Online microgrid energy generation scheduling revisited: the benefits of randomization and interval prediction

Energy generation scheduling is a fundamental problem in microgrid design that determines the on/off status and the output level of energy sources with the goal of minimizing the cost and satisfying both electricity and heat demand. The uncertainty in both renewable generation and microgrid demand makes the problem drastically different from its counterparts and in traditional power systems and brings out the essential need of online algorithm design. In the literature, an online deterministic algorithm called CHASE has achieved a competitive ratio of 3, which is the best possible among deterministic algorithms. In addition, it has been shown the accurate prediction can improve the performance. This paper revisits the problem by investigating the benefits of randomization and interval prediction, i.e., relaxing accurate prediction assumption by considering an interval of valid ranges for future demand. We propose rCHASE, a randomized algorithm that achieves competitive ratio of around 2.128, improving beyond the best deterministic algorithm. Then, we propose iCHASE, an interval prediction-aware algorithm that is built upon rCHASE and a new extension we developed for the classic ski-rental problem. Our trace-driven experiments demonstrate that iCHASE outperforms CHASE; the average cost reduction of iCHASE is 15.85%, while CHASE reduces the cost by 9.1%.