An empirical-based rainfall-runoff modelling using optimization technique

This study proposes a new hybrid biogeography-based optimization (BBO) technique to achieve a better balance between exploitation and exploration sides of BBO. The proposed hybrid metaheuristic algorithm, namely HBBPSGWO, enhances the exploration ability of BBO by combining it with the exploration side of particle swarm optimization (PSO) and grey wolf optimization (GWO) algorithms. The proposed hybrid approach is integrated with two classical machine learning models, namely artificial neural network (ANN) and adaptive neuro-fuzzy system (ANFIS), for 1-day-ahead streamflow prediction in a catchment. Daily rainfall and discharge value from 1979 to 2015 of the catchment Fal at Tregony (United Kingdom) is used to validate the performance efficiency of the proposed hybrid algorithm, the HBBPSGWO along with ANN and ANFIS separately. The results demonstrate that the HBBPSGWO-ANN/HBBPSGWO-ANFIS improves the BBO convergence by avoid the trapping it into local minima and the performance has significantly improved compare to basic BBO-based ANN (BBO-ANN)/BBO-based ANFIS (BBO-ANFIS) and PSO-based ANN (PSO-ANN)/PSO-based ANFIS (PSO-ANFIS). In testing phase, root mean square error (RMSE) of HBBPSGWO-ANN is found lowest (0.901) compare to BBO-ANN (0.949) and PSO-ANN (0.926). Whereas in the case of integrated ANFIS, the RMSE of HBBPSGWO-ANFIS is also found minimum (0.741) compare to BBO-ANFIS (0.805) and PSO-ANFIS (0.828). The finding of this research concludes that the proposed hybrid metaheuristic algorithm has better capability to predict the daily streamflow. Moreover, the convergence of HBBPSGWO requires a smaller number of iterations to run in comparison to BBO and PSO.