Towards Improving Unit Commitment Economics: An Embeddable Energy-and-Reserve Tailored Predictor

Abstract

Generally, system operators conduct the unit commitment (UC) in a predict-then-optimize process: the renewable energy source (RES) availability and system reserve requirements for the next day are first predicted; given the predictions, system operators optimize the UC model to determine the economic operation plans. However, the predictions within this predict-then-optimize process are considered raw because they are usually generated for various downstream applications, including but not limited to UC. Indeed, UC economics can benefit if the raw predictions can be further tailored to assist UC in making the best possible economic operation plans against the next-day RES availability and system reserve needs. To this end, this paper presents an embeddable RES-and-reserve tailored predictor for UC. The tailored predictor can be derived by solving a bilevel mixed-integer programming model: the upper level trains the RES-and-reserve tailored predictor based on its induced operating cost; the lower level, with given tailored predictions, mimics the system operation process and feeds the induced operating cost back to the upper level; finally, the upper level evaluates the training quality according to the operating cost. Through this training, the RES-and-reserve tailored predictor learns to customize the raw predictions into cost-oriented predictions. Moreover, the trained tailored predictor can be embedded between the raw predictor and UC blocks of the existing predict-then-optimize process, augmenting UC to a tailored-predict-then-optimize block that provides tailored predictions and operation plans with improved economics. Numerical case studies using real-world data illustrate the economic and practical advantages of the RES-and-reserve tailor over deterministic, robust, and stochastic methods.

Keywords: Unit commitmentRenewable energy sourcePrescriptive analyticsPredict-then-optimizeBilevel mixed-integer programming