| 摘要: | 水熱性質是多孔介質中，控制地下水流動與熱傳遞的重要角色，對於地熱系統尤為重要。參數估計方法常被用於地熱系統的流動、傳輸行為特徵化，許多研究指出這些估算方法的有效性取決於時空數據的品質與可得性。野外量測普遍存在限制，因此對特定場址使用模型前，制定合宜之取樣策略至關重要。本研究探討隨機方法量化水熱性質不確定性之應用，並強調觀測資料在參數估計中的角色。研究初期建立二維合成含水層模型，應用於比較跨井抽水與注入情境下，不同估計方法的有效性。合成試驗顯示，相較於蒙地卡羅模擬 (MCS)，集合卡爾曼濾波 ( EnKF ) 提供更有可靠性的水熱性質參數估計結果；在熱傳導性估計，兩者方法具有高度相關係數，分別為0.913與0.926。此外，EnKF 在此研究中，於精度與計算效率間提供良好平衡，且比MCS 的計算速度快至六倍。基於這些發現，EnKF被應用於後續受控實驗室進行之沙箱試驗時，觀測數據的參數估計貢獻探討。結果指出EnKF 對估計之參數展現極高相關係數，並成功重現觀測到的熱響應。觀測網路敏感度分析指出觀測剖面達三剖面，模型表現趨於穩定。重要結果表明，於高敏感區域布設分部式觀測網路能最大化資訊可得量，而觀測初期採納高頻量測，可有效約束近場特性。此研究證實EnKF 為進行水熱特性特徵化的有效工具，並為地熱能源開發與地下特徵提供具體之資料收集策略建議。;Hydro-thermal properties play crucial roles in controlling groundwater flow and heat transfer within porous media, particularly in geothermal systems. Parameter estimation techniques are commonly employed to characterize flow and transport behaviors in geothermal systems. As recognized in many studies, the effectiveness of these estimations depends significantly on the quality and availability of spatiotemporal data. Given the typical limitations in field measurements, identifying appropriate sampling strategies is essential before applying models to site-specific conditions. This study investigates the use of stochastic approaches to quantify uncertainties in hydrothermal properties, placing particular emphasis on the role of data observation. In the beginning, a synthetic two-dimensional aquifer model was initially developed to compare the effectiveness of different estimation methods under cross-hole pumping and injection scenarios. Results from these synthetic tests highlighted that the Ensemble Kalman Filter (EnKF) model generally provides more reliable estimates of hydrothermal properties than those obtained from the Monte Carlo Simulation (MCS) method, with extremely high correlation coefficients of 0.926 and 0.913 for thermal conductivity estimation, respectively. Moreover, EnKF offers a good compromise between accuracy and computational efficiency, up to six times faster than MCS, making it the preferred approach for our studies. Building on these findings, a sandbox experiment was conducted in a controlled laboratory setting, where only the EnKF was applied to further explore the contribution of observed data to parameter estimation. The results show that EnKF achieves high correlation coefficients for estimated parameters and successfully reproduces observed thermal responses. Monitoring network sensitivity analysis reveals performance stabilization at three observation profiles. Key findings indicate that distributed observation networks with strategic placement in high-sensitivity regions maximize information gain, while high-frequency early-stage monitoring effectively constrains near-field properties. This research establishes EnKF as an effective tool for hydrothermal characterization, offering practical guidance for data collection strategies in geothermal energy development and subsurface characterization applications.
Keywords: stochastic inversion, Monte Carlo Simulation, Ensemble Kalman Filter, hydro-thermal properties, geothermal system |
