A novel stabilized artificial neural network model enhanced by variational mode decomposing.

Area of Science:

• Environmental science
• Data science
• Machine learning

Background:

• Artificial neural networks (ANNs) struggle with complex environmental data pattern identification due to time-consuming structure optimization.
• Existing ANN optimization methods often require trial-and-error or external tools, increasing complexity and application time.

Purpose of the Study:

• To introduce a stabilized artificial neural network (SANN) for efficient ANN structure optimization.
• To enhance meteorological drought forecasting using the novel SANN model.
• To validate the SANN model's performance through practical case studies.

Main Methods:

• Proposed a stabilized artificial neural network (SANN) by incorporating an additional numeric parameter into each ANN layer.
• Developed a hybrid Variation Mode Decomposition-SANN (VMD-SANN) model by integrating VMD with SANN.
• Compared the VMD-SANN model against hybrid VMD-ANN and VMD-Radial Base Function (VMD-RBF) models using meteorological drought data.

Main Results:

• The SANN model demonstrated efficient ANN structure optimization, reducing complexity and application time.
• The hybrid VMD-SANN model achieved superior forecasting accuracy compared to VMD-ANN and VMD-RBF models.
• The VMD-SANN model reached high Nash-Sutcliffe Efficiency values of 0.945 for Burdur and 0.980 for Isparta.

Conclusions:

• The proposed SANN offers an efficient alternative for ANN structure optimization in environmental applications.
• The hybrid VMD-SANN model represents a significant advancement in meteorological drought forecasting.
• The VMD-SANN model's high accuracy validates its effectiveness for practical environmental prediction tasks.