Synthesize Data Tab |
This tab of the Long Term Adjustments window allows you to perform the MCP process on speed, direction, and temperature data. Choose how you wish to synthesize the extended dataset, and the date range you wish to synthesize, then click Extend Onsite Dataset. A drop-down box lets you choose the algorithm you wish to use to synthesize the wind speed data.
A checkbox lets you choose whether to preserve the measured onsite speed and direction data in the extended dataset, or to overwrite it with data synthesized by the MCP process. If you check this box, Windographer will create an extended dataset consisting of all of the processed onsite data, overwritten wherever possible with MCP-synthesized data. Otherwise, Windographer will create an extended dataset consisting of all of the processed onsite data, with synthesized MCP data added in times steps in which only the reference dataset contained valid data.
Tip: Windographer will apply the special synthetic data flag to all synthesized data points.
After you create the extended dataset, the lower portion of the screen will show a comparison of the onsite data before and after the MCP process, so you can see its effects. Different settings will have different effects on the extended dataset, and some may introduce considerable distortions, so it is often worth creating the extended dataset multiple times using different algorithms and settings, and comparing the results.
You can complete the MCP process in one of two ways:
Whether you extend your data to the long term, or scale it to reflect long term conditions, you have performed a long term adjustment. That means you have adjusted your measured data to reflect the long term conditions you expect at the site. But should you scale or extend?
The answer depends on what you intend to do with the resulting data. Let's look at a typical case to explore the advantages and disadvantages of scaling versus extending. In the wind power industry, a typical onsite dataset might be a met tower dataset with a 12-month period of record, multiple measurement heights, and a 10-minute time step. A typical reference dataset might be computer-modeled reanalysis data covering, say, 25 years, at one or two heights above ground, hourly time steps, and without wind speed standard deviation data. The table below compares the properties of the extended and scaled datasets in such a case:
Property | Extended dataset | Scaled dataset |
---|---|---|
Period of Record (POR) | 25 years | 1 year |
Time Step | 60 minutes | 10 minutes |
MoMM Speed Matches Long Term? | Yes | Yes |
MoMM Temperature Matches Long Term? | Yes | Yes |
Speed Data Columns | One | Multiple at different heights |
Direction Data Columns | One | Multiple at different heights |
Temperature Data Columns | One | Multiple at different heights |
Wind Shear Information | No | Yes |
Turbulence Intensity Information | No | Yes |
If you intend to use the long-term adjusted dataset for time series energy modeling (in Openwind for example), you should scale the dataset because the scaled dataset retains the turbulence and shear data that improves the accuracy of such modeling, and it retains the full time resolution because the time step remains at 10 minutes.
On the other hand, if you intend to use the long-term adjusted data to create a TAB file or other such frequency distribution versus speed and direction, then the extended dataset will work just fine. Its coarser time resolution and lack of shear and turbulence information is of no consequence in that case.
Other more subtle factors can play a role in your decision as well. Keep in mind that the scaled dataset tells a fictional story, in the sense that although it purports to describe what happened in a particular period of time, in fact it describes a modified version of what actually happened. If the onsite data covers the year 2019 for example, and the long-term adjustment process scales the wind speeds by a factor of 1.05, then the scaled data set describes not the historical 2019, but rather a fictionalized version of 2019 made 5% windier than it truly was. That distortion has no negative consequence when using that data to predict the future energy production of a wind farm. But if you are doing a grid integration study or any kind of analysis that models actual historical periods, especially if it integrates other concurrent time series data like electrical load data, that distortion may invalidate the analysis.
By contrast, the extended dataset in that case would contain unmodified historical data in 2019, preceded by 24 years of synthetic data predicted by the MCP algorithms. That fidelity to the historical 2019 may be advantageous in some situations.
Click Test Uncertainty to open the Test MCP Uncertainty window, which performs a cross-validation exercise to estimate the uncertainty in the MCP process.
See also