A Data-Driven Observer for Wind Farm Power Gain Potential: A Sparse Koopman Operator Approach

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Minor revision is required. Below are my comments:

1. In lines 66–72, while mentioning ML techniques, there is no fair comparison or discussion of why Koopman is better than deep learning or Gaussian processes.
2. The paper contains different grammatical issues.
3. A lot of space is spent reviewing AWC and wake-up control, but no gap statement specifies what other data models fail to do that this model solves.
4. There is no sensitivity study of the threshold λ in STLSQ or the number of observable functions. How does increasing p in the polynomials affect overfitting?

Comments on the Quality of English Language

1. The paper contains different grammatical issues.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This paper presents an observer framework that combines Koopman operator theory with SINDy-based sparse identification to rapidly estimate wind farm power gain under active wake control (AWC). By lifting wind direction inputs into a high-dimensional feature space and applying sparse regression, the model achieves high accuracy and low computational cost. Its effectiveness is validated on two wind farm cases, demonstrating strong potential for real-time control applications.

 

1. In the introduction section, the conventional "greedy" approach is mentioned when introducing Active Wake Control (AWC), but the term is not explained in detail. A brief explanation of what "greedy" means in the context of wind farm control would help readers better understand the motivation for AWC strategies. The following paper can be reviewed: Distributed Hybrid-Triggered Observer-Based Secondary Control of Multi-Bus DC Microgrids Over Directed Networks, Extension of pole differential current based relaying for bipolar LCC HVDC lines

 

2. In Section 1.5, only wind direction is used as the input variable, while other important environmental variables, such as wind speed and turbulence intensity, are ignored. These variables may significantly affect the power gain potential and the performance of active wake control strategies.
3. In Figure 1, the formatting of mathematical expressions is inconsistent. Some formulas are presented in proper equation format, while others appear as plain text. Additionally, mathematical symbols such as wd (wind direction) and P (power) are not consistently typeset in math notation.
4. In Section 2.2, the authors construct an observable function library for the Koopman framework, but do not explain the rationale behind selecting the specific functions included
5. In Section 3.2.1, lines 359–360, "Min ∆P for 20T", "Max ∆P for 20T", and "Avg ∆P for 20T" have not been replaced with actual numerical values.
6. The proposed method is not compared with other widely used regression or surrogate modeling techniques. The lack of such baseline comparisons weakens the persuasiveness of the results.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper explores innovative strategies to enhance the power output of wind farms. It focuses on active wake control (AWC) techniques, particularly yaw-based wake steering. The authors introduce a data-driven observer that leverages Koopman operator theory to accurately estimate potential power gains based on the direction of the wind. This approach aims to improve the economic feasibility and integration of renewable energy into the power grid by offering precise and efficient predictive models for real-time use. Two cases of study are carried out which highlight the relative advantages of the proposal.

This research introduces significant advancements in the areas of wind energy management. The work develops the IOEDMDSINDy observer, which utilizes sparse identification and Koopman operator theory to create a model that predicts power gain from AWC in wind farms.

The insights gained from this research provide an interesting start point for integrating the observer into advanced distributed control systems, potentially revolutionizing real-time AWC strategies

The manuscript is well structured, but a little bit wordy. If every section is written more concisely, the better.

The format of the references must be reviewed. Various references require more details, for example ref. 1 and 2.          

How does the data-driven observer utilize Koopman operator theory in its methodology?

Section 5. Patents?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf