Blockchain-based collaborative incentive method in computing power network

Abstract

The present invention discloses a blockchain-based collaborative incentive method in a computing network, which belongs to the field of computing network; specifically: first, a dynamic scene is built, and rendering resources are purchased for user n from the collaborative node m' side for the request node m, and the respective utility functions are obtained by respectively calculating the costs and benefits of the request / collaboration nodes; then the block producer of the current time slot t is elected, and the maintenance cost of the blockchain is calculated; the utility functions of the request node and the collaborative node are used, combined with the maintenance cost of the blockchain, to construct the objective function of maximizing the self-benefit of the request node and the collaborative node; finally, the convex optimization theory and deep reinforcement learning theory are used to solve the objective function of maximizing the self-benefit of the collaborative / request node, respectively, and the solution results constitute the Nash equilibrium of the game, and the game theory is used to analyze the competitive relationship between nodes. The present invention analyzes the competitive relationship between nodes, encourages collaborators to participate in the collaborative calculation of tasks, and ensures the privacy of nodes.

Description

Technical Field

[0001] This invention belongs to the field of computing power networks, and specifically relates to a blockchain-based collaborative incentive method in computing power networks. Background Technology

[0002] In the era of cloud computing, business data was typically processed in centralized large or super-large cloud computing centers. For a long time, the powerful storage and computing capabilities of cloud computing met the various needs of traditional businesses. However, with the development of the 6G and artificial intelligence era, various new applications are constantly emerging, and the amount of data generated is growing explosively. This places extremely high demands on network latency, while also posing differentiated requirements for data security and controllability. To address these needs, edge computing has emerged.

[0003] The emergence of edge computing has, on the one hand, met the low-latency requirements of new services and solved the congestion problem caused by large amounts of data in the backbone network; on the other hand, it has led to the widespread distribution of multi-level computing resource nodes across the network, changing the flow of network traffic. How to achieve coordinated scheduling of multi-level resource nodes and flexible deployment of applications to provide users with a consistent service experience has become crucial. Therefore, computing power networks have emerged. Computing power networks connect a large number of idle resources and manage and schedule them uniformly, thereby providing users with a unified service. [1] .

[0004] However, achieving resource collaboration among computing nodes in a computing power network still faces some challenges: First, a computing power network is a distributed network, and most of the computing resources in the network are owned and deployed by different operators and organizations. There is serious competition for interests among computing power nodes. How to incentivize nodes to contribute their valuable computing resources and achieve collaborative scheduling of network resources is a challenge for computing power networks. In addition, some key information (such as payment transactions and collaborative decisions) may be attacked by malicious nodes during node collaboration. If this information is tampered with, it will seriously reduce the efficiency of node collaboration in the computing power network.

[0005] Considering the selfishness and competition among nodes, Reference [2] studied the collaborative service placement problem in MEC-based small cell networks and proposed an efficient distributed algorithm. This algorithm combines game theory to solve the selfish behavior of nodes when performing collaborative service placement. This algorithm can effectively improve the overall utility of the network. Reference [3] studied the task offloading problem in the Internet of Things. It developed a collaborative edge computing scheme and proposed a multi-objective evolutionary algorithm to minimize the total completion delay of tasks and the energy consumption of the system. Reference [4] studied the collaborative offloading incentive mechanism of tasks in mobile edge computing networks and proposed an online resource allocation algorithm. By jointly optimizing the selection of collaborators, the allocation of communication and computing resources and the pricing strategy, the system utility is maximized. Reference [5] designed a distributed resource management mechanism for mobile edge networks and designed a pricing scheme using the Stackelberg game model. By considering the trade-off between revenue and energy consumption, MEC operators can obtain the maximum utility.

[0006] However, several issues remain to be addressed in the aforementioned work: First, in edge collaboration, nodes are inherently self-interested, leading to intense competition. On one hand, requesters do not share their decisions (e.g., pricing strategies). On the other hand, the costs of collaborators are not readily available to requesters. Designing effective collaboration strategies for nodes without this information is challenging; however, most of the aforementioned articles assume this information is readily available, an assumption that does not reflect real-world scenarios. Second, the computing power network is an open network, and the articles neglect the security of inter-node collaborative transactions. Critical information during collaboration, such as collaboration decisions and pricing information, can be tampered with by malicious nodes, leading to a decline in system performance and user service quality.

[0007] As a top-level computing power trading platform, blockchain can realize the secure management of resources and record buying and selling, thereby ensuring the security of inter-node collaboration. A few studies have proposed blockchain-based solutions to enhance system security (e.g., references [6]-[9]), but these studies are not applicable to resource collaboration scenarios in computing power networks. In addition, maintaining a blockchain platform requires a large amount of computing resources and energy. These factors should be carefully considered when designing collaboration mechanisms.

[0008] [1] Lei Bo, Chen Yunqing. Edge computing and computing power networks: new computing power platforms and network connections in the 5G+AI era [M]. Beijing: Electronic Industry Press, 2020.

[0009] [2]L.Chen,C.Shen,P.Zhou,and J.Xu,“Collaborative service placement foredge computing in dense small cell networks,”IEEE Transactions on MobileComputing,vol.20,no.2,pp.377–390,2021.

[0010] [3]G.Peng,H.Wu,H.Wu,and K.Wolter,“Constrained multiobjectiveoptimization for iot-enabled computation offloading in collaborative edge andcloud computing,”IEEE Internet of Things Journal,vol.8,no.17,pp.13 723–13736,2021.

[0011] [4]G.Li and J.Cai,“An online incentive mechanism for collaborativetask offloading in mobile edge computing,”IEEE Transactions on WirelessCommunications,vol.19,no.1,pp.624–636,2020.

[0012] [5]J.Lee,D.Kim,and D.Niyato,“Market analysis of distributed learningresource management for internet of things:A game-theoretic approach,”IEEEInternet of Things Journal,vol.7,no.9,pp.8430–8439,2020.

[0013] [5]H.Yin,Z.Zhang,L.Zhu,M.Li,X.Du,M.Guizani,and B.Khoussainov,“Ablockchain-based storage system with financial incentives for load-balancing,”IEEE Transactions on Network Science and Engineering,vol.8,no.2,pp.1178–1188,2021.

[0014] [6]R.Zhu,H.Liu,L.Liu,X.Liu,W.Hu,and B.Yuan,“A blockchainbased two-stage secure spectrum intelligent sensing and sharing auction mechanism,”IEEETransactions on Industrial Informatics,vol.18,no.4,pp.2773–2783,2022.

[0015] [7]R.Zhao,L.T.Yang,D.Liu,X.Deng,and Y.Mo,“Atensor-based truthfulincentive mechanism for blockchain-enabled space-air-ground integratedvehicular crowdsensing,”IEEE Transactions on Intelligent TransportationSystems,vol.23,no.3,pp.2853–2862,2022.

[0016] [8]X.Ding,J.Guo,D.Li,and W.Wu,“An incentive mechanism for building asecure blockchain-based internet of things,”IEEE Transactions on NetworkScience and Engineering,vol.8,no.1,pp.477–487,2021.

[0017] [9] W.Sun, J.Liu, Y.Yue, and P.Wang, "Joint resource allocation and innovative design for blockchain-based mobile edge computing," IEEETransactions on Wireless Communications, vol.19, no.9, pp.6050–6064, 2020. Summary of the Invention

[0018] To address the security concerns and selfish behaviors among nodes in inter-node collaboration, this invention proposes a blockchain-based collaborative incentive method for computing power networks. Based on game theory, it analyzes the competitive relationships between nodes and employs reinforcement learning theory to design an efficient pricing strategy for resource requesters. This strategy enables dynamic price adjustment and incentivizes collaborators to participate in collaborative task computation. This strategy does not require knowledge of other nodes' bids and computational costs, thus ensuring node privacy. Furthermore, this invention designs an optimal resource sales strategy for collaborators in the network, maximizing their benefits from the network and further incentivizing their participation in collaborative task computation.

[0019] The blockchain-based collaborative incentive method in the aforementioned computing power network comprises the following specific steps:

[0020] Step 1: Build a dynamic scenario in the computing network, including request nodes, collaborating nodes, and users;

[0021] The time slot for the dynamic scene rendering service is defined as t = {1, 2, ..., T}. and Let be the set of requesting nodes and collaborating nodes for time slot t, respectively, and the corresponding sets of users accessing the requesting nodes and collaborating nodes are defined as follows: and

[0022] Step 2: Purchase rendering resources for user n from collaborating node m' for request node m, and calculate the revenue obtained by request node m and the cost of request node m.

[0023] The benefit that node m receives through collaborative rendering is:

[0024]

[0025] Let m be the maximum benefit that request node m can obtain from user n; Let n be the total amount of data that user n can obtain from the requesting node m. The amount of rendering data required by user n.

[0026] The cost of requesting node m is expressed as:

[0027]

[0028] p m,n,m' (t) represents the price paid by the requesting node m when it helps the assisting node n purchase resources from the collaborating node m'. and These are the allocated bandwidth resources and the bandwidth cost per unit of resources, respectively. The rendering resources that request node m purchases for user n from collaborating node m'.

[0029] Step 3: Calculate the utility function of the requesting node m using its own revenue and cost.

[0030] Step 4: Nodes in the network elect the block producer for the current time slot t and calculate the maintenance cost of the blockchain;

[0031] The formula for calculating maintenance delay is:

[0032]

[0033] The time it takes for block producers to collect transaction information from the network and package it into blocks; The latency for transmitting blocks to the network; To verify the block latency; y m' (t) represents the binary scalar value indicating whether the cooperating node m' in time slot t is the current block producer;

[0034] Step 5: For each collaborating node m', calculate its revenue and cost.

[0035] The reward of the collaborating node m' is expressed as:

[0036]

[0037] R b For revenue related to blockchain maintenance.

[0038] The cost of the collaborating node m' is expressed as:

[0039]

[0040] Where α m' Let m' be the CPU energy conversion efficiency of the cooperating node. The unit energy consumption cost of the cooperating node m';

[0041] Step 6: Calculate the utility function of the collaborating node m' using its own revenue and cost.

[0042] Step 7: Using the utility functions of the requesting node m and the collaborating node m' respectively, and combining them with the maintenance cost of the blockchain, construct an objective function to maximize the self-revenue of the requesting node and the collaborating node.

[0043] The objective function for requesting node m to maximize its own profit is expressed as:

[0044]

[0045] stC9-1:p m,n,m' (t)>0

[0046] C9-2:

[0047] C9-3:

[0048] C9-4:

[0049] p represents the set of payment prices for the resources that requesting node m needs to determine. b col (t) represents the set of transmission bandwidth purchased from the collaborating node side.

[0050] Constraints C9-1 and C9-2 state that price and bandwidth allocation decisions should be positive, and C9-3 states that the total bandwidth allocated to users cannot exceed the node's remaining resource capacity.

[0051] C9-4 indicates that the amount of data rendered by a node should not exceed the user's maximum requirement.

[0052] The objective function for maximizing the self-reward of collaborating node m' is expressed as:

[0053]

[0054] stC10-1:

[0055] C10-2:

[0056] C10-3:

[0057] C10-4:

[0058] f col (t) represents the collaborative rendering resource of the collaborating node m'.

[0059] C10-1 and C10-2 indicate that collaborative resource allocation and blockchain maintenance resource decisions should be positive values, and C10-3 indicates that the total amount of all resource allocations cannot exceed the resource capacity of a node. C10-4 guarantees the performance of the blockchain, stating that its latency should be less than the duration of the current moment.

[0060] Step 8: Solve the objective function of maximizing the self-revenue of the cooperating node m' using convex optimization theory.

[0061] Step 9: Solve the objective function of maximizing the self-reward of the request node m using deep reinforcement learning theory.

[0062] Step 10: Construct the Nash equilibrium of the game by combining the policy sets of the requesting nodes and cooperating nodes obtained from the solution, and use game theory to analyze the competitive relationship between the nodes.

[0063] The advantages of this invention are:

[0064] 1) The blockchain-based collaborative incentive method in the computing power network of the present invention provides a secure computing power collaboration scheme for computing power networks. It uses blockchain technology to realize transaction records in the network and can ensure the security of computing power network collaboration in real time.

[0065] 2) The blockchain-based collaborative incentive method in the computing power network of this invention designs a pricing scheme based on reinforcement learning theory, which can effectively protect the privacy information of other nodes and realize real-time price decision-making to purchase suitable resources from the computing network;

[0066] 3) The blockchain-based collaborative incentive method in the computing power network of this invention designs an optimal selling strategy based on convex optimization theory, which can determine the optimal selling scheme for nodes with sufficient resources in the computing network. Attached Figure Description

[0067] Figure 1 This is a flowchart of the blockchain-based collaborative incentive method in the computing power network of the present invention;

[0068] Figure 2 This is a schematic diagram of a dynamic scene built in a computing power network in an embodiment of the present invention;

[0069] Figure 3 This is a comparison chart of system benefits under different schemes compared to existing algorithms;

[0070] Figure 4 This is a schematic diagram comparing the performance of the present invention with existing collaborative algorithm schemes;

[0071] Figure 5The results show the system optimization of this invention under different numbers of users. Detailed Implementation

[0072] The present invention will be further described below with reference to the accompanying drawings and examples.

[0073] This invention proposes a blockchain-based collaborative incentive method for computing power networks. To address the security of inter-node collaboration, blockchain technology is applied to the computing power network. As a top-level information management platform, the blockchain can ensure the security of resource transactions in real time. Furthermore, considering blockchain performance, this invention optimizes resource allocation to improve blockchain performance. Considering the competitive relationships between nodes in the network, game theory is used to analyze the competitive relationships between different nodes (e.g., requesting nodes and collaborating nodes). On one hand, a pricing strategy is designed for resource requesters based on reinforcement learning, thereby incentivizing collaborators to provide collaborative computing services. This strategy does not require knowledge of the bids and computing costs of other nodes in the network, thus ensuring node privacy. On the other hand, based on convex optimization theory, an optimal selling strategy is formulated for resource-sufficient collaborators. This strategy allows collaborators to obtain the maximum benefit from the network, incentivizing them to contribute their resources.

[0074] The blockchain-based collaborative incentive method in the aforementioned computing power network, such as... Figure 1 As shown, the specific steps are as follows:

[0075] Step 1: Build a dynamic scenario in the computing network, including request nodes, collaborating nodes, and users;

[0076] Requesting nodes are resource-scarce nodes that need to determine the price to purchase resources from the network; collaborating nodes are resource-sufficient nodes that can sell their resources to generate revenue.

[0077] The time slot for the dynamic scene rendering service is defined as t = {1, 2, ..., T}. and Let be the set of requesting nodes and collaborating nodes for time slot t, respectively, and the corresponding sets of users accessing the requesting nodes and collaborating nodes are defined as follows: and

[0078] Step 2: Purchase rendering resources for user n from collaborating node m' for request node m, and calculate the revenue obtained by request node m and the cost of request node m.

[0079] First, for the current rendering time slot t, calculate the amount of rendering data that request node m can provide for user n:

[0080]

[0081] Let η be the rendering resources purchased by requesting node m from collaborating node m' for user n, 'a' be the video rendering conversion rate, and η be the rendering resources purchased by requesting node m from collaborating node m'. m' Δt represents the amount of data rendered in one CPU cycle of the collaborating node m', and Δt represents the time window for data rendering.

[0082] Then, combining the downlink transmission, calculate the total amount of data that user n can obtain from the requesting node m:

[0083]

[0084] in For downlink transmission rate, For time slot t, request node m to allocate bandwidth resources to user n. Let be the signal-to-interference-plus-noise ratio (SIR) for time slot t.

[0085] Finally, the benefit obtained by request node m through collaborative rendering is calculated as follows:

[0086]

[0087] Let m be the maximum revenue that node m can obtain from user n. The amount of rendering data required by user n.

[0088] The cost of requesting node m is expressed as:

[0089]

[0090] p m,n,m′ (t) represents the price paid by the requesting node m when it helps the assisting node n purchase resources from the collaborating node m'. This refers to the bandwidth cost per unit of resource.

[0091] Step 3: Calculate the utility function of the requesting node m using its own revenue and cost.

[0092] Represented as:

[0093]

[0094] Step 4: Nodes in the network elect the block producer for the current time slot t and calculate the maintenance cost of the blockchain;

[0095] First, the nodes in the network elect the block producer at the current moment based on the node's resource capabilities; the node's resource capabilities are related to the size of its resources at the current moment.

[0096] For a collaborating node m', its resource capabilities are used to determine whether the node is a block producer;

[0097] The formula for calculating resource capacity is:

[0098]

[0099] in Let m' be the resource size of the collaborating node m' in time slot t;

[0100] Using scalar y m' (t) indicates whether the cooperating node m' in time slot t is the current block producer, then we have

[0101]

[0102] Then, the block producer collects transaction information in the network and packages it into blocks, which are then published to the network for verification; the time for packaging the blocks, the latency for transmitting the blocks, and the latency for verifying the blocks are calculated respectively.

[0103] The formula for calculating the time it takes for a node to package a block is:

[0104] S is the computing resources invested by the collaborating node m' when packaging blocks. b (t) represents the block size, C b The number of cycles required to generate a 1-bit block.

[0105] Once a block producer completes a block, the block will be further transmitted to the network for verification; the formula for calculating the transmission latency of a block is:

[0106] The transmission rate between cooperating node m' and node u;

[0107] The formula for calculating the block verification latency is:

[0108] The computational resources required to verify a block for node u. The number of cycles required for block verification;

[0109] Finally, by using the time for packaging blocks, the latency for transmitting blocks, and the latency for verifying blocks, the overall maintenance latency for blockchain transaction records is calculated.

[0110] The formula for calculating maintenance delay is:

[0111]

[0112] Step 5: For each collaborating node m', calculate its revenue and cost.

[0113] The reward of the collaborating node m' is expressed as:

[0114]

[0115] R b For revenue related to blockchain maintenance.

[0116] The cost of the collaborating node m' is expressed as:

[0117]

[0118] Where α m' Let m' be the CPU energy conversion efficiency of the cooperating node. The unit energy consumption cost of the cooperating node m';

[0119] Step 6: Calculate the utility function of the collaborating node m' using its own revenue and cost.

[0120] Finally, the utility function for the collaborating nodes can be given:

[0121]

[0122] Step 7: Using the utility functions of the requesting node m and the collaborating node m' respectively, and combining them with the maintenance cost of the blockchain, construct an objective function to maximize the self-revenue of the requesting node and the collaborating node.

[0123] The objective function for requesting node m to maximize its own profit is expressed as:

[0124]

[0125] stC9-1:p m,n,m' (t)>0

[0126] C9-2:

[0127] C9-3:

[0128] C9-4:

[0129] p represents the set of payment prices for the resources that requesting node m needs to determine. b col (t) represents the set of transmission bandwidth purchased from the collaborating node side.

[0130] Constraints C9-1 and C9-2 state that price and bandwidth allocation decisions should be positive, and C9-3 states that the total bandwidth allocated to users cannot exceed the node's remaining resource capacity.

[0131] C9-4 indicates that the amount of data rendered by a node should not exceed the user's maximum requirement.

[0132] The objective function for maximizing the self-reward of collaborating node m' is expressed as:

[0133]

[0134] stC10-1:

[0135] C10-2:

[0136] C10-3:

[0137] C10-4:

[0138] f col (t) represents the collaborative rendering resource of the collaborating node m'.

[0139] C10-1 and C10-2 indicate that collaborative resource allocation and blockchain maintenance resource decisions should be positive values, and C10-3 indicates that the total amount of all resource allocations cannot exceed the resource capacity of a node. C10-4 guarantees the performance of the blockchain, stating that its latency should be less than the duration of the current moment.

[0140] Step 8: Solve the objective function of maximizing the self-revenue of the cooperating node m' using convex optimization theory.

[0141] First, we present the Lagrange formula for the collaborative node optimization problem:

[0142]

[0143] Among them, the definition u m' (t), λ m' (t), It is a non-negative Lagrange multiplier.

[0144] Then, based on the KKT conditions, the optimal selling scheme for the collaborating nodes is expressed as:

[0145]

[0146] Next, based on the following formula, the Lagrange multipliers are updated using the subgradient optimization method.

[0147]

[0148] in and The step size can be set to 0.01.

[0149] Next, we will discuss the constructed Lagrange formula regarding... Taking the partial derivative and applying the KKT conditions, the optimal resource decision is expressed as:

[0150]

[0151] in

[0152] Step 9: Solve the objective function of maximizing the self-reward of the request node m using deep reinforcement learning theory.

[0153] First, the state space and action space of the reinforcement learning algorithm are given. The state space is defined as follows:

[0154]

[0155] The action space mainly relates to the optimization variables of the requesting node, such as bandwidth allocation decisions and resource sales decisions, and is defined as:

[0156]

[0157] Then, the system's reward is given, which is the utility function of the requesting node.

[0158] Finally, a deep reinforcement learning environment is configured, and the policy set for the requesting nodes is obtained through training.

[0159] Step 10: Construct the Nash equilibrium of the game by combining the policy sets of the requesting nodes and cooperating nodes obtained from the solution, and use game theory to analyze the competitive relationship between the nodes.

[0160] a) For the requesting node, given the optimal price p of other requesting nodes. -m (t) * At that time, the Nash equilibrium point of the collaborative nodes is and

[0161] b): For collaborating nodes, the optimal price strategy p for a given requesting node. m (t) * and optimal blockchain resource allocation strategy At that time, the optimal selling strategy for collaborative nodes is

[0162] Example:

[0163] like Figure 2 The diagram shows the blockchain-based collaborative architecture in the computing network built in this embodiment, where the blockchain is deployed as a trusted management platform for resource publishing and transactions. Taking VR services as an example, the overall architecture is described below:

[0164] Users can experience immersive VR videos using VR devices, assuming all requested VR videos are pre-cached in the MEC system. When a user requests a VR service, the MEC needs to provide computing resources to help the user render video chunks. The rendered video chunks are then transmitted to the user's side via a downlink. Considering the limited computing resources of individual nodes in the network, this embodiment implements distributed collaboration between nodes through a price incentive mechanism. Different types of nodes are considered in the computing network, such as requesting nodes and collaborating nodes. Furthermore, requesting nodes need to determine the purchase price and obtain resources from collaborating nodes to help the user complete the rendering process, thereby maximizing their profits. Collaborating nodes need to determine an optimal sales plan to obtain the greatest benefit from the market.

[0165] First, a dynamic scene is built in the system. For the rendering time slot t, in the computing power network, the requesting node needs to determine the optimal price to obtain resources from the collaborating node to help the user complete the rendering task, with the aim of obtaining revenue.

[0166] For the rendering resources purchased by requesting node m from collaborating node m' for user n, the benefit obtained by requesting node m through collaborative rendering is:

[0167]

[0168] Notice, Let m be the maximum revenue that node m can obtain from user n after local optimization. The amount of rendering data required by user n after the local optimization decision of request node m is known.

[0169] The cost of requesting a node is expressed as follows:

[0170] The utility function for the final requested node is expressed as:

[0171]

[0172] The utility functions of collaborative nodes mainly include: 1) collaboration benefits; 2) blockchain maintenance benefits; 3) collaboration rendering costs; and 4) blockchain maintenance costs.

[0173] Given the ample computing resources of the collaborating nodes, blockchain maintenance should be performed on the collaborating node side. Define y. m' (t) is a binary variable representing whether the cooperating node m' in time slot t has been elected as the current block producer; y m' (t) = 1 indicates that the cooperating node m' is the block producer at this moment;

[0174] The reward of the collaborating node m' is expressed as:

[0175] The cost of the collaborating node m' is expressed as:

[0176]

[0177] This embodiment considers two different types of transactions, such as request transactions and collaborative transactions. Therefore, the block size can be represented as:

[0178]

[0179] In the formula, K represents the size of a transaction. Represents a set Size.

[0180] Finally, the utility function for the collaborating nodes can be given:

[0181]

[0182] In a computing network, blockchain is used to record and publish important information. Blockchain requires two steps to process transactions: block generation and block consensus. Block producers need to collect transaction information from the network and package it into blocks, calculating the time it takes for a node to package a block.

[0183] Once a block producer completes a block, it further distributes the block to the network for verification. Assuming wired transmission between nodes, calculate the transmission latency of a block in the network and the verification latency required to verify a block.

[0184] Calculate the overall maintenance latency of the blockchain for recording transactions by using the time for packaging blocks, the latency for transmitting blocks, and the latency for verifying blocks:

[0185]

[0186] Nodes in the network need to elect the block producer at the current moment based on the node's resource capabilities;

[0187] The resource capabilities of a node are represented as follows:

[0188]

[0189] Therefore, we can conclude that...

[0190]

[0191] Requesting node m needs to determine the resource payment price. and transmission bandwidth The problem of maximizing one's own benefit by purchasing resources from collaborating node m′ can be represented as:

[0192]

[0193] For a collaborating node m', the goal is to determine the optimal collaborative rendering resources:

[0194]

[0195] And blockchain maintenance resources To determine how to maximize one's own benefit, the problem can be formulated as follows:

[0196]

[0197] Since the nodes are deployed and managed by different merchants and organizations, there is competition for interests among the nodes. Next, game theory will be used to analyze the competitive relationship between the nodes.

[0198] The policy sets of all requesting and cooperating nodes in the network obtained by solving the problem constitute the Nash equilibrium of the game.

[0199] In the network, collaborating nodes need to determine the optimal resource to sell based on the bids of requesting nodes. There is no competition among collaborating nodes; they can make the optimal decision solely based on the bids of requesting nodes.

[0200] The Lagrange expression for problem (10) is given below:

[0201]

[0202] To obtain the optimal resource selling decision, we first need to consider formula (11) regarding... Find the partial derivative, and it can be expressed as Based on the KKT conditions, the following conditions can be derived. Therefore, the optimal selling strategy for the collaborative nodes is expressed as:

[0203]

[0204] Next, based on the following formula, the Lagrange multipliers are updated using the subgradient optimization method.

[0205]

[0206] Next, we will discuss formula (11) regarding... Taking the partial derivative, we can obtain

[0207]

[0208] Based on the KKT conditions, the following conclusions can be drawn:

[0209]

[0210] in

[0211] because u can be obtained m' If (t) = 0, then It can be concluded that This is the same as λ m' This contradicts the statement that (t)≥0, therefore we can conclude that... because Therefore there is

[0212] Correspondingly, the optimal resource decision in blockchain can be expressed as:

[0213]

[0214] Next, we need to find the optimal pricing strategy for the requesting nodes. In the computing power network, there is competition among the requesting nodes.

[0215] Theory 1: A Nash equilibrium exists in a non-cooperative game between requesting nodes if the following conditions are met:

[0216] 1. The player pool is finite.

[0217] 2. The policy set is closed, bounded, and convex.

[0218] 3. The utility function is continuous and quasi-concave in the policy space.

[0219] Corollary 1: Among the requesting nodes, the competition between nodes exists in a Nash equilibrium.

[0220] Proof: Definition Let be the set of intermediate variables. For problem P1, there exists the following equivalent problem:

[0221]

[0222] Obviously, problem (17) is a convex optimization problem. Combining theory 1, it can be proved that there is a Nash equilibrium among the requesting nodes.

[0223] Due to privacy concerns, requesting nodes cannot know the pricing strategies of other nodes or the computational costs of collaborating nodes. Therefore, traditional optimization schemes cannot solve the pricing strategies of requesting nodes. Thus, this embodiment uses a deep reinforcement learning algorithm to explore the optimal price and uses a Markov game model to describe the pricing problem of requesting nodes.

[0224] A Markov game model can be defined as {M, S, A, P, R}, where M is the number of requesting nodes. For state space, Let P be the action space, and P be the state transition probability. This is a system report.

[0225] At time t, request node m to observe the state. And select an action based on the policy network. The node will then receive a reward from the environment. corresponding state It will be based on the state transition probability at the next moment. Transition to state For each request node, the goal is to maximize long-term returns. Among them, γ is the discount factor.

[0226] The algorithm's state space and action space are then presented. The state space mainly includes two factors: the quality of the wireless channel and the amount of rendering data provided by the node to the user. The state space can be defined as follows:

[0227]

[0228] The action space is mainly related to the optimization variables of the requesting node, such as bandwidth allocation decisions and resource sales decisions, and can be defined as follows:

[0229]

[0230] The system returns the utility function of the requesting node. Next, the Proximal Policy Optimization (PPO) algorithm is used to solve the problem. PPO is a classic reinforcement learning algorithm that uses the Actor-Critic framework. The Actor network is responsible for generating actions, and the Critic network is responsible for evaluating the quality of the actions.

[0231] The Actor network updates its parameters based on the policy gradient, and can be represented as:

[0232]

[0233] in, and These are the Actor target network and the Actor behavior network, respectively. For importance sampling probability, and

[0234] By pruning the policy gradient, the oscillation problem can be improved during training; therefore, the improved policy gradient can be expressed as: in

[0235]

[0236] The Critic network improves itself by optimizing the loss function, which can be expressed as:

[0237]

[0238] The blockchain-based collaborative incentive method proposed in this invention solves the security problem of collaboration in computing power networks. Considering the selfishness of nodes in computing power networks, a price strategy based on reinforcement learning is designed to protect the privacy of nodes and support dynamic price decisions. In addition, an optimal resource selling algorithm is designed for nodes with sufficient resources. Based on convex optimization theory, it fully considers the bids of different nodes and their own remaining resources. With the support of the price strategy and the selling strategy, nodes in the computing power network can obtain the maximum benefit from the network and improve the QoS of users.

[0239] like Figure 3 As shown, this paper compares the performance of the proposed solution with four other existing solutions. First, the four solutions are introduced: the unified bandwidth resource allocation scheme is essentially the same as the proposed solution, except that it uses a uniform allocation strategy; the unified computing resource allocation scheme is essentially the same as the proposed solution, except that it uses a uniform allocation strategy; the single purchase strategy is essentially the same as the proposed solution, except that the requesting node can only choose one collaborating node in the market to purchase resources; the random strategy does not consider resource optimization, and the user's bandwidth and computing resources are randomly allocated. As shown in the figure, the revenue of the requesting node increases with the number of users. However, compared with the unified computing resource allocation scheme and the unified bandwidth resource allocation scheme, the proposed solution always achieves the highest revenue. This is because the proposed solution designs the optimal resource management scheme by jointly optimizing bandwidth and computing resource allocation. Furthermore, the single purchase strategy has poor performance because it only allows the requesting node to purchase resources from one collaborating node in the market, which limits the choice of collaborators on the requesting node side and indirectly increases its costs.

[0240] like Figure 4As shown, this diagram compares the performance of the proposed solution with a non-cooperative solution. The non-cooperative solution does not consider node collaboration. For a given total bandwidth capacity, it can be observed that when node resource capacity is insufficient, the average revenue of the proposed solution is higher than the comparison solution. This verifies that the performance of the proposed cooperative solution is superior to the non-cooperative solution. However, when computing resources are sufficient, the performance of the proposed solution is the same as the non-cooperative solution. This is because when resource capacity is sufficient, nodes in the network do not need to purchase resources. Therefore, the strategy proposed in this invention is only applicable when resources are insufficient. Furthermore, the average revenue of requesting nodes increases with increasing bandwidth capacity. This is because higher bandwidth capacity allows requesting nodes to transmit more rendering data to users, thereby increasing the revenue of requesting nodes.

[0241] like Figure 5 As shown, the system optimization results of the proposed solution under different numbers of users are presented. The node resource capacity is set from 5GHz to 7GHz. From... Figure 5 a and Figure 5 As can be observed in b, the total price and total purchased resources increase with the number of users. Furthermore, as node resource capacity increases (from 5GHz to 7GHz), the total purchased resources and the corresponding payment price for the node decrease. This is because the higher the computing resources required by a node, the higher the bid and the corresponding resources purchased. Figure 5 In node c, the average revenue of a node increases with its resource capabilities. It can also be seen that the average revenue increases as the number of users increases. In fact, the more users there are, the higher the revenue a node can obtain from the user side.

Claims

1. Blockchain-based collaborative incentive method in computing power network, It is characterized in that The specific steps are as follows: Step 1: Build a dynamic scenario including request nodes, collaboration nodes and users in the computing network; The time slot of the dynamic scene rendering service is t = {1, 2, ..., T}, and the definition and are the sets of requesting nodes and cooperating nodes in time slot t, respectively. The corresponding sets of users accessing the requesting nodes and cooperating nodes are defined as and Step 2: Requesting node m purchases rendering resources for user n from collaborative node m', calculates the revenue and cost of requesting node m, and further obtains the utility function of the node Step 3: Similarly, for the collaborative node m', calculate the benefits and costs of the node respectively, and further obtain the utility function of the node Step 4: Nodes in the network elect the block producer for the current time slot t and calculate the maintenance cost of the blockchain; The calculation formula for maintenance delay is: The time for block producers to collect transaction information in the network and package it into blocks; is the latency of transmitting blocks to the network; Delay for verifying blocks; m' (t) a binary scalar indicating whether the cooperating node m' is the current block producer in time slot t; Step 5: Use the utility functions of the requesting node m and the cooperating node m' respectively, combined with the maintenance cost of the blockchain, to construct the objective function of maximizing the self-benefit of the requesting node and the cooperating node; The objective function of requesting node m to maximize its own benefits is expressed as: P 1 s.t.C9-1:p m,n,m' (t)>0 p is the set of payment prices for resources that the requesting node m needs to determine, p m,n,m' (t) is the price paid by the requesting node m to purchase resources from the cooperating node m' to help the node n; b col (t) is the transmission bandwidth set of resources purchased from the cooperative node side, Request node m to allocate bandwidth resources to user n for time slot t; is the maximum benefit that the requesting node m can obtain from user n, is the total amount of data that user n can obtain from requesting node m; is the amount of rendering data required by user n; Δt is the time window for data rendering; is the bandwidth cost per unit resource; The rendering resources purchased by the requesting node m for the user n from the cooperating node m'; Constraints C9-1 and C9-2 indicate that price and bandwidth allocation decisions should be positive values, and C9-3 indicates that the total bandwidth allocated to users cannot exceed the remaining resource capacity of the node. C9-4 indicates that the amount of data rendered by a node should not exceed the maximum demand of the user; The objective function of the collaborative node m' to maximize its own benefits is expressed as: P 2 f col (t) represents the collaborative rendering resources of collaborative node m', The computing resources invested by the collaborative node m' when packaging blocks; R b For the income related to blockchain maintenance; α m' is the CPU energy conversion efficiency of the cooperative node m', is the unit energy consumption cost of the cooperative node m'; S b (t) is the block size, C b The number of cycles required to generate a 1-bit block; C10-1 and C10-2 indicate that collaborative resource allocation and blockchain maintenance resource decisions should be positive values, and C10-3 indicates that the sum of all resource allocations cannot exceed the resource capacity of the node. C10-4 ensures the performance of the blockchain, and its delay should be less than the duration of this moment; Step 6: Use convex optimization theory and deep reinforcement learning theory to solve the objective function of maximizing the benefits of the collaboration node m' and the request node m respectively; Step 7: The policy sets of the requesting nodes and the cooperating nodes obtained are used to form the Nash equilibrium of the game, and the game theory is used to analyze the competitive relationship between the nodes.

2. The blockchain-based collaborative incentive method in the computing power network as claimed in claim 1, It is characterized in that The requesting node is a node that is short of resources and needs to determine a price to purchase resources from the network; the cooperating node is a node that has sufficient resources and sells its own resources to gain benefits.

3. The blockchain-based collaborative incentive method in the computing network as claimed in claim 1, It is characterized in that The step 2 is specifically as follows: First, for the current rendering time slot t, calculate the amount of rendering data that the requesting node m can provide for user n: a is the video rendering conversion rate, η m' The amount of data rendered for one CPU cycle of the cooperating node m'; Then, combined with the downlink transmission, the total amount of data that user n can obtain from the requesting node m is calculated: in is the downlink transmission rate, is the signal-to-interference-noise ratio of time slot t; Finally, the benefit obtained by requesting node m through collaborative rendering is calculated as: The cost of requesting node m is expressed as: Requesting node's utility function It is expressed as:

4. The blockchain-based collaborative incentive method in the computing power network as claimed in claim 1, It is characterized in that The specific step three is: the benefit of the cooperation node m' is expressed as: The cost of the cooperating node m' is expressed as: Utility function of collaborative nodes It is expressed as:

5. The blockchain-based collaborative incentive method in the computing power network as claimed in claim 1, It is characterized in that The step 4 is specifically as follows: First, for the collaborative node m', its resource capabilities are used to determine whether the node is a block producer; The resource capacity calculation formula is: in is the resource size of the cooperative node m' in time slot t; With scalar y m' (t) indicates whether the collaborative node m' in time slot t is the current block producer, then Then, the block producer collects the transaction information in the network and packages it into blocks, which are published to the network for verification; Calculate the time for packing blocks, the delay for transmitting blocks, and the delay for verifying blocks respectively; The time calculation formula for packing blocks is: The delay calculation formula for transmitting a block is: is the transmission rate between the cooperative node m' and the node u; The delay calculation formula for verifying a block is: The computing resources required for node u to verify a block, The number of cycles required for block verification.

6. The blockchain-based collaborative incentive method in the computing power network as claimed in claim 1, It is characterized in that The convex optimization theory is used to solve the objective function of maximizing the self-benefit of the collaborative node m', specifically: First, the Lagrangian equation of the cooperative node optimization problem is given: Among them, the definition u m' (t), λ m' (t), is a non-negative Lagrange multiplier; Then, according to the KKT condition, the optimal selling solution of the cooperative node is expressed as: Next, the Lagrange multiplier is updated using the subgradient optimization method according to the following formula: in and for different step lengths; Next, the constructed Lagrangian formula is Find the partial derivative and, based on the KKT condition, get the optimal resource decision as: in 7. The blockchain-based collaborative incentive method in the computing power network as claimed in claim 1, It is characterized in that The deep reinforcement learning theory is used to solve the objective function of maximizing the self-benefit of the requesting node m, specifically: First, the state space and action space of the reinforcement learning algorithm are given. The state space is defined as: The action space is defined as: Then, the system reward is given, which is the utility function of the requesting node Finally, the deep reinforcement learning environment is configured and the strategy set of the requesting node is obtained through training.

8. The blockchain-based collaborative incentive method in the computing power network as claimed in claim 1, It is characterized in that The game theory is used to analyze the competition relationship between nodes, specifically: a): For a requesting node, the optimal price p of other requesting nodes -m (t) * When , the Nash equilibrium point of the cooperative node is and b): For collaborative nodes, the optimal price strategy p for a given request node m (t) * And the optimal blockchain resource allocation strategy When , the optimal selling strategy of the cooperative node is

Images