Hybrid deep learning and optimized variational mode decomposition for point-interval runoff prediction.

Area of Science:

• Hydrology and Water Resources
• Environmental Engineering
• Computational Science

Background:

• Accurate runoff prediction is vital for water resource allocation and hydropower.
• Existing forecasting methods often suffer from low accuracy and high uncertainty.
• Addressing these limitations is crucial for effective water management.

Purpose of the Study:

• To develop an advanced framework for accurate and reliable runoff interval prediction.
• To improve feature extraction and uncertainty quantification in hydrological forecasting.
• To reduce prediction errors and enhance the compactness of prediction intervals.

Main Methods:

• Integration of Information Acquisition Optimizer (IAO) with Variational Mode Decomposition (VMD) for optimized decomposition (IVMD).
• Application of Convolutional Neural Network-Support Vector Machine (CNN-SVM) for enhanced point prediction using decomposed components.
• Utilizing Kernel Density Estimation (KDE) with B-spline least squares cross-validation (LSCV-B) for accurate prediction error distribution modeling.

Main Results:

• The proposed IVMD-CNN-SVM framework demonstrated significant improvements in runoff forecasting.
• Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) were reduced by approximately 40-50% compared to VMD-based methods.
• Highly reliable and compact 90% prediction intervals were achieved, indicating superior uncertainty quantification.

Conclusions:

• The integrated IAO-VMD-CNN-SVM-KDE framework offers a robust solution for improving runoff forecasting accuracy and reliability.
• The LSCV-B method enhances the precision of error density estimation, leading to more effective interval predictions.
• This approach provides a valuable tool for water resource management and hydropower planning, particularly in complex river basins like the Yangtze River Basin.