A Distributed Approach for Time-Dependent Observation Scheduling Problem in the Agile Earth Observation Satellite Constellation

Round 1

Reviewer 1 Report

This manuscript focuses on improving the distributed observation schedule in Agil Earth Observation Satellites (AEOSs). To solve this, the authors specifically developed a performance impact-based distributed method (PIDSM), which can switch between two phases and outperforms other current algorithms. In general, this manuscript is well organized, and the scientific question is clearly phrased and carefully resolved. However, there are still some improvements, please find more details in my specific comments below. Overall, I think this manuscript can be published after adequate revision.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The abstract can be improved, same with the conclusion chapter. Well written in other parts. The word "problem" appears very frequent in the text.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper studies an agile earth observation satellite constellation scheduling problem, which maximizes the total profit and the system load balance of multiple distributed satellites. A distributed performance impact based algorithm (called PIDSM) is proposed. In the experiments, the PIDSM method is compared to other distributed scheduling methods and it shows that the proposed method is effective in terms of the profit and communication times.

In my opinion, the paper is well structured and relatively well written. The efficiency of PIDSM is supported by the results. However, there are some unclear points which should be clarified. Therefore, I suggest a major revision before it could be accepted.

 

My major comments are as follows:

1.     In the literature review, the authors mention that compared with centralized scheduling on the ground, distributed scheduling on the satellite could consider the dynamic and unpredictable events better. However, we don’t see such advantage of PIDSM in the method section or in the experiment section. In the experiments, when the constellation and task number get large, the communication time could be more than 10000 (and after each communication, a lot of calculations are needed. The time should also be considered) and increases very fast when the scenario size increases. Have the authors considered the delay and packet loss of so much communication? And the authors only compare PIDSM with distributed methods, so it’s hard to see the advantage of PIDSM compared with a centralized method (on the ground and on the satellite) or a top-down method (e.g., a large constellation could be divided into multiple groups and each has a centralized master satellite). I really doubt the necessity of using distributed scheduling method in a large constellation. More discussions should be given about this.

2.     More explanations should be given for formulas (6) and (7). What’s the middle part in (6) used for? Why is “the satellite with longer average VTW gets better fitness value” good for the load balance?

3.     According to the description of PIDSM, the first part needs much calculation. Besides, from 4.4, it’s hard to tell how many iterations are needed until the algorithm could converge. I think some analysis of the running time of the algorithm should be provided in the experiment part, especially the comparison with some centralized methods.

4.     In the experiment part, we could see that PIDSM obtains a much better performance than CBBA, while CBBA and PIDSM have similar two-stage optimization frameworks. The authors only compared these two methods with some results. I suggest more deeper discussions about why such difference exists should be given, for better explaining the effectiveness of PIDSM for AEOS scheduling.

 

I also have some minor comments:

 

1.     I think for reference [17] the authors meant the following paper:

He L, Liu X, Laporte G, et al. An improved adaptive large neighborhood search algorithm for multiple agile satellites scheduling[J]. Computers & Operations Research, 2018, 100(DEC.):12-25.

The current paper listed as [17] proposes an ILS algorithm for a single AEOS scheduling problem, which doesn’t match the comment in Line 77.

2.     I don’t see the meaning of the proof of Proposition 1. First, the existence of Proposition 1 is very obvious. Besides, if the authors say “By (16) and (17), we know that r(a’)=r(a)+qc-qs”, then if (18) is satisfied, the fitness value would be increased. The discussion of four cases is redundant.

3.     In inequality (20), should it be “a_s+g”?

4.     There are many typos. The authors should read their paper carefully and avoid such mistakes.

Line 38: CEOSs and AEOSs

Line 79: “updated”, I think the authors meant “uploaded”

Line 86: a market-based method

Line 89: Li et al. [20] or Li [20] develops

Line 91: CABB should be CBBA

Line 138: “are always existed” should be “always exist”

Line 205: It is assumed that

Line 492: the strong synergies exist

Line 522: until De becomes

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

1. The use of the English language must be significantly improved.

2.  As far as I can see, several studies have adopted reinforcement learning to solve the problem. Corresponding literature should be added to the introduction.

3. The structure of this article is not easy to understand.

4. The experimental analysis is Insufficient.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Thanks for the revision. I’m satisfied with most of the changes, but still have some following concerns:

 

1.     Please add the time unit to Figs 10 and 11. I think it is ‘second’. But if so, the performance of the proposed algorithm would be disappointing: when the task number reaches 200, the running time reaches over 20000 seconds, more than five hours. In such cases, it would be much better to schedule these tasks on the ground in a centralized way and then upload them to the satellites. The authors should justify why the current performance is acceptable.

2.     I think the authors do not get my concern about Proposition 1. It is okay if they want to show it, but there is some problem with the proof. Since “r(a’)=r(a)+qc-qs”, then if “qc>qs” (18) holds, the global fitness value is increased, because “r(a’)>r(a)”. The conclusion is already proved. The discussion of the four cases becomes redundant, because no matter whether the significance and marginal significance are positive or negative, r(a) would be increased as long as “qc>qs” (18) holds. If the authors want to explain the four cases, they could do it outside the proof.

Author Response

Please see the attachment.

Author Response File: Author Response.docx