A Solar Radiation Forecast Platform Spanning over the Edge-Cloud Continuum

Round 1

Reviewer 1 Report

- The introduction and related work sections are too long. These two sections should be rearranged.

- Figure-3 quality should be improved.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Forecasting the energy generating capacity for PV farm is a very important task.

This paper proposes a hybrid edge-cloud platform that leverages both edge devices and remote cloud infrastructure. The results show the scalability improvement to infer irradiance. 

The paper is easy to read and the structure is clear. Therefore, I recommend its publication.

Author Response

We would like to thank the reviewers for their useful and constructive comments. Please find below our detailed answers including pointers to what section of the submitted manuscript has been updated to reflect them.

Thank you for the positive feedback.

Reviewer 3 Report

This paper proposed a hybrid egde-cloud platform that leverages the performance of edge devices to perform time-critical computations locally, while delegating the rest to the remote cloud infrastructure. However, there are many problems and drawbacks that affect the quality of this paper.

1.       The novelty of this paper needs to be strengthened. The topic of this paper is not new, and many previous studies have been proposed. The authors need to add more details in contribution bullet points that will clearly show readers the main contribution of this work.

2.       The selected benchmark algorithms are too simple, and authors need to compare their method with similar ones proposed in recent three years.

3.       Please clarify whether the assumptions of the model are reasonable in reality.

4.       The authors should discuss in more detail how the algorithm parameters are computed/selected and their dependence on an actual problem. The super-parameter settings for all tested algorithms need more discussions as regards how they affect the results and how sensitive are the results to these settings.

5.       Machine learning normally involves training stage and then testing stage. These are not clear of the paper

6.       There is often some noise in various data, but the authors fail to introduce how to remove the noise in the deep learning methods. The related work did not address a large body of work in noise reduction in data. For example, the following paper uses the Savitzky-Golay filter and wavelet decomposition to remove noise, e.g., Temporal Prediction of Multiapplication Consolidated Workloads in Distributed Clouds. Authors could add some discussion in their future work on how to extend current work with these noise reduction methods.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The quality of the images has been corrected.

Introduction section has been rearranged.