IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 6, NOVEMBER 2018 6501

Modeling, Tuning, and Validating SystemDynamics in Synthetic Electric Grids

Ti Xu , Member, IEEE, Adam B. Birchfield , Student Member, IEEE, and Thomas J. Overbye , Fellow, IEEE

Abstract—A synthetic network modeling methodology has beendeveloped to generate completely fictitious power system modelswith capability to represent characteristic features of actual powergrids. Without revealing any confidential information, syntheticnetwork models can be shared freely for teaching, training, andresearch purposes. Additional complexities can be added into syn-thetic models to widen their applications. Thus, this paper aims toextend synthetic network base cases for transient stability studies.An automated algorithm is proposed to assign appropriate modelsand parameters to each synthetic generator, according to fuel type,generation capacity, and statistics summarized from actual systemcases. A two-stage model tuning procedure is also proposed to im-prove synthetic dynamic models. Several transient stability metricsare developed to validate the created synthetic network dynamiccases. The construction and validation of dynamics for a 2000-bussynthetic test case is provided as an example. Simulation results arepresented to verify that the created test case is able to satisfy thetransient stability metrics and produce dynamic responses similarto those of actual system cases.

Index Terms—Power system transient stability, synthetic net-works, generator dynamics, model tuning and validation.

I. INTRODUCTION

POWER system dynamic models are essential for power en-gineers and system operators to perform transient stabilitystudies [1]–[3]. Six benchmark models with up to 16 (68) gen-erators (buses) were presented in a technical report [4] and usedfor comparisons of different stabilizer tuning algorithms. Sev-eral IEEE test cases without dynamics were established in 1962to represent a portion of the American Electric Power System (inthe Midwestern US) [5], and have been extended with generatordynamic models appropriate for performing time-domain simu-lation. A list of power system cases, some of which contain dy-namic models, are summarized in [6] and [7]. Actual large-scalenetwork models can produce realistic, insightful simulation re-sults, but access to those actual models is limited because of con-fidentiality concerns. A solution has been given in our previouswork [8], [9], by developing an automated algorithm to buildsynthetic power system models that represent the complexity

Manuscript received September 18, 2017; revised January 24, 2018; acceptedMarch 21, 2018. Date of publication April 9, 2018; date of current versionOctober 18, 2018. This work was supported by the U.S. Department of EnergyAdvanced Research Projects Agency-Energy under the GRID DATA project.Paper no. TPWRS-01442-2017. (Corresponding author: Ti Xu.)

The authors are with the Department of Electrical and Computer Engi-neering, Texas A&M University, College Station, TX 77843 USA (e-mail:,txu@tamu.edu; abirchfield@tamu.edu; overbye@tamu.edu).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TPWRS.2018.2823702

of today’s electric grids. Those synthetic models are entirelyfictitious and hence can be freely shared to the public. Oncea synthetic network base case with buses, generators, loads,transformers, and transmission lines, has a feasible ac powerflow solution, additional complexities can be added to improvethe realism of the case and include data necessary for varioustypes of studies. One application has been addressed in [10] tointegrate generators’ cost models and operational/physical con-straints into a base case for energy economic studies. This paperwill primarily focus on the construction of synthetic networkdynamics cases for transient stability studies.

Compared to [11] that proposed a methodology for auto-matically synthesizing varied sets of medium voltage distribu-tion feeders, this paper and our previous works [8]–[10] focuson large-scale high-voltage transmission systems. Bus type en-tropies and other metrics obtained using statistics from currentlyavailable cases were developed in [12] and [13] to improve themodeling of synthetic power grids. Paper [14] presented an al-gorithm for generating synthetic spatially embedded networkswith structural properties similar to the North American grids.A bottom-up method used actual demand and generation data tobuild power network models on the footprint of Singapore [15].Topological properties were used in [16] to develop syntheticnetwork graphs. However, work [16] generated only a graphthat matches the topological properties of actual grids, withoutconsidering any generation and demand data. Synthetic casesin [11]–[15], [17] were developed for steady-state power flowanalysis. The research goal of this paper is to extend syntheticnetwork base cases (for power flow studies) with appropriategenerator dynamic models, that are able to reproduce similarresponses as the actual grids. The synthetic network dynamiccases obtained using the proposed construction methodologyare primarily for transient stability studies. Model selections,parameter determination, and model tuning and validation, aremain challenges that need to be addressed to build such dynamiccases.

Previous work [18] performed detailed statistical analysison GENROU (machine), TGOV1 (governor) and SEXS (ex-citer) for coal-fueled power plants, and determined parame-ters of those three models for all synthetic generators in a200-bus case. Work [18] considered only one fuel type and onemachine/governor/exciter combination in a small-scale case.This paper aims to produce large-scale synthetic network1 dy-namic cases in consideration of multiple fuel types and various

1All synthetic network test cases are available for downloads at [19].

0885-8950 © 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistributionrequires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

https://orcid.org/0000-0002-2361-6081https://orcid.org/0000-0002-4428-782Xhttps://orcid.org/0000-0002-2382-2811mailto:txu@tamu.edumailto:abirchfield@tamu.edumailto:overbye@tamu.edu

6502 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 6, NOVEMBER 2018

Fig. 1. Extension process to include generator dynamic models.

machine/governor/exciter/stabilizer models for each fuel type.We build such dynamic cases using publicly available data andstatistics summarized from actual system cases [18]. Detailedstatistical analysis on actual system cases is performed indi-vidually for each fuel type. Commonly-used dynamic modelsare identified and then assigned to synthetic generators. Foreach parameter of every commonly-used model, a value willbe drawn from its typical distribution observed in actual sys-tem cases and then assigned to a synthetic generator. Statisticaland physical relations among parameters are used to facilitatethe parameter determination process. Several transient stabilitymetrics are adopted for validating the created cases. Those gen-erator dynamic models are then improved and validated by atwo-stage model tuning procedure such that the created caseshave satisfactory and realistic dynamic performances. This pa-per presents dynamics for a 2000-bus case on the region of theElectric Reliability Council of Texas (ERCOT) as an example.We perform dynamic simulations with selected N − 1 contin-gency events to verify that synthetic network dynamic casesare able to reproduce satisfactory dynamic responses similar tothose of actual system cases. In summary, contributions of thispaper are threefold:

� Extension of large-scale synthetic network base caseswith generator dynamics in consideration of multiplefuel types and various machine/governor/exciter/stabilizermodel types;

� Tuning model parameters to obtain satisfactory and rea-sonable dynamic responses;

� Development of proper performance indices to validate thesynthetic network dynamic cases.

The extension, tuning and validation steps together buildsynthetic network dynamic cases that are statistically andfunctionally similar to actual system cases.

Four more sections come as follows. In Section II, an algo-rithm is proposed to automatically complete the model assign-ment and parameter determination for adding dynamics to eachsynthetic generator. Transient stability validation metrics arediscussed in Section III. A model tuning process is presentedin Section IV. Section V provides simulation results using a2000-bus test case for illustration, and Section VI presents con-cluding remarks and directions for future work.

II. EXTENSION OF SYNTHETIC NETWORK BASE CASES WITHGENERATOR DYNAMICS

In order to create a synthetic network cases that could re-produce similar dynamic responses as actual grids, generatordynamic model selection/assignment and parameter determi-nation should generate synthetic network dynamic cases thatmatch statistics summarized from actual system cases. In thissection, an algorithm is developed to statistically choose ap-propriate dynamic models and parameters for each syntheticgenerator.

A. Overview of the Proposed Statistical Extension Algorithm

For simplicity, different fuel types or technologies are com-bined into five major categories: natural gas, coal, wind, nuclearand hydro. By doing so, the synthetic network creation processfocuses on modeling generators that compose the majority oftotal installed generation capacity.

Fig. 1 generalizes the steps to determine generator dynamicmodels with their associated parameters. Generation capacityand fuel type of each synthetic generator are defined in thenetwork building process. These two parameters and statisticalanalysis results obtained from actual system cases are basesto add dynamic models and set parameters for each synthetic

XU et al.: MODELING, TUNING, AND VALIDATING SYSTEM DYNAMICS IN SYNTHETIC ELECTRIC GRIDS 6503

TABLE ISTATISTICS ON GAS UNITS’ GOVERNOR TYPES IN THE EI CASE

generator’s machine, turbine-governor, exciter, and/or stabilizer.Given a synthetic generator, we first select a typical governormodel according to its fuel type and statistical analysis resultson governor model types for that fuel type. Model parametersare determined according to its capacity and statistical analysisresults on parameters of the selected governor model types forits fuel type. Upon the completion of governor model selec-tion and parameter determination, similar operations are carriedout to sequentially determine the type and parameter values ofmachine/exciter/stabilizer models. The extension process is in-terested in those models and parameter values that appear muchmore frequently than others in actual system cases. In eachmodel category (machine, governor, exciter and stabilizer) ofevery fuel category, we define one dynamic model as ”domi-nant” if power plants adopting that dynamic model have a rela-tive total capacity over 5%. By doing so, several dominant typesare selected for each model category and each fuel category.Similarly, possible ranges for model parameters are used to setmodel parameter values for each synthetic generator.

In the remaining of this section, we will discuss how to selectdominant model types and assign them to synthetic generators,and then describe how to determine corresponding model pa-rameters. Here, we use gas-fueled power plants as an illustrativeexample.

B. Model Selection and Assignment

1) Governor: Since governors are strongly dependent onfuel types, we start model selection and assignment processwith governor models. As shown in Table I, gas-fueled gener-ation units in an Eastern Interconnect (EI) case [20] have fourgovernor models, among which three of them have a combinedpercentage over 99%. Each of the GAST, GGOV1 and GAST2Amodels has a percentage over 5%. Those three models may bedefined as dominant governor types for gas-fueled units and as-signed to synthetic gas-fueled generators. The NERC reportedthat GAST is an obsolete model that should not be used ininterconnection-wide dynamics cases [21]. Thus, the GAST isnot considered for assignment to synthetic generators.

Those two dominant governor types - GGOV1 and GAST2A- are then assigned to gas units in the created synthetic network,with the same probabilities as their relative percentages. For in-stance, the percentage of gas-fueled generators with the GGOV1governor model among all synthetic gas units is 68.35%.

2) Machine: Table II shows that the GENROU machinemodel is used for the majority of gas-fueled units in the EIcase. Thus, the only dominant machine model is GENROU,

TABLE IISTATISTICS ON GAS UNITS’ MACHINE TYPES IN THE EI CASE

TABLE IIISTATISTICS ON GAS UNITS’ EXCITER TYPES IN THE EI CASE

which will be assigned to all gas-fired generators in the syn-thetic network.

3) Exciter: In the EI case, there are 38 exciters modelsadopted for natural gas power plants. However, except forESST4B, EXST1, EXPIC1 and EXAC2, each of the remain-ing exciter models appears in less than 5% of gas-fueled plants(in terms of unit generation capacity). As shown in Table III, rel-ative percentages of gas units with ESST4B, EXST1, EXPIC1or EXAC2 are 59.79%, 17.92%, 12.89% and 9.4%, respectively.

4) Stabilizer: Since there is no stabilizer model used in theEI case, we assign the IEEEST model to all synthetic generatorsof each fuel type but wind in the synthetic network. The WT3P1model is assigned to wind power plants in the synthetic network.

5) Additional Constraint: Some combinations of machine,governor, exciter and/or stabilizer models are not allowed. Forinstance, WT3G1 and WT3G2 are the dominant wind turbinemachine models, while WT12T1 and WT3T1 are the dominantwind turbine governor models. However, WT12T1 is not com-patible with WT3G1 or WT3G2. This is because WT3G1 andWT3G2 are machine models for type 3 wind generators, whichare not compatible with WT12T1 - a governor model for type 1and type 2 wind generators. Thus, only the WT3T1 is selectedas the governor model for wind turbines in the synthetic net-work dynamic case. During the model assignment process, suchincompatibility of one dominant model with another is essentialto exclude some physically infeasible combinations of machine,governor, exciter and/or stabilizer models.

6) Model Selection and Assignment Summary: As such,similar analysis and model selection/assignment are repeatedfor each fuel type. Table IV summarizes the dynamic modelcandidates for synthetic generators by fuel type.

Here, we note that Table IV simply provides a set of dom-inant dynamic models obtained from the EI case. Other fac-tors may be taken into consideration such that only a subsetof those dominant models are used to build synthetic networkdynamic cases. For instance, to build a case in PowerWorld,PSSE and PSLF formats, we only use generator dynamic modelscompatible with all three software packages.

6504 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 6, NOVEMBER 2018

TABLE IVSUMMARY ON GENERATOR DYNAMIC MODEL CANDIDATES IN THE

SYNTHETIC NETWORK

Fig. 2. The c.d.f. of Xd for the GENROU model of gas units in the EI case.

C. Model Parameter Determination

Upon the completion of dynamic model selection and assign-ment, corresponding parameters are determined individually foreach model of every synthetic generator. Here, we use a syntheticgas unit equipped with the GENROU model as an example.

Statistics obtained from actual system cases present us a pos-sible range of values for each model parameter. For any modelm with a parameter c, values are statistically selected from c’spossible range and assigned to synthetic generators equippedwith model m. For instance, as shown in Fig. 2, the cumulativedistribution function ( c.d.f. ) of parameter Xd can be approx-imated by a linear function2 with R-squared value to be 0.95.Therefore, values can be drawn from the range [1.4, 2.6] andassigned to the GENROU model of synthetic gas-fueled units.

However, parameter assignment performed independentlyfor each parameter may be oversimplified. As such, we alsoconsider the statistical and physical relations among modelparameters during the parameter determination process.

1) Correlation Analysis: Some parameters are depending onfuel type and/or generator capacity, and some other parametershave strong correlations. Given any model m with two stronglycorrelated parameters c1 and c2 , one value for c1 is statisticallydetermined first and the remaining one c2 is assigned with a

2We did not limit our statistical analysis to be linear. We adopt linear re-gression only for parameters with a relatively high correlation coefficient orR-squared value.

Fig. 3. Statistics on Xd , Xq , X ′d , X′q , X

′′d and Xl in the GENROU model

for gas units.

value computed using c1 value and their relation observed inactual system cases.

2) Additional Constraint: There are also some physical lim-itations on assigning values to model parameters. Those lim-itations are used to exclude impossible combination of modelparameters. For example, in the GENROU model, X ′′d < X

′d

and X ′d > Xl should be enforced as hard constraints.3) Example: Statistical analysis among Xd , Xq , X ′d , X

′q ,

X ′′d and Xl indicates two well-fit linear regressions, as shown inFig. 3(b) and (f) for Xd and Xq , and X ′′d and Xl , respectively.Fig. 3(c)–(e) also show the dependence of X ′q on Xq , X

′d on

X ′q , and X′′d on X

′d . Till now, with two linear relations and

three statistical dependences, only one value among Xd , Xq ,X ′d , X

′q , X

′′d and Xl is needed to determine all their values.

Here, the dependence of Xd on generator capacity for gas units[as displayed in Fig. 3(a)] is the starting point to determine thevalues of Xd , Xq , X ′d , X

′q , X

′′d and Xl .

In this way, a set of machine/governor/exciter/stabilizer mod-els and their parameters can be statistically assigned to eachsynthetic generator in every fuel type. This section uses gas unitsfor illustration. However, the proposed extension algorithm isgeneral enough for generators of other fuel types.

The proposed statistical extension process could generalizeto other system dynamics. Here, we use load dynamics as anexample. This first step is to perform statistical analysis on loaddynamic models in available actual system cases. Model selec-tion and parameter determination process similar to that in thispaper is then applied to assign proper load model types and pa-rameters to each load. One of our current research projects [22]uses public data to determine residential, commercial, and in-dustrial ratios of each synthetic load bus. Those information canalso assist in assigning load models since different load sectorsmay have distinct load dynamic model compositions. Constantimpedance loads are still quite typical for dynamic studies andthus adopted in this paper. The remaining of this paper will fo-cus on generator dynamics, and extending the process to morecomplex load models is a subject for future work.

XU et al.: MODELING, TUNING, AND VALIDATING SYSTEM DYNAMICS IN SYNTHETIC ELECTRIC GRIDS 6505

III. TRANSIENT STABILITY VALIDATION METRICS

The previous section develops a statistical approach to extendsynthetic network base cases with generator dynamics. Suchcases are statistically validated, but should also be function-ally validated. Therefore, this section discusses several transientstability validation metrics adopted in our paper.

A. Transient Stability Definition and Classification

Report [23] defines and classifies power system stability intothree primary categories: rotor angle stability, frequency stabil-ity and voltage stability.

1) Rotor angle stability refers to the ability of synchronousmachines in an interconnected power system to remain in syn-chronism after being subjected to a disturbance. Instability oc-curs when machine rotor angles exhibit poor-damped or growingoscillations. The Southwest Power Pool computes the Succes-sive Positive Peak Ratio (SPPR) using the peak rotor angle θmax,kof the kth positive peak and the minimum rotor angle, as a directrotor angle stability measure:

SPPR1 =θmax,2 − θminθmax,1 − θmin ≤ 0.950 (1)

SPPR5 =θmax,6 − θminθmax,1 − θmin ≤ 0.774 (2)

Damping ratio - an indirect rotor angle stability measure -determined by modal analysis methods is also used to quantifymachine rotor angle stability. For example, angular oscillationsare defined by Southwest Power Pool as well-damped if itsdamping ratio is larger than 0.8163% [24].

2) Frequency stability is defined as the ability of an intercon-nected power system to maintain steady frequency following acontingency event that results in a significant imbalance betweengeneration and load. Severe frequency deviation may result inload shedding, generator tripping, equipment damage or evensystem collapse. The minimum/maximum rate of change of fre-quency (RoCoF) and the minimum/maximum frequency duringthe first several seconds after disturbances are commonly usedto assess system frequency stability. For instance, the EasternInterconnection and the Western Electricity Coordinating Coun-cil set 59.7 Hz and 59.5 Hz, respectively, as their low frequencylimit3 [25]. The maximum RoCoF magnitude of 0.5 Hz/s isdefined in the Grid Code in Ireland.

3) Voltage stability refers to the ability of a power system tomaintain steady voltages at all buses. Voltage instability impliesan uncontrolled decrease in voltage, which may lead to voltagecollapse and even system blackout. Take the ERCOT as anexample. Except from the time of fault inception to the timethe fault is cleared, voltage drop is required be above 75%and 70% of pre-disturbance value for load and non-load buses,respectively.

3The highest setpoint in the interconnection for regionally approved underfrequency load shedding systems.

TABLE VIMPACTS OF INCREASES IN GAINS ON CONTROL RESPONSES AND

RECOMMENDED PERFORMANCE INDICES

B. Transient Stability Validation Metrics

This paper considers N-1 contingencies - both large (e.g.,generator outages) and small (e.g., three-phase bus fault) distur-bances - to determine transient stability metrics, which includes:

� Mr - the minimum generator rotor angle damping ratio;� Mf - the minimum/maximum bus frequency values;� Mv - the minimum ratio of bus minimum voltage to pre-

contingency voltage level.In addition to aforementioned transient stability require-

ments, each power grid has characteristic dynamic perfor-mances. Available reports provide dynamic simulation resultson several interconnected power grids. For example, report [26]presents frequency responses of three U.S. Interconnections.Those existing simulation results can also be used as referencesto verify whether synthetic dynamic models are able to repro-duce similar responses as actual system cases.

This section adopts several transient stability metrics to quan-tify frequency, voltage and angular stability. Those metrics areused to describe whether a synthetic network case has stable,well-damped dynamic responses. We refer to industrial and aca-demic documents for defining each metric and determining itsacceptable values. With different tuning and validation targets,the proposed construction methodology can consider differentmetric formulations and other stability indices.

IV. MODEL TUNING PROCEDURE

In order to achieve desirable dynamic performances and sat-isfactory transient stability metrics, a tuning procedure is pro-posed in this section to properly modify the model parametersobtained in Section II.

A. Governors and Exciters of Fossil Fuel Generators

Governors and exciters often use proportional-integral-derivative (PID) controllers. As shown in Table V, differentvalues of each gain have varying impacts on four key character-istics of control responses - rise time, overshoot, settling timeand steady-state errors. Reports [27] and [28] established typi-cal ranges of values of performance indices for speed-governingand excitation control systems, respectively, which are also pre-sented in Table V.

The Ziegler-Nichols (ZN) tuning method is a heuristic methodof tuning a PID controller [29]. Several other tuning methodsderived from the ZN method were presented in [30]. Those tun-ing methods set Kp , Ki and Kd , using the ultimate gain Ku

6506 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 6, NOVEMBER 2018

TABLE VISUMMARY OF TUNING RULES

TABLE VIIHYDROELECTRIC SPEED-GOVERNING SYSTEM SETTING [31], [32]

Fig. 4. Frequency responses without (left) and with (right) tuning stabilizers.

and the associated oscillation period Tu 4. Table VI summarizestuning rules of different methods. Those rules simply provideinitial tuned parameters, on which further adjustments are basedto obtain a set of coefficients so as to have satisfactory controlresponses. The initial dynamic case may have some unstable re-sponses caused by some generators, of which exciter or governorgains are set too high or too low. We adjust exciter and/or gov-ernor gains to stabilize this generator in both its single machineinfinite bus(SMIB) model and the full case. Such adjustmentsare done by first applying methods in Table VI as initial valueguess and further changes according to Table V. For instance,given a control system with significant overshoot issue, the lasttwo tuning rules in Table VI may be used and further adjust-ments can be done by decreasing Kp , Ki or increasing Kd ,according to Table V.

B. Hydroelectric Speed-Governing System

Hydro turbines have a peculiar response due to water iner-tia: when gate position is suddenly changed, the flow does not

4With Ki = Kd = 0, Kp is increased from zero until it reaches the ultimategain Ku , at which the output of the control system has stable and consistentoscillation with an oscillatory period Tu .

TABLE VIIISUMMARY ON BUILDING SYNTHETIC NETWORK DYNAMIC CASES

Fig. 5. Geographic footprint and one-line diagram of the 2000-bus model.

TABLE IXTRANSIENT STABILITY METRIC REQUIREMENTS

change immediately due to water inertia, but the pressure acrossthe turbine is reduced, causing the power to reduce. This pro-cess is determined by a time constant TW [1]. For stable controlperformances, a large temporary droop RT with an appropriateresetting time TR is required. As such, we consider tuning rulesdesigned for hydroelectric speed-governing systems [31], [32].Specifically, as shown in Table VII, TW and TM = 2H (H is themachine inertia) are used to set parameters of PID coefficients,temporary droop RT and its resetting time TR .

C. Stabilizer

Without properly tuned stabilizers, the synthetic network dy-namic models may have poor-damped oscillation under some

XU et al.: MODELING, TUNING, AND VALIDATING SYSTEM DYNAMICS IN SYNTHETIC ELECTRIC GRIDS 6507

Fig. 6. Computed transient stability metrics using the 2000-bus case after being subjected to N-1 contingency events.

Fig. 7. Simulation results using the 2000-bus case after a N-2 contingency event—loss of 2425 MW generation.

contingencies. The first step is to locate the oscillation source.A SMIB model is built to tune the stabilizer of each genera-tor that causes oscillations. Then, we run simulations using thefull case to carry out additional tuning operations if oscillationsare not still well damped. Tuning is very case-dependent. Anytheoretic method may or may not directly give a good solution.If not, we use those solutions as initial guess and continue totune parameters with some manual adjustments. In this paper,we perform N-1 contingency simulations and apply a practi-cal stabilizer tuning method [1], [2], [33]–[36] to appropriatelyadjust stabilizer parameters for eliminating all observed poor-damped oscillations. For illustrations, Fig. 4 presents dynamicsimulation results without and with tuning stabilizers.5

D. From A Full Model to Synthetic Network Dynamic Cases

Table VIII briefly overviews the overall procedure to builda synthetic network dynamic case. Given a synthetic network

5Simulation results are obtained by applying a generation-loss contingencyon the full model developed in Section V.

base case, the statistical extension process determines appropri-ate generator dynamic models with their associated parameters.A two-stage process is adopted to tune speed-governing andexcitation control systems, as well as stabilizer models. Thefirst stage uses a full model with all units on, while the secondstage tunes models for one specific case built from the tunedfull system model. Therefore, the proposed method outputs astatistically and functionally validated and well-tuned syntheticnetwork dynamic case.

In this section, we focus on model tuning and parameter ver-ification methods for uses in this paper, but the construction ofsynthetic network dynamic cases is general enough to considerother methods. The goal of statistical and functional validationis to have a system model that can reasonably reproduce similardynamic responses as actual system cases. Any modeling ac-tivity necessitates certain assumptions and compromises, whichare determined by a thorough understanding of the process howthe model is built and the purpose for which the model is to beused. As such, the eventual validity of a model requires bothengineering judgments and those commonly-used criteria [37].

6508 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 6, NOVEMBER 2018

The overall tuning and validation process considers both afore-mentioned methods and reasonable manual adjustments.

V. ILLUSTRATIVE EXAMPLE

In this paper, we apply the proposed method to model dy-namics for a synthetic network model - ACTIVSg2000 - on theERCOT region, as shown in Fig. 5. The 2000-bus model hasfour voltage levels (500/230/161/115 kV) and a total generationcapacity of nearly 100 GW, a portion of which is committed tosupply a load of 67 GW and 19 GVar. There are 544 generatorsand 2345 transmission lines modeled in the ACTIVSg2000. Allsynthetic network cases are available for download at [19].

Table IX presents transient stability metric requirements spec-ified in reports [25], [38]. To determine whether this case hassatisfactory transient stability metrics, three types of N-1 con-tingency events are selected and applied to disturb this case:

� generator outage;� three-phase transmission line fault, followed by line trip-

ping in 0.01 s;� three-phase bus fault (cleared in 0.01 s).First, we run simulation without any contingency event for a

time period over 100 s to verify that this case has a flat start.Then, for each N−1 contingency event, we perform full dy-namic simulations using the ACTIVSg2000 case, and computetransient stability metrics. Fig. 6 summarizes the computed tran-sient stability metrics of all contingency events. In the boxplot,the upper and the lower bars correspond to the maximum andthe minimum values for each metric, and two filled boxes covereach metric ranging from 90% to median, and median to 10%,respectively. Simulation results verify that the created syntheticnetwork dynamic case has satisfactory transient stability perfor-mances after being subjected to the given contingency events. Asan example, simulated generator rotor angles, bus frequenciesand voltages for a N−2 contingency event - loss of 2425-MWgeneration - are displayed in Fig. 7. We observe that this casehas stable dynamic responses with well-damped oscillations andacceptable voltage/frequency levels.

Therefore, simulation results demonstrate that the proposedmethodology can generate publicly available synthetic networkdynamic cases that are able to produce satisfactory dynamicresponses.

VI. CONCLUSION

In this paper, we use publicly available data and statisticssummarized from an actual system case to produce a large-scalesynthetic network dynamic case. A detailed statistical analysisperformed on selected machine / governor / exciter / stabilizermodels is presented to illustrate the statistical extension processto include generator dynamic models. Three transient stabilitymetrics are used to validate the synthetic network dynamic cases.A two-stage model tuning process is introduced to adjust modelparameters such that the created dynamic cases satisfy transientstability metric requirements.

The application and direct benefit of the concept in this pa-per are twofold. On the one hand, this paper discusses in detailabout the construction framework, consisting of modeling, tun-

ing and validation steps, for building synthetic network caseswith dynamics. Our construction framework is general enoughfor direct applications by experts in academic and industrialareas with their own modeling targets. Additional tuning andvalidation goals can also be implemented. On the other hand,the proposed construction framework generates synthetic net-work dynamic cases that are statistically and functionally similarto the actual grid, and contain no confidential data. Experts inboth academic and industrial areas are free to download thosecases that reproduce realistic, insightful dynamic performances.Although this section uses the ERCOT case to illustrate thesynthetic network creation process, the proposed methodologyis general enough for applications to other footprints of inter-est. For instance, experts from a different region may use thestatistics obtained from actual system models for that region orthose similar ones to build synthetic network cases. They couldfreely share those cases with other researchers that can provideinsights into current system conditions and offer useful opera-tion/investment advices, based on simulation results using theirown synthetic cases.

The large-scale synthetic networks with generator dynam-ics can be used for power system planning, generator siting andsome other applications related to power system transient stabil-ity. The proposed methodology is able to consider other tuningmethods and additional transient stability metrics. Comparisonamong simulation results by different software is of interest, aswell [39]. We will report these studies in future work.

REFERENCES

[1] P. Kundur, Power System Stability and Control. New York, NY, USA:McGraw-Hill, 1994.

[2] P. Sauer and M. Pai, Power System Dynamics and Stability. Champaign,IL, USA: Stripes Publ. L.L.C., 1997.

[3] M. Vaiman et al., “Risk assessment of cascading outages: Methodologiesand challenges,” IEEE Trans. Power Syst., vol. 27, no. 2, pp. 631–641,May 2012.

[4] “Benchmark systems for small-signal stability analysis and con-trol,” Aug. 2015. [Online]. Available: http://resourcecenter.ieee-pes.org/pes/product/technical-publications/PESTR18

[5] “Power systems test case archive,” Univ. Washington, Seattle, WA, USA.[Online]. Available: http://www2.ee.washington.edu/research/pstca

[6] R. D. Zimmerman, C. E. Murillo-Sanchez, and R. J. Thomas, “Mat-power: Steady-state operations, planning, and analysis tools for powersystems research and education,” IEEE Trans. Power Syst., vol. 26, no. 1,pp. 12–19, Feb. 2011.

[7] J. Bialek et al., “Benchmarking and validation of cascading failure anal-ysis tools,” IEEE Trans. Power Syst., vol. 31, no. 6, pp. 4887–4900, Nov.2016.

[8] A. B. Birchfield, K. M. Gegner, T. Xu, K. S. Shetye, and T. J. Overbye,“Statistical considerations in the creation of realistic synthetic power gridsfor geomagnetic disturbance studies,” IEEE Trans. Power Syst., vol. 32,no. 2, pp. 1502–1510, Mar. 2017.

[9] A. B. Birchfield, T. Xu, K. M. Gegner, K. S. Shetye, and T. J.Overbye, “Grid structural characteristics as validation criteria for syn-thetic networks,” IEEE Trans. Power Syst., vol. 32, no. 4, pp. 3258–3265,Jul. 2017.

[10] T. Xu, A. B. Birchfield, K. M. Gegner, K. S. Shetye, and T. J. Overbye,“Application of large-scale synthetic power system models for energyeconomic studies,” in Proc. 50th Hawaii Int. Conf. Syst. Sci., Jan. 2017,pp. 3123–3129.

[11] E. Schweitzer, A. Scaglione, A. Monti, and G. A. Pagani, “Automatedgeneration algorithm for synthetic medium voltage radial distribution sys-tems,” IEEE J. Emerg. Sel. Topics Circuits Syst., vol. 7, no. 2, pp. 271–284,Jun. 2017.

XU et al.: MODELING, TUNING, AND VALIDATING SYSTEM DYNAMICS IN SYNTHETIC ELECTRIC GRIDS 6509

[12] S. H. Elyas and Z. Wang, “Improved synthetic power grid modeling withcorrelated bus type assignments,” IEEE Trans. Power Syst., vol. 32, no. 5,pp. 3391–3402, Sep. 2017.

[13] Z. Wang, S. H. Elyas, and R. J. Thomas, “Generating synthetic electricpower system data with accurate electric topology and parameters,” inProc. 51st Int. Univ. Power Eng. Conf., Sep. 2016, pp. 1–6.

[14] S. Soltan and G. Zussman, “Generation of synthetic spatially embeddedpower grid networks,” in Proc IEEE Power Energy Soc. Gen. Meet., Jul.2016, pp. 1–5.

[15] D. Ciechanowicz, D. Pelzer, B. Bartenschlager, and A. Knoll, “A modularpower system planning and power flow simulation framework for gener-ating and evaluating power network models,” IEEE Trans. Power Syst.,vol. 32, no. 3, pp. 2214–2224, May 2017.

[16] J. Hu, L. Sankar, and D. J. Mir, “Cluster-and-connect: An algorith-mic approach to generating synthetic electric power network graphs,” inProc. 53rd Annu. Allerton Conf. Commun., Control, Comput., Sep. 2015,pp. 223–230.

[17] N. Hutcheon and J. W. Bialek, “Updated and validated power flow modelof the main continental European transmission network,” in Proc. IEEEGrenoble Conf., Jun. 2013, pp. 1–5.

[18] T. Xu, A. B. Birchfield, K. S. Shetye, and T. J. Overbye, “Creation ofsynthetic electric grid models for transient stability studies,” in Proc.IREP Symp. Bulk Power Syst. Dyn. Control, Sep. 2017, pp. 1–6.

[19] “Electric grid test case repository—Synthetic electric grid cases.” [On-line]. Available: https://electricgrids.engr.tamu.edu/

[20] K. M. Gegner, A. B. Birchfield, T. Xu, K. S. Shetye, and T. J. Overbye, “Amethodology for the creation of geographically realistic synthetic powerflow models,” in Proc. IEEE Power Energy Conf., Feb. 2016, pp. 1–6.

[21] “Phase 2 case quality metrics,” NERC, Atlanta, GA, USA, pp. 1–2. [Online]. Available: https://www.wecc.biz/Administrative/System%20Analysis%20and%20Modeling%20Subcommittee%20SAMS%20201-Phase%202%20Case%20Quality%20Metrics.pdf

[22] H. Li, A. L. Bornsheuer, T. Xu, A. B. Birchfield, and T. J. Overbye, “Loadmodeling in synthetic electric grids,” in Proc. IEEE Texas Power EnergyConf., Feb. 2018, pp. 1–6.

[23] P. Kundur et al., “Definition and classification of power system stabilityIEEE/CIGRE joint task force on stability terms and definitions,” IEEETrans. Power Syst., vol. 19, no. 3, pp. 1387–1401, Aug. 2004.

[24] “Southwest power pool disturbance performance requirements,”2016. [Online]. Available: https://www.spp.org/documents/28859/spp%20disturbance%20performance%20requirements%20(twg%20approved).pdf

[25] “Interconnection Criteria for Frequency Response Requirements,”NERC, Atlanta, GA, USA, 2011. [Online]. Available: http://www.nerc.com/docs/pc/tis/Agenda_Item_5.d_Draft_TIS_IFRO_Criteria%20Rev_Final.pdf

[26] P. Mackin R. Daschmans, B. Williams, B. Haney, R. Hunt, and J. El-lis, “Dynamic simulation studies of the frequency response of the threeUS interconnections with increased wind generation,” Ernest OrlandoLawrence Berkeley Nat. Lab., Berkeley, CA, USA, Tech Rep.LBNL-4146E, Dec. 2010. [Online]. Available: http://www.nerc.com/FilingsOrders/us/FERCOrdersRules/Dynamic_Simulation_Studies.pdf

[27] “IEEE Recommended Practice for Preparation of Equipment Specifica-tions for Speed-Governing of Hydraulic Turbines Intended to Drive Elec-tric Generators,’’ IEEE Std 125-2007 (Revision IEEE Std 125-1988),pp. c1–51, Oct. 2007.

[28] “IEEE Guide for Identification, Testing, and Evaluation of the DynamicPerformance of Excitation Control Systems,” IEEE Std 421.2-2014 (Re-vision IEEE Std 421.2-1990), pp. 1–63, Jun. 2014.

[29] J. G. Ziegler and N. B. Nichols, “Optimum settings for automatic con-trollers,” Trans. Amer. Soc. Mech. Eng., vol. 64, pp. 759–768, 1942.

[30] A. S. McCormack and K. R. Godfrey, “Rule-based autotuning based onfrequency domain identification,” IEEE Trans. Control Syst. Technol.,vol. 6, no. 1, pp. 43–61, Jan. 1998.

[31] S. Hagihara, H. Yokota, K. Goda, and K. Isobe, “Stability of a hydraulicturbine generating unit controlled by P.I.D. Governor,” IEEE Trans. PowerApp. Syst., vol. PAS-98, no. 6, pp. 2294–2298, Nov. 1979.

[32] P. L. Dandeno, P. Kundur, and J. P. Bayne, “Hydraulic unit dynamic perfor-mance under normal and islanding conditions—analysis and validation,”IEEE Trans. Power App. Syst., vol. PAS-97, no. 6, pp. 2134–2143, Nov.1978.

[33] E. V. Larsen and D. A. Swann, “Applying power system stabilizersPart I: General concepts,” IEEE Trans. Power App. Syst., vol. PAS-100,no. 6, pp. 3017–3024, Jun. 1981.

[34] E. V. Larsen and D. A. Swann, “Applying power system stabilizersPart II: Performance objectives and tuning concepts,” IEEE Trans. PowerApp. Syst., vol. PAS-100, no. 6, pp. 3025–3033, Jun. 1981.

[35] E. V. Larsen and D. A. Swann, “Applying power system stabilizers Part III:Practical considerations,” IEEE Trans. Power App. Syst., vol. PAS-100,no. 6, pp. 3034–3046, Jun. 1981.

[36] J. Shin, S.-C. Nam, J.-G. Lee, S.-M. Baek, Y.-D. Choy, and T.-K. Kim,“A practical power system stabilizer tuning method and its verification infield test,” vol. 5, pp. 400–406, 2010.

[37] “Power system model validation,” NERC, Atlanta, GA, USA, 2010. [On-line]. Available: https://www.nerc.com/docs/pc/mvwg/MV%20White%20Paper_Final.pdf

[38] “ERCOT fundamentals training manual,” ERCOT, Austin, TX, USA,pp. 290–443. [Online]. Available: http://www.ercot.com/content/wcm/training_courses/109608/ERCOT_Fundamentals_Manual.pdf

[39] K. S. Shetye, T. J. Overbye, S. Mohapatra, R. Xu, J. F. Gronquist, andT. L. Doern, “Systematic determination of discrepancies across transientstability software packages,” IEEE Trans. Power Syst., vol. 31, no. 1,pp. 432–441, Jan. 2016.

Ti Xu (S’13−M’17) received the B.S. degree from Tsinghua University, Beijing,China, in 2011, the M.S. and Ph.D. degrees from the University of Illinois atUrbana-Champaign, Urbana, IL, USA, in 2014 and 2017, respectively. He iscurrently a Postdoctoral Researcher with the Texas A&M University, CollegeStation, TX, USA.

Adam B. Birchfield (S’13) received the B.E.E. degree from the Auburn Uni-versity, Auburn, AL, USA, in 2014 and the M.S. degree from the University ofIllinois at Urbana-Champaign, Urbana, IL, USA, in 2016. He is currently work-ing toward the Ph.D. degree in electrical and computer engineering at TexasA&M University, College Station, TX, USA.

Thomas J. Overbye (S’87−M’92−SM’96−F’05) received the B.S., M.S.,and Ph.D. degrees in electrical engineering from the University of Wisconsin-Madison, Madison, WI, USA. He is currently a Distinguished Research Profes-sor with the Department of Electrical and Computer Engineering, Texas A&MUniversity, College Station, TX, USA.

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 150 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages false /GrayImageDownsampleType /Bicubic /GrayImageResolution 1200 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.00083 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages false /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages false /MonoImageDownsampleType /Bicubic /MonoImageResolution 1600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.00063 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False

/CreateJDFFile false /Description >>> setdistillerparams> setpagedevice