A combined method for wind power generation forecasting

• Published in: Archives of Electrical Engineering (Online) Vol. 70; no. 4; pp. 991 - 1009
• Main Author: Le, Tuan-Ho
• Format: Journal Article
• Language: English
• Published: Warsaw Polish Academy of Sciences 01-01-2021
• Subjects: Ambient temperature; autoregressive integrated moving average; Diameters; Error analysis; exponential smoothing method; Forecasting; Measuring instruments; Pitch (inclination); Response surface methodology; Smoothing; Statistical methods; Time series; Turbogenerators; Wind power; Wind power generation; Wind speed; Wind turbines
• ISSN: 1427-4221; 2300-2506
• DOI: 10.24425/aee.2021.138274

Most of the existing statistical forecasting methods utilize the historical values of wind power to provide wind power generation prediction. However, several factors including wind speed, nacelle position, pitch angle, and ambient temperature can also be used to predict wind power generation. In this study, a wind farm including 6 turbines (capacity of 3.5 MW per turbine) with a height of 114 meters, 132-meter rotor diameter is considered. The time-series data is collected at 10-minute intervals from the SCADA system. One period from January 04th, 2021 to January 08th, 2021 measured from the wind turbine generator 06 is investigated. One period from January 01st, 2021 to January 31st, 2021 collected from the wind turbine generator 02 is investigated. Therefore, the primary objective of this paper is to propose a combined method for wind power generation forecasting. Firstly, response surface methodology is proposed as an alternative wind power forecasting method. This methodology can provide wind power prediction by considering the relationship between wind power and input factors. Secondly, the conventional statistical forecasting methods consisting of autoregressive integrated moving average and exponential smoothing methods are used to predict wind power time series. Thirdly, response surface methodology is combined with autoregressive integrated moving average or exponential smoothing methods in wind power forecasting. Finally, the two above periods are performed in order to demonstrate the efficiency of the combined methods in terms of mean absolute percent error and directional statistics in this study.