PARTICLE SWARM OPTIMIZATION IN FEEDFORWARD NEURAL NETWORKS FOR RAINFALL-RUNOFF SIMULATION

• Main Authors: Kuok, King Kuok, Chiu, Po Chan, Md. Rezaur, Rahman, Chin, Mei Yun, Mohd Elfy, Mersal
• Other Authors: King Kuok, Kuok
• Format: Book Chapter
• Language: English
• Published: Cambridge Scholars Publishing 2024
• Subjects: T Technology (General)
• Online Access: http://ir.unimas.my/id/eprint/46907/1/Particle%20swarm.pdf; http://ir.unimas.my/id/eprint/46907/; https://www.cambridgescholars.com/product/978-1-0364-0804-6

Backpropagation neural networks have been effectively utilized by hydrologists in recent years to model various nonlinear hydrological processes due to their ability to generalize patterns in vague, noisy, ambiguous, and incomplete input and output datasets. However, the solutions may become stuck at local minima because of the slow convergence rate during the training process. To address these issues, Particle Swarm Optimization (PSO) was adopted in this study to train the feedforward neural network for modeling the rainfall-runoff relationship of the Sungai Bedup Basin in Sarawak, Malaysia. The Nash-Sutcliffe coefficient and correlation coefficient were used to evaluate the model's performance. The model's output is the current runoff, while the inputs include current rainfall, antecedent rainfall, and antecedent runoff. The results revealed that the particle swarm optimization feedforward neural network (PSONN) accurately reproduced the current runoff, achieving R=0.872 and E2=0.775 for the training dataset and R=0.900 and E2=0.807 for the testing dataset. These findings are comparable to conventional Multilayer Perceptron and Recurrent Neural Networks. Thus, PSONN successfully modeled the rainfall-runoff relationship and has the potential to be adapted for solving optimization problems in other domains.