Comparison and Evaluation of Methods for a Predict+Optimize Problem in Renewable Energy

• DOI: 10.48550/arxiv.2212.10723
• Publication Date: 2022-01-01
• Authors: Christoph Bergmeir; Frits de Nijs; Abishek Sriramulu; Mahdi Abolghasemi; Richard Bean; John T. Betts; Quang Neo Bui; Nam Trong Dinh; Nils Einecke; Rasul Esmaeilbeigi; Scott Ferraro; Priya Galketiya; Evgenii Genov; Robert E. Glasgow; Rakshitha Godahewa; Yanfei Kang; Steffen Limmer; Luis Magdalena; Pablo Montero‐Manso; Daniel Peralta; Yogesh Pipada Sun...; Alejandro Rosales...; Julian Ruddick; Akylas Stratigakos; Peter J. Stuckey; Guido Tack; Isaac Triguero; Rui Yuan
• Keywords: Benchmark (surveying); Microgrid

Algorithms that involve both forecasting and optimization are at the core of solutions to many difficult real-world problems, such as in supply chains (inventory optimization), traffic, transition towards carbon-free energy generation battery/load/production scheduling sustainable systems. Typically, these scenarios we want solve an problem depends on unknown future values, which therefore need be forecast. As problems their own right, relatively few research has been done this area. This paper presents findings ``IEEE-CIS Technical Challenge Predict+Optimize for Renewable Energy Scheduling," held 2021. We present a comparison evaluation seven highest-ranked competition, provide researchers with benchmark establish state art benchmark, aim foster facilitate The competition used data from Monash Microgrid, well weather market data. It then focused two main challenges: renewable production demand, obtaining optimal schedule activities (lectures) on-site batteries lead lowest cost energy. most accurate forecasts were obtained by gradient-boosted tree random forest models, was mostly performed using mixed integer linear quadratic programming. winning method predicted different optimized over all jointly sample average approximation method.