نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشآموخته کارشناسی ارشد علوم و مهندسی ابخیزداری، گروه مهندسی ابخیزداری، دانشکده منابع طبیعی، دانشگاه تربیت مدرس، نور، ایران
2 دانشیار گروه علوم و مهندسی ابخیزداری، دانشکده منابع طبیعی، دانشگاه تربیت مدرس ، نور، ایران
چکیده
مقدمه
مدلسازی سطح آب زیرزمینی بهدلیل ماهیت پیچیده، غیرخطی و وابسته به عوامل مکانی‑زمانی، همواره چالشی اساسی در مدیریت منابع آب زیرزمینی بوده است. در این پژوهش، هدف اصلی ارائه و ارزیابی یک رویکرد مدلسازی توزیعی مبتنی بر یادگیری عمیق برای پیشبینی همزمان تراز آب زیرزمینی در مجموعه چاههای پیزومتری آبخوان کوچصفهان در آبخیز سفیدرود است. در این چارچوب، هر چاه بهعنوان یک گره در شبکه ترسیمی در نظر گرفته شد و مدل، تراز آب زیرزمینی هر چاه را در گام زمانی ماهانه بعدی (t+1) پیشبینی میکند. با بهرهگیری از شبکههای عصبی ترسیمی ترکیبشده با ساختارهای پیچشی و حافظه کوتاهمدت‑بلندمدت، تلاش میشود وابستگیهای مکانی و زمانی رفتار آبخوان شناسایی شده و دقت پیشبینی نسبت به روشهای متداول افزایش یابد.
مواد و روشها
در این پژوهش، چهار ساختار یادگیری عمیق مبتنی بر گراف شامل شبکه عصبی ترسیمی (GNN)، شبکه توجه گرافی (GAT)، شبکه پیچشی گرافی (GCN) و مدل ترکیبی GCN‑LSTM بهکار رفت. منظور از GNN معماری پایه Graph Convolutional Network (GCN-based GNN) مبتنی بر عملگر کانولوشن گرافی استاندارد است. هر چاه پیزومتری بهعنوان یک گره در نظر گرفته شد و ماتریس مجاورت با وزندهی معکوس فاصله تشکیل شد. دادههای ورودی شامل متغیرهای هیدرواقلیمی، اطلاعات زمینشناسی، ویژگیهای پمپاژ و فواصل مکانی از عوامل هیدروژئولوژیک بودند. ارزیابی با شاخصهای R²، RMSE، MAE، NSE و KGE انجام شد. در مدل پیشنهادی، در هر گام زمانی ابتدا لایههای GCN بر روی گراف مکانی چاهها اعمال شدند تا ویژگیهای فضایی و وابستگیهای هیدرولیکی استخراج شوند. سپس بردارهای پنهان حاصل به شبکه LSTM منتقل شدند تا وابستگیهای زمانی و پویایی سطح آب زیرزمینی مدلسازی شود.
نتایج و بحث
نتایج نشان داد که مدل GNN اگرچه توانست روندهای کلی تغییرات سطح آب زیرزمینی را تا حدودی بازسازی کند، اما در نمایش رفتارهای حدی و وابستگیهای زمانی دقت کافی نداشت. مدل GAT نیز نسبت به GNN بهبود نسبی نشان داد، اما همچنان در استخراج الگوهای عمیق زمانی محدود بود. در مقابل، مدل GCN عملکرد بهتری در شناسایی وابستگیهای مکانی از خود نشان داد و موجب بهبود قابل توجه شاخصهای ارزیابی شد. بهترین عملکرد مربوط به مدل GCN-LSTM بود که توانست همزمان ویژگیهای مکانی و زمانی را بهصورت مؤثرتری بازنمایی کند و بیشترین همپوشانی را با دادههای مشاهدهای داشته باشد. بر اساس نتایج گزارششده، این مدل به مقادیر 0.336=/R2=0.896،RMSE و 0.269=MAE دست یافت که نشاندهنده دقت بالای آن در پیشبینی سطح آب زیرزمینی است.
نتیجهگیری
یافتههای این پژوهش نشان میدهد که بهرهگیری از معماریهای ترکیبی مبتنی بر گراف و توالی، بهویژه مدل GCN-LSTM, میتواند ابزار بسیار مؤثری برای مدلسازی رفتار هیدرودینامیکی آبخوانهای پیچیده باشد. این مدل توانست با دقت بالا رفتار سطح آب زیرزمینی را پیشبینی کرده و برتری خود را نسبت به مدلهای صرفاً گرافی نشان دهد. در نتیجه، استفاده از این رویکرد برای پیشبینی عملیاتی سطح آب زیرزمینی و مدیریت پایدار منابع آب توصیه میشود. این روش میتواند بهعنوان یک چارچوب نوین در تحلیل و مدیریت آبخوانها، به تصمیمگیری بهتر در حوزه منابع آب کمک کند.
کلیدواژهها
عنوان مقاله [English]
Distributed groundwater level modeling using graph, convolutional, and long short-term memory neural networks
نویسندگان [English]
- Negin Rashidi 1
- Vahid Moosavi 2
1 M.Sc. Graduate in Watershed Science and Engineering, Department of Watershed Engineering, Faculty of Natural Resources, Tarbiat Modares University, Noor, Iran
2 Associate Professor, Department of Watershed Science and Engineering, Faculty of Natural Resources, Tarbiat Modares University, Noor, Iran
چکیده [English]
Introduction
Groundwater level modeling has always been a major challenge in groundwater resource management due to its complex, nonlinear, and spatiotemporally dependent nature. The main objective of this study was to develop and evaluate a distributed deep learning-based modeling approach for the simultaneous prediction of groundwater levels in a network of piezometric wells within the Kouchesfahan aquifer located in the Sefidroud watershed. In this framework, each well was considered as a node in a graph network, and the model predicted groundwater levels for each well at the next monthly time step (t+1). By integrating graph neural networks with convolutional structures and long short-term memory architectures, the study aimed to identify the spatial and temporal dependencies governing aquifer behavior and improve prediction accuracy compared to conventional methods.
Materials and methods
In this study, four graph-based deep learning architectures, including Graph Neural Network (GNN), Graph Attention Network (GAT), Graph Convolutional Network (GCN), and the hybrid GCN–LSTM model, were employed. Here, GNN refers to the baseline GCN-based graph neural architecture using the standard graph convolution operator. Each piezometric well was represented as a graph node, and the adjacency matrix was constructed using inverse-distance weighting. Input data included hydroclimatic variables, geological information, pumping characteristics, and spatial distances related to hydrogeological factors. Model performance was evaluated using R², RMSE, MAE, NSE, and KGE indices. In the proposed model, GCN layers were first applied to the spatial graph of wells to extract spatial features and hydraulic dependencies. The resulting hidden representations were then transferred to the LSTM network to model temporal dependencies and groundwater level dynamics.
Results and discussion
The results indicated that while the GNN model could partially reconstruct the general trends of groundwater level changes, it lacked accuracy in representing extreme behaviors and temporal dependencies. The GAT model showed a slight improvement over GNN but remained limited in extracting deep temporal patterns. In contrast, the GCN model demonstrated better performance in identifying spatial dependencies, leading to a significant improvement in evaluation metrics. The best performance was achieved by the GCN-LSTM model, which effectively represented both spatial and temporal features simultaneously, showing the highest overlap with observed data. This model reached an R2 of 0.896, RMSE of 0.336, and MAE of 0.269, indicating its high accuracy in predicting groundwater levels.
Conclusion
Hybrid graph-sequential architectures, especially GCN-LSTM, are highly effective for modeling complex aquifer hydrodynamics. This model predicted groundwater levels with high accuracy and outperformed purely graph-based models. Therefore, it is recommended for operational groundwater prediction and sustainable water management, serving as a novel framework to support better decision-making in water resources.
کلیدواژهها [English]
- Adjacency Matrix
- Aquifer
- Artificial Intelligence
- Prediction
- Spatio-temporal Deep Learning