基于超前钻探测试的隧道地层智能识别方法

可靠地识别掌子面前方地层是保证隧道工程稳定与安全的重要因素之一。传统的超前地质预报方法不能同时保证有高识别精度、低实施成本和占用少的施工时间,对于不同地质情况的地层识别通用性不强。在传统超前钻孔的同时获取掌子面前方围岩钻探测试数据,实时获取不同深度岩层情况,将大大提高超前预报效率,方便快捷,不影响施工,但目前缺乏客观、准确的地层识别方法。提出了一种基于神经网络的钻探测试数据智能分析和地层识别方法,对楚大高速公路九顶山隧道超前钻探测试数据进行了深入分析,通过隧道开挖后所揭示地层对分析方法进行了验证。结果表明:单一钻进参数用于地层识别的错误率在35%左右,打击能和打击数、送水压力和送水流量的参数组合不能显著提升地层识别准确率;钻进速度、扭矩、回转数、推进力的参数组合可降低地层识别错误率至22%。在神经网络模型中引入钻进参数的标准差,可大幅降低错误率,可使地层划分错误率下降9%～12%;多参数组合下的神经网络钻探测试神经网络模型对随机抽样的地层识别错误率小于10%,对单个钻孔的地层识别错误率小于14%。

Intelligent recognition of tunnel stratum based on advanced drilling tests

The reliable recognition of strata in front of tunnel face is significant for the stability and safety of the tunnel engineering project. Traditional advanced geological forecasting methods could not ensure high identification accuracy, low cost and short construction time simultaneously, and they can’t satisfy the universality of stratum identification under different geological conditions. The advanced forecasting efficiency could be significantly enhanced if the drilling data of surrounding rocks in front of the tunnel face can be obtained while performing the conventional advanced borehole to attain the rock conditions at different drilling depths in real time, which would be convenient and efficient by not affecting the construction period. However, no objective and accurate stratum identification methods are found. In this paper, we proposed an intelligence analysis of drilling data and stratum recognition method based on neural network. It is used to analyze the advanced drilling test data of Jiudingshan Tunnel of Chuxiong-Dali highway and the analysis method was verified by the strata exposed after tunnel excavation. The results show that the error rate of stratum recognition using the single drilling parameter is about 35%. The combination of blow energy and blow number, water pressure and water flow water cannot significantly improve the accuracy of stratum recognition. The combination of drilling speed, torque, rotation speed and propulsion can reduce the error rate to 22% for stratum recognition. The error rate can be sharply decreased by 9%~12% when the standard deviation of drilling parameters is introduced into the neural network model. The error rate of stratum recognition is less than 10% for random sampled data and it is less than 14% for a single borehole using the neural network model with the combination of multiple drilling test parameters.