Meeting Corporate Renewable Power Targets

• 成果类型: Article
• 署名作者: Trivella, Alessio; Mohseni-Taheri, Danial; Nadarajah, Selvaprabu
• 署名单位: University of Twente; University of Illinois System; University of Illinois Chicago; University of Illinois Chicago Hospital
• 刊物名称: MANAGEMENT SCIENCE
• ISSN/ISSBN: 0025-1909
• DOI: 10.1287/mnsc.2022.4354
• 发表日期: 2023
• 页码: 491-512
• 关键词: power purchase agreements; corporate renewable energy; climate targets; Markov decision processes; information relaxation and duality; reoptimization

摘要

Several corporations have committed to procuring a percentage of their electricity demand from renewable sources by a future date. Long-term financial contracts with renewable generators based on a fixed strike price, known as virtual power purchase agreements (VPPAs), are popular to meet such a target. We formulate rolling power purchases using a portfolio of VPPAs as a Markov decision process, accounting for uncertainty in generator availability and in the prices of electricity, renewable energy certificates, and VPPAs. Obtaining an optimal procurement policy is intractable. We consider forecast-based reoptimization heuristics consistent with practice that limit the sourcing of different VPPA types and the timing of new agreements. We extend these heuristics and introduce an information-relaxation based reoptimization heuristic, both of which allow for full sourcing and timing flexibilities. The latter heuristic also accounts for future uncertainties when making a decision. We assess the value of decision flexibility in rolling power purchases to meet a renewable target by numerically comparing the aforementioned policies and variants thereof on realistic instances involving a novel strike price stochastic process calibrated to data. Policies with full timing flexibility and no sourcing flexibility reduce procurement costs significantly compared with one with neither type of flexibility. Introducing sourcing flexibility in the former policies results in further significant cost reduction, thus providing support for using VPPA portfolios that are both dynamic and heterogeneous. Computing near-optimal portfolios of this nature entails using our information-relaxation based reoptimization heuristic because portfolios constructed via forecast-based reoptimization exhibit higher suboptimality.