Peak flow forecasting in Mahanadi River Basin using a novel hybrid VMD-FFA-RNN approach

Abstract

Accurate prediction of peak streamflow is essential to mitigate the flood damages in watershed area. This study focuses on improving peak streamflow prediction using hybrid machine learning models in Mahanadi River Basin, India. In the hybrid model, variational mode decomposition (VMD) is used to decompose the original discharge data into various intrinsic model functions (IMFs) and firefly algorithm (FFA) is used to optimise the train/test split and hyperparameters of recurrent neural network (RNN) in two stages. The use of IMFs with original discharge data and dual stage parameter optimisation process makes this approach novel. The results show that the hybrid VMD-FFA-RNN model performed better than all other models, showing greater performance during both training and testing periods. This improved performance can be attributed to its modified structural algorithm. Furthermore, comparative analysis using statistical performance indicators, such as root mean square error (RMSE), indicates a notable 76.50% and 46.63% improvement in prediction accuracy compared to the simple RNN and VMD-RNN models, respectively. Therefore, the developed hybrid model presents a capable alternative for future time series forecasting applications, offering enhanced accuracy and reliability in peak streamflow prediction in the Mahanadi River Basin, India, and potentially in similar watershed systems.