Scholarlink

<p>Energy release caused by hydraulic fracturing in enhanced geothermal systems (EGS) is a precursor to induced seismicity. This study developed an artificial neural network (ANN) model to predict cumulative seismic moment (M0) using a dataset of 972 samples obtained from numerical inversion and sensitivity analysis of critical parameters (injection volume, rate, shear stress, normal stress, friction angle, stress drop). The ANN model showed that released energy positively correlates with injection volume, rate, shear stress, and stress drop, while it negatively correlates with normal stress and friction angle. The importance of the parameters is ranked as follows: injection volume &gt; shear stress &gt; normal stress &gt; friction angle &gt; injection rate &gt; stress drop. The ANN model demonstrates accurate predictions, with the slopes (K) of fitted lines for both original and predicted data approximating 1.0 and R<sup>2</sup> exceeding 0.96 across all datasets: training set (K = 1.0023, R<sup>2</sup> = 0.9981), test set (K = 1.0263, R<sup>2</sup> = 0.9658), extrapolation test 1 (K = 0.9015, R<sup>2</sup> = 0.9696, relative error &lt; 10%), extrapolation test 2 [predicted vs. simulated data (K = 0.9444, R<sup>2</sup> = 0.9965); predicted vs. field-measured data (2.98% relative error at the end of injection)], and Basel project M0 prediction (10.41% relative error). This study provides an essential framework for safe geothermal exploitation and seismic hazard mitigation by linking energy release parameters to induced seismicity prediction through deep learning.</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

Neural network model for prediction of water-injection-induced energy release in deep geothermal based on extended data through numerical simulation

  • 期刊论文

  • 发表日期：

    2025年7月14日

查看全文
  • Wenhang Dai,
  • Lei Zhou,
  • Yi Chen,
  • Liulin Fang,
  • Xiaocheng Li

摘要

Energy release caused by hydraulic fracturing in enhanced geothermal systems (EGS) is a precursor to induced seismicity. This study developed an artificial neural network (ANN) model to predict cumulative seismic moment (M0) using a dataset of 972 samples obtained from numerical inversion and sensitivity analysis of critical parameters (injection volume, rate, shear stress, normal stress, friction angle, stress drop). The ANN model showed that released energy positively correlates with injection volume, rate, shear stress, and stress drop, while it negatively correlates with normal stress and friction angle. The importance of the parameters is ranked as follows: injection volume > shear stress > normal stress > friction angle > injection rate > stress drop. The ANN model demonstrates accurate predictions, with the slopes (K) of fitted lines for both original and predicted data approximating 1.0 and R2 exceeding 0.96 across all datasets: training set (K = 1.0023, R2 = 0.9981), test set (K = 1.0263, R2 = 0.9658), extrapolation test 1 (K = 0.9015, R2 = 0.9696, relative error < 10%), extrapolation test 2 [predicted vs. simulated data (K = 0.9444, R2 = 0.9965); predicted vs. field-measured data (2.98% relative error at the end of injection)], and Basel project M0 prediction (10.41% relative error). This study provides an essential framework for safe geothermal exploitation and seismic hazard mitigation by linking energy release parameters to induced seismicity prediction through deep learning.