Scholarlink

<p>Peak Particle Velocity (PPV) is a key indicator for assessing blasting-induced vibration intensity and plays a critical role in ensuring safety in open-pit mining and protecting nearby structures. Although machine learning (ML) techniques have shown advantages in modeling nonlinear relationships for PPV prediction, their performance is often limited by the scarcity of high-quality blasting data, the complexity of variable structures, and high data acquisition costs. These challenges hinder the generalization and accuracy of existing models. To address this issue, this study introduces a novel conditional diffusion method designed for PPV-related data augmentation. As a pioneering exploration of diffusion models in this context, the proposed method is specifically tailored to handle hybrid discrete–continuous variables commonly present in blasting scenarios. Built upon the Denoising Diffusion Probabilistic Model framework, the method incorporates conditional embeddings, residual learning, and time-step encoding to enhance the physical consistency and structural fidelity of generated data. The implementation is developed in Python. Experimental results on a real-world blasting dataset demonstrate that the proposed method outperforms existing generative models in both data quality and downstream predictive performance. When synthetic data generated by the method is combined with real data for training, the predictive accuracy and robustness of six widely used ML models are significantly improved. In addition, interpretability analysis reveals the contribution and impact direction of each input variable on PPV predictions. The proposed method offers a new paradigm for data augmentation and feature modeling in blasting vibration analysis and provides methodological insights for broader applications in rock engineering. The source code is available at <a href="https://github.com/CUMTBLIHENG/CRTDiff">https://github.com/CUMTBLIHENG/CRTDiff</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CRTDiff: A Conditional Residual Temporal Diffusion Model for Data Augmentation to Enhance Machine Learning Prediction of PPV in Open-Pit Mining

  • 期刊论文

  • 发表日期：

    2025年10月14日

查看全文
  • Heng Li,
  • Beijing Xie,
  • Xiaoxu Li,
  • Ben Zhang,
  • Zhuo Li

摘要

Peak Particle Velocity (PPV) is a key indicator for assessing blasting-induced vibration intensity and plays a critical role in ensuring safety in open-pit mining and protecting nearby structures. Although machine learning (ML) techniques have shown advantages in modeling nonlinear relationships for PPV prediction, their performance is often limited by the scarcity of high-quality blasting data, the complexity of variable structures, and high data acquisition costs. These challenges hinder the generalization and accuracy of existing models. To address this issue, this study introduces a novel conditional diffusion method designed for PPV-related data augmentation. As a pioneering exploration of diffusion models in this context, the proposed method is specifically tailored to handle hybrid discrete–continuous variables commonly present in blasting scenarios. Built upon the Denoising Diffusion Probabilistic Model framework, the method incorporates conditional embeddings, residual learning, and time-step encoding to enhance the physical consistency and structural fidelity of generated data. The implementation is developed in Python. Experimental results on a real-world blasting dataset demonstrate that the proposed method outperforms existing generative models in both data quality and downstream predictive performance. When synthetic data generated by the method is combined with real data for training, the predictive accuracy and robustness of six widely used ML models are significantly improved. In addition, interpretability analysis reveals the contribution and impact direction of each input variable on PPV predictions. The proposed method offers a new paradigm for data augmentation and feature modeling in blasting vibration analysis and provides methodological insights for broader applications in rock engineering. The source code is available at https://github.com/CUMTBLIHENG/CRTDiff.