Hydropower preventive maintenance scheduling in a deregulated market: A multi-stage stochastic programming approach

Zhuangzhuang Li,Ping Yang,Yi Yang,Guanpeng Lu,Yufeng Tang
DOI: https://doi.org/10.1016/j.epsr.2023.109379
IF: 3.818
2023-04-14
Electric Power Systems Research
Abstract:Hydropower preventive maintenance scheduling ensures the economical and reliable operation of the plant, which is a significant and complex optimization task. The hydropower producer will schedule maintenance to pursue maximum profit in a deregulated market. The electricity price and natural inflow uncertainties have significant economic impacts and must be considered. However, the existing maintenance scheduling models were two-stage models in which the maintenance schedule for the entire planning horizon was determined before the realization of the uncertainty. This paper presents a multi-stage risk-neutral preventive maintenance scheduling model for a price-taking hydropower producer in a deregulated market. The maintenance decisions are sequentially made with the information of electricity price and inflow uncertainties being revealed gradually. To hedge against the profit risk caused by the uncertainties, a multi-stage risk-averse maintenance scheduling model is proposed based on the multi-stage Conditional Value at Risk. We reformulate the proposed multi-stage models as multi-stage mixed-integer stochastic linear programming problems and apply the Stochastic Dual Dynamic integer Programming (SDDIP) algorithm to solve them. Finally, the utility of the proposed model is verified using the data of a cascade hydropower system in Southwest China. The results show that: (1) the proposed multi-stage maintenance model outperforms two-stage model in terms of higher expected revenue (+4.95% on average) and lower risk (-10.8% on average) under both in-sample and out-of-sample scenarios. The improvements of the multi-stage model decreases with an increase in risk aversion level. (2) The risk-averse model balances the expected profit and risk according to risk preferences. More risk-averse policies lead to lower expected profits and lower risks. (3) Despite longer running time of the multi-stage model (+35% on average) than the two-stage model, it is computationally tractable by SDDIP algorithm.
engineering, electrical & electronic
What problem does this paper attempt to address?