An Open-source Model for Simulation and
Corrective Measure Assessment of the 2021 Texas
Power Outage
Dongqi Wu
Texas A&M University https://orcid.org/0000-0002-0238-6088
Xiangtian Zheng
Texas A&M University https://orcid.org/0000-0003-2884-3213
Yixing Xu
Breakthrough Energy
Daniel Olsen
Breakthrough Energy
Bainan Xia
Breakthrough Energy
Chanan Singh
Texas A&M University
Le Xie ( le.xie@tamu.edu )
Texas A&M University https://orcid.org/0000-0002-9810-948X
Article
Keywords: 2021 Texas power outage, energy demand, generation capacity outage
Posted Date: April 5th, 2021
DOI: https://doi.org/10.21203/rs.3.rs-384535/v1
License: This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
An Open-source Model for Simulation and1
Corrective Measure Assessment of the 2021 Texas2
Power Outage3
Dongqi Wu
1,†
, Xiangtian Zheng
1,†
, Yixing Xu
2
, Daniel Olsen
2
, Bainan Xia
2
,4
Chanan Singh
1
, and Le Xie
1,3,*
5
1
Department of Electrical and Computer Engineering, Texas A&M University, College Station, Texas, USA6
2
Breakthrough Energy Sciences, Seattle, Washington, USA7
3
Texas A&M Energy Institute, College Station, Texas, USA8
*
Corresponding author: le.xie@tamu.edu9
†
Co-first author10
ABSTRACT
11
Unprecedented winter storms that hit across Texas in February 2021 have caused at least 4.5 million customers to experience
load shedding due to the wide-ranging generation capacity outage and record-breaking electricity demand. While much
remains to be investigated on what, how, and why such wide-spread power outages occurred across Texas, it is imperative
for the broader research community to develop insights based on a coherent electric grid model and data set. In this paper,
we collaboratively release an open-source large-scale baseline model that is synthetic but nevertheless provides a realistic
representation of the actual energy grid, accompanied by open-source cross-domain data sets. Leveraging the synthetic gr id
model, we reproduce the blackout event and critically assess several corrective measures that could have mitigated the blackout
under such extreme weather conditions. We uncover the regional disparity of load shedding. The analysis also quantifies the
sensitivity of several corrective measures with respect to mitigating the power outage, as measured in energy not served (ENS).
This approach and methodology are generalizable for other regions experiencing significant energy portfolio transitions.
12
Main13
The extreme winter storm and associated electricity outages in February 2021 are estimated to have caused more than 70 deaths
1
14
and $200 billion economic loss
2
in the state of Texas. Besides the official brief review
3
and ongoing internal investigation on
15
the Electric Reliability Council of Texas (ERCOT), there have been preliminary reports from non-peer-reviewed articles
4–6
and
16
press interviews
7, 8
on potential causes and technical solutions for this blackout event. Given the complexity and confidentiality
17
of the actual electric grid model and relevant information, it becomes very challenging for the broader research community to
18
develop analysis and provide credible insights on such major events.19
While researchers recently have contributed to the creation of large-scale synthetic grid models
9
for analysis such as
20
macro-scope energy portfolio transition
10, 11
, cross-domain and open-source approaches to reproduce the Texas blackout event
21
and quantify impact from corrective measures against extreme frigid weather are still at a nascent stage, with several gaps in
22
existing research. First, existing open-source large-scale synthetic grid models are not ready-to-use for the event reproduction
23
without rigorous calibration. Second, the lack of aggregated and processed event timeline data prevents exhaustive simulation
24
and further investigation. Last but not least, the lack of consistent quantified criteria renders studies on the effectiveness of
25
potential corrective measures and their combined effect incomparable.26
Here, we collaboratively develop an open-source large-scale synthetic baseline grid
12
, providing a realistic representation
27
of the actual ERCOT electric grid, accompanied by the open-source data set along the event timeline. This ready-to-use
28
multi-platform synthetic grid model is calibrated based on open-source data sets, including generation by source, load by
29
weather zones, generation unit outage timeline, load shedding record, etc. To the best of our knowledge, it is the first fully
30
open-source approach to model, simulate, benchmark, as well as propose corrective measures against the 2021 Texas power
31
outage. The reproduction results transparently replay the timeline of the change of system generation capacity, load demand and
32
load shedding associated with the key events, aiding public researchers in combing the event development process, investigating
33
the event causes, and providing possible technical solutions. Additionally, we propose and evaluate multiple technical solutions
34
that can possibly mitigate the electricity scarcity under such extreme weather conditions, including energy system winterization,
35
interconnected HVDC lines, up-scaled demand response program and strategic energy storage facilities. Leveraging the
36
synthetic grid, we perform quantitative analysis on the corrective measures in the aspect of reducing the extent of blackout
37
events. Our results indicate the strong disparity among the winterization effectiveness for generation units of various source
38
types and geographical regions, the quantitative assessment of certain corrective portfolios, and the interdependence of per-unit
39
performance of corrective measures.40
Open-source Synthetic Grid Model and Data41
We first collaboratively develop an open-source, large-scale, synthetic baseline grid that provides a realistic representation of
42
the actual ERCOT electric grid, and then integrate generation and load-related data along the event timeline, which are both
43
publicly available on Github
12
. The original sources are detailed in the documentations in the Github repository (see Data and
44
Code Availability). In this paper, the synthetic model creation particularly focuses on feasibility of the direct current optimal
45
power flow (DCOPF) solution without transient stability assessment for the following reasons. First, DCOPF is commonly
46
used for generation dispatch in the real-world system operation under normal conditions, due to its computation efficiency
47
compared to the alternating current optimal power flow (ACOPF). Second, although maintaining the system transient stability
48
affects the system operation under emergency conditions, the major decisions that determine the blackout event thread, such as
49
generation dispatch and load shedding, mostly depend on the power flow solution feasibility. Finally, regulations on critical
50
energy/electric infrastructure information (CEII) limit the disclosure of information about actual power grid components, and
51
hence dynamic model parameter calibration for generators is impractical due to the lack of publicly available information.52
For the purpose of blackout event reproduction and ‘what-if’ analysis, we create a comprehensive blackout event dataset
53
via collection from publicly available sources
3, 13–16
and estimation (see Methods). This dataset integrates actual load by
54
weather zone, actual generation by source, 7-day-ahead load forecast by source, solar and wind generation forecast, generation
55
units outage, actual available generation capacity, actual load shedding and customer power outage into a single ready-to-use
56
format. Here, we define the counterfactual load as the 7-day-ahead load forecast and the simulated load shedding as the gap57
between the post-shed and counterfactual load, and introduce the
estimated generation capacity
by weather zone based on
58
rated generation capacity, thermal generation units outage and actual renewable generation (see Methods), all of which play
59
important roles in the event reproduction and what-if analysis.60
The synthetic ERCOT grid model is adapted from an existing test system
17
and rigorously calibrated in several aspects,
61
namely generator units capacity, and transmission line rating (see Methods). The geographical load distribution comes from
62
the existing grid model
9
, while their real-time magnitudes in simulation are adjusted according to the real load dataset or
63
calculated ones depending on whether the real load data are available. The generator units capacities are updated to the
64
actual available generation capacity
15
in January 2021. Without modifying the network topology, some transmission lines are
65
upgraded to ensure that the model remains feasible in the period leading up to the blackouts and that no renewable generators
66
are unreasonably curtailed due to congestion.67
Integrating the open-source datasets, the ready-to-use multi-platform synthetic grid model is the first open-source simulation
68
package dedicated to potentially provide firm interdisciplinary insights into the particular real-world blackout event. To give an
69
intuitive impression on the blackout event, we provide an event overview along with regional generation outages and customer
70
power outages (Fig. 1). The timeline of the whole blackout event (Fig. 1-a) that contains the actual total load, actual and
71
estimated generation capacity shows the electricity scarcity due to the high load demand and wide-ranging generation outage.72
The actual generation and generation outage across eight ERCOT weather zones (Fig. 1-b) show that the generation outage at
73
the darkest hour mainly consists of natural gas thermal generation outages across ERCOT and renewable generation outages in
74
the North, West, Far West and South zones. We also find the regional disparity of load shedding (Fig. 1-c) from the aggregated
75
county-level utility-reported customer outage data
16
during the "darkest" period, namely from 8 p.m. February 15 to 11 a.m.
76
February 16. Specifically, the satellite counties around Houston in the Coast zone and several counties distributed in the West
77
zone suffered the most severe outages. We notice the significant gap between the estimated generation capacity and actual
78
online capacity before February 15 and increasing mismatch between them after noon on February 16 that are in line with
79
expectations due to several reasons explained in Supplementary Note 1. We have observed that there exists a substantial
80
mismatch between actual load and either actual online or estimated generation capacity, which is beyond the reserve limit.
81
This mismatch may be attributed to multiple reasons, such as transient stability requirements, reactive power demands, and
82
capped wholesale market price
18
, which deserve more investigation but are nevertheless outside the scope of this paper. To
83
show the complex but realistic features of the synthetic grid, we visualize the topology of the whole synthetic grid (Fig. 2),
84
of which load distribution, generation units capacity and transmission lines rating are calibrated based on the static ERCOT
85
grid-related data (see Methods). In the following analysis, we will leverage the synthetic model along with the blackout event
86
dataset to reproduce the blackout event via simulation and perform quantitative assessments of multiple corrective measures
87
against extreme frigid weather.88
2/
13
Reproduction of the 2021 ERCOT Blackout Event89
To reproduce the ERCOT blackout event from February 15 to February 18, we simulate the synthetic grid model using the
90
aggregated data, where realistic load shedding allocation and DCOPF are key steps. We take the
estimated generation capacity91
(the binding constraint for load shedding) and
counterfactual load
(the ebb-flow pattern of load) as the simulation inputs. To
92
achieve the fidelity of load shedding, we mimic the guides of load shedding and restoration
19, 20
to determine the total load
93
shedding amount at any given moment, and perform appropriate spatial allocation of load shedding to reflect its regional
94
disparity (see Methods). We reproduce the blackout event by iteratively solving DCOPF given the post-shedding load (see
95
Methods), which reveals the hourly change of geographically distributed load, generation and load shedding across the ERCOT
96
in detail.97
We demonstrate the fidelity of the synthetic grid and the associated simulation methods by the reproduction results of load
98
shedding and generation composition (Fig. 3). To quantify the severity of the blackout event, we use the power system reliability
99
index
energy-not-served
21
(ENS), defined as the integral of load shedding over the event timeline, to quantitatively evaluate
100
the load shedding throughout the rest of this paper. We first demonstrate the fidelity of the geographical load distribution and
101
the designed load shedding algorithm by the good match between the actual and simulated total load shedding (Fig. 3-a) that
102
respectively represent a total of
998.8
GWh and
929.6
GWh ENS. The unavoidable mismatch attributes to the combined effects
103
of errors in synthetic grid modelling and system operation under emergency conditions (see Supplementary Note 2 for the
104
remark on the mismatch). We then validate that the simulation well captures the regional disparity of load shedding across eight
105
weather zones
22
by comparing the simulated zone-level normalized load shedding with the real one (Fig. 3-b). It shows that Far
106
West experienced the most disproportional load shedding among all zones and Coast has suffered from a significantly worse
107
condition compared to the other two most populous zones: North Central and South Central. Finally, we validate the fidelity of
108
generation units capacity by type and generation cost curves used for DCOPF by showing the almost perfect match between
109
actual and simulated generation composition throughout the event (Fig. 3-c). The reproduction results validate the synthetic
110
baseline model, the related data and the associated simulation methods, which provide a reliable basis for the following what-if
111
analysis. Additionally, the reproduction results can transparently show the change of load, generation and load shedding along
112
the timeline, aiding public multidisciplinary researchers in combing the event development process, investigating the event
113
causes and providing possible technical solutions. The transparency and reproducibility of the synthetic grid model also allow
114
public researchers to contribute to further model development and calibration.115
Quantitative Assessment of Corrective Measures against Extreme Frigid Weather116
In order to provide firm insights into potential solutions against problems revealed by extreme frigid weather, we start from
117
investigating four possible corrective measures, namely, generation units winterization, interconnection HVDC lines, up-
118
scaled demand response program, and strategic energy storage facility (see Supplementary Figure 2 for conceptual diagram).
119
Generation units winterization only refers to the adapted winterization treatments for electric energy generation units, assuming
120
no winterization is currently applied for any units. Particularly, we reserve the impacts of wide-ranging natural gas scarcity
6
121
due to failures in the non-winterized natural gas supply chain (see Methods and Supplementary Note 4). Additional HVDC
122
interconnections, besides two existing HVDC ties, respectively connect from the Western Interconnection to the West zone and
123
from the Eastern Interconnection to the Coast zone, require necessary transmission lines upgrade around the locations of their
124
converter stations (see Methods). Up-scaled demand response refers to various incentive programs distributed across ERCOT
125
that require voluntary reduction of electric energy demand. Energy storage refers to the large utility scale storage facilities
126
that absorb the excessive energy during off-peak hours and release it at high power during emergency hours. Here we treat
127
the energy storage as the first-aid measure while taking the other three as the sustained electricity supply measure, especially
128
viewing the generation units winterization as the primary preventive measure. Therefore, we conduct quantitative assessment of
129
all corrective measures from different perspectives in the following analysis.130
Taking these corrective measure settings into account, we first quantify the impacts of each single sustained electricity
131
supply measure of distinct extents in load shedding (Fig. 4). We find that 60% generation units winterization can effectively
132
reduce the ENS from 929.6 GWh to 40.8 GWh (Fig. 4-a), and about 80% generation units winterization can prevent the
133
blackout entirely, where we reserve the impacts of non-winterized natural gas supply chain. We also find that HVDC lines and
134
up-scaled demand response of equal capacity have similar but different effectiveness on mitigating the electricity scarcity (Fig.
135
4-b,c), which respectively reduces the ENS by
64.1
GWh and
67.5
GWh per 1 GW capacity (see Supplementary Note 6 for
136
more details). Since energy system winterization is the most straightforward solution against extreme frigid weather, we attach
137
additional importance to prioritizing the winterization of generation units of specific source types in different regions. We
138
perform the quantitative assessment of the effectiveness of facility winterization by source and region on the electricity scarcity
139
mitigation. The results indicate the distinct performance of per-GW generation units winterization (see Supplementary Note 7
140
for more details). Based on this, we suggest the priority of winterization for the disabled nuclear generation units located in the
141
3/13
South Central, natural gas generation units across ERCOT, coal generation units in the Coast and wind generation units in the
142
West.143
Given the quantitative assessment of single sustained electricity supply measure, we investigate the performance of several
144
sustained electricity supply portfolios and assess the first-aid capability of energy storage on the basis of sustained sources. We
145
have selected three appropriate winterization portfolios based on the foregoing priority analysis (see Supplementary Table 1) to
146
provide a quantitative assessment in terms of additional ENS reduction contributed by both HVDC and demand response (Fig.
147
5). First, we find that the performance of HVDC and demand response are slightly different but almost equivalent under certain
148
cases of winterization portfolio. Second, we find the per-GW performance of HVDC and demand response decreases as the
149
winterization capacity increases. For first-aid outage mitigation at the load shedding peak hour, we focus on load shedding
150
peak clipping by the strategic energy storage facilities on top of sustained corrective measure portfolios, each refers to one of
151
the three winterization portfolios (see Supplementary Table 1) together with HVDC and demand response of 2 GW. We find
152
that the performance of per-GWh capacity reduces as the total energy storage capacity increases, or along with increasingly
153
sufficient sustained supply corrective measures (Fig.
6).154
To summarize the key findings obtained in the foregoing quantitative analysis, we find the strong disparity of generation
155
units winterization of various source types in multiple regions, and the interdependence of per-unit performance of corrective156
measures, based on the quantitative assessment of certain corrective portfolios.157
Discussion158
We introduced an open-source cross-domain synthetic ERCOT grid model associated with the blackout event-related data,
159
which is the first-of-its-kind for broader energy researchers to provide insights into the 2021 Texas power outage event.
160
Simulation results based on this open synthetic model are shown to have captured key characteristics of the real-world event,
161
uncovering the key regional disparity of load shedding. The quantitative assessment of the corrective measures and portfolios
162
has indicated the strong disparity of winterization effectiveness among generation units of various types in multiple regions
163
and the interdependence of per-unit performance among corrective measures. It can immediately inform policy makers of
164
the priority of generation units winterization, the quantitative assessment of certain portfolios on mitigating the blackout and
165
the necessity of launching systematic investigations on the combined effects of corrective measures, which can potentially be
166
generalized for other regions around the world which are experiencing the dual challenge of energy portfolio transition and
167
extreme weather conditions.168
This open-source, cross-domain, data-driven approach to analyzing a real-world power grid during extreme events provides
169
a fresh perspective to allow broader climate and energy research communities to have high fidelity characterization of what
170
exactly happened and what could have been corrected in a large power system, as nations and regions aggressively tackle
171
energy and climate challenges. The transparency and reproducibility of the synthetic grid model will contribute its unique value
172
to the future energy policy making and technological solution discussion in the broader context of energy portfolio transition
173
and frequent extreme climate, which otherwise are mostly closed-door research in a few government agencies and transmission
174
organizations on confidential real-world models.175
For the future public research and internal ERCOT investigation, it will be fruitful to provide solid and reasonable
176
explanations to the following observations that are unanswered in this paper. First, we notice the wide-ranging natural gas
177
generation capacity outage and de-rating are not simply due to the freezing temperature, but also to natural gas scarcity and
178
interruption in the supply chain. It is particularly important to estimate and predict the impacts of interdependence between two
179
energy infrastructure systems on the overall energy system reliability and energy market stability on both sides under extreme
180
weather conditions. Second, we notice the price-ceiling-hitting whole-sale electricity price
18
at $9,000 per MW, that lasted for
181
three days ending on February 18. Its quantified impacts on generation dispatch and load restoration still remain unknown.
182
More investigation is necessary for demonstrating and developing a benign power market mechanism that can encourage
183
improving the reliability and resiliency of power grids. Last but not least, the interdependence of per-unit performance among
184
corrective measures emphasizes the necessity of systemic assessment on the combined cost-effectiveness of technical solution
185
bundles.186
Methods187
Data Aggregation188
In order to reproduce the blackout event and perform quantitative what-if analysis, we integrate the blackout related data during
189
the event period between February 14 to February 18. The original sources for all datasets are provided in the Data and Code
190
Availability section. We integrate the datasets via two ways, namely data collection from multiple resources
13–16
and data
191
estimation.192
4/13
