Synchronous Generator Stability Characterization for Gas Power Plants Using Load Rejection Tests

Round 1

Reviewer 1 Report

The article synchronous generator stability characterization for gas power plants using load rejection tests is well written and presents an interesting research. I can only suggest the following improvements.

1. Mention the contributions of the work explicitly in the introduction sections.

2. Provide the flowchart of your methodology in the article and mention respective equations wherever necessary.

3. The background color for graphs in figures 7 - 9 should be white so that the graph trend can be clearly seen.

4. The conclusion is too lengthy, it must be brief and concise.

Thanks

Author Response

Comments and Suggestions for Authors

The article synchronous generator stability characterization for gas power plants using load rejection tests is well written and presents an interesting research. I can only suggest the following improvements.

First of all, thank you very much for your time and effort spent in the revision process of this manuscript. We would also like to thank the reviewers for their participation in the process.

 

1. Mention the contributions of the work explicitly in the introduction sections.

We had commented the contributions of the work along the manuscript. However, according to your request, they are summed up in the Introduction (Section 1) in order to be highlighted.

To do it, we have included the following statements at the end of Section I: Introduction of the reviewed version of the paper (you will find them yellow underlined):

“…

With the research motivations explained, the main contributions of this study are summed up as follows:

• Performing load rejection tests is proposed to study the power plant’s and synchronous generator’s transient response. This test allow to provoke realistic fast transients emulating real faults where the AVR and the speed governor dynamics are not fast enough.
• With the load rejection tests’ data, well-known models are fitted to satisfy the dynamic response in order to provide this information to the corresponding PSO.
• To corroborate the dynamic transient response characterization, the load rejection tests have been carried out over more than 60 real power plants of the Spanish power grid. In this manuscript, three of them are shown as example.

…”

Thank you for this comment, we think the paper’s contributions are better located and clearer now.

 

2. Provide the flowchart of your methodology in the article and mention respective equations wherever necessary.

You are completely right, for an external reader, following the three models could be complicated. The addition of a flowchart will order better the conceptualization / idea of the paper. We have included a schematic flowchart (new Figure 7) including the main steps in the power plants characterization using a load rejection test at the end of “Section II: Theoretical Models” in the reviewed version of the paper as follows:

“…

Figure 7. Schematic flowchart of the power plant characterization via load rejection test.

…

Finally, an schematic flowchart that sums up the power plant characterization methodology using load rejection tests is ploted in Figure 7. …”

Please let us know if the flowchart is not clear enough and we will update it. Thank you.

 

3. The background color for graphs in figures 7 - 9 should be white so that the graph trend can be clearly seen.

As a flowchart has been added as a new figure 7, figures corresponding to the experimental tests are now Figures 8 to 10.

In first instance we considered the black background of the graphics as they were direct captures of the oscilloscope. However, it is true that its visibility decays considerably. We have changed the graphics to white background. The new figures (in the reviewed version of the paper Figures 8 to 10) stay then in “Section III: Experimental tests” as follows:

(a)	(b)	(c)

Figure 8. Plant 1 tests: (a) No load voltage set point changes; (b) No load frequency set point changes; (c) Load rejection test at P_S = 1.7 MW and Q_S = 1.4 MVAr.

(a)	(b)	(c)

Figure 9. Plant 2 tests: (a) No load voltage set point changes; (b) No load frequency set point changes; (c) Load rejection test at P_S = 5 MW and Q_S = 5 MVAr.

(a)	(b)	(c)

Figure 10. Plant 3 tests: (a) No load voltage set point changes; (b) No load frequency set point changes; (c) Load rejection test at P_S = 15 MW and Q_S = 10 MVAr.

4. The conclusion is too lengthy, it must be brief and concise.

We apologize for the inconvenience. We have tried to sum up and being more specific in the Conclusions section. We have deleted part of the conclusion that was previously commented in other parts of the paper in order to make the conclusions section shorter.

Please find it attached in the reviewed version of the paper. If you have any suggestions about it, please do not hesitate to tell it to us. Here are the modifications done in the reviewed version of the paper:

“5. Conclusions

This paper studies the characterization of power plants parameters from AVR and speed governor single tests under no-load conditions and a posterior load rejection test at reduced power in order to provide the synchronous generator (SG) parameters to the power system operator (PSO) with which simulate contingencies.

Once given the registers of voltage and frequency variations from the on-site tests the parameters’ characterization is carried out in two stages:

• In the first step, the model is simulated under no-load conditions, in order to change the controllers’ set point (steady state parameters) of the AVR and speed governor in an idle move.
• In the second step, the load rejection simulations makes possible to find the transient controllers’ parameters by fitting the parameters in an iterative process, in this case manually performed, to reach simulation curves similar to the measured in the experimental tests.

This paper presents some cases of a field work carried out in more than 60 audited gas power plants, where the response of their controllers was studied. Some of them had problems with wrong controllers adjustments creating an instable system, while others did not have proper coordination in the generator protections, resulting in inappropriate disconnections from the network. For these reasons, this type of audits have high relevance for the power system reliability.

In the audits shown in this paper, it must be noted that at the fitting process did not fit well at steady-state after-transient parts of the simulation. The accuracy of a model depends on the level of detail required for the specific application. While the GENSAL model is a simplified model, it captures the essential behavior of synchronous generators and allows the PSO to perform contingencies simulations with admissible computational times. However, for more detailed analysis, more advanced models such as the Park's model or the detailed electromagnetic transient model (EMT) could be used. The choice of model depends on the specific requirements of the analysis. On the other hand, advance fitting tools such as evolutionary algorithms should be used in further studies to perform the iterative parameters’ fitting process.”

 

Thanks

Thank you again for your comments. We hope all of them have been properly addressed. We truly believe that the revised version of the paper is clearer. Please let us know if you have any further inquiries or suggestions regarding the paper.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article presents the results of identifying the parameters of a simplified model for 3 gas power plants based on experimental data from several different transient processes. The strength of the study is the large experimental data collected by the authors. A fairly large agreement between the simulated and experimental waveforms is shown. The article may be of interest to researchers in the field of analysis of the dynamics of electrical processes. Authors should pay attention to the following points:

1) The authors write: “These models were fitted using Simulink®, with a global 231 simulation scenario that linked all the models together, as shown in Figures 1.”

Please provide more information about the algorithm used for identifying the power plant parameters listed in Table 5. It would be instructive to add a flowchart of the parameter identification algorithm used in the article.

2) Why did you not use some stable heuristic algorithm for identification, such as differential evolution, as was shown, for example, in [R1-R3]?

References:

R1. T. Marčič, B. Štumberger and G. Štumberger, "Differential-Evolution-Based Parameter Identification of a Line-Start IPM Synchronous Motor," in IEEE Transactions on Industrial Electronics, vol. 61, no. 11, pp. 5921-5929, Nov. 2014, doi: 10.1109/TIE.2014.2308160.

R2. Paramonov, A.; Oshurbekov, S.; Kazakbaev, V.; Prakht, V.; Dmitrievskii, V.; Goman, V. Comparison of Differential Evolution and Nelder–Mead Algorithms for Identification of Line-Start Permanent Magnet Synchronous Motor Parameters. Appl. Sci. 2023, 13, 7586. https://doi.org/10.3390/app13137586

R3. Guedes, J. J., Castoldi, M. F., Goedtel, A., Agulhari, C. M., & Sanches, D. S. (2018). Parameters estimation of three-phase induction motors using differential evolution. Electric Power Systems Research, 154, 204-212. https://doi.org/10.1016/j.epsr.2017.08.033

3) Not all simulation waveforms shown in Figures 10-12 coincide quite accurately with the experiment. Please comment on how critical such an inaccuracy is when analyzing the stability of the electrical system.

Author Response

Comments and Suggestions for Authors

The article presents the results of identifying the parameters of a simplified model for 3 gas power plants based on experimental data from several different transient processes. The strength of the study is the large experimental data collected by the authors. A fairly large agreement between the simulated and experimental waveforms is shown. The article may be of interest to researchers in the field of analysis of the dynamics of electrical processes. Authors should pay attention to the following points:

First of all, thank you very much for your time and effort spent in the revision process of this manuscript. We would also like to thank the reviewers for their participation in the process.

 

1) The authors write: “These models were fitted using Simulink®, with a global 231 simulation scenario that linked all the models together, as shown in Figures 1.”

Please provide more information about the algorithm used for identifying the power plant parameters listed in Table 5. It would be instructive to add a flowchart of the parameter identification algorithm used in the article.

In this work, the models used for parameters’ identification were fitted manually by visual inspection. In other words, in the same way as you set-up an automatic voltage regulator or a governor in a real power plant.

First, the load rejection test was made, and the experimental data collected. After, a first iteration was carried out in Matlab-Simulink® introducing the power plant parameters available (from power plants manuals) and initial parameters of the GAST, GENSAL and SEXS models to be fitted (commented in Section II and highlighted in a new figure in the reviewed version of the paper).

After, the simulation is carried out, and the simulated transient response is compared with the experimental one. In case of being close to the first oscillation (the important oscillation when carrying out stability studies in the grid), the fitting process is stopped. This evaluation is carried out by comparing the minimum square error feature between the simulation and the experimental curves.

If the problem does not match properly, the simulation is run again varying the parameters of the speed governor and/or AVR manually.

It is true that the procedure could be done automatically via genetic/evolutionary algorithm and could be an interesting area of research in order to improve the accuracy of the parameters’ estimation (as they avoid falling into local minimum solutions and the time is not a crucial parameter in the characterization as it can be performed offline). However, we humbly think that this improvement would be very small, as the manual fitting was carried out carefully.

To clarify this issue in the paper we have included a new figure with a flowchart that sums up the characterization process via a load rejection test and some comments (you will find them yellow underlined in the reviewed version of the manuscript).

New figure and comments in “Section II: Theoretical Models”:

“…

Figure 7. Schematic flowchart of the power plant characterization via load rejection test.

…

Finally, an schematic flowchart that sums up the power plant characterization methodology using load rejection tests is ploted in Figure 7. The iterative process to fit the parameters of GAST and SEXS models has been carried out manually in order to adjust the first transient oscillation, which is the most important for stability studies. However, other fitting tools such as evolutionary algorithms could be used to solve the problem [27-29].”

Modifications and new statements in “Section IV: Fitting Simulations”:

“After obtaining the registers, simulations are necessary to obtain the GAST, SEXS, and GENSAL model parameters. These models were fitted manually using Simulink®, with a global simulation scenario that linked all the models together, as shown in Figure 1.

…

In order to achieve more accurate solutions, heuristic fitting algorithms should be implemented in the iterative processes of GAST and SEXS models, such as genetic [¡Error! No se encuentra el origen de la referencia.-¡Error! No se encuentra el origen de la referencia.] or evolutionary algorithms [¡Error! No se encuentra el origen de la referencia.]. They would be good candidates for this application, as the computation time is not a critical variable in the characterization process and they avoid the effect of local minimum solutions.”

 

2) Why did you not use some stable heuristic algorithm for identification, such as differential evolution, as was shown, for example, in [R1-R3]?

References:

R1. T. Marčič, B. Štumberger and G. Štumberger, "Differential-Evolution-Based Parameter Identification of a Line-Start IPM Synchronous Motor," in IEEE Transactions on Industrial Electronics, vol. 61, no. 11, pp. 5921-5929, Nov. 2014, doi: 10.1109/TIE.2014.2308160.

R2. Paramonov, A.; Oshurbekov, S.; Kazakbaev, V.; Prakht, V.; Dmitrievskii, V.; Goman, V. Comparison of Differential Evolution and Nelder–Mead Algorithms for Identification of Line-Start Permanent Magnet Synchronous Motor Parameters. Appl. Sci. 2023, 13, 7586. https://doi.org/10.3390/app13137586

R3. Guedes, J. J., Castoldi, M. F., Goedtel, A., Agulhari, C. M., & Sanches, D. S. (2018). Parameters estimation of three-phase induction motors using differential evolution. Electric Power Systems Research, 154, 204-212. https://doi.org/10.1016/j.epsr.2017.08.033

As we said before, we carried out the parameters’ fitting manually. As you pointed out, of course heuristic algorithms would have provided us better results in the GAST and SEXS models’ parameters. But honestly, we did not think about it when we carried out the optimizations and we have to apologize for it. The main reason we performed the fitting manually was the experience of one of our authors that had work extensively in the field and knows how to fit them in a few iterations.

In any case, the process to implement a genetic algorithm or evolutive one is not trivial as Simulink should be automatize to:

• run the simulation
• export the simulated data
• compare it with the experiments
• variate the parameters
• import the new parameters to Simulink
• and run it again.

Unfortunately, actually we have not the computation scripts necessary to change the fitting technique.

Apart from the sentences previously exposed in the previous question, to allude to this topic and encourage further research from this paper, we have introduced the following statement in the further works part of Section V: Conclusions:

“… On the other hand, advance fitting tools such as evolutionary algorithms should be used in further studies to perform the iterative parameters’ fitting process.”

 

3) Not all simulation waveforms shown in Figures 10-12 coincide quite accurately with the experiment. Please comment on how critical such an inaccuracy is when analyzing the stability of the electrical system.

There are several reasons to have differences between the simulations and the real records in power plants.

On the one hand, the transfer functions of the real automatic voltage regulators and the real governors are not identical to the computer models. The model used for the automatic voltage regulator model (SEXS) and the governor model (GAST) are not so complex as the real ones.

On the other hand, there is other reason for the steady state conditions after the transient (load rejection) and this is related to the Droop speed control in the governor and the Voltage droop in the automatic voltage regulator.

The droop speed control reduces the power in case of the frequency increases. Therefore, in case of active power production the internal speed setpoint is higher than the rated speed. For example, in the Fig 13d you can see the rated speed corresponds to 50 Hz, however the internal set point is around 51,5 Hz.

The model was adjusted to have a similar behavior during the first oscillation. But the real governor change the setpoint to the rated speed after a certain time to have the machine ready to be synchronized to the power system.

In case of the voltage regulator (Fig 13c) the voltage at generator terminal was 100,5 % but the internal setpoint was 103%. When the steady state condition is reached, the model has a 103 % voltage. In a similar was the real AVR normally a certain time after opening the generator breaker the setpoint is set to the rated voltage or to the generator busbar voltage (matching voltage mode) in order to have the machine ready to synchronize the machine to the power system again.

The parameters of the AVR model are set to have a similar behavior than the real AVR in during the first oscillation (peak value and time). This is the reason because of in steady state conditions there are differences.

In order to clarify this issue, we have modified the end of the Section IV as:

After several iterations, we obtained the parameters shown in Table 5, which produced the transient behaviors shown in Figures 11, 12, and 13 for Plants 1, 2, and 3, respectively. As can be observed, the AVR and speed governor single fitting was performed with good accuracy. However, the load rejection test transients were only well-fitted for the first oscillation. The transfer functions of real governors and AVR are more complex than the computer models as the governor model (GAST) and the automatic voltage regulator model (SEXS). But rather the simplified models capture the essential behavior of synchronous generators in case of transient operation. It neglects some of the finer details of the regulators as the change of setpoint to have the generator ready to be synchronize again to the power system.

 

 

Thank you again for your comments. We hope all of them have been properly addressed. We truly believe that the revised version of the paper is clearer and we will deeply appreciate your suggestions in further research. Please let us know if you have any further inquiries or suggestions regarding the paper.

Author Response File: Author Response.pdf