为解决传统数值模拟方法计算耗时长、难以满足实时性需求的问题，提出一种本征正交分解与支持向量机相结合的预测方法，旨在实现复杂工况下施工隧道污染物浓度场的快速重构。首先，建立施工隧道爆破后CO运移的数值模型； 其次，针对不同炸药量和海拔的典型工况，模拟爆破后隧道内CO体积分数分布，以此来获得样本数据集；
再次，通过本征正交分解对高维度浓度场数据进行降维处理，基于能量截取前三阶模态作为特征基函数，降低计算的复杂度； 最后，采用支持向量机模型建立已知工况参数与模态系数之间的非线性映射关系，实现目标工况下不同时间隧道内CO体积分数的快速预测。研究结果表明： 该预测模型具有显著的准确性和高效性。其中，在预测未知炸药量工况下隧道爆破后的CO体积分数时，最大相对误差不大于15%，平均相对误差为2.73%； 在预测未知海拔工况下隧道爆破后的CO体积分数时，最大相对误差不大于6%，平均相对误差为1%。此外，该预测模型耗时仅为数值模拟计算耗时的1/600，使施工隧道污染物浓度场的近实时获取成为可能。
Traditional numerical simulation methods has long
calculating period for pollutant concentration in construction tunnels, this
disables real-time demand. To address this challenge, a rapid prediction method
integrating proper orthogonal decomposition (POD) and support vector machine
(SVM) is proposed, aiming to enable rapid reconstruction of pollutant
concentration fields in construction tunnels under complex working conditions.
In this approach, a numerical model is first established for CO migration in
tunnels after blasting, simulating CO concentration distributions under typical
scenarios with varying explosive quantities and altitudes, thus yielding a
sample dataset. High-dimensional concentration field
data are then subjected to dimensionality reduction via POD, where the first
three-order modes selected based on energy contribution
serve as characteristic basis functions to mitigate computational complexity.
Further, an SVM model is employed to construct a nonlinear mapping between
known operational parameters and POD modal coefficients, enabling fast
prediction of tunnel CO concentrations at target times under specified
conditions. Validation results demonstrate the model′s
notable accuracy and efficiency. For predicting CO concentrations under unknown
explosive quantities, the maximum relative error is no more than 15% with an
average of 2.73%; for unknown altitudes, the maximum relative error is no more
than 6% with an average of 1%. Critically, the computation time of the proposed
model is 1/600 of that required for traditional numerical simulations,
facilitating real-time acquisition of tunnel pollutant
concentration fields.
