Research on the Prediction Problem of Satellite Mission Schedulability Based on Bi-LSTM Model

Round 1

Reviewer 1 Report

COMMENT 1

Page 1, line 43

It is suggested to specify the acronym NP in NP-hard problem. 

COMMENT 2

Page 3, line 112

Please check that W_sj is the end time of task x_i. From figure 1, it appears to be the start time of task x_j, while W_Ei is supposed to be the end time of task x_i.

COMMENT 3

Page 3, lines 115-116

Please consider adding more information in the caption of Figure 1. For example, how the time windows relation between two tasks has been determined? Just for illustration purposes?

COMMENT 4

Page 3, line 122

Please clarify this sentence. Is B_max the maximum power that can be generated by the satellite, in accordance with EPS operations, or is it the maximum allocated power for the specific task in question?

COMMENT 5

Page 6, line 188

Please check Figure 4. Should the output layer be replaced by the input layer? 

COMMENT 6

Page 7, line 225

The caption of Table 1 is missing.  

Page 8, lines 251, 252

Please clarify the sentence. What do you intend with "battery power and storage space are consumed"?  

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper applies a Bidirectional Long Short-Term Memory (Bi-LSTM) Recurrent Neural Network (RNN) to perform satellite mission planning. The model is trained on simulated data and compared with one-way Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). The paper introduces Bi-LSTM for the satellite mission planning problem, which is able to plan for multiple observation tasks and predict whether an input task sequence is executable or not.

 

General comments:

- The methodology is difficult to follow, notation is inconsistent and the novel contributions of this paper could be more clearly highlighted.

- Too much of the paper is devoted to background introduction rather than novel aspects of the work.

- Figures from other sources should be clearly cited.

- Figures should be easily understood, using legends and captions, rather than relying on the main body of the text.

- The experimental simulation is not clearly described, both in terms of setup and analysis.

- The introduction introduces the state-of-the-art in mission planning, but the analysis appears to only compare with simple GRU and LSTM models. 

- It would be useful to mention how such an approach could be realised on an actual satellite, especially in terms of how much future task knowledge would be required.

 

Specific comments:

- Fig. 2 and 3 appear to have been copied from other sources without citation.

- Quote on p1 ln45 with no source

- Figure 1, no explination of the four categories. Also not clear Cat represents category

- Preferable to have the optimisation laid out as in [8] by Peng et al., where all the constraints are shown together

- p4 ln 136, First mention of RNN should be LSTM

- Figure 2 notation does not match text h vs s and y vs o

- Softmax introduced in Fig 4 but never mentioned in the text. In fact Eq 14 implies that Softmax is not used.

- Fig 5, column labels are highly pixelated

- Eq. 14, slightly unhelpful use of W for both time window and weight matrix. W is also used extensively as a weight matrix before Eq. 14 but only introduced after.

- Table 1 caption is inaccurate.

- p8 ln 245, Not clear what the role of the genetic algorithm is in the experimental analysis.

- How is the satellite defined in the simulation?

- Fig 6 is very messy and not helpful to the reader.

- p9 ln 266, Gated recurrent unit (GRU) is not introduced.

- Fig 7, How is accuracy calculated?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Thank you for your work in addressing the specific comments from my review. 

Point 10: p8 ln 245, Not clear what the role of the genetic algorithm is in the experimental analysis.

- It is not clear the motivation or implementation of this genetic algorithm in the paper. The author response indicated that this is not directly related to task prediction and so not described. It would still be helpful to provide more context to the reader than is currently present.

 

The authors have not responded to any of the general comments. In particular:

General comment 6 - The introduction introduces the state-of-the-art in mission planning, but the analysis appears to only compare with simple GRU and LSTM models. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 3

Reviewer 2 Report

Thank you for your responses. On the second point you highlight that the methods described in the introduction are for prediction not planning and so not relevant for comparison with the approached developed in the paper. However, the introduction describes your method as a "Bi-LSTM-based imaging satellite observation task schedulability prediction model". This distinction between prediction and planning is not clear in the final paragraphs of the introduction.

Reading the introduction, specifically:

"In this regard, Peng et al. [8] proposed a task schedulability prediction method based on long short-term memory (LSTM) model, which takes the task sequence as the input, realizes the forward and backward transfer of task timing characteristics, and further improves the prediction accuracy. However, this method only considers the impact of the forward task sequence on the prediction task, and can only decide whether to execute one observation task at a time. 

In response to this, this paper proposes a Bi-LSTM-based imaging satellite observation task schedulability prediction model, which predicts whether the input task sequence is executable or not."

I would expect a comparison between Bi-LSTM and Peng et al. LSTM model to show the improved performance of Bi-LSTM due to it taking into account whether the task sequence is executable.

At the moment the impression I get as a reader is that your method is better than the classic GRU and LSTM implementations, but I am not convinced it provides obvious benefits beyond other methods in the literature.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 4

Reviewer 2 Report

Thank you for your responses and work updating the paper.