Blockchain Consensus Algorithm for Supply Chain Information Security Sharing Based on Convolutional Neural Networks

doi:10.21203/rs.3.rs-4627597/v1

Download PDF

Article

Blockchain Consensus Algorithm for Supply Chain Information Security Sharing Based on Convolutional Neural Networks

https://doi.org/10.21203/rs.3.rs-4627597/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

To solve the problems of data silos and information asymmetry in traditional supply chain information security sharing, this article combines Convolutional Neural Networks (CNN) and blockchain consensus algorithms, analyzes data and uses blockchain for secure sharing, so that all parties can obtain and verify data in real time, improve the overall operational efficiency of the supply chain, and promote information transparency and sharing efficiency. CNN can be used to analyze data in the supply chain. Training on real digital images ensures data privacy and improves the accuracy and efficiency of data processing. Blockchain technology can be introduced into supply chain information sharing to ensure the immutability and transparency of data. This article introduces a federated learning (FL) mechanism to improve consensus algorithms, which improves the efficiency of model training. Among them, each link in the FL process is rigorously verified and recorded through the consensus mechanism of blockchain, ensuring the security and reliability of the entire process. This article adopts an improved consensus algorithm, PoDaS (Proof of Data Sharing), whose core idea is to use the computational consumption generated during FL as proof of workload. The specific steps include: local model training and uploading, model update verification shield, and model update aggregation. The PoDaS algorithm combines the advantages of PoW (Proof of Work) and PoS (Proof of Stack) to ensure the fairness of the consensus mechanism and reduce the waste of computing resources. By comparing and analyzing the block time and model accuracy of three algorithms, the superiority of PoDaS algorithm in block time and model accuracy was verified. The experimental results show that the PoDaS algorithm is significantly better than the PoW algorithm in terms of block generation time, and slightly better than the PoS algorithm. In terms of model accuracy, the PoDaS algorithm is significantly superior to traditional PoW and PoS algorithms. Its model accuracy reaches 96.00%, reflecting the effectiveness and practicality of the PoDaS consensus algorithm in the sharing of supply chain information security.

Earth and environmental sciences/Environmental sciences

Earth and environmental sciences/Environmental social sciences

Information Sharing

Proof of Data Sharing

Federated Learning

Convolutional Neural Networks

Blockchain Consensus Algorithm

With the development of the economy, the advancement of globalization, and the continuous expansion of supply chain networks [1–2], the problems of supply chain management [3–4] have become increasingly complex. How to solve the difficulties in supply chain management and achieve efficient management and sharing of supply chain information is an urgent task for enterprises in the whole society. Traditional supply chain management faces many challenges, such as data silos [5], information asymmetry, and data security, which can limit the operational efficiency of the supply chain and hinder cooperation and collaboration among enterprises. The continuous development of information technology has also brought new solutions to the challenges of supply chain management.

CNN [6] is a deep learning model that excels in processing image data. CNN consists of convolutional layers, pooling layers, and fully connected layers. The powerful feature extraction ability and good generalization performance of CNN make its application scenarios (image classification, object detection, semantic segmentation, etc.) very extensive. This article introduces CNN into supply chain data analysis, which can effectively improve the accuracy and efficiency of data processing.

FL [7] is a distributed machine learning method. By distributing the model training process among multiple participants for training, each participant trains the model locally, and then uploads the updates of the local model (non raw data) to the central server for aggregation. During this process, the original data of the participating parties is always saved locally, effectively protecting data privacy [8]. Because it only requires model updates rather than a large amount of raw data, it can significantly reduce the consumption of network bandwidth. Privacy protection and reducing bandwidth consumption are the main advantages of FL. In addition, FL utilizing the computational resources of multiple participants can significantly improve the efficiency of model training. Typical applications of FL include personalized services on mobile devices, such as predictive input, recommendation systems [9], and medical data analysis [10].

Blockchain [11] is a decentralized distributed [12] ledger technology [13–14]. The main characteristics of blockchain are decentralization, immutability [15], and transparency. Decentralization means that there is no centralized authoritative management organization [16], and all nodes work together to maintain the integrity and consistency of the ledger. Its encryption technology [17] and consensus mechanism [18–20] can ensure the immutability and transparency of data. With the development of technology, the application scenarios of blockchain technology (finance, supply chain management, Internet of Things [21], etc.) are becoming increasingly widespread.

The consensus algorithm [22–23] determines how participants in blockchain networks [24–25] reach consensus on the effectiveness of transactions, which is an extremely crucial part of blockchain technology. Due to the presence of authoritative managers in traditional centralized systems, consensus algorithms have no place to be used. However, in distributed blockchain systems, there is generally a lack of centralized managers, so consensus algorithms are needed to ensure the effectiveness and consistency of transactions, in order to prevent malicious behavior and the occurrence of dual payments [26].

PoW [27] is one of the earliest consensus algorithms applied in blockchain technology. Its basic idea is to prove the creation right of a node for a block by solving a complex mathematical problem. The node continuously tries to calculate until it finds a hash value that meets the criteria [28]. This process requires a large amount of computing resources and time, so it can effectively prevent malicious nodes [29] from attacking the blockchain network. However, the PoW algorithm requires a large amount of computation for each node to solve the problem, which leads to waste of energy and block time, resulting in high energy consumption and low efficiency. PoS [30] is another common consensus algorithm that solves the problems of energy waste and low efficiency in the PoW algorithm. The PoS algorithm no longer relies on the computing power of nodes, but determines their block creation rights based on the number of currencies held by nodes [31]. PoS algorithm can reduce energy consumption and improve block efficiency, but due to its characteristic that nodes holding more currency are more likely to obtain block rights, it can lead to the centralization of power.

PoDaS, as a new consensus algorithm, has been proposed to solve the problems existing in PoW and PoS algorithms. The PoDaS algorithm attempts to combine the advantages of PoW and PoS, ensuring the fairness of the consensus mechanism while reducing the waste of computing resources. The core solution is to use the computational consumption generated during FL as proof of workload. Nodes do not rely on computational difficulties, but on training, exchanging, validating, and aggregating local machine learning models [32–34] to prove their creation rights over blocks. This method not only reduces energy consumption and improves block efficiency, but also avoids the problem of the wealthy getting richer, ensuring the fairness of the consensus mechanism.

In the field of supply chain information security sharing, many scholars and research teams have proposed various solutions. Cole R et al. [35] explored the impact of blockchain technology on Operations and Supply Chain Management (OSCM). This confirms that blockchain can revolutionize OSCM practices in various ways, including improving quality management, promoting sustainable supply chain management. Saberi S et al. [36] proposed a distributed digital ledger technology based on blockchain, which can improve transparency, traceability, and security. A rigorous review has been conducted on blockchain technology and its relationship with sustainable supply chain management. It clarifies the four types of obstacles that blockchain adoption faces. This provides valuable background information for understanding how to integrate blockchain into supply chain management.

In the field of privacy protection, some scholars have also conducted research on this. Yang Q et al. [37] emphasized the potential of FL to achieve multi-party collaborative training while protecting data privacy. The concept of Federated Learning Chain (FLchain) proposed by Nguyen DC et al. [38] identified the main challenges and solutions for implementing FLchain, including communication costs, resource allocation, incentive mechanisms, as well as security and privacy protection. Its research provides a comprehensive framework for understanding how to combine FL and blockchain to improve supply chain management.

This article proposes a supply chain information security sharing mechanism based on CNN and blockchain consensus algorithms. By combining the powerful data processing capabilities of CNN, privacy protection and efficient training advantages of FL, as well as the security and transparency characteristics of blockchain, PoDaS consensus algorithm can be proposed by combining PoW and PoS algorithms to achieve efficient and secure information sharing and collaborative work while protecting data privacy.

3.1 Dataset Introduction

In terms of privacy protection, methods based on FL and blockchain often use image related datasets, this is because image data tends to have a large scale and rich features, and using image data sets can better protect user privacy. Compared to text data, image data is more difficult to directly identify and associate with specific personal identities, making it more conducive to protecting privacy.

In this section, the SVHN (Street View House Numbers) dataset can be used, which is commonly used in fields such as computer vision and FL. The SVHN dataset is a real image dataset from Google Street View gate sign images. The SVHN dataset contains images with over 600000 digits. Figure 1 shows the sample distribution of each digit in the SVHN dataset. It can be seen that the distribution of the 0–9 ten digits in the CVHN dataset is relatively uniform, and the actual application effect is relatively good.

The format of the SVHN dataset is shown in Fig. 2, which shows numerical data ranging from 0 to 9. Each numerical data is a sample, and the representation of the same number is different.

3.2 Blockchain

Blockchain includes "blocks" and "chains", in which data related to transaction information is stored. These blocks are connected in chronological order to form a chain structure, forming a blockchain. As shown in Fig. 3, each block in the blockchain consists of two parts:

1) Block headers store key information of blocks, and different blocks can be designed with different headers. Among them, Merkle root is a hash of the current block body information; The hash value of the previous block can be used to ensure the immutability of the data, and the hash value of the previous block can serve as a pointer to the previous block; Timestamps are used to connect blocks over time, ultimately forming a blockchain.

2) Block body stores transaction information of blocks. In block bodies, transaction data is usually recorded in the form of Merkle trees. It can be seen that the hash value of the root node is obtained by hashing all transaction information layer by layer, so any changes in transaction data in the block body can be passed up layer by layer, causing the value of the root node to change.

3.3 FL

Federated Learning (FL) aims to construct machine learning models based on datasets stored in local distributed devices, with a focus on privacy protection and data security. The core idea is to keep the data unchanged and the model moving, which means that each user device can transmit their trained models without disclosing the local training data, thus achieving the goal of multiple users collaborating to train a shared model. Therefore, FL is a collaborative machine learning method with privacy preserving properties. The specific process of FL is shown in Fig. 4:

3.3.1 Local Model Training

The first step in FL is local model training. Each participant trains their own data locally to generate local models. The specific steps are as follows:

1) Data preprocessing

Local data can be preprocessed, and the SVHN dataset can be cleaned, standardized, and segmented into training and validation sets.

2) Model initialization

It can initialize CNN model parameters and receive 28 x 28 pixel grayscale images in the input layer. After passing through the convolutional and fully connected layers, it outputs a 10-dimensional vector representing the probability that the image corresponds to a specific digit.

3) Model training

The CNN model can be trained using local data, using a Cross-Entropy Loss function and a stochastic gradient descent optimizer for parameter updates. Formula (1) is as follows:

$$\:\text{L}\text{o}\text{s}\text{s}=-\frac{1}{N}\sum\:_{i=1}^{N}\sum\:_{j=1}^{C}{y}_{ij}\text{l}\text{o}\text{g}\left({\widehat{y}}_{ij}\right)$$

Among them, $\:N$ is the number of samples, $\:C$ is the number of categories, $\:{y}_{ij}$ is the actual label, and $\:{\widehat{y}}_{ij}$ is the predicted probability.

3.3.2 Model Parameter Upload

After the local model training is completed, all participants upload the model parameters to the blockchain network. To protect the privacy of model parameters, homomorphic encryption can be used to encrypt them. The encryption process uses a public key encryption algorithm, and formula (2) is as follows:

$$\:E\left(w\right)={w}^{e}\hspace{0.33em}\text{m}\text{o}\text{d}\hspace{0.25em}n$$

Among them, $\:w$ is the model parameter, $\:e$ and $\:n$ are the public key parameters. The encrypted model parameters are uploaded to the miner nodes in the blockchain network through secure communication protocols.

3.3.3. Model Update Validation

The miner node verifies the model parameters uploaded by each participant. Firstly, use the private key to decrypt the encrypted parameters, and the decryption formula (3) is as follows:

$$\:D\left(E\left(w\right)\right)={\left(E\left(w\right)\right)}^{d}\hspace{0.33em}\text{m}\text{o}\text{d}\hspace{0.25em}n$$

Among them, $\:d$ is the private key parameter. The decrypted parameters need to undergo consistency verification. The mining node retrains some data to check the rationality of parameter updates and ensure that the model parameters submitted by each participant match their local data. After verification, the miner node records parameter updates and prepares for global model aggregation.

3.3.4 Global Model Aggregation

By aggregating the validated local model parameters at the miner node, a global model is generated. Firstly, weights are allocated based on the data volume of each participant, and the weight formula (4) is as follows:

$$\:{w}_{i}=\frac{{D}_{i}}{\sum\:_{j=1}^{N}{D}_{j}}$$

Among them, $\:{w}_{i}$ represents the weight of the $\:i$th participant, and $\:{D}_{i}$ represents the data volume of the $\:i$th participant. Then, the local model parameters are weighted and averaged according to their weights to generate global model parameters. The aggregation formula (5) is as follows:

$$\:{W}_{global}=\sum\:_{i=1}^{N}{w}_{i}{W}_{i}$$

Among them, $\:{W}_{global}$ is the global model parameter, and $\:{W}_{i}$ is the local model parameter of the $\:i$th participant.

3.3.5 Global Model Distribution

The generated global model parameters are distributed to all participants through the blockchain network. Firstly, homomorphic encryption can be used to encrypt the global model parameters, ensuring the security of the parameters during transmission. The encrypted global model parameters are broadcasted to each participant through the blockchain network, and the participants receive and decrypt the parameters, updating the local model.

4.1 PoW Consensus Algorithm

Proof of Work (PoW) consensus algorithm is one of the earliest and most common consensus mechanisms in blockchain systems, and the basic idea of the PoW consensus algorithm is to prove the workload of nodes by consuming computing resources. The specific process is shown in Fig. 5. In the supply chain information security sharing system, the PoW consensus algorithm verifies transactions and generates new blocks by requiring nodes to solve complex computing problems, thereby preventing data tampering and ensuring data transparency and credibility:

As shown in Fig. 5, the workflow of the PoW algorithm can be illustrated by a block generation process. First, the node receives the transactions in the network and records the transaction data. This transaction data will be collected and packaged into a block. The block contains the current transaction data as well as the hash value of the previous block, ensuring the continuity of the blockchain and the immutability of the data. Then it is necessary to calculate the proof of workload, during this process, the node needs to calculate the hash value. During the computation, the node checks whether the hash value is less than the preset target value. The target value is determined by the difficulty of the network; the higher the difficulty, the smaller the target value. The node needs to keep trying different random numbers until it finds a correct hash value. If the calculated hash is not less than the target value, the node increases the Nonce value and recalculates the hash value. This process is repeated until the node finds a hash value that satisfies the condition. Once found, the node will broadcast this new block to the entire network. After receiving this new block, other nodes verify it. Validation includes checking that the transaction data and Nonce values in the block are correct, and that the hash value is indeed less than the target value. If the verification is successful, the block is added to the main chain, completing a block generation process. If the verification fails, the block is rejected and the node that generated the block is notified to recalculate. Because hash computing requires a lot of computing resources, it is difficult for malicious nodes to tamper with blockchain data. Once a block is tampered with, its hash value will change, and the hash value of all subsequent blocks needs to be recalculated, which is very difficult.

Through the proof of workload mechanism, blockchain networks can ensure the uniqueness and immutability of transactions, preventing malicious nodes from conducting dual payment attacks. However, due to the fact that the PoW algorithm requires a large amount of computing resources for hash calculation, it leads to huge energy consumption and serious waste of computing resources.

4.2 PoS Consensus Algorithm

Proof of Stake (PoS) consensus algorithm is a mechanism that obtains verification block rights by holding assets (“rights”) in the blockchain system. PoS determines the validation and block out rights of nodes through the allocation of rights, which has the characteristics of low energy consumption and high efficiency.

The PoS consensus algorithm determines the block out rights of a node by the number of tokens held and the duration of token holding. Nodes that hold more tokens and hold them for a longer period of time are considered to contribute more to the system and are therefore more likely to be selected as outbound nodes. The core idea of PoS is that rights determine power, and its basic process is shown in Fig. 6:

Firstly, the equity value can be calculated, and the equity value (Stake) of each node is initialized and allocated based on the number of tokens it holds. The initial equity value is recorded in the status database of the blockchain, and corresponding blocks can be generated based on the data. Afterwards, a hash function can be calculated, using the current state of the blockchain and the hash value of the previous block as seeds to generate a random number. The selection threshold for each node can be calculated based on the node’s equity value and the total network’s equity value. The generated random numbers can be compared with the threshold of each node, and through this pseudo-random number generation algorithm, nodes that meet the conditions can be selected as block nodes. Formula (6) is as follows:

$$\:P\left(i\right)=\frac{Stak{e}_{i}}{\sum\:_{j}Stak{e}_{j}}$$

Among them, $\:P\left(i\right)$ represents the probability of node $\:i$ being selected, $\:Stak{e}_{i}$ represents the equity value of node $\:i$, and E$\:\sum\:_{j}Stak{e}_{j}$ represents the total equity value of the entire network.

The selected node is responsible for packaging unconfirmed transactions and broadcasting them to the entire network. After receiving the new block, other nodes verify the validity of the block, including verifying proof of rights, transaction legality, etc. If the verification is successful, the new block can be added to the local blockchain. After the new block is generated, the system updates the equity value of the node according to the predetermined rules. Formula (7) is as follows:

$$\:Stak{e}_{i}\left(t+1\right)=Stak{e}_{i}\left(t\right)+Reward-Penalty$$

Among them, $\:Stak{e}_{i}\left(t+1\right)$ represents the equity value of node $\:i$ at time $\:t+1$, and $\:Stak{e}_{i}\left(t\right)$ represents the equity value of node $\:i$ at time $\:t$. $\:Reward$ is the reward value, $\:Penalty$ is the penalty value.

The PoS algorithm is far superior to the PoW algorithm in terms of energy efficiency, significantly reducing the waste of computing resources and energy. Additionally, PoS enhances system security by preventing malicious node attacks through proof of stake. Nodes holding more tokens have greater economic incentives to protect the blockchain’s security, and malicious behavior results in a reduction of their stake value, serving as an economic penalty. PoS uses a pseudorandom number generation algorithm to select block validators, thus avoiding the issue of computational power centralization and promoting broader node participation and decentralization. However, PoS systems may lead to early participants accumulating a significant portion of stake, potentially impacting system fairness and decentralization.

4.3 PoDaS Algorithm

The PoW consensus algorithm realizes the security and decentralization of the blockchain system through the competition of computing power, but it also brings about a serious waste of computing resources. The PoS consensus algorithm determines the block rights of nodes by equity value, effectively solving the problem of resource waste in PoW, and improving the energy efficiency and security of the system. However, there are also problems with initial distribution and concentration of equity. PoDaS introduces a FL mechanism, which takes computational consumption as part of the proof of workload, and combines the advantages of PoW and PoS to achieve optimization of resource utilization and fairness of consensus mechanism. The following can provide a detailed introduction to the specific principles and implementation steps of the PoDaS algorithm.

The core idea of the PoDaS algorithm is to use the computational consumption in the FL process as proof of workload, achieve consensus through the contribution of computing resources of nodes, and introduce proof of equity to increase the fairness and security of consensus. The flowchart of the PoDaS consensus mechanism is shown in Fig. 7:

The PoDaS algorithm is a consensus algorithm that improves PoW and PoS algorithms and combines them with FL mechanisms. In the PoDaS algorithm, local model training and uploading are required, and each node uses CNN to train the model based on local data and update the model parameters. Then, the trained model parameters (weights, biases, etc.) can be packaged and accompanied by proof of computational consumption, and uploaded to the blockchain network. Formula (8) is as follows:

$$\:{W}_{i}^{\left(t+1\right)}={W}_{i}^{\left(t\right)}-\eta\:\nabla\:L\left({W}_{i}^{\left(t\right)},{x}_{i},{y}_{i}\right)$$

Among them, $\:{W}_{i}^{\left(t\right)}$ is the model parameter of node $\:i$ in the $\:t$ round, $\:\eta\:$ is the learning rate, and $\:\nabla\:L$ is the gradient of the loss function.

After receiving the uploaded model parameters at other nodes, it is necessary to update and verify the model to ensure the legitimacy and validity of the parameters. Formula (9) is as follows:

$$\:{L}_{val}\left({W}_{i}\right)=\frac{1}{{N}_{val}}\sum\:_{j=1}^{{N}_{val}}L\left({W}_{i},{x}_{j},{y}_{j}\right)$$

Among them, $\:{L}_{val}\left({W}_{i}\right)$ is the validation error, $\:{N}_{val}$ is the size of the validation dataset, and $\:L$ is the loss function.

After the model validation is passed, it needs to be aggregated to form a global model. It is necessary to calculate the contribution of each node, weighted average the model parameters, and update the global model parameters, broadcasting them to the entire network. Formula (10) is as follows:

$$\:{W}^{\left(t+1\right)}=\sum\:_{i=1}^{N}{\alpha\:}_{i}{W}_{i}^{\left(t+1\right)}$$

Among them, $\:{W}^{\left(t+1\right)}$ is the global model parameter, and $\:{\alpha\:}_{i}$ is the calculated contribution weight of the node.

After completing the model update aggregation, it is necessary to obtain the equity value based on the computational consumption of each node in the FL process, and calculate the contribution proof of each node. It can record the computational consumption of each node during model training, including computational time, energy consumption, etc., and then allocate equity values based on the computational consumption as part of PoS. Formula (11) is as follows:

$$\:Stak{e}_{i}=\frac{{C}_{i}}{\sum\:_{j=1}^{N}{C}_{j}}$$

Among them, $\:Stak{e}_{i}$ is the equity value of node $\:i$, and $\:{C}_{i}$ is the calculated consumption of node $\:i$.

Then, using PoW calculation and PoS weighting, a mechanism that combines PoW and PoS, block rights are allocated. Firstly, each node performs workload proof calculations to solve mathematical problems. Then, based on the equity value, the PoW calculation results are weighted to determine the block nodes. Formula (12) is as follows:

$$\:P\left(i\right)=\frac{Stak{e}_{i}\times\:Po{W}_{i}}{\sum\:_{j=1}^{N}Stak{e}_{j}\times\:Po{W}_{j}}$$

Among them, $\:P\left(i\right)$ represents the probability of node $\:i$ being selected, and $\:Po{W}_{i}$ represents the PoW calculation result of node $\:i$.

By combining PoW and PoS and introducing a FL mechanism into the consensus mechanism, the PoDaS consensus algorithm can reduce the waste of computing resources in the PoW algorithm. By allocating block rights through equity values, the fairness of the consensus mechanism is ensured, and the security of the system is enhanced by calculating contribution proofs to prevent malicious node attacks. By optimizing the consensus mechanism, block generation time can be reduced, improving the overall efficiency of the system and facilitating real-time sharing and verification of supply chain information.

To verify the performance of the PoDaS algorithm in practical applications, this article compares the experimental results of the PoDaS algorithm with traditional PoW and PoS algorithms, with a focus on analyzing their performance in terms of block generation time and model training.

5.1 Comparison of Block Production Time

This article compares the block time of PoDaS, PoW, and PoS algorithms on the SVHN dataset. Figure 8 shows the comparison of block time under the three consensus mechanisms. The overall block time of PoDaS consensus algorithm is smaller than that of PoW consensus algorithm and PoS consensus algorithm.

Figure 9 shows the average block time of PoDaS, PoW, and PoS algorithms at different nodes. This article sets three different numbers of nodes: 10, 15, and 20. It can be seen that when the number of nodes is 10, the average block time difference between the PoDaS, PoW, and PoS algorithms is very small. When the number of nodes is 15, the average block time of PoDaS and PoS algorithms is still very close. PoDaS algorithm is 0.21ms faster than PoS, but it has already opened up a slight gap with PoW algorithm. When the number of nodes is 20, there is a significant gap between PoDaS and PoS algorithms and PoW algorithm, and the average block generation time of PoDaS algorithm is 0.69ms faster than PoS algorithm. The average block time of three algorithms decreases with the increase of the number of nodes, because the increase in the number of nodes reduces the amount of data on the user end, resulting in a decrease in block time.

From Figs. 8 and 9, it can be seen that the PoDaS algorithm has significant advantages over PoW and PoS algorithms. This is because the PoDaS algorithm uses the computational consumption generated during FL as proof of workload, reducing invalid computations and improving block extraction efficiency. Specifically, the PoW algorithm relies on a large amount of computation to solve complex mathematical problems, resulting in longer block generation times; Although the PoS algorithm has optimized this process to a certain extent, there is still a problem of resource competition between nodes. PoDaS, on the other hand, introduces a computational consumption proof mechanism to make the block generation process more efficient and fair, significantly reducing the block generation time. Therefore, it can be concluded that the PoDaS consensus algorithm can effectively reduce consensus time and improve system performance compared to PoW and PoS algorithms.

5.2 Comparison of Model Training Situation

This article compares the model training performance of PoDaS, PoW, and PoS consensus algorithms on the SVHN dataset. The model accuracy changes of the three algorithms are shown in Fig. 10.

From Fig. 10, it can be seen that the PoDaS algorithm has better convergence speed and final model accuracy in the early stage of model training than PoW and PoS algorithms. This is because in the PoDaS consensus mechanism, nodes have different identities and can receive corresponding rewards based on their contributions during the training process. Based on this, nodes compete with each other to obtain accounting rights, while traditional PoW consensus mechanisms can only rely on the node’s own computing power to become the winning miner. Nodes that cannot become winning miners can not receive rewards.

The final model accuracy of PoW, PoS, and PoDaS consensus algorithms is 88.87%, 85.13%, and 96.00%, respectively. The model accuracy of PoDaS algorithm has been improved by 7.13% and 10.87% compared to PoW algorithm and PoS algorithm, respectively. From this, it can be concluded that the PoDaS algorithm can significantly improve model accuracy and system reliability compared to PoW and PoS algorithms.

This article proposes a privacy preserving blockchain data sharing method by combining CNN and blockchain technology, introducing FL mechanisms. CNN can be used to train images of real numbers to ensure data privacy. Introducing blockchain technology in supply chain information sharing ensures the immutability and transparency of data. This article introduces a FL mechanism to enable all participants to collaborate in training a global model without sharing data. This not only protects data privacy, but also ensures the security and reliability of the entire process. This article uses the consensus algorithm of PoDaS to achieve privacy protection based blockchain data sharing. The PoDaS algorithm combines the advantages of PoW and PoS to ensure the fairness of the consensus mechanism and reduce the waste of computing resources. By comparing and analyzing the block time and model accuracy of three algorithms, PoW, PoS, and PoDaS, the experimental results show that the PoDaS consensus algorithm can effectively reduce consensus time and significantly improve model accuracy compared to PoW and PoS algorithms, improving system performance and reliability.

Funding

This work was supported by The Natural Science Foundation of Hubei Province (2024AFB902): Research on the Mechanism and Utility of Supply Chain Information Governance in the Digital Ecosystem Environment) and The Philosophy and Social Science Foundation of Hubei Provincial Department of Education (22D075).

Author Contribution

Lu Cai designed the research study.Aijun Liu and Yongcai Yan analyzed the data. Lu Cai and Aijun Liu wrote the manuscript. . All authors reviewed the manuscript.

Data Availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Li Y, Chen K, Collignon S, et al. Ripple effect in the supply chain network: Forward and backward disruption propagation, network health and firm vulnerability[J]. European Journal of Operational Research, 2021, 291(3): 1117–1131.
Hajiaghaei-Keshteli M, Fathollahi Fard A M. Sustainable closed-loop supply chain network design with discount supposition[J]. Neural computing and applications, 2019, 31(9): 5343–5377.
Saragih J, Tarigan A, Pratama I, et al. The impact of total quality management, supply chain management practices and operations capability on firm performance[J]. Polish Journal of Management Studies, 2020, 21(2): 384–397.
Yildiz Cankaya S, Sezen B. Effects of green supply chain management practices on sustainability performance[J]. Journal of Manufacturing Technology Management, 2019, 30(1): 98–121.
Jones M. What we talk about when we talk about (big) data[J]. The Journal of Strategic Information Systems, 2019, 28(1): 3–16.
Wang Z J, Turko R, Shaikh O, et al. CNN explainer: learning convolutional neural networks with interactive visualization[J]. IEEE Transactions on Visualization and Computer Graphics, 2020, 27(2): 1396–1406.
Nguyen H T, Sehwag V, Hosseinalipour S, et al. Fast-convergent federated learning[J]. IEEE Journal on Selected Areas in Communications, 2020, 39(1): 201–218.
Li Q, Wen Z, Wu Z, et al. A survey on federated learning systems: Vision, hype and reality for data privacy and protection[J]. IEEE Transactions on Knowledge and Data Engineering, 2021, 35(4): 3347–3366.
Da’u A, Salim N. Recommendation system based on deep learning methods: a systematic review and new directions[J]. Artificial Intelligence Review, 2020, 53(4): 2709–2748.
Guan H, Liu M. Domain adaptation for medical image analysis: a survey[J]. IEEE Transactions on Biomedical Engineering, 2021, 69(3): 1173–1185.
Zeng S Q, Huo R, Huang T, et al. Survey of blockchain: principle, progress and application[J]. Journal on Communications, 2020, 41(1): 134–151.
Harguindeguy J B P, Cole A, Pasquier R. The variety of decentralization indexes: A review of the literature[J]. Regional & Federal Studies, 2021, 31(2): 185–208.
Kannengieber N, Lins S, Dehling T, et al. Trade-offs between distributed ledger technology characteristics[J]. ACM Computing Surveys (CSUR), 2020, 53(2): 1–37.
Gorbunova M, Masek P, Komarov M, et al. Distributed ledger technology: State-of-the-art and current challenges[J]. Computer Science and Information Systems, 2022, 19(1): 65–85.
Sharma A, Kaur P. Tamper-proof multitenant data storage using blockchain[J]. Peer-to-peer Networking and Applications, 2023, 16(1): 431–449.
Min Y A. Blockchain based Learning Management Platform for Efficient Learning Authority Management[J]. International Journal of Internet, Broadcasting and Communication, 2023, 15(3): 231–238.
Kaur M, Kumar V. A comprehensive review on image encryption techniques[J]. Archives of Computational Methods in Engineering, 2020, 27(1): 15–43.
Huang J, Kong L, Chen G, et al. Towards secure industrial IoT: Blockchain system with credit-based consensus mechanism[J]. IEEE Transactions on Industrial Informatics, 2019, 15(6): 3680–3689.
Abishu H N, Seid A M, Yacob Y H, et al. Consensus mechanism for blockchain-enabled vehicle-to-vehicle energy trading in the internet of electric vehicles[J]. IEEE Transactions on Vehicular Technology, 2021, 71(1): 946–960.
Zhang N, Sun Q, Wang J, et al. Distributed adaptive dual control via consensus algorithm in the energy internet[J]. IEEE Transactions on Industrial Informatics, 2020, 17(7): 4848–4860.
Villamil S, Hernandez C, Tarazona G. An overview of internet of things[J]. Telkomnika (Telecommunication Computing Electronics and Control), 2020, 18(5): 2320–2327.
Huang D, Ma X, Zhang S. Performance analysis of the raft consensus algorithm for private blockchains[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2019, 50(1): 172–181.
Yusoff J, Mohamad Z, Anuar M. A review: Consensus algorithms on blockchain[J]. Journal of Computer and Communications, 2022, 10(09): 37–50.
Xiao Y, Zhang N, Lou W, et al. A survey of distributed consensus protocols for blockchain networks[J]. IEEE Communications Surveys & Tutorials, 2020, 22(2): 1432–1465.
Dotan M, Pignolet Y A, Schmid S, et al. Survey on blockchain networking: Context, state-of-the-art, challenges[J]. ACM Computing Surveys (CSUR), 2021, 54(5): 1–34.
Perez-Sola C, Delgado-Segura S, Navarro-Arribas G, et al. Double-spending prevention for bitcoin zero-confirmation transactions[J]. International Journal of Information Security, 2019, 18(4): 451–463.
Ramezan G, Leung C. Analysis of proof-of-work-based blockchains under an adaptive double-spend attack[J]. IEEE Transactions on Industrial Informatics, 2020, 16(11): 7035–7045.
Pittalia P P. A comparative study of hash algorithms in cryptography[J]. International Journal of Computer Science and Mobile Computing, 2019, 8(6): 147–152.
Shang Y. Consensus of hybrid multi-agent systems with malicious nodes[J]. IEEE Transactions on Circuits and Systems II: Express Briefs, 2019, 67(4): 685–689.
Saleh F. Blockchain without waste: Proof-of-stake[J]. The Review of financial studies, 2021, 34(3): 1156–1190.
Fauzi M A, Paiman N, Othman Z. Bitcoin and cryptocurrency: Challenges, opportunities and future works[J]. The Journal of Asian Finance, Economics and Business, 2020, 7(8): 695–704.
Rodriguez-Perez R, Bajorath J. Interpretation of compound activity predictions from complex machine learning models using local approximations and shapley values[J]. Journal of medicinal chemistry, 2019, 63(16): 8761–8777.
Zheng H, Hu H, Han Z. Preserving user privacy for machine learning: Local differential privacy or federated machine learning?[J]. IEEE Intelligent Systems, 2020, 35(4): 5–14.
Kratzert F, Klotz D, Shalev G, et al. Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets[J]. Hydrology and Earth System Sciences, 2019, 23(12): 5089–5110.
Cole R, Stevenson M, Aitken J. Blockchain technology: implications for operations and supply chain management[J]. Supply chain management: An international journal, 2019, 24(4): 469–483.
Saberi S, Kouhizadeh M, Sarkis J, et al. Blockchain technology and its relationships to sustainable supply chain management[J]. International journal of production research, 2019, 57(7): 2117–2135.
Yang Q, Liu Y, Chen T, et al. Federated machine learning: Concept and applications[J]. ACM Transactions on Intelligent Systems and Technology (TIST), 2019, 10(2): 1–19.
Nguyen D C, Ding M, Pham Q V, et al. Federated learning meets blockchain in edge computing: Opportunities and challenges[J]. IEEE Internet of Things Journal, 2021, 8(16): 12806–12825.

No competing interests reported.

AuthorCont.docx

Download PDF

Reviews received at journal
25 Jul, 2024
Reviewers agreed at journal
17 Jul, 2024
Reviewers agreed at journal
15 Jul, 2024
Reviewers invited by journal
15 Jul, 2024
Editor assigned by journal
15 Jul, 2024
Editor invited by journal
11 Jul, 2024
Submission checks completed at journal
08 Jul, 2024
First submitted to journal
24 Jun, 2024

You are reading this latest preprint version

Blockchain Consensus Algorithm for Supply Chain Information Security Sharing Based on Convolutional Neural Networks

Status:

Version 1

Abstract

Figures

1. Introduction

2. Related Work

3. Blockchain Data Sharing Method Based on Privacy Protection

3.1 Dataset Introduction

3.2 Blockchain

3.3 FL

3.3.1 Local Model Training

1) Data preprocessing

2) Model initialization

3) Model training

3.3.2 Model Parameter Upload

3.3.3. Model Update Validation

3.3.4 Global Model Aggregation

3.3.5 Global Model Distribution

4. Blockchain Data Sharing Method Based on PoW and PoS Consensus Improvement

4.1 PoW Consensus Algorithm

4.2 PoS Consensus Algorithm

4.3 PoDaS Algorithm

5. Result Analysis

5.1 Comparison of Block Production Time

5.2 Comparison of Model Training Situation

6. Conclusions

Declarations

Funding

Author Contribution

Data Availability

References

Additional Declarations

Supplementary Files

Status:

Version 1