Work

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

Year: 2023

Type: article

Abstract: The operation of multi-energy systems has to be optimized repeatedly, e.g., to react to changing energy prices. Thus, operational optimization problems need to be solved in a reliably short time. Reliably short computations are challenging for optimizing multi-energy systems due to complex time-coupling constraints. These time-coupling constraints reflect effects such as component start costs or energy storage. However, time-coupling constraints increase the computational effort. Here, we present a decomposition method to solve the operational optimization using artificial neural nets efficiently. The method decomposes the operational optimization into single-time-step optimizations. The single-time-step optimizations incorporate predictions from artificial neural networks trained on long-term operational optimizations. In two case studies, the method provides high-quality solutions for all operational optimization problems in less than 2 min. The method is significantly faster up to a factor of 375 than directly solving the operational optimization problem while practically retaining the quality of the solution.

Source: Computers & Chemical Engineering

DOI: 10.1016/j.compchemeng.2023.108208

Authors: Andreas Kämper, Roman Delorme, Ludger Leenders, André Bardow

Institutions: ETH Zurich, RWTH Aachen University

Language: English

Topic: Integrated Energy Systems Optimization

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Affordable and clean energy

Open Access status: hybrid

Estimated APC paid: $4,070