A new soft computing model for daily streamflow forecasting

Accurate stream flow quantification and prediction are essential for the local and global planning and management of basins to cope with climate change. The ability to forecast streamflow is crucial, as it can help mitigate flood risks. Long-term stream flow data records are needed for hydropower plant construction, flood prediction, watershed management, and long-term water supply use. An accurate assessment of streamflow is considered as very challenging and critical tasks. A new predicting model is developed in this research, combining the technique of sunflower optimization (SFA) as an evolutionary algorithm with the multi-layer perceptron (MLP) algorithm to predict streamflow in Malaysia's Jam Seyed Omar (JSO) and Muda Di Jeniang (MDJ) stations. Principal component analysis (PCA) was performed on Q (t) (t: the number of the current day) before model creation to pick essential inputs for a maximum of 6 lags. With the classical MLP and two other hybrid MLP models (MLP-particle swarm optimization (MLP-PSO) and MLP-genetic algorithm (MLP-GA)), the results of the MLP-sunflower algorithm (SFA) were benchmarked. As compared to other models, the MLP-SFA could be able to reduce the Root Mean Square Error (RMSE) by a value of between 12 and 21% at the JSO station and between 8 and 24% at the MDJ station. In conclusion, this research found that combining MLP with optimization algorithms improved the precision of the stand-alone MLP model, with SFA integration being the most efficient.

Automated summary

Accurate stream flow quantification and prediction are crucial for basin planning and management in the face of climate change. This research developed a new predicting model using a combination of sunflower optimization and multi-layer perceptron algorithm to improve the precision of streamflow prediction in Malaysia. The MLP-SFA model outperformed other models, reducing RMSE by a significant percentage at both JSO and MDJ stations.