Estimation of the longitudinal dispersion coefficient via a fusion of optimized models

Determination of the longitudinal dispersion coefficient (LDC) is fundamental to the development of strategies for environmental management of river systems. This paper presents an integrated model for an estimation of the longitudinal dispersion coefficient by a fusion of optimized intelligent models (optimized neural network (ONN), optimized fuzzy inference system (OFIS), and optimized support vector regression (OSVR)) via committee machine (CM), with optimization done by the Bat-inspired algorithm (BA). The optimization eliminates the associated loss of accuracy of the intelligent models, which is a direct consequence of an improper adjustment of parameters (weights and biases in the neural network, membership's functions in the fuzzy inference system, and user-defined parameters in support vector regression). Data gathered from literature are employed to validate the proposed integrated model. A comparison between the optimized models and a committee machine, based on statistical parameters, shows that the committee machine model can attain high accuracy. Sensitivity analysis (SA) shows the contribution of each optimized model to the committee machine and ranks the contribution of the optimized models in ascending order as optimized neural network, optimized fuzzy inference system, and optimized support vector regression, each significantly correlated with the accuracy of longitudinal dispersion coefficient prediction.

Automated summary

This paper presents an integrated model for estimating the longitudinal dispersion coefficient by optimizing intelligent models and using a committee machine. The optimized models show high accuracy compared to the committee machine model, and sensitivity analysis ranks the contribution of each optimized model.