Deep distributional time series models and the probabilistic forecasting of intraday electricity prices

Recurrent neural networks (RNNs) with rich feature vectors of past values can provide accurate point forecasts for series that exhibit complex serial dependence. We propose two approaches to constructing deep time probabilistic models based on a variant RNN called an echo state network (ESN). The first is where the output layer ESN has stochastic disturbances and shrinkage prior additional regularization. second approach employs implicit copula Gaussian disturbances, which process space. Combining this non-parametrically estimated marginal distribution produces distributional model. resulting are functions vector also marginally calibrated. In both approaches, Bayesian Markov chain Monte Carlo methods used estimate compute forecasts. proposed suitable task forecasting intraday electricity prices. Using data from Australian National Electricity Market, we show our short term price forecasts, model dominating. Moreover, flexible framework incorporating demand as features, increases upper tail forecast accuracy significantly.