東華大學圖書館 |

Development and Application of Probabilistic Solar Power Forecasts for the Day-Ahead Unit Commitment.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Development and Application of Probabilistic Solar Power Forecasts for the Day-Ahead Unit Commitment./
作者:	Doubleday, Katharine.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	208 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-03, Section: B.
Contained By:	Dissertations Abstracts International83-03B.
標題:	Electrical engineering. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28028371
ISBN:	9798538119271

Development and Application of Probabilistic Solar Power Forecasts for the Day-Ahead Unit Commitment.
Doubleday, Katharine.

Development and Application of Probabilistic Solar Power Forecasts for the Day-Ahead Unit Commitment. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 208 p.

Source: Dissertations Abstracts International, Volume: 83-03, Section: B.

Thesis (Ph.D.)--University of Colorado at Boulder, 2021.

This item must not be sold to any third party vendors.

The electricity system is undergoing a rapid transformation, including heavy investment in carbon-free variable and uncertain renewable resources like solar photovoltaics (PV) to reduce greenhouse gas emissions. Under high PV penetrations, the ability to provide inexpensive and reliable electricity depends strongly on the quality of the PV power forecasts available to the system operator. Power system operators have historically made operating decisions using point or "deterministic" forecasts of renewable generation, but forecast errors can cause economic and reliability detriments. Recent efforts have promoted a transition to probabilistic forecasts that quantify forecast uncertainty and can be used in risk assessment. This thesis aims to advance the state-of-the-art in generating and interpreting high quality probabilistic solar forecasts, during the energy transition and in a 100% renewable energy future.This thesis begins with a foundational contribution to the academic field of probabilistic solar forecasting that standardizes methods benchmarking to promote consistency across the field. A thorough literature review indicates wide variation in the benchmarks implemented in probabilistic solar forecast studies, including many studies that neglect to include any benchmark at all. Using the literature review to identify six common benchmark classes, ten variants from these six classes are implemented and compared at two temporal scales. A case study is conducted on the seven geographically-diverse locations using open-source data to compare the benchmark methods through time-series plots, proper probabilistic metrics, and common diagnostic tools. In general, we recommend that practitioners use two benchmark methods to properly showcase state-of-the-art improvements in two key forecast features: reliability and sharpness. Based on their relative performance in the case study, we select benchmark methods from the six classes to recommend to practitioners. By recommending these benchmark methods, we aim to promote consistent and sensible methodological comparisons, making it easier for readers to discern among proposed methods. The thesis then transitions from the broader probabilistic solar forecasting field to specific methods relevant to day-ahead solar forecasting. We propose a new post-processing technique to generate high-quality PV power forecasts from numerical weather prediction (NWP) ensembles. This technique is the first demonstration of tailoring Bayesian model averaging (BMA) to utility-scale PV forecasting by modeling power clipping at the AC inverter rating. In addition, this work advances the broader field of BMA by devising a new parameterization for a beta kernel that accommodates theoretical constraints not previously addressed. We demonstrates these advances on a case study of eleven utility-scale PV plants in Texas, forecasting at hourly resolution for the complete year 2018. BMA's mixture-model approach is shown to consistently outperform a state-of-the-art parametric ensemble model output statistics (EMOS) approach from the literature by better capturing disagreements in the NWP ensemble. As a final benefit, BMA contributes the largest improvement in the lower tail of the distribution, which is of greatest benefit to power system operators due to the high cost of managing low output. Calibrating readily-available NWP ensembles with this BMA technique could provide probabilistic forecasts at the appropriate temporal resolution and in the appropriate units for direct use by power system operators. Finally, this thesis turns to an investigation of the value of probabilistic solar forecasts in power system operations, namely within the day-ahead unit commitment scheduling problem. We explore three measures to increase operational flexibility under high solar penetrations: battery storage for energy time-shifting, ancillary service provisions by solar PV, and a stochastic unit commitment formulation that prepares for a range of solar outcomes using probabilistic forecasts. These flexibility measures are compared alone and in combination for a case study system based on the Texas power grid under a high solar penetration for eight selected days through the four seasons. The case study simulations show clear benefits to the selected flexibility measures particularly in the spring, when both solar curtailment and variability are high, and on "low-tail" days when solar power is over-forecasted. In particular, allowing PV to provide ancillary services can provide a few percentage points of cost savings throughout the year, but significant cost savings in the realm of 10--20% in spring. Compared to a deterministic formulation, the stochastic formulation reduced unserved reserves by 50--100% when the other two flexibility measures were also in effect, incurring a cost increase of 1.2--3.8% on spring days, but typically 0.7% or less. These results demonstrate the practical usefulness of probabilistic solar power forecasts in supporting reliable and economic electricity service as solar PV deployments continue to grow.

ISBN: 9798538119271Subjects--Topical Terms:

649834
Electrical engineering.
Subjects--Index Terms:

Bayesian model averaging

Development and Application of Probabilistic Solar Power Forecasts for the Day-Ahead Unit Commitment.
LDR:06312nmm a2200361 4500 001 2347112
005 20220719070501.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798538119271
035 $a (MiAaPQ)AAI28028371
035 $a AAI28028371
040 $a MiAaPQ $c MiAaPQ
100 1 $a Doubleday, Katharine. $0 (orcid)0000-0002-1709-6331 $3 3686324
245 1 0 $a Development and Application of Probabilistic Solar Power Forecasts for the Day-Ahead Unit Commitment.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 208 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-03, Section: B.
500 $a Advisor: Hodge, Bri-Mathias.
502 $a Thesis (Ph.D.)--University of Colorado at Boulder, 2021.
506 $a This item must not be sold to any third party vendors.
520 $a The electricity system is undergoing a rapid transformation, including heavy investment in carbon-free variable and uncertain renewable resources like solar photovoltaics (PV) to reduce greenhouse gas emissions. Under high PV penetrations, the ability to provide inexpensive and reliable electricity depends strongly on the quality of the PV power forecasts available to the system operator. Power system operators have historically made operating decisions using point or "deterministic" forecasts of renewable generation, but forecast errors can cause economic and reliability detriments. Recent efforts have promoted a transition to probabilistic forecasts that quantify forecast uncertainty and can be used in risk assessment. This thesis aims to advance the state-of-the-art in generating and interpreting high quality probabilistic solar forecasts, during the energy transition and in a 100% renewable energy future.This thesis begins with a foundational contribution to the academic field of probabilistic solar forecasting that standardizes methods benchmarking to promote consistency across the field. A thorough literature review indicates wide variation in the benchmarks implemented in probabilistic solar forecast studies, including many studies that neglect to include any benchmark at all. Using the literature review to identify six common benchmark classes, ten variants from these six classes are implemented and compared at two temporal scales. A case study is conducted on the seven geographically-diverse locations using open-source data to compare the benchmark methods through time-series plots, proper probabilistic metrics, and common diagnostic tools. In general, we recommend that practitioners use two benchmark methods to properly showcase state-of-the-art improvements in two key forecast features: reliability and sharpness. Based on their relative performance in the case study, we select benchmark methods from the six classes to recommend to practitioners. By recommending these benchmark methods, we aim to promote consistent and sensible methodological comparisons, making it easier for readers to discern among proposed methods. The thesis then transitions from the broader probabilistic solar forecasting field to specific methods relevant to day-ahead solar forecasting. We propose a new post-processing technique to generate high-quality PV power forecasts from numerical weather prediction (NWP) ensembles. This technique is the first demonstration of tailoring Bayesian model averaging (BMA) to utility-scale PV forecasting by modeling power clipping at the AC inverter rating. In addition, this work advances the broader field of BMA by devising a new parameterization for a beta kernel that accommodates theoretical constraints not previously addressed. We demonstrates these advances on a case study of eleven utility-scale PV plants in Texas, forecasting at hourly resolution for the complete year 2018. BMA's mixture-model approach is shown to consistently outperform a state-of-the-art parametric ensemble model output statistics (EMOS) approach from the literature by better capturing disagreements in the NWP ensemble. As a final benefit, BMA contributes the largest improvement in the lower tail of the distribution, which is of greatest benefit to power system operators due to the high cost of managing low output. Calibrating readily-available NWP ensembles with this BMA technique could provide probabilistic forecasts at the appropriate temporal resolution and in the appropriate units for direct use by power system operators. Finally, this thesis turns to an investigation of the value of probabilistic solar forecasts in power system operations, namely within the day-ahead unit commitment scheduling problem. We explore three measures to increase operational flexibility under high solar penetrations: battery storage for energy time-shifting, ancillary service provisions by solar PV, and a stochastic unit commitment formulation that prepares for a range of solar outcomes using probabilistic forecasts. These flexibility measures are compared alone and in combination for a case study system based on the Texas power grid under a high solar penetration for eight selected days through the four seasons. The case study simulations show clear benefits to the selected flexibility measures particularly in the spring, when both solar curtailment and variability are high, and on "low-tail" days when solar power is over-forecasted. In particular, allowing PV to provide ancillary services can provide a few percentage points of cost savings throughout the year, but significant cost savings in the realm of 10--20% in spring. Compared to a deterministic formulation, the stochastic formulation reduced unserved reserves by 50--100% when the other two flexibility measures were also in effect, incurring a cost increase of 1.2--3.8% on spring days, but typically 0.7% or less. These results demonstrate the practical usefulness of probabilistic solar power forecasts in supporting reliable and economic electricity service as solar PV deployments continue to grow.
590 $a School code: 0051.
650 4 $a Electrical engineering. $3 649834
650 4 $a Applied mathematics. $3 2122814
653 $a Bayesian model averaging
653 $a Photovoltaics
653 $a Probabilistic forecasting
653 $a Solar power
653 $a Unit commitment
690 $a 0544
690 $a 0364
710 2 $a University of Colorado at Boulder. $b Electrical Engineering. $3 1025672
773 0 $t Dissertations Abstracts International $g 83-03B.
790 $a 0051
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28028371