A method for improving industrial Internet performance based on intelligent blockchain sharding

Abstract

The present invention discloses a method for improving the performance of the industrial Internet based on intelligent blockchain sharding, which belongs to the field of communication network technology. The present invention adopts blockchain intelligent sharding based on a reputation mechanism and an adaptive consensus protocol switching method, which can effectively protect the security and decentralization of the blockchain system while greatly improving the scalability of the blockchain system. In addition, the present invention introduces cloud-edge-end collaborative computing to provide computing power support for the sharded blockchain, and adopts parallel computing offloading and terahertz communication technology to enhance the synergy between the three-layer cloud-edge-end network. Finally, the present invention models the optimization problem as a Markov decision process and adopts a proximal strategy optimization algorithm to solve it. Simulation results show that the total system delay and blockchain transaction throughput of this method are better than those of other methods.

Description

Technical Field

[0001] The present invention relates to a method for improving Industrial Internet of Things (IIoT) performance based on intelligent blockchain sharding, and belongs to the field of communication network technology. Background Art

[0002] In today's rapidly evolving internet age, the traditional industrial sector is undergoing a profound transformation, expanding its boundaries from simple digitization to a new stage of intelligence. This historic leap has strongly driven the vigorous development of the cutting-edge Industrial Internet of Things (IIoT). As a product of the deep integration of industrial and information technology, the IIoT aims to build a highly interconnected industrial ecosystem characterized by intelligent decision-making and automated operations through the comprehensive interconnection of industrial equipment, operational scenarios, and personnel. However, while enjoying the unlimited convenience and flexibility of open networks, this architecture can expose industrial data to potential security risks and reliability challenges.

[0003] Blockchain, a hash-proof storage structure that integrates distributed storage, consensus mechanisms, and encryption algorithms, offers an innovative solution to the data security and reliability challenges that have long plagued the IIoT. Blockchain technology is unique in its decentralized architecture, which not only provides the system with a high degree of openness and transparency but also ensures data immutability. However, blockchain technology faces a severe "trilemma" in its pursuit of decentralization, security, and scalability. Furthermore, blockchain consensus protocols often require vast computing resources, a characteristic particularly pronounced in sharded blockchain systems. This leads to limited transaction throughput and significantly increased latency, severely hindering the widespread deployment and application of sharded blockchain systems.

[0004] Sharding technology can significantly enhance blockchain scalability without compromising decentralization and security, thereby cleverly resolving the "trilemma." In recent years, with the continuous advancement of technology, a variety of sharding algorithms have emerged and have been innovatively integrated into blockchain-enabled IIoT systems, aiming to fundamentally address scalability as a bottleneck. However, it is worth noting that despite the significant potential of sharding technology, its adaptability to the highly dynamic nature of the IIoT is often overlooked. Furthermore, to address the heavy computational resource consumption of blockchain consensus protocols, researchers have proposed a series of novel consensus mechanisms tailored for dynamic IIoT scenarios. These innovative efforts have undoubtedly opened up new avenues for alleviating resource pressures and improving system efficiency. However, given the constant dynamics of the IIoT, each consensus protocol has its specific application scenarios and limitations. Therefore, a single consensus protocol cannot provide a suitable solution for the IIoT.

[0005] Furthermore, industrial equipment generally faces the challenge of limited computing resources, making it difficult to independently handle the complex and arduous computation and verification tasks in sharded blockchain systems. This significantly restricts the performance of sharded blockchain-based IIoT systems. Therefore, cloud-edge-device collaborative computing architectures, with their flexibility and powerful real-time computing capabilities, have become a key strategy for handling massive data processing tasks. Currently, numerous approaches focus on leveraging cloud-edge-device collaborative architectures to enhance the computing capabilities of blockchain-based IIoT. However, given the current situation of limited computing resources and processing power on IIoT devices, methods for deeply integrating cloud-edge-device collaborative computing with sharded blockchain technology remain exploratory, with limited results. Furthermore, existing approaches still struggle to optimize computational offload efficiency and data transmission speeds between the cloud and edge layers, failing to fully tap the potential advantages of cloud-edge collaboration to achieve efficient allocation of computing resources and rapid transmission of industrial data.

[0006] In summary, given the aforementioned challenges, we propose a method for improving the performance of the Industrial Internet based on intelligent blockchain sharding. This method aims to improve the overall response speed (i.e., reduce overall latency) and transaction processing capabilities (i.e., increase transaction throughput) of sharded blockchain-based IIoT systems. To achieve this goal, this method formulates the aforementioned joint optimization problem as a Markov decision process and solves it using the Proximal Policy Optimization (PPO) algorithm. Summary of the Invention

[0007] The present invention targets the highly complex and dynamic Industrial Internet of Things (IIoT) system, which is characterized by its large scale, wide distribution of devices, and integration of multiple blockchain sharding networks and edge intelligent servers, supplemented by a single cloud server for computing support. In this complex scenario, the present invention uses a dynamic switching mechanism of intelligent blockchain sharding and adaptive consensus protocol to model and optimize the scenario problem with the joint optimization goals of minimizing the total latency of the IIoT system and maximizing the blockchain transaction throughput. Specifically, this method deeply considers the interconnection and interoperability of massive industrial equipment, the collaborative consensus between multiple blockchain sharding networks, and the collaborative computing of multiple edge intelligent servers and cloud servers. By introducing an advanced proximal policy optimization (PPO) algorithm, the constructed complex system model is deeply iteratively learned and optimized. This process not only promotes the optimization of resource allocation, but also ensures the efficient and secure sharing of industrial data in the IIoT network, while significantly reducing the overall system response delay and significantly improving the processing efficiency of blockchain transactions. In summary, this invention utilizes reputation-based blockchain intelligent sharding and adaptive switching of multi-blockchain consensus protocols, significantly improving the scalability of blockchain systems while effectively ensuring their security and decentralization. Furthermore, this method introduces a cloud-edge-end collaborative computing framework to enhance the computing power of sharded blockchain systems, and employs parallel computational offloading methods and terahertz communication technology to enhance the synergy between the three-layer cloud-edge-end network. This provides strong technical support and solutions for the intelligent transformation and sustainable development of the IIoT sector.

[0008] The IIoT model of cloud-edge-end collaborative computing enabling blockchain intelligent sharding adapted by the present invention is shown in Figure 1 .

[0009] The system operation principle flow chart of the technical solution of the present invention is shown in Figure 2 .

[0010] The relationship between the blockchain transaction throughput and the average transaction size is shown in the figure below: Figure 3 .

[0011] The relationship between the blockchain transaction throughput and the number of IIoT devices in this invention is shown in the figure Figure 4 .

[0012] The relationship between the total delay of the system of the present invention and the number of blockchain shards is shown in Figure 5 .

[0013] like Figure 1As shown in the figure, this invention constructs the IIoT industrial network as a three-layer collaborative computing network model: cloud, edge, and device. From bottom to top, this architecture consists of the device layer, edge layer, and cloud layer. At the edge layer, M edge intelligent servers are deployed. Although these servers have limited computing resources, they undertake critical tasks such as data processing and transaction consensus. M = {1, 2, ..., M} represents the set of all edge intelligent servers. Furthermore, we introduce the set H = {1, 2, ..., H} to dynamically reflect the current available computing resources of each edge intelligent server.

[0014] At the device layer, a vast number of intelligent devices are deployed. These devices, like the tentacles of the industrial network, are widely distributed throughout various production processes, collecting and transmitting crucial industrial data such as environmental conditions and production parameters in real time. To systematically manage these devices, we define their node set as N = {1, 2, ..., N}.

[0015] A blockchain network consists of edge nodes and device nodes. Given the two distinct transaction types of cross-shard transactions (CST) and intra-shard transactions (IST), data tasks in the IIoT environment can be flexibly distributed, allowing verification to be handled by either edge intelligent server nodes or device nodes. To further optimize transaction processing efficiency and security, this paper innovatively introduces a blockchain sharding mechanism based on device node reputation. Under this mechanism, all device nodes are evenly divided into multiple blockchain shards based on their reputation levels. Each shard processes IST independently and in parallel, effectively improving the concurrent processing capabilities of the blockchain system. The set of these shards is represented by K = {1, 2, ..., K}. Notably, edge server nodes play a key role in the system, specifically responsible for processing the more complex CSTs and ensuring the security and efficiency of cross-shard transactions. Furthermore, to maintain the efficiency of the sharded blockchain architecture, edge server nodes do not directly participate in the blockchain sharding. This design ensures the scalability of the blockchain system while enhancing its flexibility and robustness in handling diverse IIoT data tasks.

[0016] When the system receives an IIoT data processing task, the intelligent agent will configure the sharding model, blockchain consensus model, data transmission model, and computing model accordingly based on the real-time environmental conditions to adapt to data task characteristics such as transaction batch size and data task size. Subsequently, the intelligent agent constructs a deep reinforcement learning framework, which specifically includes designing a state space to reflect the system state, defining an action space to cover potential optimization strategies, and establishing a reward function to quantify the effectiveness of the strategy and incentivize the intelligent agent to conduct optimal exploration. In addition, based on the constructed system model, the intelligent agent will first initialize the parameters of the training network and start the iterative learning process. During this process, the deep neural network interacts with the environment to continuously update and adjust its network parameters based on environmental feedback to achieve the optimal IIoT resource optimization strategy with low system latency and high blockchain transaction throughput. Specifically, follow the following steps in sequence:

[0017] In step (1), the IIoT industrial device requests a data task and uploads the data task to the edge layer. The edge intelligent server decides the number of blockchain shards based on the transaction batch size of the data task, the size of the data task, and other information, and executes blockchain sharding based on the reputation mechanism. The sharding strategy should meet the following rules: all device nodes with the same reputation level should be evenly divided into each partition, and the master node in each partition should be elected based on the reputation level as the only indicator; all partitions should maintain similar reputation levels and similar ratios of honest nodes and malicious nodes; nodes with extremely low reputation levels will be restricted from participating in the consensus verification of data tasks.

[0018] In the present invention, the reputation level of a device node is represented by R, which is expressed as a finite state Markov process. In addition, R is split into r values ​​and expressed as

[0019] X={x1,x2,...,x r}

[0020] In addition, R will evolve to the next state according to the state transition probability matrix, which is expressed as

[0021]

[0022] Therefore, the r×r state transition matrix of the device node reputation level R is expressed as

[0023]

[0024] Before each transaction, the controller sends a sharding permission signal to the blockchain system. All device nodes are then assigned to different blockchain shards based on their reputation and await data task assignment. The specific steps are as follows.

[0025] In step (1.1), first, the intelligent agent obtains relevant environmental information to determine the number of blockchain shards.

[0026] In step (1.2), the intelligent agent obtains the reputation level of all device nodes and records it on the reputation blockchain.

[0027] In step (1.3), each device node will be assigned to the shard with the smallest current size according to its reputation level to balance the sizes of each shard and maintain the ratio of honest and dishonest nodes in each block.

[0028] In step (1.4), the bottom 20% of device nodes in the reputation ranking will be warned, and the bottom 5% of device nodes will be restricted from participating in the consensus verification process. In addition, the blockchain system will be resharded after each round of transactions.

[0029] In step (2), blockchain intelligent sharding divides the transaction processing types of the blockchain into two types: intra-shard transactions and cross-shard transactions. Intra-shard transactions are verified and processed by the IIoT device nodes in each shard, while cross-shard transactions are jointly responsible for consensus verification by the transaction parties (device nodes) and the connected edge intelligent servers.

[0030] Due to the dynamic nature of IIoT, a single blockchain consensus protocol is difficult to apply to IIoT data tasks. In this case, this paper considers a dynamic switching model of multiple blockchain consensus protocols to match the most appropriate consensus protocol for each data task. There are three blockchain consensus protocols used, as shown below:

[0031] Step (2.1), Practical Byzantine Fault Tolerance (PBFT) was proposed by M. Castro et al. in 1999 and improved in 2002. PBFT reduces the algorithm complexity to the polynomial level based on Byzantine Fault Tolerance (BFT), making it feasible in practical applications such as Hyper-ledger Fabric. In addition, PBFT has high robustness because consensus verification can be performed without error when the number of Byzantine nodes does not exceed one-third. However, due to the complex node broadcast process, PBFT consumes higher latency and computational overhead. Assume that the number of consensus nodes is N c , the signature calculation period is χ, and the message authentication code (MAC) calculation period is δ, then the PBFT consensus process is as follows:

[0032] Request: The client first submits a consensus request, then packages the transaction data into blocks, and the master node verifies the signature and MAC. Therefore, the computation cycle of this process is expressed as

[0033]

[0034] Where B(t) is the transaction batch size and A(t) is the average transaction size.

[0035] Pre-preparation: Initially, a signature and N c -1 MAC is generated and forwarded to all replica nodes participating in block verification. Subsequently, each replica node verifies the received signature and MAC. Therefore, the computation cycle of the master node is the time required to generate the signature and MAC and forward them to the replica, while the computation cycle of the replica is the time required to verify the received signature and MAC. As mentioned above, the computation cycle of the master node is

[0036] c2 p (t) = χ + (N c -1)δ

[0037] The calculation period of the replica node is

[0038]

[0039] Where ρ is the probability of correct transaction verification.

[0040] Preparation: The preparation phase involves the process of completing information broadcasting between consensus nodes. Specifically, the replica nodes will sign the verified blocks and generate N c -1 MAC to broadcast to all nodes including the master node. It is worth noting that when the number of authentication nodes is not less than 2ζ (where ζ=(N c -1) / 3), PBFT can ensure the effectiveness of the consensus verification process. Therefore, the calculation cycle of the master node in this process is

[0041] c3 p (t) = 2ζ(χ + δ)

[0042] The calculation period of the replica node is

[0043] c3 r (t)=2ζ(χ+δ)+χ+(N c -1)δ

[0044] Submit: In the submission phase, all nodes need to exchange and verify block information. Specifically, they need to generate a signature and N c -1 MAC, and verify the received signature and MAC. Therefore, the computation cycle of this stage can be given by the following formula

[0045] c4(t)=2ζ(χ+δ)+χ+(N c -1)δ

[0046] Reply: In this phase, the consensus node replies to the client with a verification message. If the client receives more than 2ζ reply messages, the consensus verification process is considered valid. At this point, the newly generated block will be appended to the blockchain. Therefore, the calculation formula for this phase is

[0047] c5(t)=2ζ(χ+δ)+(χ+δ)

[0048] In summary, the computation cycle required for PBFT is

[0049]

[0050] Step (2.2), Zyzzyva algorithm. Byzantine faults are recognized as the most serious faults in distributed systems. The BFT algorithm provides an effective method for handling Byzantine faults, but its highly redundant verification procedures consume a lot of computing resources, affecting the transaction efficiency of the blockchain. Since computer systems rarely encounter errors now, speculative BFT consensus protocols such as Zyzzyva have received widespread attention. The Zyzzyva consensus protocol can skip the call of the fault tolerance mechanism in non-faulty situations, thereby significantly improving the performance of the blockchain. Similar to PBFT, Zyzzyva can ensure the correctness of transactions when the number of honest nodes exceeds 2ζ. When all nodes are honest, Zyzzyva can skip node broadcasting and directly perform speculative execution. Otherwise, if there are malicious nodes, Zyzzyva will execute the speculative execution at t r During this time, the recovery process is started.

[0051] In the fast-case scenario, Zyzzyva’s authentication process is divided into three phases, namely, request, service request, and speculative reply. Therefore, the computation cycle can be expressed as

[0052]

[0053] In the two-phase case, Zyzzyva's authentication process is divided into five phases, namely request, service request, speculative reply, submission, and reply. Therefore, the computation cycle in the two-phase case is expressed as

[0054]

[0055] In step (2.3), the Quorum consensus protocol uses the pigeonhole principle to implement a simple communication model. Specifically, it only requires one message exchange between the client and the replica to complete the transaction verification. The consensus process of the Quorum consensus protocol consists of two steps, namely request and reply. It should be noted that the reply information includes the historical summary of the replica. If the client receives less than N replies, c If a panic mechanism is called, the verification process will be terminated. Therefore, the calculation cycle of Quorum is expressed as

[0056]

[0057] In addition, the data integrity and network robustness of the blockchain require that the proportion of malicious nodes is within the safety range of the specific consensus protocol (PBFT, Zyzzyva, and Quorum can tolerate up to one-third of malicious validators). Therefore, to ensure the security of the proposed architecture, the following constraints are imposed:

[0058] 3N p +1≤N c

[0059] Among them, N p is the number of Byzantine nodes, N c is the number of consensus nodes.

[0060] For intra-shard transactions, the consensus node consists of device nodes within a single shard, so N c =N / K. Obviously, the number of shards K must obey the following constraints:

[0061]

[0062] In step (3), terahertz communication technology is a wireless communication technology that uses electromagnetic waves in the terahertz frequency band as a carrier wave. The terahertz frequency band can support transmission rates exceeding 10Gbps and has excellent confidentiality and anti-interference capabilities. Therefore, the present invention uses terahertz communication technology to further optimize the computational offload link from the edge layer to the cloud layer.

[0063] Assuming that the average distance of the unloading link is l, when a terahertz electromagnetic wave with a frequency of ω propagates in a medium, the molecular absorption loss caused by the interaction between the electromagnetic wave and various molecules and atoms in the medium is:

[0064]

[0065] Here, τ represents the transmittance and κ(ω) is the absorption coefficient.

[0066] Propagation loss, also known as free space loss, refers to the loss caused by electromagnetic waves penetrating a medium. In terahertz communication technology, the formula for calculating propagation loss is:

[0067]

[0068] Where c = 2.9979 × 10 8 m / s represents the speed of light in a vacuum (the speed of light).

[0069] The total loss of terahertz communication technology is the sum of molecular absorption loss and free space loss, which can be expressed as:

[0070]

[0071] In the terahertz transmission link, the noise temperature of the cloud receiver mainly comes from molecular absorption noise. The equivalent noise temperature caused by molecular absorption is expressed as:

[0072] N abs T =N0 T ε

[0073] Among them, N0 T = 296 K refers to the reference temperature, ε = 1-τ represents the emissivity of the calculated unloading channel, and τ represents the transmittance.

[0074] Therefore, for a given bandwidth B, the noise power absorbed by the molecule is given by:

[0075]

[0076] According to the above analysis, the signal-to-noise ratio can be expressed as:

[0077]

[0078] Among them, P t (t) is the transmission power of the edge intelligent server. In addition, and Refers to the antenna gain of the transmitting end and the receiving end respectively.

[0079] Therefore, the data transmission rate based on terahertz communication technology can be calculated as follows:

[0080]

[0081] Step (4), in the framework proposed by the present invention, the edge intelligent server bears heavy verification and computing tasks. Therefore, in order to better assign tasks to the edge intelligent server, we need to obtain the computing resources of the edge intelligent server at any time.

[0082] Step (4.1), in this case, we model the computing resources H of the edge intelligent server as a Markov process and divide it into h discrete values, which can be expressed as:

[0083] Y={y1,y2,...,y h}

[0084] In addition, the computing resources of the edge intelligent server evolve to the next state according to the state transition probability matrix, as shown in the following formula:

[0085]

[0086] Therefore, the state transition probability matrix can be expressed as:

[0087]

[0088] In step (4.2), since the computing resources of the edge intelligent server are relatively limited, it is extremely challenging to perform multiple tasks including cross-shard consensus verification and data calculation. Therefore, in order to reduce the edge burden and improve the computing efficiency of the system, we consider a parallel offloading strategy from the edge layer to the cloud layer. Assuming that the computing offload ratio is O∈[0,1], the computing delay of the edge intelligent server and the cloud server can be expressed as:

[0089]

[0090] and

[0091]

[0092] Among them, G(t) refers to the calculation period, F m and F c denote the computing frequencies of the edge intelligence server and the cloud server, respectively. In addition, D(t) represents the data task size, and λ(t) refers to the data transmission rate.

[0093] Therefore, according to the bucket effect, the total computational delay of parallel computing can be given by the following formula:

[0094] t1(t)=max{t1 m (t),t1 c (t)}

[0095] In step (4.3), in order to efficiently execute IIoT data tasks, it is necessary to clearly classify transaction types based on blockchain sharding. As we all know, the transaction types in IIoT supported by sharded blockchains can be roughly divided into two types, namely IST and CST.

[0096] The consensus delay of IST can be expressed as:

[0097]

[0098] Among them, the value of C(t) comes from the set {CP (t),C Z 1 (t),C Z 2 (t),C Q (t)}. In addition, T I represents the block interval, F d Refers to the frequency of the device, t b Indicates the broadcast delay between consensus nodes.

[0099] The consensus delay of CST can be expressed as:

[0100]

[0101] In summary, the total delay is calculated as:

[0102] T(t)=t1(t)+t2(t)

[0103] Among them, t1(t) represents the calculation delay, and its value is t1 m (t) or t1 c (t). In addition, t2(t) is the consensus delay, which depends on the transaction type of the blockchain and is taken as t2 d (t) or t2 m (t).

[0104] In step (4.4), the present invention evaluates the transaction throughput of the blockchain system by calculating the number of transactions completed per unit time. In addition, since all shards verify IIoT data tasks in parallel, the throughput Ξ is positively correlated with the number of blockchain shards K. Specifically, the calculation formula for transaction throughput Ξ is:

[0105]

[0106] Step (5): Based on steps (1)-(4), combined with the environment and optimization objectives, set the state space, action space and reward function of the PPO algorithm. The specific steps are as follows:

[0107] In step (5.1), the state space defined in this invention consists of five key variables, namely, the data task size D(t), the transaction batch size B(t), the transmission rate λ(t), the computing resources H of the edge intelligent server, and the reputation level R of all device nodes. Therefore, the state space is set as:

[0108] S(t)={D(t),B(t),λ(t),H,R}

[0109] The data task size D(t) ranges from 4MB to 8MB, and the transaction batch size B(t) ranges from 1MB to 2MB. Due to the diversity and dynamic nature of wireless link conditions for computational offloading, the data transmission rate is modeled as a finite-state Markov decision process. The data transmission rate assigned to each computational offloading task is from the set {125, 100, 75, 50} Mbps. The probability transition matrix is ​​expressed as:

[0110]

[0111] In step (5.2), in order to obtain better long-term returns, it is necessary to flexibly adjust the action strategy according to environmental changes. Therefore, the action space is set as follows:

[0112] A(t)={S B ,T I ,O,C(t),K}

[0113] Among them, S B ={1,2,...,B} represents the block size, T I = {0.5, 1, ..., I} represents the block interval. Furthermore, O∈[0, 1] represents the computation offload ratio. C(t) represents the decision factor of the consensus protocol. Furthermore, K represents the number of shards.

[0114] In step (5.3), in order to jointly optimize transaction throughput and total latency, the reward function is defined as the weighted sum of the inverse of throughput and total latency, and its formula is:

[0115]

[0116] stC1:D(t)≤S B

[0117] C2:T(t)≤T I

[0118] C3:K≤N / (3N p +1)

[0119] Among them, α1+α2=1, ν is a reward and punishment parameter.

[0120] Step (6) completes the setting of parameters such as the PPO algorithm execution strategy and neural network training based on the state space, action space, and reward function constructed in step (5).

[0121] Propagation Poisson Optimization (PPO) is a deep reinforcement learning algorithm proposed by Open-AI in 2017 and is considered one of the most widely applicable algorithms. The key concept of PPO is to keep the policy update amplitude within a controllable range. Furthermore, the PPO algorithm updates the policy by maximizing an objective function, which consists of two parts: the policy's expected reward and a constraint on the policy update amplitude. Specifically, the PPO algorithm estimates the policy's expected reward by sampling multiple trajectories. The PPO algorithm has many advantages, including excellent stability and convergence. Furthermore, PPO can balance the policy's exploration and exploitation by adjusting the weights of the constraint terms, thereby better adapting to different environments.

[0122] Considering the dynamic characteristics of IIoT, in this method, cloud servers with super computing power are considered as intelligent agents of the PPO network. In addition, the output of the actor network is a policy π that follows a normal distribution. θ , whose mean and variance are μ(s t ) and σ(s t ). The intelligent agent follows a normal distribution policy π θ The initial action is sampled and fed back to the environment to obtain information such as the next state.

[0123] To ensure the consistency of policy optimization, the Cut Generation Objective Function (CSOF) is introduced into PPO, and then the mini-batch stochastic gradient ascent method is used to optimize it. The calculation formula of CSOF is:

[0124]

[0125] Where ε represents the shear index and r(θ) represents the ratio of the previous policy to the new policy, which is used to ensure the difference between the old and new policies, as follows:

[0126]

[0127] also, represents the generalized advantage estimation function, where V φ (s t ) represents the approximate value function. In addition, the critic network updates the strategy by gradient descent of the loss function, which can be expressed as:

[0128]

[0129] The advantage of an action at a given time t can be quantified to some extent and is defined as:

[0130]

[0131] in and Represent the value functions of actions and states respectively.

[0132] Finally, the mini-batch stochastic optimization algorithm is used to update the performer and critic parameters θ and φ respectively through gradient ascent and gradient descent as follows:

[0133]

[0134] Step (7) is to obtain the optimal strategy for the optional actions in each state based on the deep neural network trained in step (6), and use the action generated by the strategy as the optimal action in that state, and continue to execute the optimal action in each state until the execution of the instruction is completed.

[0135] The advantage of the present invention is that it integrates multiple blockchain sharding networks and edge intelligent servers, supplemented by a single cloud server as computing support, for the highly complex, large-scale, widely distributed and dynamic Industrial Internet of Things (IIoT) system. The present invention uses a dynamic switching mechanism of intelligent blockchain sharding and adaptive consensus protocol, with the minimization of the total delay of the IIoT system and the maximization of blockchain transaction throughput as joint optimization goals to model and optimize scenario problems. Specifically, this method deeply considers the interconnection and interoperability of massive industrial equipment, the collaborative consensus between multiple blockchain sharding networks, and the collaborative computing of multiple edge intelligent servers and cloud servers. By introducing an advanced proximal policy optimization (PPO) algorithm, the constructed complex system model is deeply iterated and optimized. This process not only promotes the optimization of resource allocation, but also ensures the efficient and secure sharing of industrial data in the IIoT network, while significantly reducing the overall response delay of the system and greatly improving the processing efficiency of blockchain transactions. In summary, the present invention successfully solves key issues such as how to balance data processing speed and system stability in the resource-intensive and network-heterogeneous IIoT environment, and how to achieve efficient transmission and processing of industrial data while ensuring data security. This provides strong technical support and solutions for the intelligent transformation and sustainable development of the IIoT field. BRIEF DESCRIPTION OF THE DRAWINGS

[0136] Figure 1 ,The communication scenario model includes the structural diagram of IIoT intelligent industrial equipment, ,intelligent edge server, controller, cloud computing server and blockchain system.

[0137] Figure 2 , a flow chart for designing an industrial Internet performance improvement method based on intelligent blockchain sharding.

[0138] Figure 3 , a graph showing the relationship between blockchain transaction throughput and average transaction size. In the figure, the cross represents the method described in the present invention, the diamond represents the method with a fixed number of blockchain shards, the circle represents the binary computation offloading method (non-parallel offloading), and the square represents the single blockchain consensus protocol method (non-dynamic switching of multiple blockchain consensus protocols).

[0139] Figure 4 , a graph showing the relationship between blockchain transaction throughput and the number of IIoT devices. In the figure, the cross represents the method described in the present invention, the diamond represents the method based on the A3C deep reinforcement learning algorithm, the circle represents the method with a fixed number of blockchain shards, and the square represents the non-blockchain sharding method.

[0140] Figure 5 , a graph showing the relationship between total system latency and the number of blockchain shards. In the graph, diamonds represent single blockchain consensus protocol methods (not dynamic switching of multiple blockchain consensus protocols), squares represent binary computation offloading methods (not parallel offloading), circles represent fixed blockchain shard number methods, and crosses represent the method described in the present invention. DETAILED DESCRIPTION

[0141] The following is a further explanation of the technical solution of the industrial Internet performance improvement method based on intelligent blockchain sharding with the help of accompanying drawings and examples.

[0142] This method uses blockchain intelligent sharding based on a reputation mechanism and adaptive switching of multi-blockchain consensus protocols, which can significantly improve the scalability of blockchain systems while effectively ensuring their security and decentralization. Due to the relatively scarce computing resources of sharded blockchain systems, this method introduces a cloud-edge-end collaborative computing framework to enhance the computing power of sharded blockchain systems. It also uses parallel computing offloading methods and terahertz communication technology to enhance the synergy between the three-layer cloud-edge-end network. In addition, considering the dynamic characteristics of IIoT, this method models the joint optimization problem of total system latency and blockchain transaction throughput as a Markov decision process (MDP) and solves it using the proximal policy optimization (PPO) algorithm.

[0143] The flow chart of the method of the present invention is as follows Figure 2 As shown, the following steps are included:

[0144] Step 1: Set the number of industrial device nodes and intelligent edge server nodes, and set the number of blockchain shards based on the data task size, transaction batch size, and the number of device nodes, and execute the intelligent blockchain sharding algorithm based on the reputation mechanism;

[0145] Step 2: Classify each data task into transaction types (intra-shard transactions / cross-shard transactions) and match the corresponding blockchain consensus protocol (PBFT / Zyzzyva / Quorum) to complete the consensus verification procedure for the transaction;

[0146] Step 3: Use terahertz communication technology to increase the data transmission rate of the link from the edge layer to the cloud layer for computation offload;

[0147] Step 4: Based on the data task size and the computing resources of the intelligent edge server, the ratio of computing offload from the edge layer to the cloud layer is intelligently determined. The consensus verification delay, transmission delay, and computation delay are calculated and summarized to obtain the total system delay. The blockchain transaction throughput is also calculated based on the number of blockchain shards, average transaction size, block size, and block interval.

[0148] Step 5: Establish the optimization problem and model the Markov decision process, setting the state space, action space and reward function;

[0149] Step 6: Complete the setting of algorithm execution strategy and neural network training parameters according to the PPO algorithm;

[0150] Step seven: Select the optimal action based on the optimal strategy obtained in each state to obtain the maximum benefit.

[0151] Figure 3 The figure shows the relationship between blockchain transaction throughput and average transaction size. It can be noted that as the average transaction size increases, the blockchain transaction throughput gradually decreases. In addition, when the transaction throughput of different schemes is compared vertically, under different average transaction sizes, the scheme proposed in the present invention can always achieve higher transaction throughput. For example, when the average transaction size is 200B, the transaction throughput achieved by the method of the present invention can reach 2390TPS, while the transaction throughput of the other methods can only reach a maximum of 1790B. This is due to our comprehensive consideration of parallel computing offloading and smart sharding blockchain, as well as the use of the PPO algorithm with excellent training efficiency. Parallel computing offloading can reasonably allocate the computing resources of edge intelligent servers, thereby significantly improving computing efficiency. In addition, the smart sharding blockchain method can better adapt to the data tasks of IIoT by increasing the number of independent blockchain networks (number of blockchain shards).

[0152] Figure 4The following graph shows the relationship between blockchain transaction throughput and the number of IIoT devices. As can be seen from the figure, the proposed solution achieves higher transaction throughput than other methods, and this advantage becomes increasingly apparent as the number of device nodes increases. When the number of device nodes is 70, the transaction throughput corresponding to the proposed method reaches 4446 TPS, while the transaction throughput of other methods can only reach a maximum of 3621 TPS. In addition, with the exception of the fixed sharding method and the blockchain-free sharding method, the transaction throughput of all solutions increases linearly with the number of device nodes. This is because as the number of device nodes increases, the intelligent agent tends to create more shards.

[0153] Figure 5 This is a graph showing the relationship between the total system latency and the number of blockchain shards. Figure 5 Using the number of blockchain shards as the independent variable, the total system latency achieved by different methods was compared. It can be observed that the proposed solution consistently maintains low latency across different blockchain shard numbers. For example, when the number of blockchain shards is 8, the proposed solution can achieve a total latency as low as 2.21s, while other solutions require a minimum total system latency of 3.83s. Furthermore, when using a fixed shard number method (the number of shards is manually adjusted as an independent variable), the latency decreases almost linearly with the increase in the number of shards. This is because the consensus latency consumed by the blockchain during transaction verification decreases linearly with the increase in the number of shards.

[0154] Furthermore, with the exception of the method with a fixed number of shards, the total latency of all methods did not change significantly with increasing the number of shards. This is because in the smart sharding blockchain approach, the optimal number of blockchain shards is determined by an intelligent agent rather than manually set. In this context, our proposed method can organically integrate blockchain systems with the highly dynamic IoT.

Claims

1. A method for improving industrial Internet performance based on intelligent blockchain sharding, characterized in that: The method comprises the following steps: Step 1: Set the number of industrial device nodes and intelligent edge server nodes, and set the number of blockchain shards based on the data task size, transaction batch size, and the number of device nodes, and execute the intelligent blockchain sharding algorithm based on the reputation mechanism; Step 2: Classify the transaction type for each data task, i.e., intra-shard transaction / cross-shard transaction, and match the corresponding blockchain consensus protocol to complete the consensus verification procedure for the transaction; Step 3: Use terahertz communication technology to increase the data transmission rate of the link from the edge layer to the cloud layer for computation offload; Step 4: Based on the data task size and the computing resources of the intelligent edge server, the computing offload ratio from the edge layer to the cloud layer is intelligently determined. The consensus verification delay, transmission delay, and computing delay are calculated and summarized to obtain the total system delay. The blockchain transaction throughput is calculated based on the number of blockchain shards, average transaction size, block size, and block interval; Step 5: Establish the optimization problem and model the Markov decision process, setting the state space, action space and reward function; Step 6: Complete the setting of algorithm execution strategy and neural network training parameters according to the PPO algorithm; Step seven: Select the optimal action based on the optimal strategy obtained in each state to obtain the maximum benefit.

2. The method for improving industrial Internet performance based on intelligent blockchain sharding according to claim 1 is characterized in that: In step 1, the number of industrial device nodes and intelligent edge server nodes is first set, and the number of blockchain shards is set based on the data task size, transaction batch size, and the number of device nodes. Then, the intelligent blockchain sharding algorithm based on the reputation mechanism is executed to perform blockchain sharding. The specific steps are as follows: IIoT industrial devices request data tasks and upload them to the edge layer. The edge intelligent server determines the number of blockchain shards based on the transaction batch size and data task size of the data task and executes blockchain sharding based on a reputation mechanism. The sharding strategy meets the following rules: all device nodes with the same reputation level should be evenly divided into different partitions. The master node in each partition is elected based on reputation level as the only metric. All partitions should maintain similar reputation levels and similar ratios of honest nodes to malicious nodes. Nodes with extremely low reputation levels will be restricted from participating in the consensus verification of data tasks. The reputation level of a device node is represented by R, which is expressed as a finite state Markov process; in addition, R is split into r values ​​and expressed as X={x1,x2,...,x r } Where X represents the set of different states of reputation level R that are split in the finite state Markov process, x1, x2, ..., x r They represent the first, second, and up to the rth split states of the credibility level R respectively; In addition, R will evolve to the next state according to the state transition probability matrix, which is expressed as Among them, x i and x j Indicates the different states of the device node reputation level R, Reputation level R from x i The state evolves to x j The state transition probability matrix of the state, Pr represents the probability, R(t) and R(t+1) represent the device node reputation level R at the current moment and the next moment respectively; r×r state transition matrix of device node reputation level R Expressed as Before each round of transactions, the controller sends a shard permission signal to the blockchain system. Then, all device nodes are assigned to various blockchain shards based on the reputation mechanism and wait for the assignment of data tasks. The details are as follows: First, the intelligent agent obtains relevant environmental information to determine the number of blockchain shards; second, the intelligent agent obtains the reputation level of all device nodes and records it on the reputation blockchain; each device node will be assigned to the smallest shard based on its reputation level to balance the size of each shard and maintain the ratio of honest and dishonest nodes in each block; the 20% of device nodes ranked in the bottom of the reputation level will be warned, and the 5% of device nodes ranked in the bottom will be restricted from participating in the consensus verification process; in addition, the blockchain system will be re-sharded after each round of transactions.

3. The method for improving industrial Internet performance based on intelligent blockchain sharding according to claim 2 is characterized in that: In step 2, each data task is classified into a transaction type and matched with the corresponding blockchain consensus protocol to complete the consensus verification procedure for the transaction; the details are as follows: Blockchain smart sharding divides blockchain transaction processing types into intra-shard transactions and cross-shard transactions. Intra-shard transactions are verified and processed by the IIoT device nodes within each shard, while cross-shard transactions are jointly responsible for consensus verification by the transaction parties and the connected edge intelligent servers. Due to the dynamic nature of IIoT, a single blockchain consensus protocol is difficult to apply to IIoT data tasks. In this case, a dynamic switching model of multiple blockchain consensus protocols is considered to match the most appropriate consensus protocol for each data task. There are three blockchain consensus protocols used, as shown below: Practical Byzantine Fault Tolerance (PBFT): Assume the number of consensus nodes is N. c , the signature calculation period is χ, and the message authentication code MAC calculation period is δ, then the PBFT consensus process is as follows: Request: The client first submits a consensus request, then packages the transaction data into blocks, and the master node verifies the signature and MAC. The calculation cycle is expressed as: Where, the calculation period of the signature is χ, the calculation period of the message authentication code MAC is δ; B(t) is the transaction batch size, and A(t) is the average transaction size; Pre-preparation: Initially, a signature and N c -1 MAC is generated and forwarded to all replica nodes participating in block verification; then, each replica node verifies the received signature and MAC; therefore, the computation cycle of the master node is the time required to generate the signature and MAC and forward them to the replica, while the computation cycle of the replica is the time required to verify the received signature and MAC; as mentioned above, the computation cycle of the master node is c2 p (t)=x+(N c -1)d The calculation period of the replica node is Among them, N c is the number of consensus nodes, ρ is the probability of correct transaction verification; Preparation: The preparation phase involves the process of completing information broadcasting between consensus nodes; specifically, the replica nodes will sign the verified blocks and generate N c -1 MAC to broadcast to all nodes including the master node; when the number of authentication nodes is not less than 2ζ, where ζ=(N c -1) / 3, PBFT ensures the effectiveness of the consensus verification process; the calculation cycle of the master node is c3 p (t)=2ζ(x+d) The calculation period of the replica node is c3 r (t)=2ζ(χ+δ)+χ+(N c -1)d Submit: In the submission phase, all nodes need to exchange and verify block information. Specifically, they need to generate a signature and N c -1 MAC and verifies the received signature and MAC; therefore, the computational cycle of the submission phase is given by c4(t)=2ζ(χ+δ)+χ+(N c -1)d Reply: In the reply phase, the consensus node replies with a verification message to the client. If the client receives more than 2ζ reply messages, the consensus verification process is considered valid. At this point, the newly generated block will be appended to the blockchain. The calculation formula for the reply phase is: c5(t)=2ζ(χ+δ)+(χ+δ) In summary, the computation cycle required for PBFT is Zyzzyva consensus protocol: When the number of honest nodes exceeds 2ζ, Zyzzyva ensures the correctness of the transaction; when all nodes are honest, Zyzzyva skips node broadcasting and executes directly; otherwise, if there are malicious nodes, Zyzzyva will execute the transaction at t r During this period, the recovery process is initiated; In the fast-case scenario, Zyzzyva’s authentication process is divided into three phases, namely, request, service request, and speculative reply; therefore, the computation cycle is expressed as In the two-phase case, Zyzzyva's authentication process is divided into five phases, namely request, service request, speculative reply, submission, and reply; therefore, the computation cycle in the two-phase case is expressed as Among them, t r The delay from the start of the recovery procedure to the successful recovery of the consensus process; Quorum consensus protocol: The consensus process of the Quorum consensus protocol consists of two steps, namely request and reply; the reply information includes the historical summary of the replica; if the client receives less than N replies c The panic mechanism is called and the verification process is terminated; therefore, the calculation cycle of Quorum is expressed as In addition, the data integrity and network robustness of the blockchain require that the proportion of malicious nodes is within the safety range of the specific consensus protocol; therefore, to ensure the security of the proposed architecture, the following constraints are imposed: <h2 style=";text-align:left;direction:ltr">3N<h2 style=";text-align:left;direction:ltr"> p <h2 style=";text-align:left;direction:ltr"> +1≤N<h2 style=";text-align:left;direction:ltr"> c Among them, N p is the number of Byzantine nodes, N c is the number of consensus nodes; For intra-shard transactions, the consensus node consists of device nodes within a single shard, so N c =N / K; Obviously, the number of shards K must obey the following constraints:

4. The method for improving industrial Internet performance based on intelligent blockchain sharding according to claim 3 is characterized in that: In step three, terahertz communication technology is used to improve the data transmission rate of the link for computing offload from the edge layer to the cloud layer; Apply terahertz communication to optimize the computation offloading link from the edge layer to the cloud layer; The average distance of the offloading link is When a terahertz electromagnetic wave with a frequency of ω propagates in a medium, the molecular absorption loss caused by the interaction between the electromagnetic wave and various molecules and atoms in the medium is: Where τ represents the transmittance and κ(ω) is the absorption coefficient; Propagation loss, also known as free space loss, refers to the loss caused by electromagnetic waves penetrating a medium. In terahertz communication technology, the formula for calculating propagation loss is: Where c = 2.9979 × 10 8 m / s represents the speed of light in a vacuum; The total loss of terahertz communication technology is the sum of molecular absorption loss and free space loss, which can be expressed as: In the terahertz transmission link, the noise temperature of the cloud receiver mainly comes from molecular absorption noise. The equivalent noise temperature caused by molecular absorption is expressed as: N abs T =N0 T e Among them, N0 T = 296K refers to the reference temperature, ε = 1-τ represents the emissivity of the calculated unloading channel, and τ represents the transmittance; Therefore, for a given bandwidth B, the noise power absorbed by the molecule is given by: P n =∫ B N abs T dω According to the above analysis, the signal-to-noise ratio is expressed as: Among them, P t (t) is the transmission power of the edge intelligent server; in addition, and Refers to the antenna gain of the transmitting end and the receiving end respectively; Therefore, the data transmission rate based on terahertz communication technology is calculated as follows:

5. The method for improving industrial Internet performance based on intelligent blockchain sharding according to claim 4 is characterized in that: In step 4, the ratio of computing offload from the edge layer to the cloud layer is intelligently determined based on the data task size and the computing resources of the intelligent edge server. The consensus verification delay, transmission delay, and computing delay are calculated and summarized to obtain the total system delay. The blockchain transaction throughput is calculated based on the number of blockchain shards, average transaction size, block size, and block interval, as follows: The computing resources H of the edge intelligent server are modeled as a Markov process and divided into h discrete values, expressed as: Y={y1,y2,...,y h } Where Y represents the set of different states of the computing resources H of the split edge intelligence server in the finite state Markov process, y1,y2,...,y h They represent the first, second, and finally the hth split states of the computing resource H of the edge intelligent server respectively; In addition, the computing resources of the edge intelligent server evolve to the next state according to the state transition probability matrix, as shown in the following formula: Among them, y i and y j Indicates the different states of the computing resources H of the edge intelligent server, Indicates H from y i The state evolves to y j The state transition probability matrix of the state, Pr represents the probability, H(t) and H(t+1) represent the computing resources H of the edge intelligent server at the current moment and the next moment respectively; Therefore, the state transition probability matrix is ​​expressed as: The computation offloading ratio is O∈[0,1], and the computation delays of the edge intelligent server and the cloud server are expressed as: and Among them, G(t) refers to the calculation period, F m and F c denote the computing frequencies of the edge intelligent server and cloud server respectively; in addition, D(t) represents the data task size, and λ(t) refers to the data transmission rate; According to the bucket effect, the total computational delay of parallel computing is given by the following formula: t1(t)=max{t1 m (t),t1 c (t)} Transaction types are clearly classified based on blockchain sharding. Transactions in IIoT supported by sharded blockchains can be broadly divided into two types: intra-shard transactions (ISTs) and cross-shard transactions (CSTs). The consensus delay of IST is expressed as: Among them, the value of C(t) comes from the set {C P (t),C Z 1 (t),C Z 2 (t),C Q (t)}; In addition, T I represents the block interval, F d Refers to the frequency of the device, t b Indicates the broadcast delay between consensus nodes; The consensus delay of CST is expressed as: To summarize, the total delay is calculated as: T(t)=t1(t)+t2(t) Among them, t1(t) represents the calculation delay, and its value is t1 m (t) or t1 c (t); In addition, t2(t) is the consensus delay, which depends on the transaction type of the blockchain and is t2 d (t) or t2 m (t); The transaction throughput of the blockchain system is evaluated by calculating the number of transactions completed per unit time. In addition, since all shards verify IIoT data tasks in parallel, the throughput is Positively correlated with the number of blockchain shards K; transaction throughput The calculation formula is: Among them, S B is the block size, T I represents the block interval, and A(t) is the average transaction size.

6. The method for improving industrial Internet performance based on intelligent blockchain sharding according to claim 5 is characterized in that: In step 5, we first establish the optimization problem and model the Markov decision process, setting the state space, action space, and reward function as follows: The defined state space consists of five key variables, namely, data task size D(t), transaction batch size B(t), transmission rate λ(t), computing resources H of the edge intelligence server, and reputation level R of all device nodes; therefore, the state space is set as: S(t)={D(t),B(t),λ(t),H,R} The data task size D(t) ranges from 4MB to 8MB, and the transaction batch size B(t) ranges from 1MB to 2MB. Due to the diversity and dynamic nature of wireless link conditions for computational offloading, the data transmission rate is modeled as a finite-state Markov decision process. The data transmission rate assigned to each computational offloading task is from the set {125, 100, 75, 50} Mbps. The probability transition matrix is ​​expressed as: In order to obtain better long-term returns, it is necessary to flexibly adjust the action strategy according to environmental changes; therefore, the action space is set as follows: A(t)={S B ,T I ,O,C(t),K} Among them, S B ={1,2,...,B} represents the block size, T I ={0.5,1,...,I} represents the block interval; in addition, O∈[0,1] represents the computation offloading ratio; C(t) represents the decision factor of the consensus protocol; in addition, K represents the number of shards; To jointly optimize transaction throughput Ξ and total latency T(t), the reward function is defined as the weighted sum of the inverse of throughput and total latency, as follows: s.t.C1:D(t)≤S B C2:T(t)≤T I <h2 style=";text-align:left;direction:ltr">C3:K≤N / (3N<h2 style=";text-align:left;direction:ltr"> p <h2 style=";text-align:left;direction:ltr"> +1) Among them, α1+α2=1, ν is a reward and punishment parameter.

7. The method for improving industrial Internet performance based on intelligent blockchain sharding according to claim 6 is characterized in that: In step 6, the algorithm execution strategy and neural network training parameters are set according to the PPO algorithm, as follows: Considering the dynamic characteristics of IIoT, cloud servers with super computing power are considered as intelligent agents of the PPO network; the output of the actor network is a strategy π that obeys the normal distribution. θ , whose mean and variance are μ(s t ) and σ(s t ); the intelligent agent follows a normal distribution strategy π θ Sample the initial action and feed it back to the environment to obtain the next state information; To ensure the consistency of policy optimization, the clipping objective function CSOF is introduced into PPO, and then the mini-batch stochastic gradient ascent method is used to optimize it; the calculation formula of CSOF is: Where ε represents the shear index and r(θ) represents the ratio of the previous policy to the new policy, which is used to ensure the difference between the old and new policies, as follows: also, represents the generalized advantage estimation function, where V φ (s t ) and V φ (s t+1 ) represents the approximate value function at different moments; in addition, the critic network updates the strategy by gradient descent of the loss function, which is expressed as: To quantify the advantage of an action at a given time t to some extent, it is defined as: in and Represent the value functions of actions and states respectively; Finally, the mini-batch stochastic optimization algorithm is used to update the performer and critic parameters θ and φ respectively through gradient ascent and gradient descent as follows:

8. The method for improving industrial Internet performance based on intelligent blockchain sharding according to claim 7 is characterized in that: In step seven, the optimal action is selected based on the optimal strategy obtained in each state to obtain the maximum benefit until the execution of the instruction is completed.

Images