Enabling Efficient Sizing of Hybrid Power Plants: A Surrogate-Based Approach to Energy Management System Modeling

Assaad, Charbel; Leon, Juan Pablo Murcia; Quick, Julian; Göçmen, Tuhfe; Ghazouani, Sami; Das, Kaushik

doi:https://doi.org/10.5194/wes-2024-96

Preprints

https://doi.org/10.5194/wes-2024-96

Preprints

12 Aug 2024

| 12 Aug 2024

Status: this preprint is currently under review for the journal WES.

Enabling Efficient Sizing of Hybrid Power Plants: A Surrogate-Based Approach to Energy Management System Modeling

Charbel Assaad, Juan Pablo Murcia Leon, Julian Quick, Tuhfe Göçmen, Sami Ghazouani, and Kaushik Das

Abstract. Sizing of Hybrid Power Plants (HPPs), which include wind power plants and battery energy systems, is essential to capture trade-offs among various technology mixes. To accurately represent these trade-offs, an Energy Management System (EMS) is introduced to model the operation of a battery when participating in any market, resulting in realistic operational revenues and costs. However, traditional EMS models are computationally expensive to solve, a challenge that intensifies when integrating these models into sizing processes. This research paper aims to address the critical need for a computationally efficient, accurate, and comprehensive operational model that enables quantitative assessment of HPPs. A novel methodology is introduced to approximate a state-of-the-art EMS model for HPPs involved in spot market power bidding. This approach utilizes singular value decomposition for dimension reduction and a feed-forward neural network as a regression. The accuracy of our methodology is evaluated, showing a root mean square error of 0.09 in predicting hourly operational time series. This method proves effective in accurately evaluating the operation of HPPs across various geographical locations and hence on multiple sizing problems. Furthermore, we utilized the surrogate to evaluate the profitability of several HPPs sizing, achieving a root mean square error of 0.010 on the profitability index. This shows that the developed surrogate is suitable for HPP sizing for given cost and financial assumptions.

Received: 27 Jul 2024 – Discussion started: 12 Aug 2024

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.

Download & links

Charbel Assaad, Juan Pablo Murcia Leon, Julian Quick, Tuhfe Göçmen, Sami Ghazouani, and Kaushik Das

Status: open (until 02 Oct 2024)

Post a comment Subscribe to comment alert

RC1:
'Comment on wes-2024-96', Anonymous Referee #1, 07 Sep 2024 reply
General comments
This paper present the application of a feed-forward neural network on the dispatch strategies of a wind-battery hybrid power plant. The study is motivated by the large computational power required to compute the dispatch strategy using a “realistic” energy management system, which leads to intractable design optimization problems. The topic of modelling fidelity for wind-based hybrid power plants is relevant for the wind energy community. As such, this paper is relevant for the readers of Wind Energy Science.
However, the submitted manuscript presents two major weaknesses regarding (i) its scientific relevance and (ii) its structure and clarity.
The paper focuses on computational power to motivate the use of a surrogate model to replace the EMS. However, this statement is not sufficiently backed up by data or literature. It is unclear whether the study will be relevant for an international audience. As such, the study would be significantly strengthened by shifting the focus towards comparing the studied surrogate models. A discussion on the pertinence of this type of model and the trade-off between accuracy and computational effort would be welcomed in the field.
The paper lacks clarity, structure and conciseness. The language is vague and/or convoluted in some places. As such, the text does not convey information efficiently, which in turns weakens the potential impact of the paper. This can be easily addressed by careful editing.
These two majors comments are detailed in the specific comments below, alongside additional suggestions for improvement.
Specific comments
Figure 8 reports the output time series for the high-fidelity model and for the surrogate model. It can be seen on the figure that at t=0 h the charge power is positive and the discharge power is negative. One would expect the discharge power to be zero. Furthermore, the power output of the HPP is negative between t=2 and 4 h, assuming that the HPP is able to charge from the grid. These two points raise questions on the ability of the surrogate model to represent a valid storage model. Please expand the discussion of Figure 8 to include these considerations.

The story line of the paper focuses on the field of system integration of renewable energy and hybrid power plant. In order to fit better in the scope of Wind Energy Science, consider highlighting the relevance of this study for the field of wind energy in the introduction, discussion and conclusion of the manuscript.

Motivation and research question:
l. 37: “ The importance of a realistic EMS …” : it is unclear from the literature review what is the difference between a high- and low-fidelity model for the EMS. The manuscript provides an overview of different high-fidelity models used in the field, but does not describe the low-fidelity ones. The following reference may be of interest in this context:

Stanley, A. P. J., & King, J. (2022). Optimizing the physical design and layout of a resilient wind, solar, and storage hybrid power plant. Applied Energy, 317, 119139. https://doi.org/10.1016/j.apenergy.2022.119139

Table 1 reports that 1000 “iterations [are] required to find [a] refined solution”. Please put this number in context with the literature. Is it a characteristic of HPP sizing problems to require a large number of iterations to converge?

l. 59: “HPP sizing optimization often relies on a simplified EMS representation”: the introduction of the paper does not describe the problems associated with using a simplified or low-fidelity EMS model. As such, it is unclear why one would prefer a high-fidelity model in a context where computational cost is a problem. Please describe explicitly and quantitatively why a high-fidelity model is superior to a low-fidelity one.

The list of major contributions is a good addition to the introduction. Consider writing explicitly the research question for the study.

l. 100-101: “Integration of the developed surrogate within a framework to evaluate the profitability of an HPP sizing with high accuracy.”: this statement is convoluted. Consider combining it with the previous one, e.g. : “Assessment of the surrogate model’s ability to calculate … time series and profitability of the HPP …”

Methodology
l. 90-91: “Two of which use a multivariate linear regression to establish a baseline and two others are based on Neural networks (NNs)”: please put in context the choice of surrogate models in the introduction. What are the strengths and weaknesses of these models? Have they been applied to models similar to HPP dispatch strategies? Can one expect them to perform well for this type of problem? Are there other surrogate models one could consider for HPP dispatch strategies?

In the first part of section 5, the four surrogate models are compared. However, only one model is used for the rest of the results section. This leads to a lengthy result section, where the high impact results are more difficult to identify. Consider either (i) focusing on the best surrogate model in the entirety of section 5 and moving the comparison between linear and NN models in an appendix; or (ii) compare all four models for all relevant metrics. This second approach would help the reader have insight on the trade-off between accuracy and computational time.

The surrogate models are compared to each other, with the high-fidelity EMS as a reference. However, the results would be significantly strengthened if the comparison were to include a low-fidelity model. For example, if the RMSE for the low-fidelity EMS was 10 times higher than the linear and NN models (Figure 6), this would provide an excellent motivation for the use of a surrogate.

l. 208: “applied for the 2nd and 4th surrogate models”: consider introducing a name for each surrogate, instead of referring to their number in Table 5. The names should match the labels of the figures in the results section.

l. 407: “as the linear model cannot capture the inherent non-linearities of the high-fidelity model.”: why has a linear model been chosen? This statement suggests that the choice of methodology is not appropriate for the study.

Literature review
On the topic of energy markets and subsidies for renewable energy, consider referring to the following report: European University Institute: Robert Schuman Centre for Advanced Studies, Kitzing, L., Held, A., Gephart, M., Wagner, F., Anatolitis, V., & Klessmann, C. (2024). Contracts-for-difference to support renewable energy technologies : considerations for design and implementation, European University Institute. https://data.europa.eu/doi/10.2870/379508

The literature review would be strengthened by citing literature related to machine learning and data-driven methods. An overview of methods for modelling time-series would be a good addition to the paper.

Consider including a short description of the advances done on hybrid power systems (e.g. for micro-grids). This would help contextualize better the paper since the problem of storage sizing and dispatch strategy has been addressed in this field before, even though not in relation to electricity markets.

The literature review could be complemented by citing articles related to the field of bidding strategies. For example, the works by Pierre Pinson or Kenneth Bruninx may be of interest:
Ding, H., Pinson, P., Hu, Z., & Song, Y. (2016). Integrated Bidding and Operating Strategies for Wind-Storage Systems. IEEE Transactions on Sustainable Energy, 7(1), 163–172. https://doi.org/10.1109/TSTE.2015.2472576

Toubeau, J.-F., Bottieau, J., De Greeve, Z., Vallee, F., & Bruninx, K. (2021). Data-Driven Scheduling of Energy Storage in Day-Ahead Energy and Reserve Markets With Probabilistic Guarantees on Real-Time Delivery. IEEE Transactions on Power Systems, 36(4), 2815–2828. https://doi.org/10.1109/TPWRS.2020.3046710

Structure of the paper
The last sentence of the first paragraph states that BESS are valuable to establish robust business cases. Then, the second paragraph discuss the definition of HPPs. In this case, the link between the two paragraph is not clear. Instead, consider introducing HPP in the first paragraph, and narrowing the focus of the study on HPP with storage systems.

l. 68- 71: “To evaluate the value of HPP, …”: this paragraph discussing performance metrics for HPPs does not seem relevant in the introduction. Consider moving it in a later section describing the profitability index.

l. 118-125: “In electricity trading …”: the description of the spot and balancing market does not help describing the methodology of the study. Consider moving this paragraph to the introduction.

Consider restructuring section 2 and 3 into one section describing the metrics relevant to HPPs (including the description of the relevant time series, the EMS and the profitability index) and one section describing the surrogate models.

l. 185-186 “Details on the normalization process appear later on” and l. 189 “The specific use of this method is detailed in this section”: instead of referring the reader to a later part of the paper, consider restructuring the subsection.

Consider shortening the description of the normalization steps (l.194-205) and instead state that scaling is used for all time series.

l. 213-218: the description of the shapes of the matrices does not seem relevant for the study. Consider removing the associated sentences or moving them to an appendix.

Please restructure and shorten section 2.2.3. The text mixes general statements about training neural networks, mentions of the steps of the methodology (l. 235 and l. 255 “the best-performing model … is selected”) and descriptions of the training methodology. This makes the subsection difficult to follow.

Consider moving the description of the cost model to an appendix (l. 322-344).

Clarity and conciseness:
l. 25 “power plants that combine several technologies”: please precise the type of technology. Consider using the terms “electricity generation and storage technologies”.

l. 32-35: “As HPPs transition to market-driven revenue models… throughout the power plant’s lifetime”: the start of this paragraph is vague. What are the “new possibilities and challenges” mentioned? What does the expression “navigate energy markets” mean in the context of the study? What are the characteristics of a “detailed operational strategies”?

l. 35 “Energy Management System”: please define the term, and highlight the difference between other types of “control” in the context of HPP. Consider stating the difference between EMS and adjacent terms such as bidding or dispatch strategies.

l. 95: “Development of a fast and precise surrogate”: the term “accurate” seems more pertinent in this context.

l. 98 “Assessment of the surrogate’s ability to predict hourly operational time series”: consider using the verb “compute”, “calculate”, “model” or “estimate” instead of “predict”, since the latter implies a focus on future (and unknown) data.

Please precise what the term “surrogate model” means in the context of the study. By itself, “surrogate” implies a simplified or approximation model, and does not refer to data-driven or machine learning methods specifically.

l. 404: “This RMSE provides a holistic measure of the model’s accuracy”: why is the term “holistic” used here? Consider rephrasing.

l. 413: “Table 10 contrasts the time required to execute the workflow for each surrogate model” Consider rephrasing this sentence.

l. 426: “Figure 8 shows the difference between the the surrogate’s prediction and the ideal behavior” : what does “ideal behavior” mean here? Consider rephrasing.

l. 537 “the synergistic use of SVD and FFN”: what does “synergistic” mean in the context of the study? Consider rephrasing.

l. 542: “a mere 25 seconds” and “remarkable accuracy”: Please avoid subjective terminology and use neutral language instead.

Figures
What is the information conveyed by Figure 4? Consider removing it.

Figure 5.b. : this representation of the wind distribution is unusual. A more standard representation as the probability distribution function would be more meaningful for the reader.

Please follow the journal guidelines for the captions: https://www.wind-energy-science.net/submission.html#figurestables

Consider using intelligible notation in the legend and labels when possible, instead of introducing the notation in the caption. For example, the labels of Figure 7 do not correspond to previously introduced notation.

Figure 6,7 and 10: including the equation for the RMSE in the label seems unnecessary since the notation and equations is introduced in the main text.

Figure 8: Please indicate the unit on the figure labels.

Figure 9: it is unclear why two figures are relevant here. Consider removing Figure 9 (a).

l. 443: “The mean (μ) being close to zero suggests…”: Note that Figure 9(b) indicates that the mean is equal to zero.

Section 2.1.: a figure illustrating the EMS would be relevant to support its description in the text. For example, Figure 1 could be extended to describe the time schedule for bidding and dispatch decisions.

Equations
For the presentation of the equations in the manuscript, consider introducing the relevant metrics and their notations before the equation.

Consider adding a paragraph or a subsection to introduce the notation used in the paper, since there is a wide variety of symbols, subscripts and superscripts in the manuscript.

Equations 1 to 3 are not equations since they don’t include an equal sign. Consider giving each scalar parameters a name, a symbol and describe their meaning.

The notations “SM” (l.211) and “\lambda” (Eq. 10) are used to describe the price of electricity. Please use a consistent notation throughout the paper.

Equation 8: consider introducing a specific symbol for the RMSE instead of using the abbreviation.

l. 268-271: please introduce notation in a paragraph and not as a list. This comment applies to subsequent equations as well.

Abstract:
Please describe in the abstract that the study was conducted for participation on the day-ahead market and for Denmark.

Please include in the abstract the assumption of perfect forecast.

l. 1-2: “Sizing of Hybrid Power Plants (HPPs), which include wind power plants and battery energy systems, is essential to capture trade offs among various technology mixes”: please be more specific.

l. 4 “model the operation of a battery when participating in any market”: please be more specific about the market mentioned here.

l. 5: “Traditional EMS” : what does “Traditional” mean here? Consider rephrasing.

Minor comments
Please follow the journal guidelines for references, see “In-text citations” at the following link https://www.wind-energy-science.net/submission.html#references . For example, at l. 24, the citations should be “(Dykes et al., 2020; Long et al., 2022; Paska et al.,2009)”

Please use an article before the use of abbreviations, e.g. “an EMS” instead of “EMS”.

l. 146 : please describe the characteristics of the CPU used for the work (model, RAM, etc.)

l. 204: “denoted as Z in the cited paper”: it seems unnecessary to introduce this notation.

l. 257 “Table A1 and A2 in A”: please use the term “Appendix”.

Table 9: a different precision is used for the costs of wind turbines and the cost of batteries. Please use a consistent and meaningful precision for the data.

l. 66-67 “… their applicability is limited to a subset of objective functions that are continuous and convex.”: this statement is incorrect. Gradient-based methods are used in academia and industry for non-convex problems, e.g. wind turbine design problems. Consider removing the sentence.

Reference: “Kingma, D. P. and Ba, J.: ArXiv, 2017. “ Please add the title of the paper.

Check the title for the following reference: “Huang, X., Wang, J., Huang, T., Peng, H., Song, X., and Cheng, S.: An optimal operation method of cascade hydro-PV-pumped storage generation system based on multi-objective stochastic numerical P systems, Journal of Renewable and Sustainable Energy, 13, 016 301,https://doi.org/https://doi.org/10.1063/5.0032455, 2021.”

Check the DOI link for the reference lists. Several of them indicate "https://doi.org/https://doi.org/..."

For the technical reports cited, please indicate the report number

Be aware that Wind Energy Science guidelines state that grey-literature may only be cited if there are no alternatives. The international hybrid power plant conference is grey literature, due to its lack of peer-review: “Das, K., Hansen, A. D., Koivisto, M., and Sørensen, P. E.: Enhanced features of wind-based hybrid power plants, Proceedings of the 4^th International Hybrid Power Systems Workshop, 2019.”

Reply
Citation: https://doi.org/10.5194/wes-2024-96-RC1

Charbel Assaad, Juan Pablo Murcia Leon, Julian Quick, Tuhfe Göçmen, Sami Ghazouani, and Kaushik Das

Viewed

Total article views: 188 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
114	34	40	188	11	9

HTML: 114
PDF: 34
XML: 40
Total: 188
BibTeX: 11
EndNote: 9

Views and downloads (calculated since 12 Aug 2024)

Month	HTML	PDF	XML	Total
Aug 2024	92	26	8	126
Sep 2024	22	8	32	62

Cumulative views and downloads (calculated since 12 Aug 2024)

Month	HTML	PDF	XML	Total
Aug 2024	92	26	8	126
Sep 2024	22	8	32	62

Viewed (geographical distribution)

Total article views: 178 (including HTML, PDF, and XML) Thereof 178 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 13 Sep 2024

Short summary

This research develops a new method for assessing Hybrid Power Plants (HPPs) profitability, combining wind and battery systems. It addresses the need for an efficient, accurate, and comprehensive operational model by approximating a state-of-the-art Energy Management System (EMS) for spot market power bidding using machine learning. The approach significantly reduces computational demands while maintaining high accuracy. It thus opens new possibilities in terms of optimizing the design of HPPs.


Total:	0
HTML:	0
PDF:	0
XML:	0