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

^{*}

## Abstract

**:**

## 1. Introduction

- The Quantile Regression (QR) uses a GBRT methodology and generates an ensemble of forecasts based on the IFS forecasts of the ECMWF model and the plant power output measured previously under similar conditions.
- The Ensemble Prediction System (EPS) uses the results of the EPS weather forecast ensemble of the ECMWF model, processed through the regression model used in deterministic prediction, and no calibration or ensemble post-processing technique.
- The Variance Deficit (VD) applies the variance deficit calibration technique to the results of the EPS model.
- The Quantile Ensemble (QE) still uses the EPS forecasting ensemble and the plant model, but applies a post-processing procedure based on quantile regression with the GBRT methodology to the predictions obtained.
- The Quantile Ensemble Plus (QE+) is similar to the previous one, but uses both the IFS and EPS forecasts of the ECMWF model as features in quantile regression, in order to make the most of the information that can be extracted from both weather models.

- the use of the power index to efficiently forecast solar power generation even with temporally-coarse weather forecasts;
- the effectiveness of the gradient-boosted quantile regression trees for the post-processing of the ensemble of predictions from the EPS;
- the relative importance of the EPS and IFS forecasts of the ECMWF depending on the forecast horizon considered;
- the combined use of the IFS and EPS forecast of ECMWF for an accurate probabilistic and deterministic PV power forecast.

## 2. Methods

#### 2.1. Gradient-Boosted Regression Trees

#### 2.2. Deterministic Forecasting

#### 2.3. Quantile Regression

#### 2.4. Ensemble Probabilistic Forecasting

#### 2.5. VD Calibration

#### 2.6. Quantile Ensemble

- The deterministic model is applied to the ECMWF EPS forecasts, generating an ensemble of 51 power forecasts for the plant
- For each time step, 9 quantiles [0.1, 0.2, …, 0.9] are used to describe the distribution of the ensemble and constitute a set of new features.
- Fifty one GBRT quantile regressors are trained, choosing the quantile of the target function, in the quantile range [1/52, 2/52, …, 51/52], using as input features the 9 quantiles with which the distribution of power forecasts for the training period is represented and, as the training variable, the power actually measured.

#### 2.7. Quantile Ensemble Plus

#### 2.8. Implementation

## 3. Forecast Verification

## 4. Results

## 5. Discussion

## 6. Conclusions

## Author Contributions

## Acknowledgments

## Conflicts of Interest

## Abbreviations

PF | Probabilistic Forecasting procedures |

GBRT | Gradient-Boosted Regression Trees |

PV | Photo-Voltaic |

ECMWF | European Centre for Medium-Range Weather Forecasts |

IFS | Integrated Forecasting System |

EPS | Ensemble Prediction System |

QR | Quantile Regression |

VD | Variance Deficit |

NWP | Numerical Weather Predictions |

AnEn | Analog Ensemble |

PeEn | Persistence Ensemble |

WRF | Weather Research and Forecasting regional model |

EMOS | Ensemble Model Output Statistics |

QE | Quantile Ensemble |

QE+ | Quantile Ensemble Plus |

DNI | Direct Normal Irradiation |

DHI | Diffuse Horizontal Irradiation |

GHI | Global Horizontal Irradiation |

POA | Plane Of Array |

RTM | Radiative Transfer Models |

CF | Control Forecast |

ME | Mean Error |

MAE | Mean Absolute Error |

RMSE | Root Mean Square Error |

nME | normalized Mean Error |

nMAE | normalized Mean Absolute Error |

nRMSE | normalized Root Mean Square Error |

SS | Skill Score |

ROC | Receiver Operating Characteristic |

AUC | Area Under the ROC Curve |

BS | Brier Score |

UNC | Uncertainty |

REL | Reliability |

RES | Resolution |

CRPS | Continuous Ranked Probability Score |

CDF | Cumulative Distribution Function |

## References

- Kraas, B.; Schroedter-Homscheidt, M.; Madlener, R. Economic merits of a state-of-the-art concentrating solar power forecasting system for participation in the Spanish electricity market. Solar Energy
**2013**, 93, 244–255. [Google Scholar] [CrossRef] - Luoma, J.; Mathiesen, P.; Kleissl, J. Forecast value considering energy pricing in California. Appl. Energy
**2014**, 125, 230–237. [Google Scholar] [CrossRef] - Nonnenmacher, L.; Kaur, A.; Coimbra, C.F. Day-ahead resource forecasting for concentrated solar power integration. Renew. Energy
**2016**, 86, 866–876. [Google Scholar] [CrossRef] - Yuan, S.; Kocaman, A.S.; Modi, V. Benefits of forecasting and energy storage in isolated grids with large wind penetration—The case of Sao Vicente. Renew. Energy
**2017**, 105, 167–174. [Google Scholar] [CrossRef] - Hong, T.; Pinson, P.; Fan, S.; Zareipour, H.; Troccoli, A.; Hyndman, R.J. Probabilistic energy forecasting: Global Energy Forecasting Competition 2014 and beyond. Int. J. Forecast.
**2016**, 32, 896–913. [Google Scholar] [CrossRef][Green Version] - Antonanzas, J.; Osorio, N.; Escobar, R.; Urraca, R.; Martinez-de Pison, F.; Antonanzas-Torres, F. Review of photovoltaic power forecasting. Solar Energy
**2016**, 136, 78–111. [Google Scholar] [CrossRef] - Van der Meer, D.; Widén, J.; Munkhammar, J. Review on probabilistic forecasting of photovoltaic power production and electricity consumption. Renew. Sustain. Energy Rev.
**2017**, 81, 1484–1512. [Google Scholar] [CrossRef] - Bacher, P.; Madsen, H.; Nielsen, H.A. Online short-term solar power forecasting. Solar Energy
**2009**, 83, 1772–1783. [Google Scholar] [CrossRef][Green Version] - Lorenz, E.; Hurka, J.; Heinemann, D.; Beyer, H.G. Irradiance forecasting for the power prediction of grid-connected photovoltaic systems. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.
**2009**, 2, 2–10. [Google Scholar] [CrossRef] - Monteiro, C.; Santos, T.; Fernandez-Jimenez, L.A.; Ramirez-Rosado, I.J.; Terreros-Olarte, M.S. Short-term power forecasting model for photovoltaic plants based on historical similarity. Energies
**2013**, 6, 2624–2643. [Google Scholar] [CrossRef] - Bracale, A.; Caramia, P.; Carpinelli, G.; Di Fazio, A.R.; Ferruzzi, G. A Bayesian method for short-term probabilistic forecasting of photovoltaic generation in smart grid operation and control. Energies
**2013**, 6, 733–747. [Google Scholar] [CrossRef] - Almeida, M.P.; Perpiñán, O.; Narvarte, L. PV power forecast using a nonparametric PV model. Solar Energy
**2015**, 115, 354–368. [Google Scholar] [CrossRef][Green Version] - Alessandrini, S.; Delle Monache, L.; Sperati, S.; Cervone, G. An analog ensemble for short-term probabilistic solar power forecast. Appl. Energy
**2015**, 157, 95–110. [Google Scholar] [CrossRef][Green Version] - Huang, J.; Perry, M. A semi-empirical approach using gradient boosting and k-nearest neighbors regression for GEFCom2014 probabilistic solar power forecasting. Int. J. Forecast.
**2016**, 32, 1081–1086. [Google Scholar] [CrossRef] - Pierro, M.; Bucci, F.; De Felice, M.; Maggioni, E.; Moser, D.; Perotto, A.; Spada, F.; Cornaro, C. Multi-Model Ensemble for day ahead prediction of photovoltaic power generation. Solar Energy
**2016**, 134, 132–146. [Google Scholar] [CrossRef] - Sperati, S.; Alessandrini, S.; Delle Monache, L. An application of the ECMWF Ensemble Prediction System for short-term solar power forecasting. Solar Energy
**2016**, 133, 437–450. [Google Scholar] [CrossRef][Green Version] - Rana, M.; Koprinska, I. Neural network ensemble based approach for 2D-interval prediction of solar photovoltaic power. Energies
**2016**, 9, 829. [Google Scholar] [CrossRef] - Ahmed Mohammed, A.; Aung, Z. Ensemble learning approach for probabilistic forecasting of solar power generation. Energies
**2016**, 9, 1017. [Google Scholar] [CrossRef] - Owens, R.G.; Hewson, T.D. ECMWF Forecast User Guide; Technical Report; European Centre for Medium-Range Weather Forecasts: Reading, UK, 2018. [Google Scholar]
- Kopp, G.; Lean, J.L. A new, lower value of total solar irradiance: Evidence and climate significance. Geophys. Res. Lett.
**2011**, 38. [Google Scholar] [CrossRef][Green Version] - Inman, R.H.; Pedro, H.T.; Coimbra, C.F. Solar forecasting methods for renewable energy integration. Prog. Energy Combust. Sci.
**2013**, 39, 535–576. [Google Scholar] [CrossRef] - Goswami, D.Y.; Kreith, F.; Kreider, J.F. Principles of Solar Engineering; CRC Press: Boca Raton, FL, USA, 2000. [Google Scholar]
- Ineichen, P.; Perez, R. A new airmass independent formulation for the Linke turbidity coefficient. Solar Energy
**2002**, 73, 151–157. [Google Scholar] [CrossRef][Green Version] - Ineichen, P. Comparison of eight clear sky broadband models against 16 independent data banks. Solar Energy
**2006**, 80, 468–478. [Google Scholar] [CrossRef] - Rigollier, C.; Bauer, O.; Wald, L. On the clear sky model of the ESRA—European Solar Radiation Atlas—With respect to the Heliosat method. Solar Energy
**2000**, 68, 33–48. [Google Scholar] [CrossRef] - Massidda, L.; Marrocu, M. Use of Multilinear Adaptive Regression Splines and numerical weather prediction to forecast the power output of a PV plant in Borkum, Germany. Solar Energy
**2017**, 146, 141–149. [Google Scholar] [CrossRef] - Friedman, J.H. Greedy function approximation: a gradient boosting machine. Ann. Stat.
**2001**, 29, 1189–1232. [Google Scholar] [CrossRef] - Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V.; et al. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res.
**2011**, 12, 2825–2830. [Google Scholar] - Pinson, P. Very-short-term probabilistic forecasting of wind power with generalized logit–normal distributions. J. R. Stat. Soci. Ser. C (Appl. Stat.)
**2012**, 61, 555–576. [Google Scholar] [CrossRef] - Taillardat, M.; Mestre, O.; Zamo, M.; Naveau, P. Calibrated ensemble forecasts using quantile regression forests and ensemble model output statistics. Mon. Weather Rev.
**2016**, 144, 2375–2393. [Google Scholar] [CrossRef] - McKinney, W. Data structures for statistical computing in python. In Proceedings of the 9th Python in Science Conference, Austin, TX, USA, 28 June–3 July 2010; Volume 445, pp. 51–56. [Google Scholar]
- Holmgren, W.F.; Groenendyk, D.G. An open source solar power forecasting tool using PVLIB-Python. In Proceedings of the 2016 IEEE 43rd Photovoltaic Specialists Conference on Photovoltaic Specialists Conference (PVSC), Portland, OR, USA, 5–10 June 2016; pp. 972–975. [Google Scholar]
- Murphy, A.H. Skill scores based on the mean square error and their relationships to the correlation coefficient. Mon. Weather Rev.
**1988**, 116, 2417–2424. [Google Scholar] [CrossRef] - Wilks, D.S. Statistical Methods in the Atmospheric Sciences; Academic Press: Cambridge, MA, USA, 2011; Volume 100. [Google Scholar]
- Murphy, A.H. A new vector partition of the probability score. J. Appl. Meteorol.
**1973**, 12, 595–600. [Google Scholar] [CrossRef] - Hersbach, H. Decomposition of the continuous ranked probability score for ensemble prediction systems. Weather Forecast.
**2000**, 15, 559–570. [Google Scholar] [CrossRef]

**Figure 1.**Flowchart showing the steps required to realize the QE and QE+ forecast starting from the EPS and IFS power forecasts.

**Figure 2.**Rank histograms for forecast ensembles produced by the five methods examined for the 24–48-h forecast horizon: (

**a**) Quantile regression; (

**b**) Uncalibrated EPS (full scale); (

**c**) Uncalibrated EPS; (

**d**) EPS with variance deficit calibration; (

**e**) Quantile ensemble with EPS forecast; (

**f**) Quantile ensemble with IFS and EPS forecasts. The calibration of the EPS method is not satisfactory, and the VD method has a better performance. The QR and QE methods show a good calibration.

**Figure 3.**ROC curves for a threshold equal to half the clear sky power and for a forecast interval of 24–48 h. In the legend within the figure, the values of the area under the curve are also shown.

**Figure 4.**CRPS results as the lead time forecast varies for the probabilistic forecast methods examined: (

**a**) CRPS values for times of forecast in 0–24-h time range; (

**b**) same as before for 24–48-h; (

**c**) same for 48–72-h.

Jan–Feb | Mar–Apr | May–Jun | Jul–Aug | Sep–Oct | Nov–Dec | Total | |
---|---|---|---|---|---|---|---|

Samples (-) | 5664 | 5856 | 5856 | 5952 | 5856 | 5856 | 35,040 |

mean value (kW) | 68.4 | 204.2 | 261.6 | 276.5 | 150.3 | 49.1 | 169.2 |

std. deviation (kW) | 161.8 | 309.1 | 344.0 | 361.3 | 262.2 | 128.2 | 289.9 |

max value (kW) | 957.4 | 1187.9 | 1326.0 | 1276.2 | 1111.5 | 824.0 | 1326.0 |

Parameter | Search Values | Deterministic | QE | QE+ |
---|---|---|---|---|

learning rate | 0.05, 0.02, 0.01 | 0.05 | 0.05 | 0.01 |

maximum depth of trees | 3, 4, 5 | 3 | 3 | 3 |

minimum number of samples in leaf | 3, 5, 7 | 7 | 7 | 5 |

number of estimators | 100, 200, 500 | 100 | 200 | 500 |

**Table 3.**Feature importance as the relative occurrence of the feature in the GBRT regression trees for the prediction of ${k}_{pv}$.

Feature | Relative Importance |
---|---|

k | 56.9% |

Precipitation | 3.9% |

Temperature 2 m | 11.3% |

Zenith | 9.5 % |

Azimuth | 18.4 % |

Forecast (h) | R2 (-) | ME (kW) | MAE (kW) | RMSE (kW) | nMAE (%) | nRMSE (%) | SS(%) |
---|---|---|---|---|---|---|---|

0–24 | 0.785 | $-3.3$ | 93.3 | 146.4 | 7.18 | 11.26 | 43.61 |

24–48 | 0.687 | 6.9 | 114.3 | 176.5 | 8.80 | 13.57 | 35.62 |

48–72 | 0.636 | 7.5 | 124.8 | 190.4 | 9.60 | 14.65 | 27.78 |

**Table 5.**Probabilistic error measures with different forecast lead times. The Brier Score (BS), its components reliability (REL), Resolution (RES) and Uncertainty (UNC) and the Area Under the ROC Curve (AUC) are calculated for a threshold of $0.5\xb7{p}_{cs}$. The cumulative ranked probability score is the mean value for the test set in the daylight hours only.

Forecast (h) | Model | BS (-) | REL (-) | RES (-) | UNC (-) | AUC (-) | CRPS (kW) |
---|---|---|---|---|---|---|---|

0–24 | QR | 0.140 | 0.008 | 0.117 | 0.249 | 0.882 | 65.7 |

EPS | 0.204 | 0.043 | 0.089 | 0.249 | 0.814 | 87.5 | |

VD | 0.179 | 0.014 | 0.085 | 0.249 | 0.808 | 82.3 | |

QE | 0.162 | 0.006 | 0.094 | 0.249 | 0.845 | 74.9 | |

QE+ | 0.143 | 0.006 | 0.113 | 0.249 | 0.876 | 67.7 | |

24–48 | QR | 0.170 | 0.008 | 0.087 | 0.249 | 0.827 | 79.2 |

EPS | 0.199 | 0.027 | 0.077 | 0.249 | 0.803 | 86.3 | |

VD | 0.188 | 0.012 | 0.074 | 0.249 | 0.794 | 85.3 | |

QE | 0.173 | 0.007 | 0.083 | 0.249 | 0.821 | 78.4 | |

QE+ | 0.165 | 0.007 | 0.091 | 0.249 | 0.836 | 76.8 | |

48–72 | QR | 0.184 | 0.007 | 0.073 | 0.249 | 0.800 | 85.9 |

EPS | 0.199 | 0.026 | 0.077 | 0.249 | 0.796 | 86.2 | |

VD | 0.191 | 0.014 | 0.072 | 0.249 | 0.789 | 86.3 | |

QE | 0.176 | 0.006 | 0.079 | 0.249 | 0.815 | 81.2 | |

QE+ | 0.172 | 0.008 | 0.086 | 0.249 | 0.825 | 80.7 |

**Table 6.**Error measures for the deterministic forecast based on the median of the quantile ensemble probabilistic model with the IFS and EPS forecasts.

Forecast (h) | R2 (-) | ME (kW) | MAE (kW) | RMSE (kW) | nMAE (%) | nRMSE (%) | SS (%) |
---|---|---|---|---|---|---|---|

0–24 | 0.784 | −5.1 | 93.5 | 146.8 | 7.19 | 11.29 | 43.46 |

24–48 | 0.721 | 0.6 | 110.0 | 166.7 | 8.46 | 12.82 | 39.18 |

48–72 | 0.704 | 0.7 | 117.1 | 171.6 | 9.00 | 13.20 | 34.92 |

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Massidda, L.; Marrocu, M.
Quantile Regression Post-Processing of Weather Forecast for Short-Term Solar Power Probabilistic Forecasting. *Energies* **2018**, *11*, 1763.
https://doi.org/10.3390/en11071763

**AMA Style**

Massidda L, Marrocu M.
Quantile Regression Post-Processing of Weather Forecast for Short-Term Solar Power Probabilistic Forecasting. *Energies*. 2018; 11(7):1763.
https://doi.org/10.3390/en11071763

**Chicago/Turabian Style**

Massidda, Luca, and Marino Marrocu.
2018. "Quantile Regression Post-Processing of Weather Forecast for Short-Term Solar Power Probabilistic Forecasting" *Energies* 11, no. 7: 1763.
https://doi.org/10.3390/en11071763