Kusiak.qxd_Outsourcing_book_temp.qxd 26/02/2010 10:16 Page 66
Short-term Prediction of Windfarm Power – A Data-driven Approach
Table 1: Prediction Performance Comparison for Six Different Data-Mining Algorithms
Power Output Prediction Rotor Speed Prediction
Algorithm Average Average Mean Mean Average Average Mean
Observed Predicted Absolute Relative Observed Predicated Absolute
Power Power Error (kW) Error (%) Rotor Rotor Error (rpm)
Ouput Output Speed Speed
Random forest 573.90 573.90 28.60 21.07 15.66 15.67 0.18
Neural network 576.03 576.03 8.03 4.95 15.64 15.64 0.18
Boosting tree 575.65 575.75 34.55 54.71 15.62 15.62 0.27
Support vector machine 547.47 588.82 23.71 50.82 15.76 13.81 2.67
Generalised additive 576.03 576.03 11.13 20.21 15.64 15.64 0.19
k-nearest neighbours 574.47 573.97 28.94 9.59 15.76 15.75 0.54
Figure 3: Short-term Prediction of Wind Power Using a Virtual Turbine Figure 1A shows the first 200 observed and predicted (10 minutes
ahead) wind power values. It is easy to see that the observed and
Non-controllable parameters
predicted wind farm power values are almost identical. Figures
WS(t), WS(t-1), WS(t-2), WS(t-6)
1B–1D show the first 200 observed and predicted wind farm power
Virtual model
PO(t)
(power output)
Controllable parameters values over 40-minute, one-hour and three-hour future time
BPA(t), BPA(t-1), BPA(t-2)
intervals, respectively.Parameters of
interest
GT(t), GT(t-1), GT(t-2)
Virtual model
Integrated Model
Performance parameters RS(t)
(rotor speed)
PO(t-1), PO(t-2) Integrating the kNN model with the wind speed time series model for
power prediction was inspired by wind industry practices. The
RS(t-1), RS(t-2)
prevailing approach to windfarm power prediction is to forecast
the wind speed and use it to compute power based on a pre-defined
Figure 4: Prediction Performance Using a Virtual Turbine
power curve function.
A
1,400
The important predictors {y(t), y(t-T), y(t-2T), y(t-3T), y(t-4T), y(t-5T)}
of the time series model for wind speed were selected. Five data-
mining algorithms {y(t), y(t-T), y(t-2T), y(t-3T), y(t-4T), y(t-5T)} that
1,200
appeared the most promising were used to construct the 10-minute
time series prediction model for wind speed. The SVM algorithm
outperformed the other four algorithms. The MLP and REP tree
1,000ower output (kW)
P
algorithms performed worst. The SVM algorithm was selected to
build the 10-minute time series wind speed prediction model.
800
The kNN is a machine-learning algorithm predicting the unknown value
1 11213141516171
for an instance (here power) using the kNN of the data supporting that
Observed power output Predicted power output
instance. The predicted value is associated with the majority votes of
these k neighbours. Euclidean distance is often used to measure the
B
20.5
closeness of the data points. Previous research has shown that the kNN
model is quite accurate for computing windfarm power given the wind
speed as input.
20
The computational results reported in this article have shown that
the 10-minute time series model for wind power prediction
19.5
Rotor speed (rpm)
outperforms the integration model. Figure 2 illustrates the observed
and predicted wind power using integrated time series models 10 to
60 minutes ahead.
19
1 11213141516171
Although the kNN model and the 10-minute wind speed time series
Observed power output Predicted power output
prediction model perform well individually, the integrated model
A: Observed and predicted wind power (10-second intervals); B: Observed and predicted rotor speed
produces a larger error when predicting future power. This could be
(10-second intervals).
due to the fact that the power is a cubic function of the wind speed.
of the hourly time series model. Here, the MLP algorithm outperformed Additionally, the wind speed in the kNN model is too sensitive as a
the other four algorithms. The REP algorithm performed the worst. Based predictor for wind farm power, and thus it might lead to a worse
on its performance, the MLP algorithm was selected to build the hourly prediction for the integration model. The integration of the two
time series model 1 for the mean hourly wind farm power prediction. models did not improve prediction accuracy.
MODERN ENERGY REVIEW VOLUME 2 ISSUE 1
66
Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52  |  Page 53  |  Page 54  |  Page 55  |  Page 56  |  Page 57  |  Page 58  |  Page 59  |  Page 60  |  Page 61  |  Page 62  |  Page 63  |  Page 64  |  Page 65  |  Page 66  |  Page 67  |  Page 68  |  Page 69  |  Page 70  |  Page 71  |  Page 72  |  Page 73  |  Page 74  |  Page 75  |  Page 76  |  Page 77  |  Page 78  |  Page 79  |  Page 80  |  Page 81  |  Page 82  |  Page 83  |  Page 84  |  Page 85  |  Page 86  |  Page 87  |  Page 88  |  Page 89  |  Page 90  |  Page 91  |  Page 92  |  Page 93  |  Page 94  |  Page 95  |  Page 96  |  Page 97  |  Page 98  |  Page 99  |  Page 100  |  Page 101  |  Page 102  |  Page 103  |  Page 104  |  Page 105  |  Page 106  |  Page 107  |  Page 108