Garbage in, garbage out. A phrase that underscores the relationship between the quality of input data and the quality of output data. This is a fundamental principle in many fields, including production cost modelling.

A goal when operating a production cost model (PCM) is to use input data that is as high quality as possible, with the hope that it will produce output data representative of reality. Not only can it be difficult to obtain high quality data, but it can be difficult defining what “high quality” even means. There is no perfect dataset, and importantly, whatever dataset that is chosen will propagate its biases forward into the results of the PCM. The difficulty in determining what “high quality is” is not to imply that “all datasets are okay”, but it does mean that it depends on what qualities you believe are most important to look for in a dataset.

In a PCM, while you can run any kind of scenario, the goal is to often to represent a “p50” or “normal” outcome. One of the core things to get correct is generation profiles.

Below are variety of ways to choose wind profiles where we note the downside to each:

• Take a “normal” weather year and run it through a power curve to create a synthetic generation profile. Downsides: this is costly, and unless you’re conscientious to use the same year throughout the simulation, you break the correlations between weather, generation and load, as well as the inter-regional weather correlations. Note: ideally, weather year should be considered for most approaches. In addition to this, you’ll probably want some kind of regional aggregate; for example, the weather conditions at the coordinates (561688, -117.882815) cannot be used for a broader zone, which is comprised of many points on a map.  On the plus side, some things about synthetic profiles are easier to control than profiles derived from historical data – like ensuring that you have the correct capacity factor.
• Average the wind shapes between several years. Downside: this makes your new wind shape lose its volatility. There is no upside that makes this worthwhile. Don’t do this. An example of the issue is shown in figure 1 – a week in August 2025 shown in red, and week in August 2024 shown in blue, and the average between the two shown in dark green. The more years averaged together, the flatter the resulting profile. This is also visible in figure 2 even in 12x24 form – look at November 2023 & 2024, whose gen profiles mirror each other.
• Choose a year to represent “normal” and just use its generation. Downsides: it is typical for a year to have a couple “unusual” events. You can think about it as a Poisson process. While it might be rare to have a “weather event” on February 13^th for example, with enough coin flips, it is likely that there will be some weather event sometime during a longer timeframe. By choosing a full historical year as your 8760, you’re unavoidably choosing to systematically include the weather events of that year in every simulation year. Looking at Figure 2, for example, even accounting for capacity growth, March had pretty high wind generation. If 2025 was to be chosen as the representative year, we’d now have unusually high wind generation every March.
• Create a composite year, for example, by splicing together months that approximate normal. Downsides: this underestimates risk by biasing results towards the mean. The resulting capacity expansion will be more conservative. It can also be difficult to maintain generation’s correlation with other variables.

Figure 1 | PACE Wind Generation – One Week of August 2024 & 2025, and Average

Figure 2 | Historical PACE Wind Generation 12x24

Wind profiles are one of many inputs that go into Production Cost Modeling. At Energy GPS, the attention to detail around every input profile is valued by our client base.  For more information tied to our Consulting Services, please fill out the Contact Us form on the Energy GPS website or email us at [email protected].