نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی آب، دانشکده مهندسی آب و خاک، دانشگاه علومکشاورزی و منابع طبیعی گرگان، گرگان، ایران
2 گروه مهندسی آب، دانشکده مهندسی آب و خاک، دانشگاه علوم کشاورزی و منابع طبیعی گرگان، گرگان، ایران
3 گروه مهندسی آب، دانشگاه زابل، زابل، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The present study was conducted with the aim of evaluating the efficiency of the multilayer perceptron artificial neural networks in estimating the pump rotation of drinking water wells located in Bandar Turkman using the key parameters of pump discharge, pump installation depth, and groundwater level. This study involves an integrated approach of artificial neural networks modeling and its sensitivity analysis as well as Monte-Carlo simulation. The neural networks results showed that this model has an acceptable performance for predicting the pump rotation speed of wells. Furthermore, it was reveled that this model performance is highly affected by seasonal changes. The best performance of this model was achieved in May with R2=0.98 and RMSE=15.15 and in March with R2=0.99 and RMSE=15.13 for training and testing phases, respectively. The sensitivity analysis of input variables indicated that the groundwater level with 48% importance has the greatest effect on the pump rotation speed, followed by the pump discharge and pump installation depth with 37% and 15%, respectively. The reliability index analysis (β) showed that the seasonal changes have high influence on studied pumping system so that in winter season, it has excellent reliability (β=1.7-2.2) and through summer season, it has negative reliability which indicates the possibility of system failure or excessive energy consumption by pumps. The proposed approach in this study provides a practical framework for water and wastewater managers to optimize pumping operations, reduce energy costs, and enhance water management in arid regions.
کلیدواژهها [English]
EXTENDED ABSTRACT
In recent years, optimal management of water resources has become one of the main challenges, particularly in arid and semi-arid regions where droughts and water scarcity pose significant challenges for sustainable development and public health. This issue is of increasing importance due to many influential factors such as climate change, increased human interference in watersheds, population growth and the consequent increase in the need for drinking water. This reality is also true on a smaller scale, as the pumping systems in drinking water wells, despite their important role in water supply infrastructure, are currently facing numerous operational challenges in terms of efficiency, reliability, and energy consumption. These stations face problems such as reduced efficiency, increased energy consumption and operating costs. Currently, many of these systems are managed traditionally without sufficient attention to the variables affecting the efficiency and stability of pump performance. While pumping water out of the wells to the surface distribution systems needs an optimal performance of pumping system, the lack of accurate approaches to analyzing and evaluating the efficiency of these systems has reduced their reliability and increased the likelihood of devices failure and energy waste. Recent publications have shown that optimizing these systems could improve their supply reliability in addition to reducing their operating costs by up to 30%. Approximately 10%-20% of global electrical energy is consumed by water pumping systems, with significant potential for efficiency improvement. Furthermore, this percentage can even increase to 25-30% of water energy use in developing countries. Existing studies gave demonstrated that optimization of these systems presents compelling opportunities for energy conservation and operational cost reduction while ensuring reliable water supply for communities. Comprehensive analytical approaches (e.g. probabilistic methods, artificial intelligence techniques, etc) are required to deal with the complex interaction between different operating parameters in water pumping systems. Specifically, in regions with significant seasonal variations in water demand and availability, traditional methods of pump system analysis fail to capture the dynamic nature of these interactions. The present research builds on these advances by combining probabilistic methods with artificial intelligence techniques.
This study focuses on 17 water wells in the Bandar Torkaman region, a coastal city in Golestan Province, Iran. These wells serve as the primary source of drinking water for the urban and rural populations of Bandar Torkaman and its neighbor, Gomishan. In water wells, five parameters are measured, including well depth, pump installation depth, pump discharge, pump speed, and water level. Initially, four probability distributions are fitted to these data (Normal, Lognormal, Gamma, and Weibull), using model selection criteria to determine the best-fitting distribution. In this study, there are two parallel analytical approaches including convolutional neural networks (CNN) modeling and reliability analysis. CNN models simulate the using output variable of pumping system (pump rotation) based on three more effective input variables (pump discharge, pump installation depth, and groundwater level) through an optimal multi-layer architecture. While the reliability analysis branch calculates the reliability index through Monte Carlo simulation, resulting in a seasonal performance evaluation. Ultimately, both analytical branches converge to provide a comprehensive evaluation of pump performance and reliability. This integrated approach combines deep learning and probabilistic reliability assessment to provide a robust framework for analyzing and optimizing drinking water well systems.
MATLAB environment was used to perform a comprehensive statistical analysis of well data on a monthy basis. The Weibull probability distribution showed the best fit for water levels in 10 out of 12 months, while the Lognormal distribution dominated flow rate patterns in 7 months. Operational data from 17 wells showed distinct patterns across the three key parameters of pump rotation, groundwater level, and pump flow discharge. During early months (March-April), Gamma distribution provided the best. During May-June, the distribution shifted to Lognormal while maintaining a high goodness of fit as temperatures increased. In the middle of the year (July-September), the model selection criteria (AIC and BIC) have undergone significant changes, suggesting that operational variability has increased. There was a remarkable consistency in water level measurements for 10 out of 12 months, with Weibull distribution providing the best fit. The AIC and BIC values of these fits (385-860) were higher than those of pump rotation distributions, suggesting a more complex behavior of water levels. Normal distributions provided better fits in September and March, possibly reflecting specific hydrological conditions during those months. Flow discharge analysis revealed that Lognormal distribution was the dominant pattern, which was optimal for 7 months of the year. AIC and BIC values for these fits were lower (80-1010) than those for pump rotation and water level, which suggests simpler underlying patterns. As a result of seasonal variations in water consumption patterns, Gamma distributions held the best fit in March and October, while Normal distributions held the best fit in July. A significant correlation was observed between the distributions of pump rotation and flow discharge during certain months. For example, both showed Gamma distributions in March and Lognormal distributions in May-June, suggesting a strong relationship between pump performance and water flow. Among the three parameters, pump rotation demonstrated the greatest variability in optimal distribution type, indicating greater sensitivity to operational and environmental factors. Flow discharges exhibited moderate stability with notable seasonal variations, while water level exhibited the most stability. Based on four years of operational data, these findings provide valuable insights for managing and optimizing drinking water wells. They highlight the complex interactions between pump performance, water levels, and flow discharges in drinking water wells, and they suggest seasonal adjustments to operational strategies.
The CNN model's performance also exhibited significant seasonal variations. Optimal accuracy was achieved in February, with a substantial decline in performance observed during autumn (September and October). The model's performance fluctuated during spring and summer seasons, with the best accuracy in June and July. Pump rotation range analysis revealed distinct seasonal patterns, with wider ranges in summer and early autumn indicating increased operational variability. A narrower range in late autumn and winter suggested more stable well conditions or reduced activity. To improve the model's overall performance, seasonal adjustments are crucial. By considering the specific characteristics of each season, the model's predictive capabilities can be significantly enhanced. The R² metric further supports these findings, with the highest value (0.95) in February and the lowest (0.44) in September and October. This highlights the importance of addressing seasonal factors in developing robust machine learning models for predicting pump rotation.
Monte Carlo reliability index (beta) analysis also shows significant variations in pump performance and energy consumption throughout the year. In winter (January, February, and March) and early spring (April), the beta index shows high values (between 1.72 and 2.19), indicating optimal and reliable pump performance. As a result of favorable weather conditions and balanced demand, energy consumption may be more efficient. A sharp decline in the beta index occurs during summer months (June and July), with negative values in July and August (-0.05 and -0.15), indicating serious performance challenges for pumps and likely increased energy consumption. A gradual improvement is observed in the index from late summer to autumn, suggesting the system gradually returns to a more optimal state. During the warmer months, specific strategies are required to optimize energy consumption and pump performance, which emphasizes the importance of seasonal management of the pumping system.
This comprehensive study of drinking water well systems in Bandar Torkaman, demonstrated the effectiveness of combining CNN modeling with traditional statistical analysis to optimize pump performance and energy efficiency. Based on the results of the study, significant seasonal variations have been observed in system behavior, ranging from peak accuracy in January to reduced reliability in July. The statistical analysis also revealed distinct distribution patterns for key operational parameters (Weibull distribution of water levels and Lognormal distribution of flow discharges). In light of these findings, it is important to implement seasonal-adjusted operational strategies, with optimal flow rates of 7.5-8.5 m3/h and groundwater levels of 45-65 meters, whereas pump rotation requirements vary significantly between winter (1100-1200 RPM) and summer (2350-3040 RPM). As a result of the research, a foundation has been laid for enhancing drinking water well operations through adaptive management strategies, with future improvements focusing on incorporating environmental parameters, developing season-specific models, and implementing hybrid optimization approaches to address the complex interactions between operational efficiency and seasonal variability.
Conceptualization, J.Piri and A.Zahiri; methodology, J.Piri; software, A.Ghooli and J.Piri; validation, A.Zahiri; formal analysis, A.Zahiri, and J.Piri; investigation, A.Ghooli; resources, A.Ghooli; data curation, A.Ghooli and J.Piri; writing-original draft preparation, A.Zahiri; writing-review and editing, A.Ghooli, A.Zahiri, and J.Piri; visualization, A.Zahiri and J.Piri; supervision, A.Zahiri and J.Piri; project administration, A.Zahiri and J.Piri; funding acquisition, A.Zahiri. All authors have read and agreed to the published version of the manuscript.
Data available on request from the authors.
The authors avoided data fabrication, falsification, plagiarism, and misconduct.
The author declares no conflict of interest.
)