Medium-Term Wind Speed Prediction using Bayesian Neural Network (BNN)

Abstract

Renewable energy has become an emerging source of daily energy utilization in recent years. Non-conventional sources are extensively applied in the smart grid due to their environment friendly and relatively easy maintenance. Wind energy unlike other conventional sources has drawn attention in terms of clean energy production. Due to unpredictable nature of wind, it is difficult to trade energy to the smart grid without any power loss. Variations in wind energy affect power scheduling, wind power generation, and energy management. Therefore, wind speed forecasting is an important tool to address such problems. Machine learning approaches have always been considered for accurate wind speed prediction. To evaluate the performance of machine learning algorithms, several models have been tested to achieve precise prediction. Amongst these several models, Neural Networks perform best and optimizes the prediction at its maximum. Apropos, in this paper, Bayesian Neural Network (BNN) is used for predicting medium-term wind speed on different time horizons. The input for training purposes is taken from Numerical Weather Prediction (NWP) model and sifted as per the model’s requirement. After successive training, it is evident from the percentage Mean Absolute Percentage Error (MAPE) and Normalized Mean Absolute Error (NMAE) criterion that BNN has achieved good accuracy as compared to Least Absolute Shrinkage and Selection Operator (LASSO). Ultimately, the proposed model has shown that it can bring precision and accuracy for prediction and can be applied for other renewable sources as solar and water as well.