Considering Uncertainty in Energy System Optimisation

Abstract

In energy system optimisation, there is a tendency towards working with models that feature perfect foresight. On the other hand, devices in today´s energy systems have been operational for a long time, e.g. one third of home heaters in Germany are older than 20 years, some German coal power plants from the 1950s are still in operation, and - more anecdotally - according to the company that built and installed it in the Zurich town hall, the world´s oldest water-water heat pump from 1936 is still operational. Thus, it is rather optimistic that a prognosis over the complete operational period can be made with good accuracy. In fact, in research it is often correctly stated that scenarios are created rather than prognoses are made. However, if decisions are to be taken, a prognosis is needed, and the further that prognosis looks into the future, the more uncertainty it has.

The issue can be addressed by considering multiple scenarios for the future, instead of just a single one. Options include a sensitivity analysis to test the dependency of the result on the input parameters, sometimes in the flavour of Monte-Carlo simulation. Stochastic programming and robust optimisation are two commonly used ways to do so already in the optimisation process.

This contribution gives examples on how to implement these types of multiscenario approaches into energy system optimisation that is based on linear programming. It also comments on the generation and selection of scenarios, that can than be used for these methods. Besides manually picking optimistic and pessimistic extremes, we discuss the generation of self-consistent time-series using Markov chains. While the methods can be applied in multiple fields, we use charging of battery electric vehicles (EV) as an example.