WDS-Logo   Deutsch   English 
 
 
You are here: ⇒ Examples ⇒ MSSCP

Example For MSSCP (Multiple Synoptic Scale Correlate Predict)

WindDataSuite ("WDS" in the following) implements the new MCP method MSSCP (Multiple Synoptic Scale Correlate Predict). MSSCP has been developed by WindDataSuite and is based on an evaluation of the wind field variability on the synoptic scale.

All in the following described (and not described) separate processing steps, which are necessary for an MSSCP calculation, will be performed by WDS in each MSSCP session automatically in the background.

The time series data of the reference stations (long-term and short-term) will be processed with a digital low-pass filter thus eliminating all wave signals below the synoptic scale. Clusters will be determined from the resulting wind field dynamics according to the Navier-Stokes equations. The clusters comprise the entire measured dynamics and enter into the MSSCP technique as the matrix basic dimension.

MSSCP can use multiple reference stations, which can be weighted for the long-term extrapolation appropriate to their distances, and/or in the particular wind direction sectors appropriate to their positions relative to the site of the short-term wind measurement ("measurement station" in the following), and/or in the particular wind direction sectors appropriate to their correlations with the measurement station data. The long-term time range and the short-term time range common with the measurement time range at the measurement station can be selected for each reference station individually. Long-term time range and short-term time range may include each other, overlap, or exclude each other.

MSSCP searches in each reference wind direction sector within each cluster for the one phase shift between the time series which shows the highest positive correlation. Thus, the real time delay of the wind transport as well as the different correlations within the wind field due to the actual wind field dynamics will be captured. The regression parameters will be calculated for each cluster and each reference wind direction sector individually. Inappropriate/invalid regression parameters will be conveniently replaced/interpolated.

Every MCP method (including MSSCP) produces more or less erroneous predicts. This is an inevitable problem inherent to the system of every statistical model. The main problem of, e.g., regression-based MCP methods is that the long-term extrapolations at the measurement station mostly produce wind speed frequency distributions with an overestimated Weibull shape parameter. Hence, the resulting mean wind power density (WPD) will be underestimated, though the error in the resulting mean wind speed may be less by far. In contrary, matrix-MCP methods, which are based on a direct transformation of the respective short-term and long-term wind speed frequency distributions, may produce long-term extrapolated wind speed frequency distributions at the measurement station with a underestimated Weibull shape parameter. The resulting wind power density then will be clearly overestimated.

WindDataSuite has performed a lot of hindcast tests with different MSSCP variants and their combinations: regression with orthogonal distances instead of with vertical, regression through the origin of co-ordinates, including standard deviations, different regression models in different wind speed classes, and many more. All these statistical variations have not improved the results really (actually, most of them have deteriorated the results). Instead, they just cause confusion due to the numerous degrees of freedom and confront the user with the problem to have to make a decision (on basis of what?) on which of the methods to be used.

A really essential improvement of the extrapolation results, however, is achieved by the MSSCP technique in that MSSCP analyzes and embeds the actual wind field dynamics on the synoptic scale and thereby resolves the variability of a physical process which is essential for the accuracy of the regression model.

With reference to the above-mentioned "more or less erroneous", MSSCP definitely contributes to the "less": the reduction of the extrapolation errors by MSSCP is enormous and is for the wind power density resulting from the extrapolations compared to a pure directional MCP method about 50% (averaged over numerous hindcast tests).

In the following example, an hindcast was performed with MERRA-2 (Modern Era Retrospective-analysis for Research and Analysis - Version 2, NASA GEOS-5 model) re-analysis data at MERRA-2-point (J287,I306), 11.25E, 53.5N, as measurement station. As the short-term measurement, the time series data of the time range from 2015/08/01 to 2016/07/31 were selected. As long-term reference stations, the time series data of two DWD (Deutscher Wetterdienst) weather stations, Boltenhagen, 11.19E, 54.00N, and Schwerin, 11.39E, 53.64N, were selected, from 1995/01/01 to 2015/12/31 for the long-term time range and from 2015/08/01 to 2016/07/31 for the short-tem time range common with the short-term measurement.
The example illustrates the robustness of the MSSCP technique: area-averaged (and temporal smoothed) quantities will be predicted from point measurements. So, the correlations with the reference stations in particular wind direction sectors are very weak (coefficient of determination minimum R2min = 0.32 and R2min = 0.30 for Boltenhagen and Schwerin, respectively).
In the following figures, the real "measured" long-term time range from 1995/01/01 to 2015/12/31 at MERRA-2-point (J287,I306) is depicted in black, the long-term extrapolation with the directional MCP method in blue, and the long-term extrapolation with MSSCP in red.

Fig.1 Frequencies v Fig.2 Percentages WPD
Fig.1:  Frequency distributions of the wind speed
Fig.2:  Percentaged distributions of the wind power density
Fig.3  Sectorial frequencies v Fig.4 Sectorial percentages WPD
Fig.3:  Sectorial frequency distributions
of the wind speed
 
Fig.4:  Sectorial percentaged distributions
of the wind power density
 
  real measured directional MCP MSSCP
mean wind speed v (m/s) 6.9 6.9 6.9
mean wind power density WPD (W/m2) 321 303 320
 
;
Data range : 1 year for the short-term measurement (adjustable in WDS), sampling rate 1 hour
21 years for both of the long-term references (adjustable in WDS), sampling rate 1 hour
Processed channels : 1 height level of wind speed and wind direction, respectively
(arbitrary height levels are adjustable in WDS when height profiles are available)
Matrix : 2 reference sites (adjustable in WDS)
25 clusters for the synoptic scale per reference site (adjustable in WDS)
12 reference wind direction sectors per cluster (adjustable in WDS)
12 measurement station wind direction sectors per reference wind direction sector (adjustable in WDS)
4 wind speed classes per reference wind direction sector (adjustable in WDS)
Output:
31 measurement station wind speed bins (adjustable in WDS)
12 measurement station wind direction bins (adjustable in WDS)
Computing time : approx. 10 seconds
Processing time : approx. 1 minute