WHALE OPTIMIZATION NEURAL NETWORK FOR DAILY WATER LEVEL FORECASTING CONSIDERING THE CHANGING CLIMATE

The influence of climate change is crucial to ensure effective planning and management of water resources in the future. Researchers previously adopted conventional artificial neural network (ANN) models for solving optimization problems. However, conventional ANN models exhibited a tendency for local optima trapping, rendering them ineffective as the complexity of optimization problems increased. This issue can be addressed using metaheuristic-based ANN models, such as Whale Optimization Neural Networks (WONN), for developing a water level model at the Batu Kitang Submersible Weir (BKSW). Hyper-parameter tuning was conducted to determine the optimal configuration of WONN using the GFDL-CM3 Global Circulation Model (GCM) under the RCP4.5 scenario. A total of 2,555 daily data points were used, with 70% allocated for training and 30% for testing. The models' performance was evaluated based on average mean absolute error, average root mean square error, and average correlation coefficient. Results revealed that the optimal configuration of WONN was determined to be 18 hidden nodes, 30 search agents, 250 maximum iterations, and 2,500 epochs, yielding an average mean absolute error of 0.1425, an average root mean square error of 0.1989, and an average correlation coefficient of 0.9097. The future long-term daily weir water level is forecasted to increase over the years, necessitating further efforts and measures to control water downstream for flood mitigation purposes.