军事知识服务平台
基于特征选择的MSCOA-CNN-SVM隧道爆破线性超挖预测
The linear over-excavation prediction of tunnel blasting based on feature selection using MSCOA-CNN-SVM
摘要 为解决隧道爆破效果难以精确预测的问题,提出了一种基于特征选择的MSCOA-CNN-SVM预测模型.以蟠龙山隧道爆破施工数据为基础,首先采用随机森林和皮尔逊相关系数分析对12个最大线性超挖影响因素进行降维,筛选出9个关键特征,作为模型的输入特征.利用改进的龙虾算法(MSCOA)优化卷积神经网络与支持向量机(CNN-SVM)模型的超参数,并将最优参数输入模型进行训练、测试与评估.研究结果表明,随机森林和皮尔逊相关系数分析结合的特征降维方法有效降低了冗余信息,从而提高了预测精度;MSCOA优化算法在超参数调整方面克服了人工选择可能导致的过拟合问题.最终,基于特征选择的MSCOA-CNN-SVM模型的预测决定系数(R2)达到0.962 8,优于其他3种预测模型,表现出更高的准确性,为多因素影响下的爆破超挖预测提供了有效参考.
Abstract:
To address the challenge of accurately predicting tunnel blasting effects,this study proposes a feature selection-based MSCOA-CNN-SVM prediction model.Using blasting data from the Panlong Mountain Tunnel construction as a case study,the model first applies random forest and Pearson correlation analysis to reduce the dimensionality of 12 factors influencing maximum linear over-excavation,selecting 9 key features as the model's input.An improved lobster algorithm(MSCOA)is used to optimize the hyperparameters of the Convolutional Neural Network and Support Vector Machine(CNN-SVM)model.These optimized parameters are then input into the model for training,testing,and evaluation.The results show that the combined feature reduction method using random forest and Pearson correlation analysis effectively minimizes redundant information,thus improving prediction accuracy.The MSCOA optimization algorithm overcomes the overfitting issues that may arise from manual parameter selection.Ultimately,the feature-selected MSCOA-CNN-SVM model achieves a prediction coefficient of determination(R2)of 0.9688,outperforming three other prediction models,demonstrating higher accuracy.This approach provides an effective reference for predicting over-excavation in blasting under multi-factor influences.
刊 名 工程爆破 ISTICPKU
年,卷(期): 2026, 32(4)
分类号: TD235
机标分类号: TP391.41 TP181 TN98
在线出版日期: 2026年9月9日
基金项目: 52308495:国家自然科学基金QN2025278:河北省教育厅青年基金资助项目,