the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Oct 2025
A Decision Support System for the Continuous Economic Evaluation of Wind Farms
Abstract. This paper presents a decision support system that integrates digital twin technology with advanced economic evaluation tools to enable continuous and holistic assessment of wind farm investments. The framework combines real-time turbine health data, including remaining useful life estimates derived from SCADA and condition monitoring systems, with financial models employing discounted cash flow, scenario testing, Monte Carlo simulations, and real-options valuation. Implemented through the open-source DigiWind platform, the system adheres to FAIR data principles and provides a flexible, interoperable environment for asset management. A case study on an 8 MW wind turbine in Germany demonstrates the framework’s ability to guide decisions such as life extension, repowering, decommissioning, or sale under volatile market conditions. Results highlight the importance of coupling technical reliability forecasts with market-based financial outlooks to capture both risks and upside potential, offering a scalable and transparent tool for investors, operators, and policymakers navigating the evolving wind energy sector.
- Preprint
(1057 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 24 Dec 2025)
- RC1: 'Comment on wes-2025-175', Anonymous Referee #1, 21 Nov 2025 reply
-
RC2: 'Comment on wes-2025-175', Anonymous Referee #2, 02 Dec 2025
reply
General Comment:
The paper addresses the integration of economic evaluation into the digital twin framework for decision support regarding the economic assessment of wind farms. The presented approach uses remaining useful lifetime estimates generated from SCADA data and provided by the DigiWind digital twin platform as input to a pipeline of economic evaluation tools to provide recommendations on financial decisions regarding a wind farm. The approach is demonstrated on an 8MW wind turbine, where the decision is made between keeping and selling the asset.
The approach of integrating economic decision models and remaining useful lifetime estimates in a digital twin approach is very relevant and timely, and the authors present a good grasp of the topic. Nonetheless, I believe that some parts of the paper deserve more attention to clearly communicate the approach, its strengths, and remaining challenges/future extensions.
In summary, based on the abstract and introduction, the method is designed to cover a range of complex cases, but the case study is fairly simplistic in comparison. The economic decision model incorporates several methods, but information on the functioning of the economic decision model and its components is needed. Another paper currently under review is cited, but it is difficult to estimate to what extent it covers the economic decision model or overlaps with the presented work. Integration of the remaining useful lifetime is pointed out as a potential future work, but the implication of omitting it in the current version is not addressed. While the approach of integrating economic decision models and remaining useful lifetime estimates into a digital twin is commendable, I recommend a major revision before publication to address these concerns.Main concerns:
- The paper cites an unpublished work that is currently under review for conference proceedings publication (Kerr & Carriveau, Demonstration of an Integrated Framework to Support Major Wind Energy Investments), and which I was not able to find a published preprint of. Additionally, in the introduction, it is stated, "This paper builds on those efforts by combining two proven processes ...". Does that imply that the economic decision model is already/is being published/to be published in another paper, and that the novelty of this work is the combination of that model with the DigiWind platform? Based on the current description in the paper, I would naively assume that the combination of these two methods is as simple as using the output of the first method (RUL of each component) as input into the second. If the novelty of the paper is the combination of both existing approaches, I believe it should be explained in more detail which challenges there are in this coupling and how they are solved to create novelty.
- The remaining useful lifetime estimates based on SCADA data likely have substantial uncertainty, especially since RUL is integrated based on damage increments. However, the current approach treats them as fixed. This will likely result in significantly overconfident recommendations by the economic model and possibly compromise the usefulness of the model. While the potential for extension to incorporate RUL uncertainty is mentioned in the discussions, the implications of not having RUL uncertainty in the current model should at least be discussed as well. Additionally, how does the implementation of all RUL uncertainties affect the number of iterations required by the MC simulation, since several more input probability distributions need to be sampled?
- Section 2.2 "Economic Decision Support Framework" currently describes the economic evaluation model only superficially through the figures, but lacks information on the details of the individual components and how outputs of one method are utilized in the next. Without the unpublished paper from Kerr & Carriveau on the economic decision model available yet, it is difficult to judge if this section in its current form in combination with the second paper, is sufficient to understand and reproduce the economic decision model
- As a key component of the paper, the flow through Figure 4 should be explained in more detail the text.
- Depending on whether the target audience has a financial background or covers a broader wind energy community (which I would expect based on the journal), it may be beneficial to at least briefly explain DCF, ROV, and NPV.
- It should be explained in more detail in the text how exactly DCF, scenario testing, ROV, and MC interact, and why the methods were chosen.
- Considering that the final aim is a decision support tool, the "key financial metrics" and their impact on the final decision recommendation should be clearly defined. In particular, several options are mentioned in the text (lifetime extension, repowering, decommissioning, and selling), but Figure 4 shows the final output as only ROV/NPV. How is this used to recommend diverse actions?
- It is mentioned that tools like Markov Decision Process could be incorporated to adapt to changing conditions over time, but it is not mentioned which conditions are covered or where in the pipeline it should be implemented.
- Figure 4: It is said that the key integration with the DigiWind platform occurs at the RUL Data and Turbine Data(Age) nodes. Does turbine age simply describe years in operation and then use a fixed end of lifetime based on insurance/warranty/average life expectancy, or does it incorporate the component RULs?
- Figure 4, the Markov Chain box only covers the AEP calculation. I assume that the MC simulation is run over a larger part of the full figure to include also other uncertainties? If not, given that RUL is estimated as constant and the Monte Carlo simulation box currently only covers AEP per remaining year of turbine lifetime, does this imply that the only uncertainty in the Monte Carlo simulation comes from the annual energy production? Why is a Monte Carlo simulation needed, and how could RUL uncertainty and other uncertainties be incorporated fully, as stated in the discussion, if they also affect blocks outside of the MC?
- The approach is presented as a holistic tool to evaluate complex financial cases on the farm level that can account for uncertainty on most parameters and provides decision support considering several decision outcomes (lifetime extension, repowering, decommissioning, selling). However, in comparison, the case study only investigates a single turbine without uncertainty on most parameters (including RUL, component prices, etc.), only a single risk factor of changing cap price, and only two decision stages(keep/sell). Additionally, the real-time integration with the DigiWind platform was pointed out as a major contribution of the paper, but the case study is evaluated as a single event where it treats the RUL as fixed inputs and does not consider their fluctuation over time. I believe that a more complex case study would significantly benefit the paper by demonstrating the strength of the approach that is presented in the abstract, introduction, and discussion.
Minor comments:
- Section 2.2 appears twice (2.2 Remaining Useful Life Calculation and 2.2 Economic Decision Support Framework)
- While the remaining useful lifetime estimates combined with the economic decision model provide a first approximation, other factors, such as maintenance costs, maintenance weather windows, and lost power production, complicate the decision process further by introducing more variability into the system. Could these factors be incorporated into the presented approach through probability distributions and additional simulation components?
- How simple is it to sell the wind turbine/farm for a given price when the financial outlook suggests selling? It seems to me that if the simulation suggests a negative NPV, the only way to sell is to find a buyer who believes that the farm has a better NPV and that the financial decision model the vendor is using is not accurate? Furthermore, assuming the RUL can be evaluated through the platform dynamically and automatically, do these market parameters still have to be continuously evaluated and updated manually?
- "If the decision making were fixed, say if the owner were certain that they wanted to keep the farm, then the operation could be optimized to create the most economically profitable scenario within this mindset." Could the authors elaborate on what such optimization would look like?
- Typo on Capacity_{Nameplat} -> Nameplate
- The acronym NPV is not included in the Abbreviations or spelled out in the text.
Citation: https://doi.org/10.5194/wes-2025-175-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 206 | 39 | 17 | 262 | 14 | 13 |
- HTML: 206
- PDF: 39
- XML: 17
- Total: 262
- BibTeX: 14
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General Comment:
The manuscript addresses an relevant and timely topic: integrating digital-twin-based asset health information with long-term economic evaluation for wind farms. The conceptual framework presented is sound, and the overall vision of linking SCADA/CMS-based RUL estimation with investment decision support is compelling.
My main concern lies in the simplicity and limited depth of the case study. Although the earlier sections of the paper emphasize detailed RUL modelling, physics-informed degradation assessment, and component-level trajectories, the case study ultimately incorporates RUL only as fixed replacement times. As a result, the economic model is driven by binary electricity-price scenarios rather than integration of dynamic or uncertain RUL estimates. A more comprehensive case study, that incorporates RUL uncertainty, component-specific degradation paths and connected interventions would strengthen manuscript.I therefore recommend major revision.
Individual Comments:
Abstract:
The abstract could be improved by providing a more detailed picture of the study.
- how are health metrics derived
- main components/functionality of the financial model
- which kind of scenarios are tested through monte carlo analyisis
- A brief summary of key results
Introduction:
Line 30: I would challenge the statement that methods have not progressed in the last 20 years. Please specify.
Line 37: Please specify the "varying needs" of each turbines
Line 43 and following : It would be useful to give an overview of key methodologies for RUL estimation. What are main assumption and limitations? Also it would be worth mentioning that RUL estimation from SCADA data is connected to high uncertainty as lifetime may not be determined by fatigue.
Section 2.1:
The description of the DigiWind digital twin framework is somewhat generic and does not convey clearly how the architecture supports the specific integration tasks discussed later. The text lists several software concepts (microservices, MQTT, Docker, BPMN) but does not explain their functional relevance in the context of the decision-support workflow. I suggest revising the section and potentially integrating an overview figure.
Section 2.2:
feels too high-level and does not provide enough methodological detail to understand how RUL is actually computed. Which components and failure modes are targeted by the surrogate model?
How are load–damage relationships derived and how is partial-operation scaling is applied. As the economic evaluation highly depends on the RUL information the manuscript would benefit from more detail.
Section 2.2 (section numbering is duplicated). The authors list several methods (DCF, Monte Carlo simulation, scenario testing, real-options valuation), but the manuscript does not describe how these techniques are implemented. For example, which variables are sampled in the Monte Carlo simulation, how uncertainties are parameterized,or how RUL estimates enter the cash-flow model. Likewise, the Real-Options methodology is mentioned but not explained.
Additionally, the reference to future integration of Markov Decision Processes should be moved to future work section.
Section 2.3:
The authors introduce the WESgraph ontology, a SPARQL query, and a BPMN workflow diagram, but I find it difficult to understand how these concepts interact operationally. Are SPARQL queries are executed at runtime to determine connections? Is the ontology used at runtime? The section would benefit from a clearer explanation of the operational workflow.
Also please explain how FAIR data principles are ensured.
Section 3: The case study tests different stochastic scenarios describing the transition from feed-in tariff market price while the market price is assumed to be a constant value. All other parameters are deterministic and fixed, particularly also the RUL estimates. So, in my understanding, the case study is merely a financial sensitivity test on future electricity pricing, not really an integrated techno-economic RUL decision problem. Uncertainty in RUL estimation, the degradation trajectory, or some updating through the digital twin is not considered.
Results: Please provide a more detailed explanation of how the real option value is calculated. Some interpretation on the double-peak shape of the NPV distribution would be helpful as well. Legends for for NPV/ROV level lines would be helpful in the plots.