Previsioni di fonti rinnovabili non programmabili: un modello - PowerPoint PPT Presentation

Mercati energetici e metodi manuele.aufiero@milanomultiphysics.com quantitativi nicholas.bonfanti@milanomultiphysics.com 18 ottobre 2018 Previsioni di fonti rinnovabili non programmabili: un modello integrato per l'idroelettrico ad acqua fluente Milano Multiphysics -- ENTSO-E Smart solutions for complex problems

Main goals of the developed approach ➔ Assess suitability of SMHI data for the production of hydro databases ➔ Investigate the possibility of modeling hydro plant inflows without geographic information ➔ Study an automated procedure for automated regression of RoR inflow ➔ Generation of synthetic hydro time series for adequacy studies MMP - ENTSO-E SMHI data for Hydro database

Main goals of the developed approach ➔ Perform additional analyses: ◆ Automatic identification of regulated inflows impact on RoR ◆ Non-linear correction of inflow/power curve ◆ Robust detection of plant maintenance effects MMP - ENTSO-E SMHI data for Hydro database

Main steps of the statistical analysis ➔ Pre processing ◆ Selection of complete datasets of plant production data MMP - ENTSO-E SMHI data for Hydro database

Considered datasets: RoR plants (Italy) MMP - ENTSO-E SMHI data for Hydro database

Main steps of the statistical analysis ➔ Pre processing ◆ Selection of complete datasets of plant production data ◆ SMHI inflow data normalization MMP - ENTSO-E SMHI data for Hydro database

Considered datasets: SMHI inflows (Italy) MMP - ENTSO-E SMHI data for Hydro database

Main steps of the statistical analysis ➔ Pre processing ◆ Selection of complete datasets of plant production data ◆ SMHI inflow data normalization ◆ Preliminary verification of the SMHI/hydro spatial correlation MMP - ENTSO-E SMHI data for Hydro database

Verification of the SMHI/hydro correlation Simple sanity check for high quality RoR plant dataset with high correlation Case study: Italy Testing SMHI data for hydro database

Italy - NORD SMHI/RoR correlation Zone-wise aggregated RoR correlation MMP - ENTSO-E SMHI data for Hydro database

Main steps of the statistical analysis ➔ Pre processing ➔ Dimensionality reduction ◆ Proper orthogonal decomposition of SMHI inflow data Reduction of the input dimensionality from ~1000 to ~50-150 variables ➔ Drastic reduction of cpu requirement ➔ Avoid regression overfitting MMP - ENTSO-E SMHI data for Hydro database

SVD of SMHI inflow data Derivation a reduced set of input variables as linear combination of original SMHI inflows MMP - ENTSO-E SMHI data for Hydro database

Model order reduction (SVD) of SMHI data Case study: Italy Testing SMHI data for hydro database

Model order reduction (SVD) of SMHI data Case study: Italy Testing SMHI data for hydro database

Model order reduction (SVD) of SMHI data Case study: Italy Testing SMHI data for hydro database

Model order reduction (SVD) of SMHI data Case study: Italy Testing SMHI data for hydro database

Model order reduction (SVD) of SMHI data Case study: Italy Testing SMHI data for hydro database

Main steps of the statistical analysis ➔ Pre processing ➔ Dimensionality reduction ➔ Transfer function derivation ◆ Plan by plant SMHI/hydro power regression by least squares minimization MMP - ENTSO-E SMHI data for Hydro database

Plant-by-plant SMHI/power transfer function MMP - ENTSO-E SMHI data for Hydro database

Main steps of the statistical analysis ➔ Pre processing ➔ Dimensionality reduction ➔ Transfer function derivation ◆ Plan by plant SMHI/hydro power regression by least squares minimization ◆ p-value based elimination of least significant regressors MMP - ENTSO-E SMHI data for Hydro database

p-value based regression simplification Output example of the linear fit function, removing the least significant regressors. MMP - ENTSO-E SMHI data for Hydro database

Main steps of the statistical analysis ➔ Pre processing ➔ Dimensionality reduction ➔ Transfer function derivation ➔ Post-processing ◆ Identification of spurious RoR (influenced by regulated hydro) MMP - ENTSO-E SMHI data for Hydro database

Fourier analysis The production data has been analysed with a Fourier transform to show its spectrum. ➔ Yearly dynamics ➔ Seasonal dynamics ➔ Weekly dynamics (expected for regulated plants, but not for “true” RoRs) MMP - ENTSO-E SMHI data for Hydro database

Identification of spurious RoR Fourier transform of hydro power production MMP - ENTSO-E SMHI data for Hydro database

Identification of spurious RoR Fourier transform of hydro power production MMP - ENTSO-E SMHI data for Hydro database

Identification of spurious RoR Fourier transform of hydro power production MMP - ENTSO-E SMHI data for Hydro database

Identification of spurious RoR Fourier transform of hydro power production MMP - ENTSO-E SMHI data for Hydro database

Main steps of the statistical analysis ➔ Pre processing ➔ Dimensionality reduction ➔ Transfer function derivation ➔ Post-processing ◆ Identification of spurious RoR ◆ Zone-wise aggregation of RoR power ◆ Regression validation with independent input dataset MMP - ENTSO-E SMHI data for Hydro database

Validation set Part of the data is not used in the regressor training. This data is used to validate the regressors and check the presence of overfitting. MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - SMHI/RoR correlation MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - RoR Model verification MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - RoR Model verification MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - Past climatic year evaluation MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - RoR statistics MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - ``Basin´´ statistics* The subdivision in ``Basin´´ and ``Reservoir´´ plants was provided by TERNA in the input data MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - ``Reservoir´´ statistics The subdivision in ``Basin´´ and ``Reservoir´´ plants was provided by TERNA in the input data MMP - ENTSO-E SMHI data for Hydro database

Italy - NORD - Total storage statistics MMP - ENTSO-E SMHI data for Hydro database

Italy - CENTRO SUD - RoR Model verification MMP - ENTSO-E SMHI data for Hydro database

Italy - RoR aggregated results MMP - ENTSO-E SMHI data for Hydro database

Italy - CSUD - Total storage statistics MMP - ENTSO-E SMHI data for Hydro database

Italy - Mean Regression Errors Mean daily RoR power errors: - Italy: 0.168 GW [4.0% 3.2%] - NORD: 0.161 GW [4.3% 3.8%] - CSUD: 0.0199 GW [3.7% 2.5%] Mean weekly Storage power errors: - Italy: 0.135 GW [3.0% 1.8%] - NORD: 0.128 GW [3.0% 2.2%] - CSUD: 0.0172 GW [3.0% 2.3%] [absolute error / maximum power] [absolute error / installed power] MMP - ENTSO-E SMHI data for Hydro database

France - Blind test This reduced analysis on France hydro data has been performed in collaboration with RTE thanks to the invaluable support of Frédéric Bréant and Pierre Goutierre For testing purposes, automatic regression on confidential, non-disclosable RoR time series has been kindly performed by RTE adopting the algorithm code provided by Milano Multiphysics. No data or hydro plant information were provided to MMP or ENTSO-E by RTE for this test. The automatic regression code only produced graphical, aggregated results for testing scopes. MMP - ENTSO-E SMHI data for Hydro database

France - Blind test - Zone 5 MMP - ENTSO-E SMHI data for Hydro database

France - Blind test - Zone 7 MMP - ENTSO-E SMHI data for Hydro database

France - Blind test - Zone 9 MMP - ENTSO-E SMHI data for Hydro database

Conclusion ➔ SMHI inflow data proved to be a good dataset for the modeling of most RoR and Storage plants ➔ The developed modeling approach allows for the derivation of reliable and accurate transfer functions ➔ France ``blind´´ test confirms the possibility to adopt the procedure without geographical information on hydro plants ➔ Developed transfer functions can be applied to reanalysis climate data to provide synthetic hydro time series (correlated with other climatic variables) for adequacy studies MMP - ENTSO-E SMHI data for Hydro database

Parallel activities and ongoing development ➔ Extension of the approach to produce hydro time series databases for all ENTSO-E members ➔ Adoption of derived transfer functions for short-term (week ahead) RoR hydro forecast with robust uncertainty estimates MMP - ENTSO-E SMHI data for Hydro database

Parallel activities and ongoing development ➔ Extension of the approach to produce hydro time series databases for all ENTSO-E members ➔ Adoption of derived transfer functions for short-term (week ahead) RoR hydro forecast with robust uncertainty estimates Thank you for the attention! MMP - ENTSO-E SMHI data for Hydro database

Backup MMP - ENTSO-E SMHI data for Hydro database