A Blockchain-Based Architecture for Energy Trading to Enhance Power Grid Stability

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Here a blockchain-based energy trading system architecture is proposed for enabling a self-regulating, sustainable, and resilient grid on the Ethereum Blockchain and it was applied it to a microgrid-scale distributed automated trading environment. Introduction and background studies are inclusive. Section 3 has described the proposed model scientifically well. Unique part of the proposed model is it has integrated bottom-up EMS-driven self-management with top-down blockchain-based incentive coordination, thereby exhibiting strong distributed control capability to maintain grid stability and results have shown that it can effectively support self-management at grid endpoints, achieving peak-shaving improvements of up to 25.8%, and it can maintain supply–demand balance through distributed regulation, with a minimum deviation of only 1–5%. Future works are well defined.

Now as without security any model is void so now add a section before conclusion named as Security analysis and here discuss which attacks like MiTM, DDoS, 51% and what other attacks this model is able to protect.

Also can you highlight in discussion section more on training efficiency of lightweight models embedded in heterogeneous edge devices to make this model more adaptive?

 

Author Response

Comments #1: Now as without security any model is void so now add a section before conclusion named as Security analysis and here discuss which attacks like MiTM, DDoS, 51% and what other attacks this model is able to protect.

Response 1: We sincerely thank the reviewer for highlighting the importance of security in evaluating the validity of any system design. Following this suggestion, we have added a dedicated section titled Security Analysis in the revised manuscript before the evaluation (see Lines 418-500).

In this new section, we first formalize a realistic threat model for our architecture, reflecting the permissioned blockchain environment and the motivation of grid endpoints and the DSO/LPD conducting rational attacks. We then analyze the security properties of the proposed architecture under various adversarial behaviors. Specifically, we discuss attacks such as falsifying consumption data, tampering with predictions, delaying submissions, Sybil-style identity manipulation, and DDoS attempts on the ISC contract. We also outline why PoP’s security fundamentally relies on the integrity of consumption data and the correctness of score computation, both enforced through identity-bound registration and zero-knowledge proofs. Furthermore, we analyze threats arising from a malicious or colluding DSO, including fabricated smart meters or manipulated meter readings, and show how the design (i.e., meter identity registration, ZK-based verification of consumption records, and contract-level proof validation) prevents such attacks from influencing leader selection.

To support this section with greater technical clarity, we have added two appendices: Appendix A (Lines 778-810), which describes our ZK-proof assumptions and the abstract computation relations used for meter and DSO verification; and Appendix B (Line 811), which provides the detailed algorithms for supervised consumption-sequence generation and leader selection with ZK verification. These additions ensure that the security analysis is both rigorous and fully transparent.

Comments #2: Also can you highlight in discussion section more on training efficiency of lightweight models embedded in heterogeneous edge devices to make this model more adaptive?

Response 2: We thank the reviewer for the valuable suggestion. Following this comment, we have expanded the discussion in the revised manuscript to explicitly address the training efficiency and adaptiveness of the lightweight forecasting models deployed at EMSs.

In particular, we now describe the training configuration used in our prototype. Each endpoint trains a compact LSTM-based model using one year of its own load and generation traces, split into 80% training and 20% validation. The model comprises four LSTM layers with 50 hidden units internally and is trained for 50 epochs with a batch size of 32 and a learning rate of 0.001. Despite its small size, the model converges efficiently, achieving an average training loss of approximately 0.0005 and a validation loss around 0.0006, demonstrating both computational efficiency and strong generalization across heterogeneous endpoints. These details have been incorporated into the revised manuscript (Lines 526-540). The added discussion clarifies how lightweight local models can be effectively integrated into the proposed architecture and highlights their suitability for execution on resource-constrained EMSs.

Reviewer 2 Report

Comments and Suggestions for Authors

The paper presents a distinctive combination of blockchain technology and AI-based forecasting for grid stability, addressing a pressing challenge in the energy sector. The hierarchical structure (EMS, BTP, LEMMP) is well-structured and conceptually clear. The implementation details, including consensus logic, forecasting models, and experimental evaluation, are thorough. The proposed Proof-of-Prediction mechanism is an innovative contribution linking forecasting accuracy to blockchain consensus incentives. The research goals and system design stages (Stage I–III) are logically defined and provide a good conceptual roadmap.

Meanwhile, although the simulation environment is described, real-world data validation is limited to the Sundance dataset. Field testing or comparison against other benchmark systems (e.g., Proof-of-Stake or PoA-based energy systems) would strengthen the empirical evidence.

The paper briefly mentions that blockchain latency is mitigated by a permissioned approach, but lacks quantitative analysis of transaction throughput or delay.

Given the topic of sustainable energy, a discussion of the computational and energy costs of the blockchain operations should be included.

Figures 2–5 contain dense labels and overlapping text, which hinders readability. Consider redrawing or simplifying.

Some abbreviations (e.g., EMS, TC, SC) appear early before formal definition. A table of symbols or an expanded nomenclature list would improve readability.

The discussion section could benefit from a comparison with other blockchain-based grid control systems, such as those using Proof-of-Authority, Practical Byzantine Fault Tolerance, or Federated Learning–based coordination.

While the paper mentions peak-shaving effectiveness (up to 25.8%), it does not discuss how these results compare to existing smart grid optimization methods.

The proposed model employs zero-knowledge proofs but does not specify computational overhead or data privacy trade-offs. Clarify how scalability and privacy are balanced.

Potential vulnerabilities in smart contract design or identity management (PKI-based) should be discussed.

Abstract is clearly written; however, add one numerical performance highlight (e.g., “reduced supply–demand deviation by 5%”).

The introduction section provides excellent contextual background; consider shortening some parts describing conventional energy systems to improve focus.

Proof-of-Prediction consensus in Section 3.4 is novel—add flowchart or pseudocode for clearer understanding.

In Section 4, provide quantitative performance tables in addition to figures for better comparison.

References are comprehensive and current; ensure all URLs and DOIs are active and formatted per MDPI style.

Comments on the Quality of English Language

While generally clear, there are minor grammatical inconsistencies (“in the following section builds upon…” should read “in the following section we build upon…”). Thus, thorough proofreading is recommended.

Author Response

Comments #1: The paper briefly mentions that blockchain latency is mitigated by a permissioned approach but lacks quantitative analysis of transaction throughput or delay.

Response 1: We thank the reviewer for the insightful comment. We fully agree that permissioned and permissionless blockchains differ substantially in transaction latency and throughput due to their distinct consensus requirements. Traditional permissionless blockchains operate in an open environment and must tolerate Sybil attacks, which forces the use of resource as the very proof to achieve consensus such as Proof-of-Work, leading to slow confirmation times and limited throughput. Although recent advances, such as Proof-of-Stake, have significantly improved the performance of permissionless systems, permissioned blockchains still provide more predictable latency due to authenticated participation and lightweight consensus protocols.

In our work, we adopt a permissioned blockchain primarily because the proposed PoP consensus mechanism requires identity-restricted participation: only authorized grid endpoints should be allowed to propose or validate blocks. We have revised the manuscript (Lines 343-349) to clarify this design choice more explicitly.

Furthermore, performance optimization is not the main focus of this paper. From a consensus perspective, PoP consensus mechanism consists of two components:

1. Leader selection through an incentive mechanism.
2. Validation of blocks proposed by the elected leader.

Our intention is to use the incentive-driven leader election to motivate high-quality self-regulation at the grid endpoints, thereby reducing single-point volatility and lowering global coordination overhead—for example, through improved forecasting of consumption and production.

The reviewer’s insight regarding blockchain latency is highly valuable. As noted in the Conclusion and Future Work section, improving system performance is an important next step, especially given the time sensitivity of energy trading. A natural extension is to generalize the leader-selection component of PoP into a multi-leader design, enabling integration with high-performance consensus families such as DAG-based multi-leader protocols (Lines 384-390, Lines 735-738).

Comments #2: Given the topic of sustainable energy, a discussion of the computational and energy costs of the blockchain operations should be included.

Response 2: We thank the reviewer for this important suggestion. In response, we have expanded the manuscript to include a clearer discussion of the computational and energy costs associated with blockchain operations. The revised text appears in Lines 655-666. Specifically, we now explain how the gas expenditure in our prototype is computed and interpreted. We further clarify that our on-chain costs remain moderate even in peak time because our smart contract is intentionally designed in a lightweight, information-oriented form that performs only simple data-registration operations.  The updated discussion highlights how this architectural separation ensures low blockchain overhead while maintaining transparency and tamper-resistance, aligning the system with the sustainability goals of modern energy infrastructures.

Comments #3: Figures 2–5 contain dense labels and overlapping text, which hinders readability. Consider redrawing or simplifying.

Response 3: We thank the reviewer for pointing out the readability issues in Figures 2–5. We have carefully revised all figures to improve clarity and visual consistency.

• Figures 2 and 3: The original workflow and data-flow diagrams included several diagonal lines and overlapping text elements, which indeed affected readability. We have redrawn both figures to eliminate label intersections and adjusted the layout to ensure that arrows and labels no longer overlap. In Figure 2, we also removed the redundant repetition of abbreviations and full names to reduce visual density.

• Figure 4 and 5: These two figures illustrate the complete multi-stage processing workflow within EMSs. To maintain smooth logical continuity across stages, certain elements (e.g., the consumer ID) appear more than once by design. While some overlapping or repeated components are therefore intentional, we have refined the layout by improving spacing, alignment, and label positioning to enhance overall readability.

Comments #4: Some abbreviations (e.g., EMS, TC, SC) appear early before formal definition. A table of symbols or an expanded nomenclature list would improve readability.

Response 4: We thank the reviewer for the helpful suggestion regarding abbreviation usage. We agree that clarity of terminology is important for readability. In the current version of the manuscript, each abbreviation (including Endpoint Mini Server (EMS), Transaction Controller (TC), and Self-Controller (SC) etc.) is introduced with its full name upon first occurrence. Additionally, we have an abbreviation list/nomenclature section later in the manuscript for ease of reference (Lines 774-777 of the revised manuscript). To further improve readability, we carefully reviewed the manuscript again to ensure that all abbreviations are introduced consistently at first use and remain aligned with the abbreviation list. Minor adjustments were made to enhance consistency.

Comments #5: Although the simulation environment is described, real-world data validation is limited to the Sundance dataset. Field testing or comparison against other benchmark systems (e.g., Proof-of-Stake or PoA-based energy systems) would strengthen the empirical evidence. The discussion section could benefit from a comparison with other blockchain-based grid control systems, such as those using Proof-of-Authority, Practical Byzantine Fault Tolerance, or Federated Learning–based coordination. While the paper mentions peak-shaving effectiveness (up to 25.8%), it does not discuss how these results compare to existing smart grid optimization methods.

Response 5: We thank the reviewer for this insightful comment. We fully agree that broader empirical validation—including additional datasets, field testing, and comparison with other blockchain-based grid-control systems—would further strengthen our evaluation.

First, public access to household-level renewable generation data and real-world distribution-network traces remains extremely limited due to privacy regulations and geographic constraints. As a result, most widely used datasets, including the Sundance dataset adopted in our work, only provide consumption and generation profiles but lack corresponding network-level parameters (e.g., distribution topology, line impedance, node voltage). Therefore, our empirical evaluation focuses on load-level behavior within a simulated microgrid environment. To address this limitation, we have expanded the Discussion section (Lines 610-619) to explain how our load-level peak-shaving results theoretically translate to real-world grid-level phenomena, including voltage stability and peak-load reduction, supported by recent literature.

Second, responding to the reviewer’s suggestion, we have surveyed existing blockchain-enabled grid-management systems and added a comparative discussion in the evaluation (Lines 670-678). These systems differ substantially in architecture, control objectives, and required hardware capabilities, making direct quantitative comparison difficult. Nevertheless, our examination highlights that PoP’s incentive-driven design provides a distinct advantage: it enables fine-grained user-side behavioral shaping without requiring heavy on-chain control logic or high-performance blockchain platforms.

Third, we acknowledge the importance of more comprehensive empirical validation. In the expanded Discussion and Future Work section (Lines 738-755), we emphasize the need for extensive real-world experiments, including integration with additional datasets, deployment on heterogeneous edge hardware, comparison with different blockchain-based energy platforms, and full power-flow simulations once appropriate distribution-network data becomes available. These steps are part of our ongoing research plan.

Comments #6: The proposed model employs zero-knowledge proofs but does not specify computational overhead or data privacy trade-offs. Clarify how scalability and privacy are balanced.

Response 6: We thank the reviewer for raising this important point regarding the computational overhead and privacy–scalability trade-offs of employing zero-knowledge proofs. In our architecture, ZK proofs are used specifically for supervising hardware-generated consumption data and verifying the correctness of the DSO’s score computation, rather than for continuous on-chain constraint enforcement. As a result, the ZK components incur limited overhead and do not impose significant scalability bottlenecks on the trading or consensus workflow.

To address the reviewer’s concern, we have added Appendix A and Appendix B, which formally describe (i) the abstract ZK relations for smart-meter supervision and PoP score computation, and (ii) the corresponding detailed algorithms. These appendices clarify how ZK proofs allow verification of correctness without revealing users’ raw consumption traces, thereby preserving privacy while ensuring trust in the computation.

Moreover, because ZK proofs are generated off-chain and only the proofs (not the raw data or intermediate states) are submitted on-chain, the blockchain footprint remains compact. This design ensures that scalability is maintained while still providing strong integrity guarantees for meter readings and DSO computation results.

Comments #7: Potential vulnerabilities in smart contract design or identity management (PKI-based) should be discussed.

Response 7: We sincerely thank the reviewer for this valuable suggestion. In the revised manuscript, we have added a dedicated Security Analysis section (see Lines 418-500), where we systematically examine potential vulnerabilities arising from smart-contract logic and identity management in a permissioned blockchain environment. In this section, we formalize a realistic threat model and analyze how different grid roles, grid endpoints and the DSO/LPD, may behave adversarially under rational economic incentives. We discuss possible contract-level vulnerabilities (e.g., incorrect verification logic, malformed submissions, or contract-level denial-of-service attempts) as well as identity-related risks such as forged smart-meter identities or Sybil-style impersonation. We explain how our architecture mitigates these threats through identity-bound meter registration, role-restricted contract interfaces, and zero-knowledge–based verification of consumption data and DSO computations. These mechanisms collectively ensure that malicious behaviors cannot influence PoP score calculation or leader election.

Comments #8: Abstract is clearly written; however, add one numerical performance highlight (e.g., “reduced supply–demand deviation by 5%”).

Response 8: We thank the reviewer for the constructive suggestion. In the revised manuscript, we have updated the abstract to include a quantitative performance highlight (Lines 16-19 of the revised manuscript). Specifically, we now report the improvement achieved by our method, including the peak-shaving ability and prediction deviation, which provides a clearer numerical summary of our contribution.

Comments #9: The introduction section provides excellent contextual background; consider shortening some parts describing conventional energy systems to improve focus.

Response 9: Thank you for the constructive suggestion. We agree that the introduction can be more concise. In the revised manuscript, we have shortened the part describing traditional centralized energy systems and streamlined the transition to the problem statement and blockchain-based P2P energy trading. Specifically, redundant explanations of conventional LPD/DSO energy flow and general background paragraphs were removed or condensed, improving clarity and focus. The revised text appears in Lines 22-44.

Comments #10: Proof-of-Prediction consensus in Section 3.4 is novel—add flowchart or pseudocode for clearer understanding.

Response 10: We thank the reviewer for the constructive feedback on the PoP consensus. To improve the clarity and reproducibility of our design, we have added a detailed pseudocode description of the PoP leader-selection process in the Appendix B (Line 811).  And the conceptual design is now explicitly introduced in the main text (Lines 364-390). We believe these additions significantly improve the readability of the protocol and help clarify how PoP integrates with the overall architecture.

Comments #11: In Section 4, provide quantitative performance tables in addition to figures for better comparison.

Response 11: We thank the reviewer for this helpful suggestion. In response, we have added the full numerical performance tables corresponding to the evaluation results, including the day-by-day predicted consumption, actual consumption, and deviation metrics used in Pattern 1. Due to space limitations in the main manuscript, these quantitative details are included in Appendix C of the revised version. This addition provides a clearer and more comprehensive basis for comparison alongside the figures presented in the Evaluation section.

Comments #12: References are comprehensive and current; ensure all URLs and DOIs are active and formatted per MDPI style.

Response 12: We thank the reviewer for the positive feedback and for highlighting the formatting issue. The entire reference list has now been revised to fully comply with MDPI reference style requirements. Specifically, we have:

• Converted all journal names to their official MDPI/ISO4-standard abbreviated forms.
• Added DOIs and formatted them as full links (https://doi.org/...) where available.
• Updated conference references to include full venue information (location and publisher).
• Reformatted online resources using the MDPI-recommended format:
  “Available online: \url{...} (accessed on DD Month YYYY).”

These revisions are reflected in the updated reference section (Lines 815-917 of the revised manuscript).

Comments #13: While generally clear, there are minor grammatical inconsistencies (“in the following section builds upon…” should read “in the following section we build upon…”). Thus, thorough proofreading is recommended.

Response 13: We thank the reviewer for the careful reading and the suggestion regarding phrasing consistency. The sentence has been revised as follows (Lines 231-232 of the revised manuscript):

“Since our design in the following section builds upon the existing hardware infrastructure, we classify smart meters as public grid assets.”

In addition, we carefully performed a thorough proofreading of the entire manuscript to enhance readability.

Reviewer 3 Report

Comments and Suggestions for Authors

This paper presents blockchain-based architecture for decentralized energy trading that explicitly addresses grid stability. It combines blockchain, AI-driven forecasting, and incentive mechanisms (Proof-of-Prediction) in a modular framework compatible with existing smart grid infrastructure. The work demonstrates both conceptual clarity and experimental validation at a microgrid scale. Some comments to the authors to boos the paper

• Condense background sections and strengthen the “Related Work” discussion.
• Section 2.2 - Stage II: Limited Distributed Control Mechanism. Discuss Recent ML-based approaches that have refined demand-side modeling through consumer segmentation and load aggregation frameworks check MAS-DR: An ML-Based Aggregation and Segmentation Framework for Residential Consumption Users to Assist DR Programs.
• While peak-shaving rates and mean absolute percentage errors are provided, there is no statistical evaluation of grid voltage stability or economic efficiency. The results section would benefit from comparisons with baseline methods
• Discuss security aspects and the DSO’s trust model.
• Provide pseudocode or flow diagrams for key processes (e.g., PoP leader selection) to improve clarity

 

Author Response

Comments #1: Condense background sections and strengthen the “Related Work” discussion.

Response 1: We thank the reviewer for the valuable suggestion. We have revised the Introduction section to make the background more concise by removing redundant descriptions of conventional energy systems and streamlining the transition to our problem statement (Lines 22-44).

In the revised manuscript, we have strengthened the Related Work section by explicitly highlighting the limitations of existing research at different stages of blockchain adoption in energy systems. Specifically, we emphasize the drawbacks in prior work as follows:

• In the first stage, existing studies focused excessively on designing blockchain-based trading models for local energy markets.
  However, these works did not consider the stability challenges introduced by the large-scale integration of renewable energy sources (RESs) and distributed energy resources (DERs). (Lines 116-122)
• In the second stage, several studies attempted to use blockchain platforms to achieve distributed control at the smart-meter or household level.
  Due to the inherent design and performance limitations of blockchain systems, these approaches could not support fine-grained distributed control at the household level. Moreover, many of these works completely removed the role of the regulator or coordinator, which is impractical in real-world grids where clean energy production heavily depends on external and uncertain conditions. (Lines 150-155)

Based on these observations, our work introduces an incentive-based consensus that encourages each household to perform self-regulation and self-control, thereby reducing endpoint volatility. This approach not only decreases the system load on blockchain operation but also supports large-scale endpoint participation. In addition, we preserve the regulator role to enable a soft upgrade of the grid architecture and enhance global coordination capability.

Comments #2: Section 2.2 - Stage II: Limited Distributed Control Mechanism. Discuss Recent ML-based approaches that have refined demand-side modeling through consumer segmentation and load aggregation frameworks check MAS-DR: An ML-Based Aggregation and Segmentation Framework for Residential Consumption Users to Assist DR Programs.

Response 2: We sincerely thank the reviewer for the valuable suggestion and for highlighting recent cutting-edge machine-learning-based advances such as MAS-DR. These works offer important insights into how consumer segmentation and load aggregation can refine demand-side modeling, and they are highly relevant to distributed energy management. We have expanded the Related Work section to include a detailed discussion of recent ML-based demand-side modeling approaches, including multi-agent DR frameworks as well as other consumer-segmentation and load-profiling methods (Lines 128-147).

The main objective of our paper is to achieve distributed control through an incentive-driven mechanism, encouraging all endpoints to perform accurate self-management of their production and consumption activities in exchange for rewards. This design relies on autonomous local decision-making to reduce global coordination overhead, rather than depending on a fully centralized or blockchain-embedded control layer, and we have emphasized this design intent accordingly in the manuscript (Lines 295-300). As acknowledged in the discussion section, the forecasting model used in our evaluation remains relatively traditional. The ML-based frameworks highlighted by the reviewer provide valuable opportunities to further improve the accuracy and adaptability of endpoint-level self-management and could also be naturally incorporated as local forecasting agents within our proposed architecture (Lines 732-734).

Comments #3: While peak-shaving rates and mean absolute percentage errors are provided, there is no statistical evaluation of grid voltage stability or economic efficiency. The results section would benefit from comparisons with baseline methods.

Response 3: We sincerely thank the reviewer for this valuable comment. Our current evaluation is based on household-level consumption and generation time series, and the available dataset does not include distribution-network models (e.g., feeder topology, line impedances, or reactive power flows). As a result, the present study focuses on load-level performance, including peak-shaving rate and forecasting accuracy to verify that the proposed architecture can fundamentally influence and regulate endpoint behavior through the incentive mechanism, thereby demonstrating its ability to shape demand at the source before introducing grid-level constraints.

To address the reviewer’s concern, we have now added corresponding theoretical discussion in the revised manuscript (Lines 610-619) explaining how the observed peak-shaving rates relate to grid-level indicators according to the well-established linearized DistFlow model for radial distribution networks, which pointed out reductions in peak demand are expected to proportionally mitigate voltage drops and relieve feeder loading. We also highlight the economic implications documented in prior work, where peak-load reduction contributes to lowering capacity charges and deferring network reinforcement.

We fully agree with the reviewer’s insightful comment that grid-level indicators, such as voltage stability and network constraints, are essential for evaluating the broader impact of demand-side control mechanisms. In light of this, we have expanded the Conclusion and Future Work section to clearly state that the absence of network-level data is a current limitation of public datasets, including the Sundance dataset used in our experiments, and to explain that we are actively working toward integrating richer, real-world datasets and practical testbeds in the future works (Lines 738-755).

Comments #4: Discuss security aspects and the DSO’s trust model.

Response 4: We sincerely thank the reviewer for this valuable suggestion. In the revised manuscript, we have added a dedicated Security Analysis section (see Lines 418-500), where we systematically examine potential vulnerabilities arising from smart-contract logic and identity management in a permissioned blockchain environment. In this section, we formalize a realistic threat model and analyze how different grid roles, grid endpoints and the DSO/LPD, may behave adversarially under rational economic incentives. We discuss possible contract-level vulnerabilities (e.g., incorrect verification logic, malformed submissions, or contract-level denial-of-service attempts) as well as identity-related risks such as forged smart-meter identities or Sybil-style impersonation. We explain how our architecture mitigates these threats through identity-bound meter registration, role-restricted contract interfaces, and zero-knowledge–based verification of consumption data and DSO computations. These mechanisms collectively ensure that malicious behaviors cannot influence PoP score calculation or leader election.

Comments #5: Provide pseudocode or flow diagrams for key processes (e.g., PoP leader selection) to improve clarity

Response 5: We thank the reviewer for this helpful suggestion. To improve the clarity and reproducibility of our design, we have added a detailed pseudocode description of the PoP leader-selection process in the Appendix B (Line 811).  And the conceptual design is now explicitly introduced in the main text (Lines 364-390). We believe these additions significantly improve the readability of the protocol and help clarify how PoP integrates with the overall architecture.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Now all the necessary changes are done and it's a good contribution.

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for thoroughly addressing all my comments and concerns. The revisions have strengthened the manuscript, and I now consider it suitable for publication.