A future demand prediction based approach for the design of pelton turbines on irrigation channels

Abstract

In the study, a sequence-to-one regression methodology utilizing the predictive capabilities of the Long Short-Term Memory (LSTM) algorithm was employed to provide forecasts of flow rate values in the future time periods based on the two-year discharge data of Akbas HEPP Irrigation Canal. Thus, the aim is to evaluate the algorithm's forecasting ability for different timeframes beyond the existing dataset. Predictions for the future working conditions were sought by comparing the results of the pelton-type turbine, designed using traditional methods based on the current dataset, with the anticipated outcomes. Additionally, a study was conducted to optimize the performance of the LSTM design. After selecting the appropriate architecture, it was observed that the Long Short-Term Memory (LSTM) architecture, used for flow prediction in irrigation systems, achieved approximately 0.96, 0.92, and 0.90 R2 values, confirming its effectiveness in prediction tasks. LSTM models were trained multiple times, consistently demonstrating strong performance with diverse values for different prediction horizons. © 2024, Begell House Inc. All rights reserved.