A Moving Window Method for Time Series Optimisation, with Applications to Energy Storage and Hydrogen Production

Temporal decomposition methods aim to solve optimisation problems by converting one problem over a large time series into a series of subproblems over shorter time series. This paper introduces one such method where subproblems are defined over a window that moves back and forth repeatedly over the length of the large time series, creating a convergent sequence of solutions and mitigating some of the boundary considerations prevalent in other temporal decomposition methods. To illustrate this moving window method, it is applied to two models: an energy storage facility trading electricity in a market; and a hydrogen electrolyser powered by renewable electricity produced and potentially stored onsite. The method is simple to implement and it is found that for large optimisation problems, it consistently requires less computation time than the base optimisation algorithm used in this study (by factors up to 100 times). In addition, it is analytically demonstrated that decomposition methods in which a minimum is attained for each subproblem need not attain a minimum for the overall problem.