Abstract:
In this masters thesis, two type of Artificial Neural Networks, Feed Forward Back Propagation
Neural Network (FFBP NN) and Nonlinear Autoregressive Neural Network (NAR NN) are
used for wind speed estimations and predictions. Feed Forward Back Propagation Neural
Network is used to estimate wind speed based on three meteorological parameters namely,
Temperature, Pressure and Humidity. While Nonlinear Autoregressive Neural Network is used
to predict upcoming wind speeds without any input parameters. Feed Forward Back
Propagation Neural Network with Levenberg training algorithm needs input and target
parameters for its training. In this study an additional input parameter known as humidity is
considered and results are compared with existing study. Input parameters that are used for
training are Temperature, Pressure and Humidity while target values are of wind speed.
Networks output value is compared with real time data of wind speed values for computation
of performance parameter. 70% of data is used as training data and remaining 30% data is used
for validation and testing. In case of Nonlinear Autoregressive Neural Network, the network
is trained by taking previous wind speed as input and data point next to those inputs as target.
The number of previous data points that are taken as input are dependent on the set delay value.
Both networks are trained with real data and results showed an improvement in accuracy due
to consideration of additional input parameter. This study used multilayered Artificial Neural
Network for medium to long term wind speed predictions. Daily, monthly and yearly wind
speed predictions are part of this study. Weibull analysis that is most common and popular
wind speed estimation approach is also discussed in this study. This study contains a brief
review of Weibull curve analysis and its limitations. Also the improvement in results due to
consideration of an additional parameter in comparison with an existing study is discussed.
Matlab is used for building Neural Architecture and running algorithms.
