Scholarlink

<p>The prediction of mining-induced ground movements, as addressed in subsidence engineering is characterized by the high heterogeneity of geological structures and operational conditions.</p><p>Existing models—whether empirical, physical, or stochastic—typically require local calibration and are therefore only transferable to a&#xa0;limited extent.</p><p>This contribution presents a&#xa0;novel methodological approach in which a&#xa0;neural network (NN) is used to predict surface subsidence caused by underground mining activities or cavern construction. A&#xa0;key innovation lies in the training data: Instead of relying on real, often inaccessible measurement data, the network is trained using synthetically generated subsidence troughs based on empirically grounded influence functions from subsidence engineering. These synthetic troughs allow for a&#xa0;controlled variation of geological and technical parameters, thereby covering a&#xa0;wide range of possible scenarios.</p><p>The aim is to develop a&#xa0;flexible, generalizable prediction model that does not require real-world data, yet is capable of capturing complex patterns of mining-induced ground movements. The presented approach demonstrates how an established domain knowledge can be systematically integrated to enable data-independent and adaptive modeling. As such, this article contributes to the advancement of innovative methods in subsidence engineering and provides a&#xa0;foundation for future applications and research efforts.</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

Ein konzeptioneller Ansatz zur Prognose bergbaulicher Senkungen auf Basis synthetischer Daten und neuronaler Netze

  • 期刊论文

  • 发表日期：

    2025年8月29日

查看全文
  • Selmin Sakiri

摘要

The prediction of mining-induced ground movements, as addressed in subsidence engineering is characterized by the high heterogeneity of geological structures and operational conditions.

Existing models—whether empirical, physical, or stochastic—typically require local calibration and are therefore only transferable to a limited extent.

This contribution presents a novel methodological approach in which a neural network (NN) is used to predict surface subsidence caused by underground mining activities or cavern construction. A key innovation lies in the training data: Instead of relying on real, often inaccessible measurement data, the network is trained using synthetically generated subsidence troughs based on empirically grounded influence functions from subsidence engineering. These synthetic troughs allow for a controlled variation of geological and technical parameters, thereby covering a wide range of possible scenarios.

The aim is to develop a flexible, generalizable prediction model that does not require real-world data, yet is capable of capturing complex patterns of mining-induced ground movements. The presented approach demonstrates how an established domain knowledge can be systematically integrated to enable data-independent and adaptive modeling. As such, this article contributes to the advancement of innovative methods in subsidence engineering and provides a foundation for future applications and research efforts.