An hourly-resolution capacity sharing market for generation-side clustered renewable-storage plants

Abstract

With the increasing penetration of renewable energy on the generation side, their volatility greatly challenges power balancing in the power grids. Deploying energy storage in wind farms, solar stations, and collection stations allow renewable plants to sell energy guided by the electricity price signal and increase their market revenues. This paper considers a representative scenario on the generation side. Wind farms and solar stations managed by different entities sell energy to a market through a collection station, aiming to maximize individual profits. Each renewable plant is equipped with a local battery in order to store energy and wait for a higher price. They can also rent some capacity from a shared energy storage unit at the collection station for better profitability. This paper designs a day-ahead hourly-resolution capacity rental market for the shared energy storage in the collection station and proposes an online operation policy for individual renewable plants. In the day-ahead market, renewable plants bid their needs of storage capacity in each time period based on the rental price and a batch of renewable power scenarios in the next day, and then the market is cleared at the Stackelberg equilibrium where the shared storage acts as the leader. Given the capacity obtained from the day-ahead market, each renewable plant obtains reference storage level trajectories in the pre-specified scenarios as experiences. In the real-time stage, the dispatch of local and shared storage units is determined from the conditional expectation of experiences, where the conditional distribution is generated by kernel regression using dynamic time warping as the distance measure. This proposed method does not rely on renewable power forecasts and is easy to implement. Numerical results validate the economy of the proposed method. Compared to the autarky mode, the profit of a renewable plant is increased by 40.6% on average. Compared to the ideal optimum, the optimality gap of the proposed method is 1.4% on average.