Test sensitivity and weather risk
A sensitivity study asks how a saved design’s result changes when an assumption changes. A weather ensemble asks how that result varies across the defined weather samples. Keep those two sources of change distinct.
Use this after a baseline Plant simulation. A sensitivity plot cannot correct an unexplained load profile or an unsuitable weather year.
Prepare the study
You need a saved design version, an axis that applies to its installed components, a range of values and the required weather years. The workbench offers applicable axes for the chosen version. Do not infer support for a parameter simply because it exists elsewhere in the editor.
Run a sweep
- Open Sensitivity explorer from the plant overview.
- Select the design and version.
- Choose the axis and review its range and steps. State the engineering question, such as whether a decision changes over a credible fuel-cost range.
- In Weather, choose the mode, start and end years, trial count and seed where required.
- Check that the required years are cached. Acquire missing years in Weather.
- Read the stated simulation budget. A larger grid and more samples increase the work.
- Select Run sweep and wait for the completed run.
- Inspect the verdict, chosen metric and the response across the axis. Pin points for comparison and use the comparison CSV when needed.
Interpret the risk figures
| Figure | Meaning within the selected sample set |
|---|---|
| Mean | The arithmetic average. |
| P50 | The median: half the values are at or below it. |
| P90 cost | A high cost percentile; it is not a 90% energy-exceedance statement. |
| CVaR | A mean over the selected adverse tail, as defined by the model. |
| Unmet-load metric | Supply shortfall under the modeled dispatch and inputs. |
Percentiles describe the supplied model and sample population. They are not a guarantee of future weather, prices or service reliability. State the axis, window, trials and seed when sharing a figure.
Check the shape of the result
A flat response can be real: the varied component may not bind in that design. First confirm that the axis applies and the range changed. A discontinuity can indicate a dispatch or feasibility threshold. Inspect nearby points and the operating results before interpreting it as an error.
Do not treat failed samples as valid zero-cost or zero-shortfall outcomes. Read the failure information. If required years are missing, fetch them rather than silently shortening the comparison window.
The explorer supports pinned point comparison. Deferred chart types such as hourly fan charts and boxplots are not prerequisites for this workflow and are not promised here.
Retain the evidence
Record the baseline version, sweep run, axis values, weather window, seed and assumptions. Keep the baseline and alternatives on a consistent basis.
Use Sizing and optimization to search designs under an objective. Use Financial model to assess a selected design’s business assumptions. This guide follows the September 2026 development build.