Optimal bidding strategy of a virtual power plant in day-ahead energy and frequency regulation markets: A deep learning-based approach

Abstract

Various challenges and opportunities are recognized by increasing the penetration of distributed energy resources (DERs) in power systems. In this regard, the concept of virtual power plants (VPPs) has been proposed to tackle the imposed challenges and exploit the offered opportunities. In this paper, an optimal bidding strategy of a VPP participating in the day-ahead frequency regulation market (FRM) and the energy market (EM) is proposed. A comprehensive form of a VPP that contains various DERs has a high potential in FRM due to its fast response. In this study, to strengthen the VPP performance in FRM, which has strict rules with steep penalties, a deep learning-based approach known as bi-directional long short-term memory (B-LSTM) network is employed. Forming a precise estimation on the VPP internal components (i.e. renewable energy resources (RESs) generation, VPP load demand, and electric vehicles (EVs) behavior) and market signals (i.e. different electricity prices, regulation signals) is a crucial factor in the VPP optimal bidding strategy in the day-ahead markets. This accurate forecasting obtained by B-LSTM helps VPP operators to fulfill the day-ahead awarded bids and avoid substantial penalties in the real-time market. The CAISO market rules are considered to form the test environment. The numerical results illustrate the success of the proposed method in handling the uncertainty of the various parameters by providing accurate results (3.75% error in comparison with real data); which, will increase the VPP’s profit significantly. Furthermore, the diversification of the VPP resources through implementation of distributed generations (DGs), energy storages (ESs), and EVs and their mobilization in FRM yields 470.76 $, 550 $, and 33.58 $ profit, respectively—which constitutes 24.41% of the VPP's total profit for real data.