LSTM forecasting for long-term MHP performance in irrigation: A feasibility study

Abstract

This study presents a feasibility assessment of hydroelectric energy generation using an inline-pipe Pelton turbine installed on the Akbaş Irrigation Channel in the Denizli region. The design flow rate and turbine operating characteristics were determined using flow-duration curves derived from measured hourly data. To examine future operational conditions, a Long Short-Term Memory (LSTM) deep learning model was trained with the historical time series to predict 700 days of future flow values. The prediction-based feasibility results indicate that the future discharge remains below the present-day average, leading to an estimated reduction of approximately 20% in annual energy production and revenue. Despite this decrease, the Pelton turbine is still capable of operating above the minimum discharge requirement for most of the year, confirming the continued feasibility of electricity generation under predicted conditions. The findings demonstrate that combining traditional hydrological analysis with deep learning-based forecasting provides a preliminary practical decision-support tool for evaluating the long-term performance of micro-hydropower installations on irrigation and water conveyance systems. While the methodology provides a robust framework, further validation across diverse geographical locations and longer multi-year datasets is essential to confirm its generalizability and to transition this proof-of-concept into a broad-scale planning tool.