Wind power bidding strategy by harmonizing neural network and genetic algorithm

ABSTRACT

As the importance of wind power as a renewable energy resource increases, the need for strategic management has increased for wind power producers (WPPs). However, many countries are still beginning in the renewable market, such as South Korea. Therefore, in this study, we discuss a bid determination model for a WPP that can be easily applied with limited prices and forecast information. Most previous studies attempted to minimize forecast error and then decide bids based on the predicted wind power generation. However, our model aims to maximize the total profit of a WPP directly. In our profit-focused neural network (NN) model, the general NN predicts generation forecasts at first. Next, based on the forecasts, WPP’s bid for the next 24 hours is determined by a genetic algorithm to maximize WPP’s profit. The bid based on the profit-focused NN model achieves 0.3–4.6% more gains than that of a general NN’s bid under a variable ratio of a day-head and balancing market prices and penalty cost. These practical ranges signify that our model remains profitable and robust in various conditions.

KEYWORDS:

• Forecasting
• profit maximization
• renewable energy
• wind power producer

Acknowledgments

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 2019R1F1A105461512).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data subject to third party restrictions

The data that support the findings of this study are available from Korea South-East Power Company.

Restrictions apply to the availability of these data, which were used under license for this study. Data are available at https://www.koenergy.kr/kosep/en/main.do with the permission of Korea South-East Power Company.

Notes

³ The “feedforwardnet,” “configure,” “net.performFcn,” “net.trainParam.goal,” and “net.trainParam.epochs” are MATLAB functions regarding the Neural Network (Xy 2000).

⁴ Refers to the data not used to build NN model.

⁵ The “gafunction,” “gaoptimset,” and “ga” are MATLAB functions regarding the Genetic Algorithm (Xy 2000).

⁶ Refers to the data not used to build NN model.

⁸ $\frac{\mathrm{M}\mathrm{S}\mathrm{E}\left(\mathrm{o}\mathrm{u}\mathrm{t}\mathrm{p}\mathrm{u}\mathrm{t}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{e}\mathrm{o}\mathrm{f}\mathrm{N}\mathrm{N}-\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{e}\mathrm{t}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{e}\mathrm{o}\mathrm{f}\mathrm{N}\mathrm{N}\right)}{\mathrm{V}\mathrm{A}\mathrm{R}\left(\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{g}\mathrm{e}\mathrm{t}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{e}\mathrm{o}\mathrm{f}\mathrm{N}\mathrm{N}\right)}\le \omega$. This means the $(1-\omega$) % of a random test target subset variance is successfully modeled. In our study, we set $\omega$ as 0.3 due to the high volatile weather of the region we collected the data..