MCP
The MCP (Measure-Correlate-Predict) module is for long-term correction of measured wind data on site and based on the correlation with long-term reference data. The module includes slicing and dicing, automatic time offset, comparison graphs and an uncertainty calculation.
-
Long-term correction and more
Long-term correct local measurements with a long-term reference, automatically applying time-shifts and filling gaps -
Multiple Calculation Models
Not all data is similar. Different calculation models makes it possible to select the calculation model that best suits your data -
Easy comparison of choices
See how different data sources and calculation models influences the long-term correction and pick the best setup -
Pick the best long-term reference
Evaluate different long-term references, scale multiple long-term references, distance weight multiple long-term references -
Create wind statistics and time series
Use the results to calculate AEP in PARK, or to create a resource map
Long-term correction
Long term correction is an essential part of calculating a realistic long-term AEP estimate. The MCP module in windPRO (Measure-Correlate-Predict) offers long-term correction of local measurements, predicting the future wind with ample support tools, tables and graphs. The module was massively updated in windPRO 3.2 and 3.3, introducing many handy tools such as:
- Uncertainty calculation based on selected data (correlation, length of concurrent data, wind index in concurrent period and variability)
- Automatic filtering of time offset of long-term datasets
- Automatic veer correction of local or reference time series.
- Slice data and test and train different calculation models to see which setup performs best.
- Ability to resample reference datasets to the temporal resolution of the local dataset. For instance, turn a 1-hour resolution long-term dataset into a 10-minute dataset, matching the local measurements. This increases the amount of data for training the model, improving prediction results.
- Create a distance weighted average of multiple long-term references using a Scaler.
- Downscale multiple Mesoscale datasets to the same point as the local measurements before creating a long-term corrected wind statistic. This is useful when the reference data points are located in areas with different roughnesses or located on different parts of a hill.
- Scale local measurements to be long-term representative. This maintains the direction distribution and dynamic nature of the local data, compared to doing a regular long-term correction.
Multiple Calculation Models
Four different models are available in the MCP module:
Simple Speed scaling is the default reference model, taking the ratio between local and reference mean wind speed and multiplying each reference wind speed time step. It is the simplest model, useful for comparing the improved performance of other models.
Regression analysis of the data establishes regression lines for each direction sector. Inspect the fit directly through an animated graph and adjust a wide range of parameters to provide a better fit. The regression tool handles both linear regression, and also higher order polynomials for modelling wind speed and wind veer.
Matrix – models the changes in wind speed and wind direction through a joint distribution fitted on the ‘matrix’ of observations of wind speed and wind direction changes. Use either polynomials fitted to the data statistics or – where appropriate – use the measured samples directly when doing the Matrix MCP. Compared to the Regression model, Matrix handles turning of the wind smother as it allows for transporting data from one direction sector to another.
Neural network – The Artificial Neural Network (ANN) MCP method establishes transfer functions from reference data to measurement data by using ANN networks trained with reference data as input and measurement data as output. Once the networks are trained (with backpropagation algorithm) they can be used to predict wind speed and direction based on the reference data.
Compare choices
It’s easy to compare consequences of the selected dataset, models and model configurations on different levels:
Session concept: Makes it possible to organize MCP calculations in distinct sessions, allowing for comparing results across sessions with completely different setups. Each setup may use different short term and long term data, different models, filters and settings. The consequences of these choices can be inter-compared in terms of the predicted Mean Wind Speed, AEP Uncertainty, Concurrent Wind Index and Wind Speed Correlation. This is also documented in a report.
Model comparison overview: Within each session, different MCP models (Matrix,Neural Network etc) can be compared against each other to find the best setup for the selected data. This includes metrics like Mean Bias Error, Mean Absolute Energy Error, Root Mean Square Energy Error, Correlation, Kolmogorov-Smirnov and more.
Compare measured with predicted within each model: View e.g Diurnal and Monthly Wind Speeds, Wind Speed and Direction Frequencies in graphs.
Pick the best long-term reference
With the vast number of long-term reference data-sets available today, picking the right one is increasingly important when long-term correcting measurement data.
The MCP module offers tools for comparing the selected long-term reference with other alternative long-term references, helping you identify trending behavior and ultimately selecting the best long-term reference dataset. The impact on AEP caused by choosing an alternative long-term references dataset can be evaluated using the “Compare to other references” tool, which also takes the local data measurement period into account.
You can easily download alternative 20-year reference datasets directly within MCP through windPRO Online data service. Alternatively, data can be loaded from your own METEO object or as a wind energy index from a PERFORMANCE CHECK session.
Create wind statistics and time series
The output from the MCP analysis is a wind statistic generated with WAsP or a time series of long-term corrected wind data. The time series are automatically loaded into a new Meteo object, which can be used directly in a PARK calculation or for a wind resource map calculation.
The reporting consists of two parts:
Session overview– The different session results are compared based on selected model within each session. The uncertainties based on selected data for each session is a part of the report, where each parameter included in the uncertainty calculation is shown.
Session details– For each session, the tested models are reported for comparison.