New study in @iScience_CP with #PyPSA-Eur combining modelling-to-generate-alternatives (MGA) with global sensitivity analysis (GSA) to show many ways to design cost-effective renewable electricity systems with robustness to uncertain technology costs.
We added technology cost projections as parametric uncertainty to the structural exploration of near-optimal solutions.
Like a previous study by @timtroendle, we used multi-fidelity surrogate modelling techniques using sparse polynomial chaos expansions and low-discrepancy sampling to manage over 50,000 optimisation runs effectively.
That approach allowed us to merge results from one simpler and another more detailed model. One covers 37 regions at 4-hourly and the other 128 regions at 2-hourly resolution over a full year.
Results and visualisations made possible with the applied methodology include ...
... the distributions of system cost, generation, storage, and transmission expansion in least-cost solutions.
... the sensitivity of capacities towards their own technology cost with uncertainty bands for the influence of the remaining cost parameters.
Note the varying width of uncertainty bands, the low uncertainty for transmission expansion needs, and the shape of the battery plot!
... sensitivity indices attributing output variance in capacities build and system cost to individual technology costs.
These show that the uncertainty of *electricity* system cost is largely driven by the investment cost of wind.
... fuzzy near-optimal cones identifying feasible alternatives common to all, few or no cost samples.
... the probabilistic near-optimal feasible space in two technology dimensions simultaneously (cross-section of a near-optimal cone for given epsilon). These outline constraints around extremising the capacity expansion of combinations of two technologies simultaneously.
The robust finding of our study is that there is consistent investment flexibility in shaping fully renewable power systems, even without availing of the many flexibility options offered through sector coupling (electricity-only) and perfect technology cost foresight.
This opens the floor to discussions about social trade-offs and navigating around issues, such as public opposition toward wind turbines or transmission lines. We offer a method to present a wide spectrum of expansion options that are feasible and within a reasonable cost range.
All code is, of course, open-source and available here:
🐍🧑🏫💻This semester I taught a new course about Data Science for Energy System Modelling, for which I built a website with energy-focused Python tutorials:
It includes hands-on introductions to various libraries useful for modelling energy systems and processing data: Python, numpy, matplotlib, pandas, geopandas, cartopy, rasterio, atlite, networkx, pyomo, pypsa, plotly, hvplot, and streamlit.
Topics covered include:
- time series analysis (e.g. wind and solar, prices load)
- tabular geographical data (e.g. location of power plants, LNG terminals, industrial sites)
- converting reanalysis weather data to renewable generation (e.g. ERA5)
Narrowly following least-cost energy system optimisation results risks inequitable outcomes.
In a paper on near-optimal trade-offs, I show for a renewable European power system that more regionally balanced expansion plans can be achieved at little extra cost below 4% ...
... and that completely autarkic solutions, without power transmission, appear much more costly.
Providing just a single least-cost solution underplays an immense degree of freedom when planning future energy systems.
There are many near-optimal alternatives with attractive properties like social acceptance due to less onshore wind capacity or limited grid reinforcement.
We systematically explored the decision space of a European power system model based on wind and solar that co-optimises generation, storage and grid infrastructure.
We look at how the capacities of each technology can deviate if the costs are epsilon % away from the optimum.