Model-based deep reinforcement learning for wind energy bidding

Wind energy is an important source of clean energy. Due to the common trade through bidding, many attempts have been made to apply deep reinforcement learning techniques to generate appropriate bidding policies to maximize profits. However, these algorithms are based entirely on a model-free strategy. The present study aims to develop a dynamic model capable of strategic bidding for wind energy. Thus, the model MB-A3C is implemented and proves to be quite resilient. Herein, Nord Pool'', a conventional benchmark that comprises six datasets representing each wind power site in Denmark and Sweden is duly investigated. Results show that the policies generated by MB-A3C are less costly than those produced by both previous model-free and model-based algorithms i.e. Conv-A3C, DPPO, DDPG, and MBPG. The optimal bidding approach demonstrated in this study can be utilized to optimize profits and overcome the uncertainties in both the energy and reserve markets.

Automated summary

This study develops a dynamic model for strategic bidding in wind energy, aiming to optimize profits and overcome uncertainties in energy and reserve markets. The results show that the policies generated by this model are less costly compared to previous algorithms.