A VAE-Bayesian Deep Learning Scheme for Solar Generation Forecasting based on Dimensionality Reduction

Abstract

The advancement in distributed generation technologies in modern power systems has led to a widespread integration of renewable power generation at customer side. However, the intermittent nature of renewable energy pose new challenges to the network operational planning with underlying uncertainties. This paper proposes a novel Bayesian probabilistic technique for forecasting renewable power generation by addressing data and model uncertainties by integrating bidirectional long short-term memory (BiLSTM) neural networks while compressing the weight parameters using variational autoencoder (VAE). Existing Bayesian deep learning methods suffer from high computational complexities as they require to draw a large number of samples from weight parameters expressed in the form of probability distributions. The proposed method can deal with uncertainty present in model and data in a more computationally efficient manner by reducing the dimensionality of model parameters. The proposed method is evaluated using pinball loss, reconstruction error, and other forecasting evaluation metrics. It is inferred from the numerical results that VAE-Bayesian BiLSTM outperforms other probabilistic deep learning methods in terms of forecasting accuracy and computational efficiency for different sizes of the dataset.

A Bayesian Deep Learning Approach With Convolutional Feature Engineering to Discriminate Cyber-Physical Intrusions in Smart Grid Systems

“…These distributions define the uncertainty in weights and can further be used to estimate variability in predictions. Bayesian networks are trained using variational inference and instead of learning deterministic weight values directly, distribution parameters are learned [37], [38], [39].…”

A Bayesian Deep Learning Technique for Multi-Step Ahead Solar Generation Forecasting

“…Recently, authors in [19] have compared various DL methods such as RNN, long-short term memory (LSTM), gated recurrent unit (GRU), and convolutional neural networks (CNN) for solar photovoltaic (PV) power forecasting in both SSA and MSA manner and claimed LSTM to be best performing DL model amongst all. Furthermore, BiLSTM as a bidirectional component of LSTM serves as a more effective model due to its ability of back-and-forth learning [20]. In this regard, Gairaa et al also utilized multiple linear regression (MLR) and DL-based MSA approach for six-hours ahead solar irradiance prediction but the accuracy has not been investigated for sites with high variability [21].…”

“…In this regard, Bayesian inference integrated with neural layers [27], specifically LSTM [28] and BiLSTM [20] has emerged as a potential solution to deal with the problem of weight and prediction uncertainty in the energy forecasting domain. Contrary to the standard DL models, these methods can generate prediction intervals (PI), which can provide lower and upper bounds for the future prediction values with a given probability.…”