Effect of High-Resolution Data Input on Wind Speed Prediction using Machine Learning Algorithms

Wind speed anticipation plays an indispensable role in the operation and planning of wind generators in evolving power systems. An appropriate wind speed anticipation model can be an economical option to increase the use of sustainable wind energy. Nowadays, Machine Learning algorithms are frequent in wind prediction models, as described in the existing literature. However, a gap is identified in comparing the performance of wind prediction models for different temporal resolutions of input data sets. The effect of temporal resolution on the model’s performance is investigated in this paper using Recurrent Neural Network-Long Short Term Memories (RNN-LSTMs) and Auto-Regressive Integrated Moving Average (ARIMA). The performance is evaluated by different estimation parameters and correlation analysis among the predictor variables. Obtained results show that RNN-LSTMs outperform the ARIMA models and multivariate LSTMs are more effective than univariates on off-shore and on-shore locations. Increasing temporal resolutions of input data enhance the model’s performance.