Maximum energy entropy: A novel signal preprocessing approach for data-driven monthly streamflow forecasting
Alireza B. Dariane, Mohammad Reza M. Behbahani
Abstract
In recent years, the application of Data-Driven Models (DDMs) in ecological studies has garnered significant attention due to their capacity to accurately simulate complex hydrological processes. These models have proven invaluable in comprehending and predicting natural phenomena. However, to achieve improved outcomes, certain additive components such as signal analysis models (SAM) and input variable selections (IVS) are necessary. SAMs unveil hidden characteristics within time series data, while IVS prevents the utilization of inappropriate input data. In the realm of ecological research, understanding these patterns is pivotal for grasping the ecological implications of streamflow dynamics and guiding effective management decisions. Addressing the need for more precise streamflow forecasting, this study proposes a novel SAM called “Maximum Energy Entropy (MEE)” to forecast monthly streamflow in the Ajichai basin, located in northwestern Iran. A comparative analysis was conducted, pitting MEE against well-known methods such as Discreet Wavelet (DW) and Discreet Wavelet-Entropy (DWE), ultimately demonstrating the superiority of MEE. The results showcased the superior performance of our proposed method, with an NSE value of 0.72, compared to DW (NSE value of 0.68) and DWE (NSE value of 0.68). Furthermore, MEE exhibited greater reliability, boasting a lower Standard Deviation value of 0.13 compared to DW (0.26) and DWE (0.19). The utilization of MEE equips researchers and decision-makers with more accurate predictions, facilitating well-informed ecological management and water resource planning. To further evaluate MEE's accuracy using various DDMs, we integrated MEE with Artificial Neural Network (ANN) and Genetic Programming (GP). Additionally, GP served as an IVS method for selecting appropriate input variables. Ultimately, the combination of MEE and GP within the ANN forecasting model (MEE-GP-ANN) yielded the most favorable results.