Comparison Among ARIMA, ANN, and SVR Models for Wind Power Deviation Charge Reduction

Abstract

Adverse climatic changes and exponentially increasing electric power demand have been forcing various countries to increase the percentage share of renewable generation in their total generation mix. This leads to very high penetration of wind energy in various power systems due to its low-cost technologies and widespread availability. Uncertain variability of such high wind penetration resulted in various challenges such as wind power curtailment, load shedding, and voltage collapse to economic and secure system operation. These challenges increase total system operation cost as such challenges are met by various balancing products like flexible ramp products and additional operating reserves. Such increased operating costs are usually transferred to wind power producers as penalties, called Deviation Charges (DCs). This reduces wind power producers’ profit. DCs are estimated based on the difference between actual generation and wind power forecasts. This paper compares ARIMA, ANN, and SVR forecasting models for their capability to reduce DCs. Performance analysis is conducted on data collected from the BPA balancing area. Seasonal analysis shows that the ANN model is the best option to reduce deviation charges compared to other models, even though the SVR and ARIMA models show the least absolute percentage in the spring, winter, and fall seasons respectively.