Enhancement of Texas wind turbine power predictions using fractional order neural network by incorporating machine learning models to impute missing data

Abstract

In real-world datasets, missed data is often expected due to sensor errors, environmental conditions, communication errors, and other technical limitations. These factors can affect the accuracy of power predictions, particularly in wind speed and direction parameters. Enhancing the accuracy of wind energy forecasts and maintaining the electrical grid's stability are essential in addressing related challenges. This paper proposes a bidirectional long short-term memory (Bi-LSTM) model to forecast missed wind speed and direction data to address these issues. In addition, a novel fractional-order neural network (FONN) with various developed fractional activation functions presents enhanced wind power prediction using forecasted missing wind speed and direction data. The performance of the FONN model is demonstrated in four comprehensive case studies on the Texas wind turbine dataset. The first case study uses wind speed, direction, pressure, and air temperature data. The second case study uses pressure, air temperature, wind speed, and forecasted wind direction data. Similarly, pressure, air temperature, wind direction, and forecasted wind speed data are used in the third case study. The final case study uses pressure, air temperature, and forecasted wind speed and direction to predict power. The proposed models’ performance is evaluated using mean square error and coefficient of determination metrics.