XinHao (2025) A novel hybrid DOA-PSO-enhanced LSSVM model for monthly runoff forecasting in the upper Heihe river basin

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-10-14
• Authors: Zhang XinHao
• DOI: 10.1038/s41598-025-19865-z

Research Groups

Taiyuan University of Technology, Taiyuan, China

Short Summary

This study proposes a novel hybrid DOA-PSO-LSSVM model for monthly runoff forecasting, integrating the Dream Optimization Algorithm (DOA) for global exploration, Particle Swarm Optimization (PSO) for local refinement, and Least Squares Support Vector Machine (LSSVM) for nonlinear learning. Applied to the upper Heihe River Basin, the model demonstrates superior predictive accuracy and robustness compared to conventional and single-optimizer models, effectively addressing the challenges of nonlinear and non-stationary hydrological processes.

Objective

• To develop a novel hybrid DOA-PSO-LSSVM model for accurate monthly runoff forecasting by integrating global and local optimization algorithms to effectively tune LSSVM hyperparameters.
• To enhance prediction accuracy, reduce mode mixing and subjectivity in parameter selection, and improve computational efficiency compared to conventional models.
• To evaluate the model's performance, stability, and spatial/temporal generalization capabilities using hydro-meteorological data from the upper Heihe River Basin.

Study Configuration

• Spatial Scale: Upper Heihe River Basin, China, specifically the area above the Yingluoxia hydrological station, with a watershed area of approximately 9900 km². Meteorological data from Qilian, Yeniugou, and Zhangye stations.
• Temporal Scale: Monthly runoff forecasting. Data spans 30 years (January 1, 1980, to December 31, 2009). Training set: 1980-2000 (252 months); Testing set: 2001-2009 (108 months).

Methodology and Data

• Models used:
  • Hybrid DOA-PSO-LSSVM (Dream Optimization Algorithm, Particle Swarm Optimization, Least Squares Support Vector Machine)
  • Comparative models: LSSVM, PSO-LSSVM, DOA-LSSVM
• Data sources:
  • Digital Elevation Model (DEM) data: 30-meter spatial resolution, from Geospatial data cloud. Used for subbasin delineation and river network extraction.
  • Meteorological station records: Daily records (1980-2009) of precipitation, average temperature, wind speed, and dew point temperature from Qilian, Yeniugou, and Zhangye stations. Obtained from the National Climatic Data Center (NOAA, USA) and China Meteorological Administration (CMA). Input variables for the model: average temperature, dew point temperature, and precipitation.
  • Hydrological station streamflow data: Monthly records (1980-2009) from Yingluoxia Hydrological Station. Obtained from the Gansu Provincial Hydrology Bureau. Missing values addressed through linear interpolation.

Main Results

• The hybrid DOA-PSO-LSSVM model consistently achieved the highest predictive accuracy across all experimental groups and stations.
• Compared to single-optimizer models (DOA-LSSVM, PSO-LSSVM, LSSVM), DOA-PSO-LSSVM demonstrated RMSE reductions ranging from 4% to 23%.
• Correlation coefficients (r) for DOA-PSO-LSSVM consistently exceeded 0.95 across all stations and scenarios, indicating a strong linear relationship between predicted and observed runoff.
• DOA-PSO-LSSVM2 (using Yeniugou station data) exhibited the best performance on the testing dataset with a Nash-Sutcliffe Efficiency (NSE) of 0.914, Root Mean Square Error (RMSE) of 13.16, and Pearson correlation coefficient (r) of 0.961.
• The model showed strong capability in simulating peak flows and water balance, with generally lower Volume Error (Ve), Relative Error (RE), and Peak Flow Relative Difference (RD) values compared to other models.
• The narrow performance gap between training and testing datasets indicated high generalization capability and effective avoidance of overfitting.
• The model demonstrated robustness and adaptability across different meteorological input scenarios and temporal scales, maintaining high accuracy (r values above 0.93) even when validated on different time periods.

Contributions

• Development of a novel hybrid framework that integrates the Dream Optimization Algorithm (DOA) for global exploration and Particle Swarm Optimization (PSO) for local refinement to optimize Least Squares Support Vector Machine (LSSVM) hyperparameters, significantly improving prediction accuracy and avoiding issues associated with mode decomposition.
• Design of a systematic hyperparameter tuning strategy that enhances model stability and reduces sensitivity to variations in input data, providing a more robust approach to hydrological modeling.
• Application of the proposed model to the upper reaches of the Heihe River using real-world meteorological station data, accompanied by a multidimensional analysis of runoff simulation results to evaluate the suitability of different datasets for runoff prediction and provide practical implications for water management.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{XinHao2025novel,
  author = {XinHao, Zhang},
  title = {A novel hybrid DOA-PSO-enhanced LSSVM model for monthly runoff forecasting in the upper Heihe river basin},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-19865-z},
  url = {https://doi.org/10.1038/s41598-025-19865-z}
}