日本地球惑星科学連合2023年大会

講演情報

[E] オンラインポスター発表

セッション記号 M (領域外・複数領域) » M-GI 地球科学一般・情報地球科学

[M-GI25] Near Surface Investigation and Modeling for Groundwater Resources Assessment and Conservation

2023年5月26日(金) 10:45 〜 12:15 オンラインポスターZoom会場 (22) (オンラインポスター)

コンビーナ:Jui-Pin Tsai(National Taiwan University, Taiwan)、Ping-Yu Chang(National Central University)、Hwa-Lung Yu(National Taiwan University)、谷口 真人(総合地球環境学研究所)

現地ポスター発表開催日時 (2023/5/25 17:15-18:45)

10:45 〜 12:15

[MGI25-P01] Developing a Multi-wells Groundwater Level Simulation Model Using Artificial Neural Network and Superposition Principle

★Invited Papers

*Liang Cheng Chang1You Cheng Chen2、Chia Szu Yen2、Tzu hua Lai2 (1.National Chiao-Tung University、2.National Yang Ming Chiao Tung University)

キーワード:artificial neural network, numerical experiment, superposition principle

A groundwater simulation model is an important tool for groundwater analysis and management. At present, numerical models, such as MODFLOW and FEMWATER, are the most commonly used groundwater simulation models. However, building a groundwater numerical model is complicated and time-consuming and required to collect a large amount of hydrogeological parameters data. In responding to this situation, this study uses Artificial Neural Network to establish a general groundwater level model for a single pumping well. Based on this general single-well groundwater level model, this study applied the linear superposition principle and parameter optimization to develop a multi-wells groundwater level simulation model.
The general single-well groundwater level model is an ANN model. This study uses the groundwater numerical model MODFLOW to generate training data, in which the pumping volume, hydraulic conductivity, storage coefficient, and previous groundwater drawdown are the input variables, and the future groundwater drawdown is the output variable. In generating the training data, the pumping rate is time-varying. Therefore, the trained ANN model can simulate the groundwater level variation under the time-varying pumping rate.
Moreover, this study uses the general single-well groundwater level model as a kernel to develop a multi-well groundwater level simulation model by applying the superposition principle and parameter optimization. The numerical experiments for confined and unconfined aquifers with homogeneous or heterogeneous hydrogeology parameters were conducted to verify the multi-wells groundwater level simulation model. The results show that the model can effectively predict the groundwater level variations. The root mean square error of the groundwater level in all observation wells is less than 1 meter, and the simulation time is greatly reduced compared to the numerical model. In the general single-well groundwater level model, the aquifer is assumed to be homogeneous and isotropic. Therefore, in the case of heterogeneous aquifers, this study uses the Simulated Annealing Algorithm in SPOTPY to optimize the equivalent hydraulic conductivity and storage coefficient for each pair of pumping versus observation wells. The simulation results show that the groundwater prediction based on the verified parameters can conform to the actual groundwater level change, and the root mean square error of the groundwater level prediction in observation wells is reduced to less than 0.85 meters.
The research results show that the proposed multi-well groundwater level simulation model can effectively and accurately simulate the groundwater level, and does not need to retrain the model due to the change in wells locations or well number. The model can apply to real-time operation problems and can also be a computing kernel for a complicated groundwater management model.