Data-Driven Virtual Power Plant Bidding Strategy in Electricity Markets Integrating Hybrid Forecasting Model and Customized Incentive Demand Response
Hyung Joon Kim, Mun-Kyeom Kim
Abstract
The advancement of Internet of Things technologies has accelerated the development of virtual power plants; however, uncertainties within these systems can jeopardize their operational flexibility and profitability in electricity markets. To address these challenges, this study proposes a novel data-driven bidding strategy for virtual power plants in both day-ahead and real-time electricity markets, incorporating a hybrid forecasting model and an unsupervised learning-based customized incentive demand response. First, unlike single forecasting models that naively train original input features, a hybrid decomposition and deep learning-based forecasting model is proposed to efficiently handle irregular and volatile data sequences, thereby improving forecasting accuracy and reducing mismatch costs. Second, to address the inherent inflexibility of current incentive mechanisms for demand response, a new customized incentive demand response program is proposed to efficiently motivate demand response by offering customized incentive rates that simultaneously consider the cus-tomer diversity, market prices, and various stakeholder preferences. To bid in electricity markets, a multi-objective optimization is formulated in the day-ahead stage to account for interests of various stakeholders, while a deep learning-based rolling horizon optimization is executed in the real-time stage to minimize mismatch costs. Simulation results using real data from the Pennsylvania-New Jersey-Maryland market validate the superiority of the proposed data-driven bidding strategy for virtual power plants in both day-ahead and real-time electricity markets.