Forecasting renewable energy for environmental resilience through computational intelligence.

Mansoor Khan,Muhammad Rashid Naeem,Hyun-Koo Kang,Hyeong-Jin Choi,Essam A. Al-Ammar,Wonsuk Ko,Zaher Mundher Yaseen

doi:10.1371/journal.pone.0256381

Mansoor Khan, Muhammad Rashid Naeem + Show 5 more

Open Access

https://doi.org/10.1371/journal.pone.0256381

Copy DOI

Abstract

Wind power forecasting plays a key role in the design and maintenance of wind power generation which can directly help to enhance environment resilience. Offshore wind power forecasting has become more challenging due to their operation in a harsh and multi-faceted environment. In this paper, the data generated from offshore wind turbines are used for power forecasting purposes. First, fragmented data is filtered and Deep Auto-Encoding is used to select high dimensional features. Second, a mixture of the CNN and LSTM models is used to train prominent wind features and further improve forecasting accuracy. Finally, the CNN-LSTM deep learning hybrid model is fine-tuned with various parameters for reliable forecasting of wind energy on three different offshore Windfarms. A state-of-the-art comparison against existing models is presented based on root mean square error (RMSE) and mean absolute error (MAE) respectively. The forecasting analyses indicate that the proposed CNN-LSTM strategy is quite successful for offshore wind turbines by retaining the lowest RMSE and MAE along with high forecasting accuracy. The experimental findings will be helpful to design environment resilient energy transition pathways.

Highlights

Renewable energy has been studied for a long time due to high electricity generation costs and global warming
The results showed that the Long Short-Term Memory (LSTM) auto-encoder based preprocessing can perform better compared to simple LSTM wind power forecasting
The orange curve indicates the predicted power by Deep Auto-Encoders (DAC)-Convolution Neural Network (CNN)-LSTM strategy while the grey curve shows the actual wind power generated from US offshore turbines

Summary

Introduction

Renewable energy has been studied for a long time due to high electricity generation costs and global warming. Power sectors of different countries use various wind power forecasting methods to generate sufficient wind energy. The mathematical models use historical wind data to estimate potential power generation [4,5] and may take advantage of various hybrid frameworks such as machine learning and deep learning-based power forecasting. It would be very difficult to develop an efficient numerical framework without any in-depth analysis of the systems engineering as well as the wind area atmosphere These types of models need to construct certain variables with the aid of illustrative variables, census data algorithms, to overcome the association between empirical evaluation and extracted wind properties. For short-term forecasting, minutes and hourly time series data are more accurate as they can be used for stochastic wind signals and have been used in many models such as Kalman filters [17], Box-Jenkins [18], as well. As per empirical analysis, the proposed method demonstrated excellent forecasting accuracy with a low error ratio at various intervals, making it more suited approach for offshore wind power forecasting

Literature review

Results and discussion

Conclusion