Session: 11-01: Alternative Energy Converstion Technology (including Wind, Geothermal, Hydro, and Ocean)

Paper Number: 168713

168713 - Levelized Cost of Energy Analysis of Wind Power Plants Using a Cnn-Lstm Hybrid Prediction Model 

Abstract: 

Wind energy has emerged as a leading renewable energy source, with significant potential to reduce carbon emissions and contribute to global energy needs. However, accurately assessing the economic viability of wind energy projects is challenging due to the variability of wind generation. This study investigates the effectiveness of a CNN-LSTM hybrid neural network in improving the accuracy of wind energy predictions for Levelized Cost of Energy (LCOE) calculations. Wind power's economic viability hinges on precise energy production forecasts, yet traditional prediction methods often fall short due to wind's inherent variability. Using a comprehensive dataset of 50,530 wind speed records spanning a full year, we developed and compared three prediction models; an ARIMA linear regression approach, a standalone LSTM network, and our novel CNN-LSTM hybrid to forecast wind energy production.

Our models utilized LV active power, wind direction, and theoretical power as input variables to predict wind speed as the target output variable. The comparative analysis demonstrated a clear performance hierarchy, with the CNN-LSTM hybrid model achieving superior accuracy (RMSE = 0.439 m/s) compared to both the standalone LSTM (RMSE = 0.5057 m/s) and the ARIMA model (RMSE = 1.1744 m/s). The CNN component effectively extracted spatial features from the input data, which complemented the LSTM's temporal pattern recognition capabilities, resulting in enhanced prediction precision.

When integrated into LCOE calculations for a wind farm installation, the CNN-LSTM hybrid model produced an LCOE value of 28.893 $/MWh, closely aligning with real-life measurements (29.01 $/MWh). This represented a significant improvement over both the standalone LSTM model (28.363 $/MWh) and especially the ARIMA model (25.66 $/MWh). This improved accuracy has significant implications for investment decision-making in renewable energy projects, as even small deviations in LCOE estimates can substantially impact financial projections over a wind farm's 25-year lifecycle.

These findings demonstrate that advanced hybrid machine learning architectures can substantially enhance the economic assessment accuracy of wind power projects, potentially reducing investment risk and improving renewable energy adoption rates. Future work will incorporate additional variables and explore further neural network architectures to improve prediction accuracy for wind energy applications.

Presenting Author: Sara Mouafik Mississippi State University

Presenting Author Biography: Sara Mouafik is a Ph.D student in Michael W. Hall School of Mechanical Engineering at Mississippi State University, currently doing research in the renewable energy area, specifically introducing machine learning in renewable energy predictions.