particle swarm algorithm

SWAT model is a suitable tool for simulating hydrological processes. This model requires many inputs that often cannot be measured directly and is considered one of the main sources of uncertainty in these models. The recalibration process can reduce the uncertainty in the model results by adjusting and adapting these inputs. The researches showed that calibrating a hydrological model by using the common automatic CV calibrating algorithms will not provide proper accuracy in the prediction of hydrological variables during the validation period, so PSO algorithm was used to calibrate the SWAT model. Since there is no mathematical and logical rule to determine the best combination of PSO algorithm parameters and these combinations are selected based on trial and error and among many different combinations, therefore trial and error based methods are very time-consuming and sometimes impossible. In this research, Taguchi method was used to determine the best combination of PSO algorithm parameters.

In this research, the ability to use the SWAT model to simulate monthly runoff in the Javanmardi Watershed, one of the main sub-basins of the Lordegan Watershed with an area of 380 square kilometers, was investigated. In this study, the PSO algorithm parameters, including the number of simulations (A), the number of repetitions (B), the speed calculation weight (C) and the movement parameter (D), were defined in four levels. Then, these parameters were designed and implemented according to the experiments in the L16 orthogonal array (using the Taguchi experiments design method). The performance scale used to evaluate the algorithms was RPD (Relative Percentage Deviation). Considering the variable nature of the response in this study, the S/N index "the lower the better" was used to analyze the Taguchi test results. The selection of arrays and calculations were done in Minitab 16 software.

In the sensitivity analysis stage, which was performed before the model recalibration, among the 28 parameters studied in this research, the model showed sensitivity to the changes of 22 parameters, and they were identified as variables influencing the simulation of runoff in Javanmardi Watershed. The results showed that the parameter of the runoff curve number (CN) is the most important factor and the parameters of soil apparent density in the wet state (SOL_BD) and average water usable by the plant (SOL_AWC) are among the most important factors controlling the flow rate in the study basin, respectively. Based on the results simulated by the PSO algorithm, it was found that the SWAT model has an acceptable accuracy for estimating the monthly runoff in the study area. So, in the recalibration phase, the r-factor and p-factor indices were 1.23 and 0.88, respectively, and the explanatory and Nash-Sutcliffe coefficients were 0.77 and 0.75, respectively. In the validation stage, the r-factor and p-factor indexes were 1.31 and 0.84, respectively and the explanatory and Nash-Sutcliffe coefficients were 0.72 and 0.73, respectively. In this study, the best combination resulting from the application of Taguchi method for the parameters of the number of simulations, the number of repetitions, the speed calculation weight and the appropriate parameters in the PSO algorithm were determined as 40, 100, 0.2 and 0.15 respectively (A4B4C4D3).

The results show that the SWAT model has an acceptable accuracy for estimating the monthly runoff in the Jawanmardi Watershed, and the PSO method is an effective algorithm in calibrating and determining the uncertainty of the model in this basin, and the use of the Taguchi test design method is a suitable way to determine the best combination of PSO algorithm parameters is for researchers who use this method to optimize the SWAT model.

Groundwater and water resource management play key roles in water resource sustainability in arid and semi-arid areas. Forecasting groundwater level is very important for water resource management and planning. In this study, an artificial neural network and a particle swarm algorithm based on artificial neural network models have been used to estimate groundwater level in the Ardebil plain. Water table level data for the 1972 -2011 period was used as our data in this study. Model inputs were water table level of various months. Results of both models were evaluated by root-mean-square error, the correlation coefficient and Nash-Sutcliffe coefficient. Results showed the performance of the particle swarm algorithm based on artificial neural network models to be superior. Root-mean-square error results for the particle swarm algorithm model in spring, summer, autumn and winter were 0.476, 0.507, 0.309, and 0.386 respectively. These results show that the hybrid structure of the network in training leads to increased accuracy. Thus, the particle swarm algorithm based on artificial neural network models can be used to estimate groundwater level in the Ardebil plain with acceptable accuracy.