Robust Wind Speed Forecasting: A Deep Spatio-Temporal Approach

• Type: Article in International Conference Proceedings Book
• Date: 2021

Authors

Saffari, M (Author) Other; Williams, M (Author) Other; Khodayar, M (Author) Other; Shafie khah, M (Author) Other; Catalao, JPS (Author) FEUP

Conference proceedings International

Title: 2021 21ST IEEE INTERNATIONAL CONFERENCE ON ENVIRONMENT AND ELECTRICAL ENGINEERING AND 2021 5TH IEEE INDUSTRIAL AND COMMERCIAL POWER SYSTEMS EUROPE (EEEIC/I&CPS EUROPE)

21st IEEE International Conference on Environment and Electrical Engineering / 5th IEEE Industrial and Commercial Power Systems Europe (EEEIC/I and CPS Europe)

Politecnico Bari, Bari, ITALY, SEP 07-10, 2021

Indexing

ISI Web of Knowledge - 6 Citations

Scopus - 7 Citations

Other information

Authenticus ID: P-00W-BDR

DOI: 10.1109/eeeic/icpseurope51590.2021.9584602

Abstract (EN): With the significant increase in wind speed usage as a clean source of energy, an accurate wind speed forecasting system is a must for more effective utilization of this energy. Failure to consider the inherent spatio-temporal features of wind speed time series leads to the lack of generalization capacity for current wind speed forecasting approaches. This paper proposes an end-to-end deep neural network framework, i.e., convolutional rough long short-term memory (ConvRLSTM), to extract spatio-temporal wind correlations and mitigate the inherent uncertainties in wind time series by incorporating the Rough set theory into a combination of convolution neural network (CNN) and LSTM units. Our proposed model receives the historical data of wind speed for a 20x20 array of wind turbines in North Carolina, US. Several ConvRLSTM layers extract the most relevant features for the forecasting task, and finally, fully connected layers predict 400 wind speed values using the spatial features obtained by the CNN and temporal features computed by the LSTM. Through analyzing the numerical forecasting results, it can be inferred that the proposed approach outperforms the mainstream and recently published forecasting strategies in terms of the RMSE metric.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 6