How do different optimisation strategies affect the development of transformation paths in myopic optimisation?

Kurzfassung

Energy-intensive industries face strong economic and ecological pressure, e.g. through the transition to low CO2 emissions energy systems. These challenges are exemplified by the metalworking industry. Transformation paths, which are defined as investment decisions made over an time horizon, can support this transition. Energy system optimisation models (ESOMs) enable the development of such transformation paths while considering both economic and ecological objectives. This work investigates how different optimisation strategies influence the development of transformation paths.

ESOMs can be classified by the degree to which future events are taken into account. In myopic optimisation, the time horizon is divided into shorter optimisation periods, and only information from the current and previous periods is considered. This better reflects the short-term investment behaviour of investors. At the same time, a distinction can be made between single-objective optimisation, which focuses on one optimisation objective, and multi-objective optimisation, which considers several optimisation objectives. In this work, four optimisation strategies for the development of transformation paths are compared. The comparison is carried out qualitatively and quantitatively with respect to total costs, total CO2 emissions, deviation of transformation paths, and computational time. The optimisation strategies considered are: linear optimisation, linear optimisation under consideration of the cap-and-trade system of the Emissions Trading System, - constraint-based Pareto optimisation, and evolution-based Pareto optimisation. Three companies from the metalworking industry are analysed as exemplary case studies. To this end, a Python based myopic optimisation algorithm built on MTRESS and oemof.solph has been developed. A time horizon of 25 years with optimisation periods of five years are considered, and projections for the relevant parameters are specified for this period.

Linear optimisation with consideration of the Emissions Trading System has only a minor impact on the transformation paths compared to the cost optimisation of the linear optimisation. The end date of the free allocation of CO2 certificates, whether it is 2038 or 2050, does not influence investment decisions; it only affects the number of CO2 certificates bought or sold. When CO2 emissions are included as an optimisation objective in multi-objective optimisations, the flexibility in choosing technology combinations increases, and the selection via multi-criteria decision making methods strongly affects the optimisation outcomes. For two optimisation objectives and linear cost assumptions, the -constraint-based Pareto optimisation yields a substantial reduction in computational time. At the same time, evolution-based Pareto optimisation offers more flexibility: non-linear cost curves and three optimisation objectives were successfully implemented. The comparison of the non-dominated sorting particle swarm optimization (NSPSO) and non-dominated sorting genetic algorithm-II (NSGAII) algorithms shows that a Pareto front could only be obtained with NSGA-II. Future work may focus on run-time reduction, additional optimisation objectives, validation of the models, and applications to further energy systems.