Batch Acquisition for Parallel Bayesian Optimization—Application to Hydro-Energy Storage Systems Scheduling

Bayesian Optimization (BO) with Gaussian process regression is a popular framework for the optimization of time-consuming cost functions. However, joint exploitation BO and parallel processing capabilities remains challenging, despite intense research efforts over last decade. In particular, choice suitable batch-acquisition process, responsible selecting promising candidate solutions batch-parallel evaluation, crucial. Even though some general recommendations can be found in literature, many its hyperparameters remain problem-specific. Moreover, limitations existing approaches terms scalability, especially moderately expensive objective functions, are barely discussed. This work investigates five algorithms based on different processes, applied to optimal scheduling Underground Pumped Hydro-Energy Storage stations classical benchmark Efficient management such energy-storage units requires able find very restricted time comply responsive energy markets. Our experimental results show that considered methods, batch four candidates good trade-off between execution speed relevance candidates. Analysis each method’s strengths weaknesses indicates possible future directions.