Forecasting of wind speed is essential for the effective design, management, and maintenance of wind energy systems, which are a key component of renewable energy sources. The difficulty of precisely predicting wind speeds is addressed in this study, as wind patterns are often erratic and unpredictable. Four predictive models have been used in this study for comparative analysis. It has been shown that ARIMA model helps to capture temporal dependencies (linear) & the Kalman Filter very efficiently helps to reduce noise in wind speed data. However, non-linear dependencies have been modeled efficiently by LSTM and proposed Bi-LSTM model. Data sets are gathered from two sites encompassing a period of one year, i.e., June 2022 onwards.
In comparison to the LSTM, the Bi-LSTM model consistently outperformed the other methods, with higher error rates for the ARIMA and Kalman Filter, a root mean square error of 0.1280 m/s, mean absolute percentage of 0.0531 m/s, symmetric mean absolute percentage error of 0.0108 m/s. Improvement of Bi-LSTM’s performance is due to its ability to deal with time-series data (in forward as well as backward directions), that helps it to capture the stronger correlations (temporal) in the dataset of wind speed. This work is novel because it makes use of Bi-LSTM for wind speed prediction, a method not completely explored in the previous research, and it further integrates statistics and deep learning models. The results make demonstration of the fact that Bi-LSTM significantly improves wind speed prediction accuracy, further improving the reliability as well as efficiency of wind energy systems.
By demonstrating improved prediction accuracy for a wide range of datasets, this work significantly contributes to the renewable energy systems and highlights the benefits of using advanced deep learning approaches for prediction problems.