Abstract
The deployment of smart grids and renewable energy dispatch centers motivates the development of forecasting techniques that take advantage of near real-time measurements collected from geographically distributed sensors. This paper describes a forecasting methodology that explores a set of different sparse structures for the vector autoregression (VAR) model using the least absolute shrinkage and selection operator (LASSO) framework. The alternating direction method of multipliers is applied to fit the different LASSO-VAR variants and create a scalable forecasting method supported by parallel computing and fast convergence, which can be used by system operators and renewable power plant operators. A test case with 66 wind power plants is used to show the improvement in forecasting skill from exploring distributed sparse structures. The proposed solution outperformed the conventional autoregressive and vector autoregressive models, as well as a sparse VAR model from the state of the art. Copyright (c) 2016 John Wiley & Sons, Ltd.

Abstract
The strong growth that is felt at the level of photovoltaic (PV) power generation craves for more sophisticated and accurate forecasting methods that could be able to support its proper integration into the energy distribution network. Through the combination of the vector autoregression model (VAR) with the least absolute shrinkage and selection operator (LASSO) framework, a set of sparse VAR structures can be obtained in order to capture the dynamic of the underlying system. The robust and efficient alternating direction method of multipliers (ADMM), well known for its great ability dealing with high-dimensional data (scalability and fast convergence), is applied to fit the resulting LASSO-VAR variants. This spatial-temporal forecasting methodology has been tested, using 1-hour and 15-minutes resolution, for 44 microgeneration units time-series located in a city in Portugal. A comparison with the conventional autoregressive (AR) model is performed leading to an improvement up to 11%.

Abstract
Wind power probabilistic forecast is being used as input in several decision-making problems, such as stochastic unit commitment, operating reserve setting and electricity market bidding. This work introduces a new on-line quantile regression model based on the Reproducing Kernel Hilbert Space (RKHS) framework. Its application to the field of wind power forecasting involves a discussion on the choice of the bias term of the quantile models, and the consideration of the operational framework in order to mimic real conditions. Benchmark against linear and splines quantile regression models was performed for a real case study during a 18 months period. Model parameter selection was based on k-fold cross-validation. Results showed a noticeable improvement in terms of calibration, a key criterion for the wind power industry. Modest improvements in terms of Continuous Ranked Probability Score (CRPS) were also observed for prediction horizons between 6 and 20 h ahead.

Abstract
In order to reduce the curtailment of renewable generation in periods of low load, operators can limit the import net transfer capacity (NTC) of interconnections. This paper presents a probabilistic approach to support the operator in setting the maximum import NTC value in a way that the risk of curtailment remains below a pre-specified threshold. Main inputs are the probabilistic forecasts of wind power and solar PV generation, and special care is taken regarding the tails of the global margin distribution (all generation all loads and pumping), since the accepted thresholds are generally very low. Two techniques are used for this purpose: interpolation with exponential functions and nonparametric estimation of extreme conditional quantiles using extreme value theory. The methodology is applied to five representative days, where situations ranging from high maximum NTC values to NTC=0 are addressed. Comparison of the two techniques for modeling tails is also comprised.

Abstract
Wind power probabilistic forecast is a key input in decision-making problems under risk, such as stochastic unit commitment, operating reserve setting and electricity market bidding. While the majority of the probabilistic forecasting methods are based on quantile regression, the associated limitations call for new approaches. This paper described a new quantile regression model based on the Reproducing Kernel Hilbert Space (RKHS) framework. In particular, two versions of the model, off-line and on-line, were implemented and tested for a real wind farm. Results showed the superiority of the on-line approach in terms of performance, robustness and computational cost. Additionally, it was observed that, in the presence of correlated data, the optimal on-line learning may cause unreliable modelling. Potential solutions to this effect are also described and implemented in the paper.