Simulation of monthly river flow using improved support vector machine regression model using gray wolf optimization algorithm

Document Type : Research Paper

Authors

• Safoura Pirouzmehr ¹
• saeid shabanlou ²
• fariborz yosefvand ³
• Behrouz Yaghoubi ³
• ahmad rajabi ³
• mohammad ali izadbakhsh ³

¹ Ph.D. Candidate, Department of Water Engineering, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran

² Department of Water Engineering, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran.

³ Department of Water Engineering, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran

Abstract

Keywords

• Monthly river flow
• support vector regression
• gray wolf optimization algorithm
• water resources management

Main Subjects

• منابع آب

EXTENDED ABSTRACT

Introduction

Scientists are always trying to use accurate methods to measure monthly river flow and model its time series. Artificial intelligence-based methods are one of the best and most accurate tools for this work, among various artificial intelligence methods, SVR is one of the most useful methods for modeling monthly river flow. One of the most important parts of this model is the activator function. Different activation functions can be used to develop this model. Each activator function has one or more parameters. Determining the optimal value of these parameters is one of the most important parts of developing any SVR model. The aim of this study is to evaluate the efficiency of the gray wolf optimization algorithm for simulating the parameters of the SVR model's activation functions and, as a result, to improve the accuracy of the monthly river flow simulation using this model.

Methods and Materials

For this purpose, the monthly information of monthly river flow, temperature and rainfall was used for a 15-year period (1385 to 1400). The best input variables to the model were selected using the trial and error method. Therefore, variables Q_(t-1) R_t, T_t were used to develop all models. After determining the best variables for simulating the monthly flow of the river, all information was normalized between zero and one to avoid model bias. Then, 80% of all data were used to train SVR and GWO-SVR models, and other data were used to evaluate the models. In this study, linear, radial base, simgoid and polynomial activator functions were used to train the SVR model. In this way, the efficiency of each of the SVR and GWO-SVR models was evaluated with each of these activation functions. Finally, RMSE, R^2 and NSE indicators were used to evaluate different models and compare their results. According to the results, SVR and GWO-SVR models with polynomial activation function have the best performance and linear activation function have the worst performance in the training and validation stage. Also, more precisely, among the various models, the GWO-SVR model with the polynomial activation function has the best performance and the SVR model with the linear activation function has the worst performance. Based on the results of this study, the GWO algorithm is a more suitable tool than the trial and error method for simulating the monthly river flow using the SVR model.

Results and Discussion

Based on the results of this study, the SVR model with polynomial activation function has the best performance in the training and validation phase, and with the linear activation function, it has the worst performance in the training and validation phase. Next, the GWO algorithm was used to determine the parameters of the activator functions. Based on the final results, the SVR model performs better with the GWO algorithm. Therefore, to simulate the monthly flow of river water using this model, it is better to use the GWO algorithm instead of the trial and error method.

Conclusion

After developing the SVR model with different activator functions, the results of these models were compared with each other. Based on the results of that study, the GWO-SVR(LKF) model performs better than the SVR(LKF) model. Also, the GWO-SVR(RBF) model has a better performance than the SVR(RBF) model. The GWO-SVR(SKF) model also performs better than the SVR(SKF) model. Like other models, the GWO-SVR(PKF) model also performs better than the SVR(PKF) model. Also, based on the comparison of the error indices of different models, the GWO-SVR(PKF) model performs better than all models in the training and validation stage, and the SVR(LKF) model has the worst performance. After the end of the training and testing process of different models, their sensitivity to different variables was evaluated. Based on the results of the sensitivity analysis, all models have the most sensitivity to the variable Q_(t-1) and the least sensitivity to the variable E.

Author Contributions:

For research articles with several authors, a short paragraph specifying their individual contributions must be provided. The following statements should be used “Conceptualization, Safora Pirouzmehr and Saeid Shabanlou; methodology, Safora Pirouzmehr and Saeid Shabanlou; software, Safora Pirouzmehr and Saeid Shabanlou and Fariborz yosefvand and Behrouz Yaghoubi and Ahmad Rajabi and Mohammad ali Izadbakhsh; validation, Safora Pirouzmehr and Saeid Shabanlou; formal analysis, Safora Pirouzmehr and Saeid Shabanlou; investigation, Safora Pirouzmehr and Saeid Shabanlou; resources, Safora Pirouzmehr and Saeid Shabanlou; data curation, Safora Pirouzmehr and Saeid Shabanlou; writing—original draft preparation, Safora Pirouzmehr and Saeid Shabanlou; writing—review and editing, Safora Pirouzmehr and Saeid Shabanlou and Fariborz yosefvand and Behrouz Yaghoubi and Ahmad Rajabi and Mohammad ali Izadbakhsh; visualization, Saeid Shabanlou; supervision, Saeid Shabanlou; project administration, Saeid Shabanlou; All authors have read and agreed to the published version of the manuscript.” Please turn to the CRediT taxonomy for the term explanation. Authorship must be limited to those who have contributed substantially to the work re-ported.

Data Availability Statement:

Data available on request from the authors

Acknowledgements

The authors would like to thank all participants of the present study.

Ethical considerations

The authors avoided data fabrication, falsification, plagiarism, and misconduct.

Conflict of interest:

The author declares no conflict of interest