نوع مقاله : مقاله پژوهشی

نویسندگان

• زینب رحیمی آتانی ¹
• محمد بی جن خان ²
• هادی رمضانی اعتدالی ³
• حامد مازندرانی زاده ⁴
• سارا فکوری ⁵
• سکینه کوهی ⁶
• مینا قلی زاده ⁷

¹ دانشجوی دکتری علوم و مهندسی آب/ گرایش آبیاری و زهکشی، دانشگاه بین المللی امام خمینی، قزوین، ایران

² دانشیار،گروه علوم و مهندسی مهندسی آب، دانشگاه بین‌المللی امام خمینی(ره)، قزوین، ایران

³ استاد، گروه علوم و مهندسی مهندسی آب، دانشگاه بین‌المللی امام خمینی(ره)، قزوین، ایران

⁴ دانشیار، گروه علوم و مهندسی مهندسی آب، دانشگاه بین‌المللی امام خمینی(ره)، قزوین، ایران

⁵ دانشجوی دکتری منابع آب، گروه علوم و مهندسی آب، دانشگاه بین المللی امام خمینی(ره)، قزوین، ایران

⁶ دکتری منابع آب، گروه علوم و مهندسی مهندسی آب، دانشگاه بین‌المللی امام خمینی(ره)، قزوین، ایران

⁷ کارمند شرکت آب منطقه ای استان قزوین

چکیده

در سال های اخیر، مدیریت برداشت آب زیرزمینی به عنوان یکی از چالش های اساسی در مناطق خشک و نیمه خشک مطرح بوده است. با توجه به هزینه بر بودن نصب و نگهداری کنتورهای هوشمند آب، استفاده از داده های در دسترس کنتورهای برق میتواند به عنوان یک رویکرد جایگزین برای برآورد غیرمستقیم دبی چاه های کشاورزی مورد توجه قرار گیرد. در این پژوهش، امکان تخمین دبی چاه های کشاورزی بر بررسی شده است. دادههای مورد استفاده (ANN) اساس داده های کنتور برق و ویژگی های فیزیکی چاه با استفاده از شبکه عصبی مصنوعی شامل اطلاعات توان مصرفی ، مشخصات هندسی چاه (عمق و قطر)، موقعیت جغرافیایی و شرایط بهره برداری از ۳۵ حلقه چاه در منطقه مطالعه میباشد.  مدل توسعه یافته از نوع پرسپترون چندلایه (MLP) بوده و با استفاده از الگوریتم لونبرگ–مارکوارت آموزش داده شد. عملکرد مدل بر اساس معیارهای آماری مختلف از جمله ضریب تعیین(R²)،میانگین مربعات خطا (MSE) و درصد خطای نسبی ارزیابی گردید. نتایج نشان داد که مدل ANN توانایی مناسبی در برآورد دبی چاه ها بر اساس داده های کنتور برق دارد. نوآوری اصلی این پژوهش در ترکیب داده های توان مصرفی کنتور برق با ویژگی های فیزیکی و مکانی چاه، ارزیابی سناریوهای مختلف ترکیب متغیرهای ورودی و ارائه یک روش کم هزینه و قابل اجرا برای برآورد غیرمستقیم دبی چاه ها بدون نیاز به تجهیزات اندازه گیری مستقیم است. این رویکرد می تواند به عنوان ابزاری کاربردی در پایش برداشت آب زیرزمینی و پشتیبانی از تصمیمگیری های مدیریتی در مناطق دارای محدودیت تجهیزات اندازه گیری مورد استفاده قرار گیرد.

کلیدواژه‌ها

• آبدهی چاه '
• 'شبکه عصبی مصنوعی '
• 'کنتور هوشمند آب '
• 'تکنیک‌های یادگیری ماشین

موضوعات

• منابع و مصارف آب

عنوان مقاله [English]

Development of an Artificial Neural Network Model for Estimating Agricultural Well Discharge Using Electricity Meter Data (Case Study: Qazvin Plain)

نویسندگان [English]

• Zeynab Rahimi Atani ¹
• Mohammad Bijnkhan ²
• Hadi Ramezani Etedali ³
• Hamed Mazandaranizadeh ⁴
• Sara Fakuri ⁵
• Sakineh Kohi ⁶
• Mina Gholizadeh ⁷

¹ PhD student in Water Science and Engineering/Irrigation and Drainage, Imam Khomeini International University, Qazvin, Iran

² Associate Professor, Department of Water Engineering Science and Engineering, Imam Khomeini International University, Qazvin, Iran

³ Professor, Department of Water Engineering Science and Engineering, Imam Khomeini International University, Qazvin, Iran

⁴ Associate Professor, Department of Water Engineering Science and Engineering, Imam Khomeini International University, Qazvin, Iran

⁵ PhD Student in Water Resources, Department of Water Engineering Science and Engineering, Imam Khomeini International University, Qazvin, Iran

⁶ PhD in Water Resources, Department of Water Engineering Science and Engineering, Imam Khomeini International University, Qazvin, Iran

⁷ Employee of the Regional Water Company of Qazvin Province

چکیده [English]

Extended Abstract
Introduction
Groundwater resources play a vital role in sustaining agricultural activities in arid and semi-arid regions. In Iran, over 75% of irrigation water is supplied from aquifers (Yazdanpanah et al., 2019), making groundwater the primary source of agricultural production. However, continuous groundwater level decline due to overexploitation, reduced natural recharge, and climate change has made accurate monitoring of groundwater abstraction an essential requirement for
sustainable water resource management. Direct measurement of well discharge faces numerous challenges, including high equipment costs, operational difficulties, spatial distribution of wells, and the presence of unauthorized wells. While smart water meters have been introduced as a solution for abstraction control, their widespread implementation is constrained by high installation and maintenance costs. In contrast, most agricultural wells are already equipped with electricity meters that continuously record power consumption data, presenting a cost-effective alternative for indirect discharge estimation (Wang et al., 2020; Alam et al., 2023). This study investigates the feasibility of estimating agricultural well discharge using electricity meter data and physical well characteristics through an Artificial Neural Network (ANN). The novelty of this research lies in developing a cost-effective approach that eliminates the need for expensive flow measurement equipment, relying solely on accessible electricity consumption data.
Methodology
Study Area and Data Collection
The research was conducted in Qazvin Province, central Iran, characterized by diverse climatic conditions and intensive agricultural activities. Field data were collected from 70 agricultural wells, including geographical coordinates, well depth, pipe diameter, electricity consumption parameters, and measured discharge values. Due to incomplete power records, 35 wells with reliable data were selected for modeling. Discharge measurements obtained using an ultrasonic flowmeter served as reference data.
Artificial Neural Network Modeling
A multilayer perceptron (MLP) neural network was implemented in the MATLAB environment using the Levenberg–Marquardt training algorithm. Eight scenarios were defined to evaluate the sensitivity of model performance to various combinations of input parameters, including power consumption, well depth, pipe diameter, geographical coordinates, and aeration condition. The dataset was randomly divided into training (70%), validation (15%), and testing (15%) subsets. Model performance was evaluated using the coefficient of determination (R²), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE). Different numbers of neurons (5 to 10) were tested to identify the optimal network architecture.
 
Results and Discussion
Model Performance and Scenario Selection
The ANN successfully estimated agricultural well discharge using electricity meter data and well characteristics. Among the evaluated scenarios, Scenario 7 comprising power consumption, well depth, pipe diameter, and geographical coordinates—demonstrated the most stable and reliable performance across training, validation, and testing stages. Although Scenario 8, which included aeration condition, produced slightly lower error during training, it required operational parameters difficult to obtain in practice. Consequently, Scenario 7 was selected as the optimal model. The optimal ANN architecture consisted of one hidden layer with 10 neurons, achieving the highest coefficient of determination (R²=0/9204) and lowest prediction error (MAPE=13/96%). The model achieved average errors of approximately 15%, 13%, and 14% during training, validation, and testing phases, respectively.
Analysis of Input Variables
Power consumption was identified as the most influential variable, consistent with the hydraulic power equation (P = ρ·g·Q·H/η), which establishes a direct physical relationship between power and discharge. Well depth affects pumping head (H), while pipe diameter influences friction losses. Geographical coordinates were incorporated into the model to account for spatial heterogeneity in hydrogeological conditions across the study area, as variations in aquifer thickness, sediment texture, hydraulic conductivity, and water table depth are not uniformly distributed throughout the Qazvin Plain.
Comparison with Previous Studies
The results align with findings by Alam et al. (2023), who achieved R²≈0/92 in alluvial aquifers, and Wang et al. (2020), who reported errors below 20% for energy-to-water conversion. The model performance (R²=0/9204) is comparable to Salimi et al. (2023), who obtained R²=0/825 using Random Forest, and Salamat et al. (2023), who achieved 0/945 accuracy using SVM-PSO. Although the dataset (35 wells) is limited compared to previous studies (e.g., 236–359 samples), the model demonstrates competitive performance, confirming the feasibility of the proposed approach.
Limitations and Sources of Error
The main limitations include: (1) limited sample size (35 wells), increasing the risk of overfitting, mitigated by early stopping and simple network architecture; (2) lack of comparison with other machine learning algorithms (e.g., Random Forest, SVM, XGBoost); (3) absence of hydrogeological variables such as water table level, dynamic head, and pump specifications; (4) limited temporal sampling (August–November 2023), not capturing seasonal variations; and (5) regional specificity, requiring local calibration for application in other areas. Key sources of error include pump efficiency variations, groundwater level fluctuations, voltage instability, aeration, and differences in maintenance conditions.
Conclusions
This study demonstrates that combining electricity meter data with artificial neural networks provides a practical, cost-effective approach for estimating agricultural well discharge. The proposed method eliminates the need for expensive smart water meters while maintaining acceptable accuracy (R²=0/9204, MAPE=13/96%). Power consumption, well depth, pipe diameter, and geographical location were identified as key predictors. The approach can support groundwater monitoring programs, improve water resource management, and assist policymakers in implementing sustainable abstraction strategies. Future research should expand the dataset, incorporate temporal variations, compare multiple algorithms (Random Forest, SVM, XGBoost), and include hydrogeological variables (water table, dynamic head, pump characteristics) to enhance model accuracy and generalizability.
Conflict of Interest
The authors declare that they have no conflict of interest regarding the preparation and publication of the materials and findings presented in this study.

Funding
The authors received no financial support for the research, authorship, or publication of this article.

Data Availability Statement
The datasets generated and analyzed during this study are not publicly available due to applicable restrictions, but are available from the corresponding author upon reasonable request.

Author Contributions
Conceptualization: Z. Rahimi Atani, M. Bijankhan, H. Ramazani; Methodology: Z. Rahimi Atani, H. Mazandaranizadeh; Software: Z. Rahimi Atani, H. Mazandaranizadeh; Validation: M. Bijankhan, H. Ramazani; Formal analysis: Z. Rahimi Atani, M. Bijankhan; Investigation: Z. Rahimi Atani, M. Gholizadeh, S. Koohi, Fakouri; Resources: Z. Rahimi Atani; Data curation: M. Bijankhan, H. Ramazani, M. Gholizadeh, S. Koohi, Fakouri; Writing—original draft: Z. Rahimi Atani; Writing—review and editing: M. Bijankhan; Visualization: Z. Rahimi Atani; Supervision: M. Bijankhan, H. Ramazani; Project administration: M. Bijankhan. All authors have read and agreed to the published version of the manuscript.

Acknowledgement
The authors would like to express their sincere gratitude to the Qazvin Regional Water Company for providing access to well data and facilitating field measurements. This research was supported by the Imam Khomeini International University (IKIU), Qazvin, Iran. The authors also acknowledge the cooperation of local farmers and agricultural well owners who participated in this study. This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
 
 
 

کلیدواژه‌ها [English]

• Well discharge'،'Artificial neural network
• '،'Smart water meter '،'Machine learning