Developing and Comparing Different Strategies for Combining Probabilistic Photovoltaic Power Forecasts in an Ensemble Method

Abstract

Accurate probabilistic forecasts of renewable generation are drivers for operational and management excellence in modern power systems and for the sustainable integration of green energy. The combination of forecasts provided by different individual models may allow increasing the accuracy of predictions; however, in contrast to point forecast combination, for which the simple weighted averaging is often a plausible solution, combining probabilistic forecasts is a much more challenging task. This paper aims at developing a new ensemble method for photovoltaic (PV) power forecasting, which combines the outcomes of three underlying probabilistic models (quantile k-nearest neighbors, quantile regression forests, and quantile regression) through a weighted quantile combination. Due to the challenges in combining probabilistic forecasts, the paper presents different combination strategies; the competing strategies are based on unconstrained, constrained, and regularized optimization problems for estimating the weights. The competing strategies are compared to individual forecasts and to several benchmarks, using the data published during the Global Energy Forecasting Competition 2014. Numerical experiments were run in MATLAB and R environments; the results suggest that unconstrained and Least Absolute Shrinkage and Selection Operator (LASSO)-regularized strategies exhibit the best performances for the datasets under study, outperforming the best competitors by 2.5 to 9% of the Pinball Score.

A Novel Ensemble Algorithm for Solar Power Forecasting Based on Kernel Density Estimation

“…In Reference [9], different strategies for combining forecasts of solar photovoltaic (PV) generation were presented. In this study, the ensemble prediction was obtained by combining different probabilistic models rather than an ensemble of results of the same model.…”

Bayesian bootstrap quantile regression for probabilistic photovoltaic power forecasting

“…Simple QR (SQR): the first benchmark [10,16] is introduced to be used as a reference in which each predictive quantile is directly provided as a single value, rather than by passing through the bootstrap. This allows assessing whether the bootstrap is effective or not in improving forecasts.…”

“…Probabilistic PV power forecasting systems range from pure statistical models to hybrid physical-statistical models. High-performance solutions are based on Quantile Regression (QR) models [9][10][11], machine learning approaches (such as gradient boosting [12], quantile regression forests [10,13,14] and k-nearest neighbors [15]). It is worth noting that, although the analytic formulation of QR models is much simpler than machine learning approaches, QR predictions still are somehow competitive in most cases.…”

“…Usually several weather variables are available to forecasters, and thus model selection (i.e., the selection of the most informative predictors for the final model) is typically applied to discard uninformative inputs. To further maximize the exploitation of the available input data, ensemble approaches [7] have been applied with success in probabilistic energy forecasting [10,[16][17][18][19]. These approaches are commonly divided into boosting, stacking, and bagging.…”

Averaging Predictive Distributions Across Calibration Windows for Day-Ahead Electricity Price Forecasting

“…To address this gap, here we consider two novel extensions of the averaging across calibration windows concept to probabilistic forecasting: one based on Quantile Regression Averaging (QRA) [7] and one using the Quantile Regression Machine (QRM) [8]. As the underlying statistical technique both use quantile regression [9], which has recently become the workhorse of probabilistic energy forecasting [10][11][12][13][14]. Moreover, both apply it to a pool of point forecasts obtained for calibration windows of different lengths and yield predictions for the 99 percentiles of the next day's price distribution for each hour.…”