithm（SSA）optimized deep belief network （DBN） was used.Using theselected eight parameters thatafect theblastingoutcomes as inputindicators，and theaverageabsoluteerror E_MA ，mean square error E_MS ，and determination coefficient of R² as evaluation indicators，a comparative evaluation was conducted on theoutput indicators（maximum linear over excavation，under excavation and fragmentation）of DBN model， principal component analysis（PCA）optimized DBN model（PCA-DBN），and SSA-DBN model.The results show that R² （204 of the maximum linearover excavation，under excavation，and fragmentationof SSA-DBNmodel is O.9973，0.9977，and 0.998 1，respectively. E_MA is 0.461 0， 0.338 0，and 0.360 2，respectively. E_MS is 0.297 5， 0.178 2，and 0.175 3，respectively.SSA-DBN model has the highestfiting degree between predicted values andactual values，folowed by DBN model，and PCA-DBN model has the lowest. The sensitivity index r² of input parameters to blasting outcomes is mainly between 0.6 and O.7. The accuracy and stability of SSA-DBN model have been verified.

[KEYWORDS]blasting engineering；DBN neural network；sparow search algorithm；prediction of blasting outcome

0 引言

山岭隧道常采用钻爆法进行掘进施工，爆破效果是隧道掘进效率和安全性的重要体现[1-3]。其中，超、欠挖和破碎块度是对爆破效果评价的主要指标[4-9]。雷明锋等[6分析了山岭隧道爆破超、欠挖过大和渣体块度不适等问题，确定了隧道爆破开挖效果评价指标体系。王赞等[7基于深度神经网络预测模型分析了爆破参数和岩石破碎度之间的关系；通过对爆破参数进行敏感性分析，得到炸药单耗和孔距是主要的影响因素，并分别建立了炸药单耗和孔距与岩石破碎度之间的关系。宋战平等[8]将非线性算子引入传统的模糊层次分析法，基于非线性模糊层次分析法评价隧道掘进机施工的风险。张万志[9以拱顶沉降、超挖与欠挖和块石大小为输出，采用粒子群优化算法（particleswarmoptimiza-tion，PSO）与深层反向传播（deep back propagation，DBP）神经网络耦合模型PSO-DBP优化了水平层状岩体隧道光面爆破参数。目前，爆破领域对神经网络的运用主要表现在爆破振动预测和爆破参数优化等方面，而通过麻雀探索算法（sparrowsearchalgo-rithm，SSA）优化深度置信网络（deepbeliefnetwork，DBN）模型SSA-DBN对山岭隧道爆破效果预测的研究和应用较少[10-13]

爆破参数和爆破效果之间的关系具有模糊性和复杂性[10-17]。为了通过岩石参数和爆破参数对爆破效果进行有效预测，建立了SSA优化DBN的SSA-DBN山岭隧道爆破效果预测模型。该模型借助SSA的全局寻优能力优化DBN模型的权重与值，改善DBN模型易陷入局部寻优的缺陷，提高寻优速度。对比SSA-DBN模型和主成分分析（princi-palcomponentanalysis，PCA）优化DBN的PCA-DBN模型，SSA-DBN模型的预测精度更高。……