Adaptive Rolling Horizon for operational optimization of multi-energy systems

Abstract

Operational optimization problems of multi-energy systems have to be solved repeatedly, e.g., to react to changing energy prices. Thus, operational optimization problems should compute fast. The computation time of operational optimization problems is, therefore, often reduced by the Rolling-Horizon method. The Rolling-Horizon method decomposes the original optimization problem heuristically into smaller subproblems. Each subproblem optimizes only the next few time steps of the whole time series. While reducing computation time, the Rolling-Horizon solutions are, in general, suboptimal for the original problem. For example, suboptimal decisions arise for start-ups and shut-downs of components in a multienergy system, if the subproblems are too short to contain sufficient information about future energy prices. Longer subproblems increase solution quality, but also computation time.To resolve this trade-off between solution quality and computation time, we propose the Adaptive-Rolling-Horizon approach. Adaptive Rolling Horizon adjusts the size of each subproblem based on the energy-price predictions and the current system state. Adaptive Rolling Horizon aims to keep the subproblems as short as possible to minimize computation time, but as long as necessary to resolve optimal decisions about start-ups and shut-downs. In a real-world case study, Adaptive Rolling Horizon reduces the computation time by up to two orders of magnitude while achieving a near-optimal solution. At the same time, Adaptive Rolling Horizon can improve the solution quality by more than 15 % compared to common Rolling Horizon. The results show that Adaptive Rolling Horizon can significantly accelerate the operational optimization of multi-energy systems while retaining high solution quality.

Identifying key environmental objectives for integrated process and fuel design

“…Note that the normalization variants lead to different ranges of values regarding the normalized solutions and, consequently, resulting δ -errors. To enable comparability between δ -errors calculated with different normalization variants, we use the relative δ -error δ rel introduced by Postels et al : 55 For each normalization variant, the δ -error is referenced to its maximum value that occurs if all but one, i.e. , 16 out of 17, objectives are omitted.…”

“…For energy system optimization, Postels et al analyzed the influence of normalization variants based on mathematical and environmental reference values on the identified key objective subset. 55 The authors show that, in their case study, mathematical normalization variants lead to similar results while the identified key objective subsets differed substantially for the environmental reference values.…”

Boosting operational optimization of multi-energy systems by artificial neural nets

“…An easy-to-implement approach to reduce computation time is the Rolling-Horizon method (Cao et al, 2019). For this reason, Rolling Horizon is often used in practice to reduce computation time (Bischi et al, 2017;Shin and Zavala, 2020;Kämper et al, 2021a) or to consider uncertainties (Gupta et al, 2016;Wang et al, 2015;Kopanos and Pistikopoulos, 2014). Rolling Horizon heuristically decomposes the original optimization problem into smaller subproblems along the time axis.…”

“…However, applying Rolling Horizon requires the selection of values for the parameters step size and foresight, which strongly influences the computation time and solution quality (Marquant et al, 2015). The https://doi.org/10.1016/j.compchemeng.2023.108208 Received 7 January 2023; Received in revised form 22 February 2023; Accepted 27 February 2023 following main trends occur (Kämper et al, 2021a): The larger the step size, the fewer subproblems to solve, but the longer the computation time of each subproblem. Thus, the choice of the step size results mainly in a trade-off in computation time.…”