An improved framework to predict river flow time series data.

Area of Science:

• Hydrology
• Time Series Analysis
• Signal Processing

Background:

• River flow time series data often exhibit non-stationary and noisy characteristics, posing challenges for accurate prediction.
• Existing pre-processing methods struggle with the multi-scale and noise complexity inherent in such data.

Purpose of the Study:

• To propose an improved framework for pre-processing and predicting non-stationary and noisy river flow time series data.
• To enhance the reliability of river flow predictions for applications in water resource management and power generation planning.

Main Methods:

• Developed a hybrid framework combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise-Empirical Bayesian Threshold (CEEMDAN-EBT).
• Decomposed river flow data into Intrinsic Mode Functions (IMFs), separating noise-dominant and noise-free components.
• Applied empirical Bayesian thresholding for denoising IMFs and utilized data-driven and stochastic models for prediction before aggregating results.

Main Results:

• The proposed CEEMDAN-EBT-MM framework demonstrated superior performance in predicting river flow time series.
• Achieved the smallest Mean Absolute Percentage Error (MAPE) across all four case studies compared to other methods.
• Successfully handled the non-stationary and noisy characteristics of the river flow data.

Conclusions:

• The CEEMDAN-EBT-MM framework is an efficient tool for reliable prediction of non-stationary and noisy time series data.
• The findings support its application for policymakers in areas like power generation planning and water resource management.
• The hybrid approach effectively integrates signal decomposition and predictive modeling for improved hydrological forecasting.