【摘要】 为了提升爆破振动峰值质点速度预测的准确性，满足爆破振动引发边坡岩体失稳的防控需求，提出了一种融合特征筛选-数据降维-参数优化的路堑边坡岩体爆破振动峰值质点速度智能预测模型。以首都地区环线高速公路（G95）承德（李家营）—平谷（冀京界）段TJ-2标为工程背景，通过相关性分析剔除低相关性影响因子，筛选出影响爆破振动峰值质点速度的关键爆破参数；采用主成分分析（PCA）消除特征间多重共线性，实现数据降维；引入Runge-Kutta优化器（RUN）算法对极端梯度提升（XGBoost）模型初始超参数进行优化，构建了PCA-RUN-XGBoost爆破振动峰值质点速度预测模型。研究结果表明，采用PCA对爆破参数进行预处理，不仅完整保留了原始信息，而且显著降低了数据冗余程度；引入RUN优化算法对XGBoost超参数进行全局优化搜索，有效地避免了人工调参产生的主观性误差，提升了模型的收敛效率与稳定性；在5种对比模型中，PCA-RUNXGBoost预测模型的拟合度最高，平均相对误差仅为7.02%，可为复杂环境下边坡岩土工程爆破振动峰值质点速度的预测提供可靠参考。更多还原

【Abstract】 In order to improve the accuracy of predicting the particle velocity of peak blasting vibration and meet the prevention and control needs of slope rock instability caused by blasting vibration, an intelligent prediction model for blasting vibration peak particle velocity of road cut slope rock mass is proposed, which integrates feature screeningdata dimensionality reduction-parameter optimization. Taking the TJ-2 section of the Chengde(Lijiaying)-Pinggu(Hebei Beijing border) section of the Capital Ring Expressway(G95) as an example, low correlation influencing factors were removed by correlation analysis, and key blasting parameters affecting the peak particle velocity of blasting vibration were screened out; Multicollinearity between features were eliminated by principal component analysis(PCA) to achieve data dimensionality reduction; The Runge-Kutta optimizer(RUN) algorithm was introduced to optimize the initial hyperparameters of the Extreme Gradient Boosting(XGBoost) model, and a particle velocity prediction model of peak blasting vibration of PCA-RUN-XGBoost was constructed. The research results indicate that preprocessing blasting parameters by PCA not only preserves the original information completely, but also significantly reduces the degree of data redundancy; The RUN optimization algorithm was introduced for global optimization search of XGBoost hyperparameters and effectively avoided the subjective errors caused by manual parameter tuning, and improved the convergence efficiency and stability of the model; Among the 5 comparison models, the PCA-RUN-XGBoost prediction model had the highest fitting degree, with an average relative error of only7.02%. It can provide a reliable reference for predicting the peak velocity of blasting vibration of slope geotechnical engineering in complex environments.更多还原