Precipitation forecasting using multilayer neuralNetwork and support vector machine optimization based on flow regime algorithm taking intoAccount uncertainties of soft computing models
Drought, climate change, and demand make precipitation forecast a very important issuein water resource management. The present study aims to develop a forecasting model for monthlyprecipitation in the basin of the province of East Azarbaijan in Iran over a ten-year period using themultilayer percep...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
2020
|
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Drought, climate change, and demand make precipitation forecast a very important issuein water resource management. The present study aims to develop a forecasting model for monthlyprecipitation in the basin of the province of East Azarbaijan in Iran over a ten-year period using themultilayer perceptron neural network (MLP) and support vector regression (SVR) models. In thisstudy, the flow regime optimization algorithm (FRA) was applied to optimize the multilayer neuralnetwork and support vector machine. The flow regime optimization algorithm not only identifies theparameters of the SVR and MLP models but also replaces the training algorithms. The decision treemodel (M5T) was also used to forecast precipitation and compare it with the results of hybrid models.Principal component analysis (PCA) was used to identify effective indicators for precipitation forecast.In the first scenario, the input data include temperature data with a delay of one to twelve months,the second scenario includes precipitation data with a delay of one to twelve months, and the thirdscenario includes precipitation and temperature data with a delay of one to three months. The meanabsolute error (MAE) and Nash-Sutcliffe error (NSE) indices were used to evaluate the performanceof the models. The results showed that the proposed MLP-FRA outperformed all the other examinedmodels. Regarding the uncertainties of the models, it was also shown that the MLP-FRA modelhad a lower uncertainty band width than other models, and a higher percentage of the data will fallwithin the range of the confidence band. As the selected scenario, Scenario 3 had a better performance.Finally, monthly precipitation maps were generated based on the MLP-FRA model and Scenario3 using the weighted interpolation method, which showed significant precipitation in spring andwinter and a low level of precipitation in summer. The results of the present study showed thatMLP-FRA has high capability to predict hydrological variables and can be used in future research. © 2019 by the authors. Licensee MDPI, Basel, Switzerland. |
|---|