Ensemble Learning Approach for Probabilistic Forecasting of Solar Power Generation

Abstract

Probabilistic forecasting accounts for the uncertainty in prediction that arises from inaccurate input data due to measurement errors, as well as the inherent inaccuracy of a prediction model. Because of the variable nature of renewable power generation depending on weather conditions, probabilistic forecasting is well suited to it. For a grid-tied solar farm, it is increasingly important to forecast the solar power generation several hours ahead. In this study, we propose three different methods for ensemble probabilistic forecasting, derived from seven individual machine learning models, to generate 24-h ahead solar power forecasts. We have shown that while all of the individual machine learning models are more accurate than the traditional benchmark models, like autoregressive integrated moving average (ARIMA), the ensemble models offer even more accurate results than any individual machine learning model alone does. Furthermore, it is observed that running separate models on the data belonging to the same hour of the day vastly improves the accuracy of the results. Getting more accurate forecasts will help the stakeholders come up with better decisions in resource planning and control when large-scale solar farms are integrated into the power grid.

Quantile Regression Post-Processing of Weather Forecast for Short-Term Solar Power Probabilistic Forecasting

“…The next day's probability forecast for three PV plants in Australia is presented in [18], which shows the advantage of using a combination of machine learning methods for greater accuracy. The error metric pinball loss is adopted in their work, and the authors in their conclusions refer to "the need" to evaluate the performance of the method with CRPS.…”

“…GBRT is a machine learning technique for regression and classification problems. This is a particularly effective regression technique for datasets with a limited number of samples, unlike recurrent neural networks and has proven effective in predicting PV production; see, for example, [14] and [18].…”

“…The results for short-term forecasting in comparison with benchmark methodologies are promising. In [18], Mohammed and Aung proposed to realize the prediction of the production of three solar farms 24 h in advance, using the deterministic meteorological forecast of the ECMWF model, with a total of seven different regression methodologies and combining them with three different ensembles of methods. The results showed that the individual methods all offer superior performance to traditional time series analysis techniques and that their evaluation as an ensemble can guarantee even greater accuracy.…”

Comparison of statistical post-processing methods for probabilistic NWP forecasts of solar radiation

“…They used only past observations of global radiation and the CSI as predictors. Mohammed and Aung (2016) made a comparison between deterministic forecasts using decision trees, nearest neighbours, gradient boosting decision trees, random forests and lasso and ridge regression. They used ECMWF output variables as predictors in the regression methods.…”

Recent Advances in Energy Time Series Forecasting

“…Thus, an ensemble learning approach for probabilistic forecasting can be found in [14] and a neural network ensemble for solar photovoltaic power 2-D interval forecasting in [15].…”