Risk-averse contextual predictive maintenance and operations scheduling with flexible generation under wind energy uncertainty

Ensuring resiliency and sustainability of power systems operations under the uncertainty of the intermittent nature of renewables is becoming a critical concern while considering the integration of flexible generation resources that provide additional adjustability during planning. To address this emerging issue, this study proposes a risk-averse contextual predictive generator maintenance and operations scheduling problem with traditional and flexible generation resources under wind energy uncertainty. We formulate this problem as a two-stage risk-averse stochastic mixed-integer program, where the first-stage determines the maintenance and unit commitment related decisions of the traditional generation units, whereas the second-stage determines the corresponding decisions for flexible generators along with the production related plans of all generators. To integrate contextual information and the uncertainty around the wind power, we propose a Gaussian Process Regression approach for predicting wind power generation, which is then leveraged into this stochastic program. Since this problem is computationally challenging to solve with a mixed-integer recourse due to the second-stage decisions involving flexible generation resources, we provide two versions of a progressive hedging based solution algorithm by first utilizing the classical progressive hedging approach and then leveraging the Frank–Wolfe algorithm for improving the solution quality. In both versions, we extend these algorithms to the risk-averse setting and present various computational enhancements. Our results on the IEEE 118-bus instances demonstrate the impact of adopting a risk-averse approach compared to risk-neutral and deterministic alternatives with a better worst-case performance, and highlight the value of integrating flexible generation and contextual information with resilient maintenance and operations schedules leading to cost-effective plans with less component failures. Furthermore, our solution algorithms provide good quality solutions in significantly less time compared to the off-the-shelf solver, where the Frank–Wolfe version of the algorithm is capable of finding optimal solutions in majority of the test instances.