A runoff prediction method based on hyperparameter optimisation of a kernel extreme learning machine with multi-step...

Area of Science:

• Hydrology
• Computational Intelligence
• Time Series Analysis

Background:

• Accurate runoff forecasting is crucial for water resource management and flood control.
• Existing forecasting models often struggle with the complex, non-linear dynamics of hydrological data.
• The need for advanced hybrid models integrating signal processing and machine learning techniques is evident.

Purpose of the Study:

• To develop and evaluate a novel hybrid model for enhancing the accuracy of daily runoff forecasting.
• To compare the performance of the proposed hybrid model against individual components and other established methods.
• To validate the model's effectiveness using real-world daily runoff data from multiple hydrological stations.

Main Methods:

• Signal decomposition using Variational Modal Decomposition (VMD) and Complementary Ensemble Empirical Modal Decomposition (CEEMD).
• Optimization of the Kernel Extreme Learning Machine (KELM) model using the Butterfly Optimization Algorithm (BOA).
• Integration of VMD, CEEMD, BOA, and KELM to create a combined forecasting framework (VMD-CEEMD-BOA-KELM).

Main Results:

• The VMD-CEEMD-BOA-KELM model demonstrated superior performance in runoff time series prediction.
• Achieved low average absolute errors (e.g., 23.72 m³/s) and root mean square errors (e.g., 18.66 m³/s).
• High decision coefficients (above 90%) and Nash efficiency coefficients (above 90%) confirm the model's predictive power.

Conclusions:

• The hybrid VMD-CEEMD-BOA-KELM model offers a significant advancement in runoff forecasting accuracy.
• The combined approach effectively captures complex hydrological patterns, outperforming individual methods.
• This methodology provides a robust tool for operational hydrological forecasting and water resource management.