Skip to content

Performance Metrics

This page summarizes the benchmarking results for WEC-Grid, evaluating scalability and performance across different grid sizes and WEC farm configurations using the PSS®E and PyPSA modelers.

Test Environment

  • Operating System: Windows 10
  • Processor: Intel i7 dual-core, 4 logical CPUs @ 2.6 GHz
  • Memory: 7.9 GB RAM
  • Threading: Linear algebra threading disabled (MKL/OPENBLAS/NUMEXPR = 1)

Each datapoint represents the average of 10 runs, with warm-up iterations discarded. This ensures stable and reproducible results.

Test Cases

  • IEEE standard grids: 14, 24, 30, 39, 96 (RTS), 118, and 300 buses. These are widely used benchmark systems in power systems research.
  • PSS®E: limited to IEEE 14–39 bus systems (software constraint).
  • PyPSA: ran the full set up to IEEE 300 bus system.

What this means: "Buses" represent electrical nodes in the grid (connection points for generators, loads, and lines). Scaling tests with different bus counts evaluate how runtime grows as the network size increases.

Results

Scaling with Grid Size

Simulation time grew approximately linearly with the number of buses for both modelers: - PSS®E: ~0.50 seconds per bus, intercept ~8.9 s, R² = 0.97.
- PyPSA: ~0.64 seconds per bus, intercept ~101.3 s, R² = 0.99.
- Memory demand: ~0.0002 GB per bus (linear scaling).
- CPU utilization: consistently >98%, confirming efficient single-core use.

What this means: Doubling the number of buses roughly doubles the runtime. Both solvers behave predictably, which is important for planning large-scale studies.

Simulation time vs. grid size

Scaling with WEC Farms

For the IEEE 39-bus system, simulations were run with 1–3 synthetic WEC farms: - PSS®E: ~30% fixed runtime overhead when enabling WECs, with little change beyond the first farm.
- PyPSA: smoother incremental growth (+3–6% across 1–3 farms).
- Memory overhead: negligible (<5 MB per farm).

What this means: Adding WEC farms increases runtime slightly, but overhead remains modest. The results show that scaling up the number of WECs is computationally feasible.

Simulation time vs. number of WEC farms

Regression Fits

Runtime fits the linear model:

\(T(N) \approx aN + b\)

where: - a = slope (runtime per bus) – how much extra time is needed for each additional bus.
- b = fixed overhead – baseline runtime even for the smallest system (startup and initialization cost).
- R² = fit quality – a value close to 1.0 means the linear model explains the data very well.

Backend Slope a (s/bus) Intercept b (s) R²
PSS®E 0.50 8.9 0.97
PyPSA 0.64 101.3 0.99

What this means: For example, with PSS®E, adding 10 more buses adds about 5 seconds to the runtime, while PyPSA adds about 6 seconds plus a higher startup cost. The very high R² confirms that scaling is highly predictable.

Conclusions

  • WEC-Grid introduces modest runtime costs compared to baseline power flow studies.
  • PyPSA: scales well to large grids (up to 300 buses), with small proportional WEC overheads.
  • PSS®E: faster for small grids, but shows a larger fixed overhead when WEC models are enabled.
  • Memory demand and CPU use remain efficient across both solvers.

Overall, WEC-Grid provides predictable, scalable performance for integration studies, supporting both small and large network sizes with minimal overhead from WEC modeling.