Nearest prototype methods can be quite successful on many pattern classification problems. In these methods, a collection of prototypes has to found that accurately represents the input patterns. The classifier then assigns classes based nearest in this collection. paper, we first use standard particle swarm optimizer (PSO) algorithm find those prototypes. Second, present new algorithm, called adaptive Michigan PSO (AMPSO) order reduce dimension search space and provide more flexibility than...

Deep neural networks (DNN) are becoming increasingly relevant for probabilistic forecasting because of their ability to estimate prediction intervals (PIs). Two different ways estimating PIs with stand out: quantile estimation posterior PI construction and direct estimation. The former first estimates quantiles, which then used construct PIs, while the latter directly obtains lower upper bounds by optimizing some loss functions, advantage that width is considered in optimization process thus...

Abstract Wind and solar energy forecasting have become crucial for the inclusion of renewable in electrical power systems. Although most works focused on point prediction, it is currently becoming important to also estimate forecast uncertainty. With regard methods, deep neural networks shown good performance many fields. However, use these comparative studies probabilistic forecasts energies, especially regional forecasts, has not yet received much attention. The aim this article study...

Abstract A methodology, aimed to be fully operational, for automatic cloud classification based on the synergetic use of a sky camera and ceilometer is presented. The random forest machine learning algorithm was used train classifier with 19 input features: 12 extracted from images 7 ceilometer. method developed tested set 717 collected at radiometric stations Univ. Jaén (Spain). Up nine different types clouds (plus clear sky) were considered (clear sky, cumulus, stratocumulus, nimbostratus,...

Abstract Machine learning is routinely used to forecast solar radiation from inputs, which are forecasts of meteorological variables provided by numerical weather prediction (NWP) models, on a spatially distributed grid. However, the number features resulting these grids usually large, especially if several vertical levels included. Principal Components Analysis (PCA) one simplest and most widely-used methods extract reduce dimensionality in renewable energy forecasting, although this method...

Time series analysis using nonlinear dynamics systems theory and multilayer neural networks models have been applied to the time sequence of water level data recorded every hour at ‘Punta della Salute’ from Venice Lagoon during years 1980–1994. The first method is based on reconstruction state space attractor delay embedding vectors characterisation invariant properties which define its dynamics. results suggest existence a low dimensional chaotic with Lyapunov dimension, DL, around 6.6...

Summary This article addresses two issues in solar energy forecasting from the numerical weather prediction (NWP) models using machine learning. First, we are interested determining relevant information for task. With this purpose, a study has been carried out to evaluate influence on accuracy of number NWP grid nodes used as input model, well their relative importance. Several learning (support vector machines and gradient boosting) feature selection algorithms (linear, ReliefF, local...

Recent research has shown that the integration or blending of different forecasting models is able to improve predictions solar radiation. However, most works perform model point forecasts, but probabilistic not received much attention. In this work estimation prediction intervals for four Global Horizontal Irradiance (GHI) (Smart Persistence, WRF-solar, CIADcast, and Satellite) addressed. Several short-term horizons, up one hour ahead, have been analyzed. Within context, aims article study...

Accurate solar radiation nowcasting models are critical for the integration of increasing energy in power systems. This work explored benefits obtained by blending four all-sky-imagers (ASI)-based models, two satellite-images-based and a data-driven model. Two approaches (general horizon) (linear random forest (RF)) were evaluated. The relative contribution different forecasting blended-models-derived was also explored. study conducted Southern Spain; provide one-minute resolution 90...

This paper shows the performance of binary PSO algorithm as a classification system. These systems are classified in two different perspectives: Pittsburgh and Michigan approaches. In order to implement approach algorithm, standard dynamic equations modified, introducing repulsive force favor particle competition. A neighborhood, adapted problems, is also defined. Both classifiers tested using reference set where both achieve better than many techniques. The classifier clear advantages over...

Machine Learning techniques are routinely applied to Brain Computer Interfaces in order learn a classifier for particular user. However, research has shown that classification perform better if the EEG signal is previously preprocessed provide high quality attributes classifier. Spatial and frequency-selection filters can be this purpose. In paper, we propose automatically optimize these by means of Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The technique been tested on data...