Analysis and Modelling of Meteorological Sensory Signals for One-Hour-Ahead Wind Speed Forecasting using Dynamic Neural Networks.

Predicting the expected wind speed can help to estimate the wind power generation. Hence, it is important to obtain forecasting as much as close to actual wind speed. One of the importance of estimating wind power generation is to inform the consumers of the corresponding wind farm area to control their electricity consumption. To model and predict the wind speed, it is also required to include other weather-based factors such as relative humidity, temperature, and so on due to their strong relationship between wind speed. Therefore, in this paper, the first meteorological sensors’ data are analysed by computing the mutual information score of the sensors and as a result of this irrelevant sensors’ data are removed from the system inputs. Then, the sensors’ data were reduced by compressing these data by using principal component analysis. After obtaining the final processed input features, four different dynamic neural networks are used to model the temporal relationship between processed sensors’ data and one-hour-ahead wind speed. Consequently, the time delay radial basis function network outperformed other dynamic networks.