Analysing Uncertainties in Offshore Wind Farm Power using Measure Correlate Predict Methodologies.

This study builds on recent analysis (Mifsud, et al., 2018) in which uncertainties in predicting long term wind speeds using different Measure Correlate Predict methodologies were compared. These included the Simple Linear Regression, Artificial Neural Networks, Decision Trees and Support Vector Regression. The current work focuses on the uncertainties resulting from the different MCP methods to predict the power and energy yield from a wind farm. The analysis is based on a case study that utilizes short-term data acquired from a LiDAR system deployed in a coastal site in the northern part of the island of Malta and long-term measurements from the island’s airport. LiDAR measurements at various heights, ranging from 10 to 200m are considered. A hypothetical offshore wind farm layout consisting of 50 turbines rated at 5-MW each is assumed. The wind farm faces the north-west direction, which is the prevailing wind direction for the island archipelago. The turbines are placed in rows of ten wind turbines each, with a cross-wind inter-turbine spacing set at five rotor diameters and a downwind spacing of eight rotor diameters between successive rows. The predicted power output from the various MCP methodologies is compared to the actual power output obtained directly from the input of LiDAR data to establish which MCP methodology best predicts the output power. The different MCP methods, that were implemented in Matlab®, are used to predict the absolute wind speed, while wind direction is predicted using the same regression methodology but, using the vector components of the wind. The wind direction is then calculated from the predicted vector components. The power output from the wind farm is predicted by inputting wind speed and direction into WindPro®, using a Risø/EMD (N.Ø. Jensen) wake model and a standard 5-MW wind turbine. The wake model introduces a more realistic value of the power output for the wind farm, as it includes the shadowing effects on the wind turbines and the wind direction. The power is predicted for a range of wind turbines (from 10 to 50, in steps of 10) and at hub heights of 30, 80, 100, 150 and 200m. The predicted power is compared to the power output generated from the actual wind and direction data by using the Mean Squared Error (MSE) and the Mean Absolute Error (MAE). This methodology will establish which combination of MCP methodology and wind farm configuration will have the least prediction error