Probabilistic prediction of wind speed using an integrated deep belief network optimized by a hybrid multi-objective particle swarm algorithm

Abstract

An improvement in wind speed prediction is highly necessary for estimating the accuracy as well as stability of wind power. In this work, we proposed probabilistic forecasts of wind speed for predicting the short-term wind speed intervals. The optimal model has been designed by considering three different modules such as data decomposition, prediction, and optimization. Variational-Mode-Decomposition (VMD) is utilized for decomposing the primary time series data into a suitable number of modes followed by the Deep Belief Network (DBN) for probabilistic wind sped prediction. Here the Gauss-Bernoulli restricted Boltzmann machine (GBRBM) and Bernoulli-Bernoulli RBM (BBRBM) are combined in the DBN where the GBRBM is utilized as the initial RBM to convert the continuity feature of the source data into a binomial distribution feature. Multi-kernel-random-vector-functional-link-network (MKRVFLN) is employed here as supervised learning in DBN to avoid long execution time and get the model into local optima. In the model optimization, a hybrid multi-objective Sine-Cosine particle-swarm-optimization (MOSCPSO)is used to optimize the DBN parameters for obtaining high accuracy and strong stability output simultaneously. It determines the wind speed at 95%, 90%, 85%, and 80% prediction interval nominal confidence (PINC). To validate the proposed model and comparing with other benchmark prediction techniques, the data are taken from the wind farm located at Sotavento, Spain, at different time horizons (30 min–1 h) in different seasons. The results obtained demonstrate that the proposed technique outperforms the further existing model on the basis of prediction accuracy and stability.