Comparison of wavelet neural network models, support vector machine and gene expression programming in estimating the amount of oxygen dissolved in rivers

Water-soluble oxygen is one of the most effective parameters in determining the water quality of rivers and its control in the rivers is one of the most important factors for the development of water resources in each region. For this reason, in this study, the performance of wavelet neural network models, support vector machines and gene expression scheduling was investigated to estimate the oxygen dissolved in the water of the Cumberland River in Tennessee. For this purpose, the monthly time series DO of the Cumberland River index were simulated using flow and temperature flow parameters over a 10-year period (2006-2016). The correlation coefficient, root mean square error and mean absolute error value were used for evaluation and performance of the models. The results showed that hybrid structures in all three models offer better performance than other structures. Also, the results of the evaluation criteria showed that the wavelet neural network model with the highest correlation coefficient (0.960), the lowest root mean square error (0.668), and the lowest mean error of error (0.519) in the verification stage were among the applied models. In total, the results showed that in terms of the high ability of wavelet neural network and the elimination of time series noise in the estimation of river water quality parameters, this model could be a suitable and fast way to manage the quality of water resources and ensure the results of quality monitoring and cost reduction.