一种隧道基坑多维度时变预测EPS模型的应用

Application of a multidimensional time-varying predictive EPS model for tunnel pit

摘要:

如何能准确地掌握到正在施工的隧道深基坑所发生形变的态势,就可以进一步实现动态预测并采取有效的扼制举措,才能有效保障施工安全。据此提出一种基于信号分析法,通过耦合经验模态分解法(Empirical Mode Decomposition)、鸟群觅食算法(PSO)与单隐层前馈神经网络SLFNs学习算法,结合成专为非线性情况下的基坑施工作多维时变预报模型EMD-PSO-SLFNs(简称EPS)。其先将隧道形变的深坑序列分解时的EMD进行多尺度原生模态函数(IMF)；并引入IMF、PSO-SLFNS变量序列进行预测,对其进行叠加预测,用模型的进行最终结果的运算预测,再用耦合PSO与SLFNs量化算法的作末端处理、变量序列进行预测。下文以南宁某隧道基坑施工为例,经过深层次透析得出,单凭EMD分解模型预测的相对误差为值在0.22%至0.42%之间,值δ=0.32%实际均差值；而进行EMD-PSO-SLFNs组合型作多维度时变分解模型预测的相对误差为0.31%至0.75%之间,值δ=0.64%,该预测精度明显高于前者,而且能在非平稳线性、变序情况下预测,为隧道基坑形变预测提供了一种实用新型的方法。

Abstract:

How to accurately grasp the deformation of the deep pit in the tunnel under construction, can further achieve dynamic prediction and take effective containment measures to effectively guarantee construction safety. Accordingly, a signal-based analysis method is proposed to combine the coupled empirical mode decomposition (EMD-PSO-SLFNs), bird feeding algorithm (PSO) and single cryptic feedforward neural network SLFNs learning algorithm into a multidimensional time-varying prediction model EMD-PSO-SLFNs (referred to as EPS) for pit construction under non-linear conditions. It first decomposes the EMD of the tunnel deformation in the deep pit sequence for multi-scale native modal function (IMF); and introduces IMF and PSO-SLFNS variable sequences for prediction, superimposes prediction on them, and predicts the final results using the model, and then uses the end-processing and variable sequences of the coupled PSO and SLFNs quantization algorithm for prediction. The following is an example of a tunnel pit construction in Nanning, after in-depth analysis, the relative error predicted by the EMD decomposition model alone is between 0.22% and 0.42%, with δ=0.32% actual mean difference; while the relative error predicted by the EMD-PSO-SLFNs combination for multi-dimensional time-varying decomposition model is between 0.31% and 0.75%, with δ=0.64%, the prediction accuracy is significantly higher than the former, and can be predicted under non-stable linear, variable order, providing a practical and novel method for tunnel pit deformation prediction.