نوع مقاله : پژوهشی
نویسندگان
1 دانشگاه یزد
2 گروه آب، دانشکده مهندسی عمران، دانشگاه یزد، یزد، ایران
3 استاد گروه مرتع و آبخیزداری دانشگاه یزد
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
In this study, the efficacy of several AI models, including the fuzzy-neural adaptive inference system (ANFIS), support vector machine (SVM), artificial neural network (ANN), as well as hybrid models integrating ANN with optimization algorithms such as whale optimization algorithm (ANN-WOA) and particle swarm optimization algorithm (ANN-PSO), has been evaluated for predicting the daily flow discharge of the Poldokhtar station, situated at the outlet of the Kashkan basin. The research aims to assess the performance of different AI models and optimization algorithms in daily flow discharge prediction, while also identifying optimal input parameters. By comparing the performance of these models, researchers can gain insights into which AI approaches are most effective for river flow prediction tasks. Additionally, determining the optimal input parameters can further enhance the accuracy and reliability of flow discharge forecasts, contributing to improved water resource management and decision-making processes.The use of ACF and PACF analyses revealed notable correlations in the context of daily flow and precipitation patterns. Specifically, the daily flow exhibited its strongest correlations with lag times of one day (Q-1), two days (Q-2), and three days (Q-3). Additionally, precipitation showed a significant correlation when lagged by one day (P-1). The results showed that all models have an acceptable ability in predicting the flow. The findings reveal that across all models, variables such as flow rate with one (Q-1), two (Q-2), and three (Q-3) days delay, along with precipitation with a one-day delay (P-1), exhibited the strongest correlation with daily flow rate in the Kashkan watershed. Overall, the models demonstrated acceptable accuracy in flow rate modeling. Among them, the ANN-WOA model stood out with the highest accuracy, boasting an R2 value of 0.896, NE of 0.803, and the lowest RMSE of 0.0186. Following closely, the SVM model, employing a radial base kernel function structure with parameter values of C=4, γ=1, and ε=0.001, showcased commendable performance. It yielded an explanation coefficient of 0.895, Nash Sutcliffe coefficient of 0.801, and an error value of 0.0187. The ANN-PSO and ANN models secured third and fourth positions, respectively. The integration of meta-engineering optimization algorithms notably enhanced the ANN model's accuracy, making it viable for network traininging. Furthermore, the exploration of various ANFIS structures revealed that triangular and Gaussian functions generally exhibited superior capabilities in modeling flow rate within the study area. However, despite its competence, the ANFIS model yielded slightly higher errors compared to other models, with an R2 of 0.88, NE of 0.76, and RMSE of 0.023.
کلیدواژهها [English]
)