Strategic retail pricing and demand bidding of retailers in electricity market: A data-driven chance-constrained programming
Dawei Qiu,Zihang Dong,Guangchun Ruan,Haiwang Zhong,Goran Strbac,Chongqing Kang
DOI: https://doi.org/10.1016/j.adapen.2022.100100
2022-09-01
Advances in Applied Energy
Abstract:This paper proposes a novel bi-level optimization model to study the strategic retail pricing and demand bidding problems of an electricity retailer that considers the interactions between demand response and market clearing process. In order to accurately forecast the day-ahead demand bids submitted by the retailer, a novel deep learning framework based on convolutional neural networks and long short-term memory is proposed that can capture both local trends and long-term dependency of the forecasting data. In addition, uncertainties about the retailer’s served demand, rivals’ demand bids, and wind power generation are incorporated using the data-driven uncertainty set constructed from data. We further propose chance-constrained programming that introduces a set of chance constraints to represent the operational risk associated with the market uncertainties. To solve this problem, we first reformulate chance-constrained programming as a tractable second-order conic programming and then convert it into a single-level mathematical program with equilibrium constraints by using its Karush Kuhn Tucker conditions. The scope of the examined case studies is four-fold. First, they evaluate the benefits of the proposed forecasting framework in terms of higher accuracy and expected profit compared to the conventional forecasting methods. Second, they demonstrate how demand flexibility affects the retailer’s strategies and its business cases. Third, they highlight the added value of the proposed bi-level model capturing the market clearing process by comparing its outcomes against the state-of-the-art bi-level model with exogenous market prices. Finally, they analyze the retailer’s strategies and business cases at different confidence levels regarding the imposed chance constraints.