Real-Time AI-Based Data Prioritization for MODBUS TCP Communication in IoT-Enabled LVDC Energy Systems

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a very interesting manuscript, which is, in general, well structured and organized. The text is clear and concise. The references used are relevant and duly cited and are up to date.

To improve the quality and relevance of this research the authors should clarify and emphasize the main purpose/reason for choosing MODBUS over TCP/IP and emphasize its main contribution to the field.

Also, for clarity, explain the designation of the three algorithms used: RF (Random Forest ?) an MLP (Multi-Layer Perceptron ?) and DT (Decision Tree ?) and include references for each one.

The explanation for those algorithms/Classifiers appear in Fig 6 but that is too far from the text where they are first used.

In page 15, include an expression for the F1-score metric... I rather suggest including a more general expression for the Fbeta score and detail it in particular for the F1-score referreing that it corresponds to an equal relevance for precision and recall. Include also a reference for that metric

For clarity and easy of reading modify the sizes of figures 7 and 8 according to roughly the size of Fig. 9 and harmonize the size of Fig 10, 11 and 12 accordingly. A column layout is advisable for Fig. 7 and 8 with a) on top and b) at the bottom.

In lines 779 and 780 instead of the acronyms ICS and EMS I suggest using their designations: Industrial Control Systems and Energy Management Systems 

Comments on the Quality of English Language

The quality of English is in general ok. The text is clear and concise.

Just a few notes on the text in Abstract suggesting to avoid text with brackets within brackets, which is the case in lines 3 to 5. Why not use "... consumers, e.g., Electric Vehicles (EV) chargers, Power Distribution units (PDUs), or as sources ..." or, alternatively, "... consumers - e.g., Electric Vehicles (EV) chargers, Power Distribution units (PDUs) - or as sources ..."

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The comments of this reviewer can be found below:

1. The results reported in the paper need to be compared to TinyML models.
2. The paper has used synthetic data while validating the method using a small, captured real-world dataset is necessary to improve the credibility.
3. It is required that the latency analysis clearly state whether feature extraction overhead is included in the presented inference times. This is mainly significant for real-time applications.
4. Although the scalability has been validated, analysis of the RevPi5's theoretical breaking point with more concurrent models is also required.
5. It is highly necessary to review recently published papers on adaptive sampling and event-triggered communication in the literature review section.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

1-The introduction is informative but too broad. It mixes generic IoT/AI context with the specific MODBUS problem, which dilutes the focus.
2-The research gap lack of contextual data prioritization in MODBUS TCP for LVDC systems is relevant but should be more sharply articulated against state-of-the-art ML-in-edge systems.
3-Missing references: recent AI-based microgrid control and optimization works that directly link to your IoT gateway approach. For example:
Akarne, Y., Essadki, A., Nasser, T., et al. Optimized control of grid-connected photovoltaic systems: Robust PI controller based on sparrow search algorithm for smart microgrid application. Global Energy Interconnection, 2025.
4-The problem definition is sound, but the methodology lacks rigor in several areas:
The feature extraction process (rolling slopes, flags, normalization) is described, but no formal mathematical formulation is provided. Equations should be added to clarify the transformations.
The binary prioritization (0/1) is oversimplified. Real LVDC systems often require multi-level prioritization (critical, high, medium, low). This limitation should be acknowledged.
The choice of RF, DT, and MLP is justified, but the rationale for finally preferring MLP over RF/DT needs quantitative justification (latency vs. F1 trade-off).
The simulation setup should specify sampling rate synchronization, solver, hardware constraints in more detail.
5-The classification metrics (Accuracy, Precision, Recall, F1) are informative, but:
No comparison with baseline ML models (SVM, k-NN, Logistic Regression) is given. Even lightweight methods could serve as benchmarks.
Results are entirely simulation-based. No preliminary hardware-in-the-loop (HIL) or small-scale deployment validation is shown.
The generalization issue (missed events in temp_acdc and soc_bat) is an important weakness. The final labeling simplification helps, but a quantitative before/after comparison (numerical gain in F1, recall) should be clearly highlighted.
6-Figures (e.g., Figs. 7–11) are overloaded with annotations (TP/FP/FN markers) that make them visually dense. Consider:
Using zoomed subplots for critical transitions.
Larger fonts for labels and legends.
Standardizing axis units (°C, %, V) across figures.
Table formatting (Tables 3–6) could be improved with clearer separation of parameters and results.
7-Future work should explicitly mention planned hardware deployment in SHIFT2DC demonstrators and potential extension to event-driven protocols (MQTT, OPC-UA).

Comments on the Quality of English Language

The manuscript is clear but wordy. Sentences often exceed 40 words, reducing readability.

Common issues: redundant phrases (“it should be noted that”), inconsistent tense usage, and awkward figure captions.

Professional English editing is recommended.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The paper is fine.

Reviewer 3 Report

Comments and Suggestions for Authors

The revised manuscript shows clear improvements compared to the initial submission. The authors have carefully addressed the reviewers’ comments, clarified the methodology, and enhanced the presentation of the results. The contributions are now well-structured, and the work provides valuable insights into AI-based data prioritization for MODBUS TCP in IoT-enabled LVDC energy systems. Overall, the paper has reached a publishable standard.