A method for dynamic entry and exit mechanism based on blockchain cloud computing services

Abstract

This invention discloses a dynamic entry and exit mechanism method based on blockchain-based cloud computing services, belonging to the field of cloud computing technology. The method includes: establishing a dynamic task offloading and resource allocation system model, including: a local execution model, an edge server computing model, and a communication model; analyzing the constraints of the dynamic task offloading and resource allocation problem and the formulation of the system problem; and designing a dynamic task offloading and resource allocation strategy based on an entry and exit mechanism combining blockchain and cryptography. This invention employs the aforementioned dynamic entry and exit mechanism method based on blockchain-based cloud computing services, considering the randomness of task arrival and the potential variability of the number, location, and resource supply of cloud computing service providers. It can learn and optimize its performance according to changes in environmental conditions, minimize system latency and energy consumption through the dynamic Dueling DDQN algorithm, and can perceive dynamically changing environments, providing more flexible and stable performance.

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, and in particular to a dynamic entry and exit mechanism method for blockchain-based cloud computing services. Background Technology

[0002] With the development of cloud computing and the Internet of Things (IoT), edge computing IoT networks (ECINs) are an emerging computing architecture that plays a crucial role in various fields. By deploying edge nodes on IoT devices, ECINs enable distributed computing for data processing and task execution. Compared to traditional cloud computing, ECINs offer lower latency, higher real-time capabilities, lower network load, lower bandwidth requirements, and enhanced data privacy and security, driving innovation and development in IoT applications. However, with the rapid proliferation and increasing intelligence of IoT devices, the amount of data generated is growing dramatically. Such a massive number of devices generates enormous amounts of data, and processing this data solely using cloud servers will lead to severe latency and energy consumption, presenting challenges such as ubiquitous communication and computing resource demands, as well as data privacy issues.

[0003] Blockchain-based Cloud Computing Internet of Things (BECIN) has emerged as a promising solution for providing secure and fast communication and computing services due to its flexible deployment, high security, and ease of scalability. However, existing research on compute offloading in cloud computing largely ignores the random arrival of computing tasks and the potential variability in the number, location, and resource availability of cloud computing service providers. Therefore, a dynamic, self-regulating BECIN framework is needed to provide long-term, stable, efficient, and secure cloud computing data offloading services for terrestrial users in specific regions, supporting the dynamic entry and exit of cloud computing service providers. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic entry and exit mechanism method based on blockchain cloud computing services. It takes into account the randomness of task arrival and the potential variability of the number, location and resource supply of cloud computing service providers. It can learn and optimize its performance according to changes in environmental conditions, minimize system latency and energy consumption through the dynamic Dueling DDQN algorithm, and can sense the dynamically changing environment to provide more flexible and stable performance.

[0005] To achieve the above objectives, this invention provides a dynamic entry and exit mechanism method based on blockchain cloud computing services, comprising the following steps:

[0006] S1. Establish a dynamic task unloading and resource allocation system model, including: local execution model, edge server computing model and communication model;

[0007] S2. Analyze the constraints of dynamic task unloading and resource allocation problems, as well as the description of system problems;

[0008] S3. Based on the combination of blockchain and cryptography, a dynamic task unloading and resource allocation strategy is designed.

[0009] Preferably, in step S1, the system model is established, and the local execution model is implemented as follows:

[0010] S111, Local Delay The calculation formula is as follows:

[0011] ;

[0012] in, Represents the MSD set. ; Representing the BECs set, ; For the device MSD n In the time slot Computational power; CPU cycle frequency; Indicates time slot From MSD n To BEC m The proportion of computing tasks that are unloaded; Each time slot The computational tasks generated in the process;

[0013] S112, Local Execution Capability The calculation formula is as follows:

[0014] ;

[0015] in, It is a constant power coefficient;

[0016] S113, Local Energy Consumption The calculation formula is as follows:

[0017] .

[0018] Preferably, in step S1, the computational model of the edge server in establishing the system model is as follows:

[0019] S121, Time Slot From edge server BEC m To MSD n The allocation of computing resources is as follows:

[0020] ;

[0021] in, Allocate computing resources for the whole; The resource allocation coefficient represents the time slot. BEC from BEC server m To MSD device MSD n Resource allocation, subject to ∈P c,m (t) constraint; For time slots BEC m The set of MSDs within the wireless coverage area;

[0022] S122, Time Slot BEC m Total computational task latency As shown below:

[0023] ;

[0024] in, BEC m Computational capabilities when processing incoming tasks;

[0025] S123, BEC m In the time slot The energy consumption is as follows:

[0026] ;

[0027] in, BEC power consumption for executing one CPU / GPU cycle.

[0028] Preferably, in step S1, the communication model established in the system model is as follows:

[0029] S131, in time slot From the edge server BEC m To MSD n The allocated bandwidth is:

[0030] ;

[0031] Overall communication bandwidth resource allocation As shown below:

[0032] ;

[0033] in, =0,1 ∈ , ∈M; when =1 indicates MSD n AND sub-channel Related, when =0 indicates no association;

[0034] S132, from arrive Channel power gain As shown below:

[0035] ;

[0036] in, Indicates the path loss coefficient. This represents the path loss index. This represents the channel power gain caused by small-scale fading. express Location and Location In the time slot The distance between them;

[0037] S133, Sub-channel From MSD on time slot t n To BEC m The signal-to-noise ratio (SINR) is shown below:

[0038] ;

[0039] in, MSD n Transmission power, Represents the noise power spectral density. This indicates that in time slot t, when BEC m For MSD n Inter-cell interference when allocating sub-channels for transmission tasks;

[0040] S134, in time slot BEC m and MSD n The transmission rates between them are as follows:

[0041] ;

[0042] MSD n To BEC m The transmission delay between them is as follows:

[0043] ;

[0044] S135, Time Slot From MSD nTransmitted to BEC m The energy consumption is as follows:

[0045] .

[0046] Preferably, in step S2, the constraints of the dynamic task unloading and resource allocation problem are as follows:

[0047] S211, The unloading decision constraint is:

[0048] ;

[0049] in, MSD n In the time slot Unloading decision at that time;

[0050] S212, The local processor CPU cycle frequency constraint is:

[0051] ;

[0052] in, Indicates the CPU cycle frequency of the local processor;

[0053] S213, The computing resource constraints of the edge processor are:

[0054] ;

[0055] in, This refers to BEC m MSD allocated within the coverage area n Computing resources;

[0056] S214, Bandwidth constraint is:

[0057] ;

[0058] in, This refers to BEC m The coverage area is allocated to MSD n Communication resources;

[0059] S215, The wireless channel power constraint is:

[0060] ;

[0061] in, MSD n exist The wireless channel power of the time slot, MSD n exist The maximum permissible wireless channel power during time-slotted wireless communication;

[0062] S216. Sub-channel constraints, as detailed below:

[0063] Each subchannel is assigned to only one MSD, satisfying the following constraints:

[0064] ;

[0065] Each MSD is assigned to a single subchannel, which obviously satisfies the following condition:

[0066] ;

[0067] in, This indicates the bandwidth of the BEC server in time slot t. m To MSD device MSD n The allocation coefficient.

[0068] Preferably, in step S2, based on constraints, the task unloading and resource allocation problems are combined as follows:

[0069] S221, Given time slot The task is performed on the local processor MSD. n and edge server BEC m The total latency for transmission and computation is as follows:

[0070] ;

[0071] S222, Given time slot Local Processor MSD n and edge server BEC m The total energy consumption for task transmission and computation is as follows:

[0072] ;

[0073] S223. Based on total delay and total energy consumption, the optimization objectives are as follows:

[0074] ;

[0075] S224. The BECIN problem, a blockchain-based cloud computing IoT network, is as follows:

[0076] .

[0077] Preferably, in step S3, a dynamic task unloading and resource allocation strategy is designed, and the specific steps are as follows:

[0078] S31. Introduce a join / exit mechanism to update service strategies based on dynamic task arrivals and resource changes;

[0079] S32. The Duel DDQN algorithm is adopted to dynamically update the task unloading and resource allocation strategy in a dynamic environment based on the real-time arrival of tasks and resource changes, so as to achieve the optimal task unloading and resource allocation strategy.

[0080] Preferably, in step S31, a dynamic task unloading and resource allocation strategy is designed based on the joining and leaving mechanism combining blockchain and cryptography, as shown below:

[0081] S311, System Initialization: Let D be the registration authority of the BEC system, responsible for system registration and exit management, including key generation, secret distribution, and data deletion upon exit; BEC m This represents the m participants currently added to the system, with the threshold set to... t b Shared secret set to S b ;

[0082] S312, Generation of secret sharing and verification information, as detailed below:

[0083] (1) The secret sharing is jointly generated by the system registry D and the MBEC node;

[0084] (2) Each BEC i Calculate your own verification information, denoted as and As shown below:

[0085] ;

[0086] ;

[0087] S313. Verification of secret sharing, as shown below:

[0088] ;

[0089] S314, Key Recovery.

[0090] Preferably, in step S32, the Duel DDQN algorithm is used to achieve the optimal task unloading and resource allocation strategy, as follows:

[0091] S321. The formula for calculating the state value function is:

[0092] ;

[0093] in, Indicates the state given The probability of making the next action. Represents network parameters, The parameters represent the state-value function;

[0094] S322. The formula for calculating the action advantage function is:

[0095] ;

[0096] in, These are the parameters of the action advantage function;

[0097] S323. Based on steps S321 and S322, calculate the state value function and action value function, as shown below:

[0098] .

[0099] Therefore, the present invention adopts the above-mentioned dynamic entry and exit mechanism method based on blockchain cloud computing services, which takes into account the randomness of task arrival and the potential variability of the number, location and resource supply of cloud computing service providers. It can learn and optimize its performance according to changes in environmental conditions, minimize system latency and energy consumption through dynamic Dueling DDQN algorithm, and can perceive dynamically changing environments to provide more flexible and stable performance.

[0100] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0101] Figure 1 This is a data offloading scenario in the BECIN architecture;

[0102] Figure 2 It is a dynamic self-adjusting BECIN framework based on Duel DDQN;

[0103] Figure 3 It is a comparison of average system rewards under different learning rates;

[0104] Figure 4 This is achieved by offloading data to all BEC servers, resulting in a convergence of average expected overhead and satisfaction utility for users.

[0105] Figure 5 This is a comparison of different DRL approaches under changes in external resources; where (a) is the cost expenditure in scenario 1, (b) is the loss in scenario 1, (c) is the cost expenditure in scenario 1, and (d) is the loss in scenario 1.

[0106] Figure 6 The performance of different algorithms during the iteration process; where (a) is the system reward of the four algorithms within the time slot; (b) is the energy consumption of the four algorithms within the time slot; and (c) is the delay of the four algorithms within the time slot.

[0107] Figure 7 The comparison shows the performance of different algorithms on average data size (MB); (a) is a comparison of the total cost of the five algorithms; (b) is a comparison of the energy consumption of the five algorithms; and (c) is a comparison of the latency of the five algorithms.

[0108] Figure 8 The performance of different algorithms varies with the computing power (MHz) of the user equipment; where (a) is the total cost of the offloading system; (b) is the energy consumption of the offloading system; and (c) is the latency of the offloading system.

[0109] Figure 9 The performance of different algorithms at the maximum edge server CPU cycle frequency; where (a) is the total cost of the edge processor at the maximum CPU cycle frequency; (b) is the energy consumption of the edge processor at the maximum CPU cycle frequency; and (c) is the latency of the edge processor at the maximum CPU cycle frequency.

[0110] Figure 10 The performance of different algorithms on each BEC maximum bandwidth (MHz); where (a) is the total cost in different BECs; (b) is the energy consumption in different BECs; and (c) is the latency in different BECs.

[0111] Figure 11 The performance of different algorithms on the maximum transmission power MSD (mW) is shown, where (a) is the impact of maximum transmission power on total cost; (b) is the impact of maximum transmission power on energy consumption; and (c) is the impact of maximum transmission power on latency. Detailed Implementation

[0112] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0113] This invention discloses a dynamic entry and exit mechanism method based on blockchain cloud computing services, comprising the following steps:

[0114] S1. Establish a dynamic task unloading and resource allocation system model, including: local execution model, edge server computing model and communication model;

[0115] S2. Analyze the constraints of dynamic task unloading and resource allocation problems, as well as the description of system problems;

[0116] S3. Based on the combination of blockchain and cryptography, a dynamic task unloading and resource allocation strategy is designed.

[0117] Example 1

[0118] S1. Establish a dynamic task unloading and resource allocation system model, including: local execution model, edge server computing model and communication model.

[0119] like Figure 1 The diagram illustrates a blockchain-authorized cloud computing IoT scenario, where cloud computing nodes (i.e., base stations) participate in the blockchain network as blockchain nodes, providing users with secure, reliable, and environmentally friendly services. It outlines the potential actions that multiple cloud computing nodes might take within a given time period under different circumstances. This scenario involves N mobile smart devices (MSDs) and M blockchain-based cloud computing servers (BECs), denoted as follows: and .

[0120] The system is modeled as a discrete, time-slot-based environment, as shown in Table 1, which provides a list of key variables. Time is divided into multiple time slots, denoted as... The length of each time slot is For each mobile smart device MSD n Each time slot The computational task generated in the process is represented as We assume that tasks arriving in different time slots are independent and identically distributed. To capture the randomness of task arrivals, we use a Poisson distribution as the mathematical model in the simulation, as shown below:

[0121] ;

[0122] in, For MSD n In each time slot The average arrival rate of the internal calculation task The maximum number of tasks is defined here. To limit the upper limit of the number of tasks, a [missing information] is introduced. As the upper limit of the number of tasks, that is By appropriately setting the Poisson distribution Parameters control the average arrival rate of tasks, and use Limit the maximum number of tasks.

[0123] Table 1 List of key variables

[0124]

[0125] S11, Local Execution Model;

[0126] MSDs themselves have some data processing capabilities, so they can be considered for local processing of offloaded data. Therefore, consider the MSD of each device. n In different time slots Processing power and power management, device MSD n In the time slot The computing power is denoted as This indicates the device's computational efficiency when processing inbound tasks. Additionally, the device can adjust the CPU cycle frequency. To manage the power consumption of the device, The maximum computing speed of the device must not be exceeded. This is to ensure that the equipment can process tasks at maximum computing speed.

[0127] S111, Local Delay As shown below:

[0128] ;

[0129] in, Indicates time slot From MSD n To BEC m The proportion of computing tasks that are unloaded.

[0130] S112, Local Execution Capability The calculation formula is as follows:

[0131] ;

[0132] in, It is a constant power coefficient;

[0133] The allocation of compute and radio resources to the MSD remains unchanged from one time slot to the next in the system.

[0134] S113, Local Energy Consumption The calculation formula is as follows:

[0135] .

[0136] S12, the computing model of edge servers;

[0137] like Figure 1 As shown, in the data offloading scenario of the BECIN architecture, multiple cloud computing servers based on blockchain technology are merged. For each time slot... Each MSD device MSD n The resulting computational tasks can be offloaded to different BEC servers. m Up, uninstall ratio This indicates that each BEC server has BEC. m Having a specific number of computing resources f m and bandwidth resources In addition, the BEC system supports dynamically adjusting the CPU cycle frequency to reduce power consumption.

[0138] In order to allocate computing resources according to task requirements, the overall computing resource allocation is defined as follows: To facilitate resource allocation, a resource allocation coefficient is introduced. It indicates that in a time slot BEC from BEC server m To MSD device MSD n Resource allocation, and subject to ∈ constraint.

[0139] S121, Time Slot From edge server BEC m To MSD n The allocation of computing resources (such as CPU cycle frequency) is as follows:

[0140] ;

[0141] Based on the observations, when , At that time, the edge server BEC m It will distribute computing resources evenly to available MSDs. Representing time slots At that time, BEC m The set of MSDs within the wireless coverage area.

[0142] S122, Time Slot BEC m Total computational task latency As shown below:

[0143] ;

[0144] in, BEC m The computing power when processing incoming tasks is measured in cycles per second.

[0145] S123, BEC m In the time slot The energy consumption is as follows:

[0146] ;

[0147] in, The BEC energy consumption for executing one CPU / GPU cycle (the expected number of CPU / GPU cycles required by the task) is set to 9 × 10. -5 Wh.

[0148] S13, communication model;

[0149] In the communication model, communication between the MSD and BECs is based on OFDMA, a widely adopted multiple access scheme. OFDMA connects the edge server BECs... m Total bandwidth resources are divided into There are 3 equal orthogonal sub-channels, each with a bandwidth of 1. This is to support communication between MSDs. The sub-channel set used for MSD communication is represented as follows: .

[0150] It is important to emphasize that each sub-channel can be allocated to at most one terminal device. This design allows multiple terminal devices on different sub-channels to communicate simultaneously, thereby improving the overall capacity and efficiency of the system. By dividing the available spectrum into orthogonal sub-channels, different terminal devices can transmit data in parallel within the same time interval, which is beneficial for the efficient use of spectrum resources. In addition, using orthogonal sub-channel allocation helps to mitigate interference problems, because each terminal device has its own dedicated sub-channel, avoiding mutual interference between devices.

[0151] S131, in time slot From the edge server BEC m To MSD n The allocated bandwidth is:

[0152] ;

[0153] In order to allocate computing resources according to task requirements, the overall communication bandwidth resource allocation is defined as follows:

[0154] ;

[0155] in =0,1 ∈ , ∈M. When =1 indicates MSD n AND sub-channel Related to BEC m Transmitting data; when =0 indicates no association. Within each time slot, bandwidth resources are allocated according to demand; the total bandwidth allocation is denoted as... , m∈M. Indicates time slot From BEC server BEC m To MSD device MSD n The bandwidth allocation coefficient is affected by ∈ The constraints. Additionally, in equation (8), when At that time, BEC m Communication resources in MSD n They are evenly distributed between them.

[0156] ;

[0157] S132, from arrive Channel power gain Divide by the sub-channel k, as shown below:

[0158] ;

[0159] in, Indicates the path loss coefficient. This represents the path loss index. This represents the channel power gain caused by small-scale fading. express Location and Location In the time slot The distances between them are as follows:

[0160] ;

[0161] in, It refers to the vector norm, which represents the length of a directed line segment in two-dimensional space.

[0162] S133. During communication, signals are affected by various interferences and noises, such as interference signals from other users, multipath interference from multipath propagation, and environmental noise. These interferences and noises significantly degrade the quality of the received signal, thus affecting the overall performance of the communication system. To evaluate signal quality, the signal-to-interference-plus-noise ratio (SINR) is used as a metric. It quantifies the ratio of the received signal power to the combined power of interference and noise.

[0163] Therefore, sub-channel In the time slot From MSD n To BEC m The SINR is as follows:

[0164] ;

[0165] in, MSD n Transmission power, Represents the noise power spectral density. Indicates in Time slot, when BEC mFor MSD n Inter-cell interference when allocating sub-channels for transmission tasks;

[0166] SINR is a key parameter for estimating channel quality and calculating transmission rate. Transmission rate calculation primarily revolves around determining channel capacity, which represents the maximum data rate achievable under given channel conditions. In an ideal, interference-free environment, channel capacity can be derived by applying Shannon's formula:

[0167] C = B·log2(1+SINR);

[0168] Where C represents channel capacity and B represents channel bandwidth.

[0169] S134, in time slot BEC m and MSD n The transmission rates between them are as follows:

[0170] (14);

[0171] Based on the aforementioned transmission rate, MSD n To BEC m The transmission delay between them is as follows:

[0172] ;

[0173] S135, Time slot t from MSD n Transmitted to BEC m The energy consumption is as follows:

[0174] ;

[0175] S2. Analyze the constraints of dynamic task unloading and resource allocation problems, as well as the description of system problems.

[0176] S21. The constraints of the dynamic task unloading and resource allocation problem are as follows:

[0177] S211, Unloading decision constraints: Let... MSD n In the time slot When the uninstallation decision is made, =0 or 1 indicates that the task is fully allocated to the local MSD processor or the edge processor, respectively. Furthermore, assuming each task is divisible, specifically, when... This indicates that the task is assigned to multiple processors (the local MSD processor and edge processors) for processing. Clearly, the task offloading constraint needs to meet the following conditions:

[0178] ;

[0179] S212, Local Processor CPU Cycle Frequency Constraint: This indicates the CPU cycle frequency of the local processor, corresponding to its processing speed. Specifically, the local processor can dynamically adjust its CPU cycle frequency to reduce power consumption. The local processor can only process one task at a time, and its computing resources are limited. The computing configuration must meet the following conditions:

[0180] ;

[0181] S213. Computational resource constraints of edge processors: This refers to BEC m MSD allocated within the coverage area n The computing resources of edge processors are limited. Unlike local processors, edge processors can handle multiple tasks. Similarly, the computing resources of edge processors are limited. Clearly, the following constraints need to be met:

[0182] ;

[0183] S214, Bandwidth Constraints: This refers to BEC m The coverage area is allocated to MSD n The communication resources of the edge server are limited. Therefore, the following constraints need to be met:

[0184] ;

[0185] S215, Wireless Channel Power Constraints: MSD n exist Wireless channel power in time slots. MSD n exist The maximum permissible wireless channel power during time-slot wireless communication. This constraint is imposed to limit the wireless channel power of each MSD in a specific time slot, in order to avoid interference with other devices or exceeding the system's power limits. This constraint can be formally expressed as:

[0186] ;

[0187] S216, Sub-channel constraints: This indicates the bandwidth of the BEC server in time slot t. m To MSD device MSD n The allocation coefficient is subject to ∈ P b,m tThe constraints are as follows: Each sub-channel is assigned to only one MSD, therefore the following constraints must be satisfied:

[0188] ;

[0189] Furthermore, each MSD is assigned to a single subchannel, which clearly satisfies the following conditions:

[0190] ;

[0191] S22. Based on constraints, the joint task unloading and resource allocation problem is as follows:

[0192] Based on the constraints of step S21, the joint task unloading and resource allocation problem in a dynamic environment requires simultaneous consideration of task unloading strategies and resource allocation strategies to optimize system performance.

[0193] In a constantly changing environment, dynamic determination is required:

[0194] (1) Tasks that should be offloaded to edge servers for processing;

[0195] (2) How to allocate limited computing and communication resources to maximize system efficiency.

[0196] S221. In the system, the total cost is defined as the weighted sum of task latency and processing energy consumption. Task latency includes transmission latency and computation latency. For a given time slot... t The task is performed on the local processor MSD. n and edge server BEC m The total delay of transmission and computation can be expressed as:

[0197] ;

[0198] S222, For a given time slot Local Processor MSD n and edge server BEC m The total energy consumption for task transmission and computation can be expressed as:

[0199] ;

[0200] Based on the task delay shown in Equation (24) and the energy consumption shown in Equation (25), the optimization objective is as follows:

[0201] ;

[0202] Therefore, the BECIN problem can be expressed as:

[0203] ;

[0204] S3. Based on the combination of blockchain and cryptography, a dynamic task unloading and resource allocation strategy is designed.

[0205] A join / leave mechanism combining blockchain and cryptography is used to address the problem of updating service policies in BECIN based on dynamic task arrivals and resource changes. Furthermore, a state-of-the-art DRL algorithm called Dynamic Duel (DDQN) is introduced to approximately optimize task offloading and resource allocation strategies. The entire process is as follows: Figure 2 As shown.

[0206] S31. Add an exit mechanism;

[0207] Based on the dynamic characteristics of task arrival and resource fluctuations, BECIN's service strategy is adjusted, leveraging the immutability and decentralization of blockchain technology to facilitate the dynamic joining and leaving of edge service providers. When an edge service provider joins, its relevant information is recorded on the blockchain, allowing other participants to access it and update their service strategies accordingly. Similarly, when an edge service provider leaves, its information is removed from the blockchain, ensuring the system only contains valid service providers. To increase the protocol's flexibility, a joining and leaving mechanism is introduced. The entire process is coordinated by a smart contract.

[0208] Add: If a new BEC m To join the protocol, you can call the join function defined in the smart contract. Once authentication is complete, the new user will be immediately authorized.

[0209] First, an auditable and secure multi-party computation for BEC nodes is implemented based on a verifiable threshold secret-sharing scheme using the Chinese Remainder Theorem (CRT). An improved CRT Paillier scheme is employed to ensure secure key transmission between BEC nodes during the auditing process. The specific steps are as follows:

[0210] S311, System Initialization: Let D be the registration authority of the BEC system, responsible for system registration and exit management, including key generation, secret distribution, and data deletion upon exit. BEC m This represents the m participants currently added to the system, with the threshold set to... t b Shared secret set S b This threshold acts similarly to the consensus mechanism in blockchain nodes, meaning a new node can only join the network after completing its computation and receiving verification from a certain number of other nodes.

[0211] D chooses a larger prime number. , As a finite field The elements in, where Nis an integer that satisfies the hardness of integer factorization, greater than the integer . is an element of order in. Additionally, select a sequence of positive integers that satisfies the following conditions:

[0212] (1) is strictly increasing;

[0213] (2) ;

[0214] (3) ;

[0215] (4) ;

[0216] [[ID=3​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​

[0221] ;

[0222] ;

[0223] After calculating their respective verification information, and Public disclosure. Finally, it can be seen that after completing the above procedures, BEC... i The secret share received by (i=1,2,…,M) is ={ =1,2,..., }

[0224] S313. Authentication of Secret Sharing: During the preceding system initialization phase, the threshold was set to... This means at least... The nodes approve the addition of a new node. If M BEC nodes approve the addition of a new BEC node, it indicates their willingness to share secret shares with other agreeing parties. If the number of secret shares that can be shared is less than... Then D notifies the relevant potential edge service providers, informing them that they are not allowed to join the blockchain network. Let... The nodes are {BEC1, BEC2, ..., BEC} tb To prevent dishonesty among BEC nodes in the blockchain network, BEC... k Upon receiving from other BECs j Secret Sharing S j Then, use the following verification formula (i=1,2,…,M) to verify n components. :

[0225] (31);

[0226] If equation (31) holds, then it is considered that share S j Valid; otherwise, consider S to be valid. j Invalid, indicating BEC j Fake shares were provided.

[0227] S314, Key Recovery: Regarding the above t... b After the se-cret shares are verified, the system registry D securely sends these secret shares to any participant. Each participant receives t b The following are the shares of se-cret:

[0228] ;

[0229] Each participant is based on these The secret shares can be constructed using M sets of congruence equations, as follows:

[0230] ;

[0231] According to the CRT, solving these M sets of equations will yield X1, X2, ..., X... M The unique value. This is determined by calculating X = X1 + X2 + ... + X... M Using the formula X=S b +A·p d You can obtain the shared secret S b For S b =X−A·p d .

[0232] Obtain the shared secret S b This means you have qualified to join the BEC network, marking a successful registration.

[0233] Exit: If the existing BEC m The process of requesting to exit the protocol is similar to that of joining. It requires constructing a transaction to trigger the smart contract's exit function. All related data existing in the blockchain will be promptly deleted.

[0234] Theorem 1: Adding an exit mechanism can verify user identity and ensure the security of the signature authentication scheme, provided that:

[0235] ;

[0236] Theorem 2: Participants only need to possess at least Secret shares can be used to reclaim the secret.

[0237] Theorem 3: It can prevent deception from the system's secret distributor D.

[0238] Theorem 4: It is possible to identify mutual deception among M BEC nodes in a BEC network.

[0239] Theorem 5: The verification process is safe.

[0240] S32. The Duel DDQN algorithm is adopted to achieve the optimal task unloading and resource allocation strategy;

[0241] To achieve the optimal task offloading and resource allocation strategy, the Duel-based DDQN algorithm is employed. This algorithm trains a dual-q network to approximate the optimal strategy, with the two q networks competing against each other to improve the stability and convergence speed of the strategy. Utilizing the parallel DDQN algorithm, the task offloading and resource allocation strategy can be dynamically updated in a dynamic environment based on real-time task arrivals and resource changes.

[0242] The Origin of Duel DDQN: Traditional reinforcement learning algorithms typically use tabular representations to store the state-value function V(s) or state-action-value function Q(s, a). However, this approach has some limitations. For example, real-world reinforcement learning tasks often require continuous state spaces characterized by an infinite number of states, making tabular storage infeasible. To overcome these limitations, solutions for approximating the state-value function or state-action-value function by directly fitting the function have emerged. This approach reduces storage requirements and effectively solves the problem.

[0243] S321. In the context of continuous state space, approximation functions can be used. To compute value functions In mathematics:

[0244] ;

[0245] here, and Representing states respectively and actions The vector representation of the function. Typically parameterized It is usually implemented as a neural network that outputs a real value and is often called a q-network.

[0246] The core of the Deep Q-network (DQN) algorithm lies in preserving the Q-function and using it for decision-making. Representation Strategy The action value function is defined below. Upon reaching the state... Then, the algorithm traverses the entire action space and selects actions that make... Maximizing actions as a strategy:

[0247] ;

[0248] Then, DQN updates using the Bellman equation. :

[0249] (35);

[0250] In simpler tasks, fully connected neural networks are typically used to approximate the task. However, for more complex tasks, such as playing Atari games, convolutional neural networks are used to approximate the mapping from images to value functions. Due to its limited ability to handle a finite number of action values, DQN is typically used for tasks with discrete action spaces. In the original DQN algorithm, when updating the q-value using the Bellman equation, the target q-value is estimated using the maximum action value in the current state. However, this estimation introduces a high bias, leading to unstable algorithm performance.

[0251] Dual DQN proposes a modification to mitigate this bias. The core idea of ​​Dual DQN is to separate the calculation of the target q-value from the action selection process.

[0252] In calculating the target q-value, the optimal action is first selected using the current q-network. Then, the target q-network is used to estimate the q-value corresponding to the action. Therefore, the formula for calculating the target q value is:

[0253] ;

[0254] in, These are the parameters of the target q-network. The purpose of this method is to avoid overestimating the current q-value when calculating the target q-value. During training, the target q-value is calculated using the target q-network, and the estimated q-value is calculated using the current q-network.

[0255] Then, the difference between these two values ​​is calculated using the mean squared error, and the parameters of the current q-network are updated accordingly. The main improvement of dual DQN lies in reducing the overestimation of the target q-value, thereby improving the stability and performance of the algorithm.

[0256] Duel-based DQN is an improved approach to DQN that separates the state-value function from the action advantage function. This separation allows for a better estimation of the relative advantage of different actions in each state, thus improving the algorithm's performance and efficiency. Combining Duel-based DQN with these two improved methods, utilizing the network structure of Duel-based DQN and the target q-value estimation method of Dual-DQN, further enhances the algorithm's performance. Specifically, traditional DQN uses a single q-network to output a q-value action for each network, which does not fully utilize the information in the state. Duel-based DQN introduces a network structure that separates state values ​​and action advantages, more effectively modeling the relationship between the state-value function and the action-value function.

[0257] The Duel-based DQN network structure consists of two subnets: a value function and an advantage function. The value function represents the value of a state, while the advantage function represents the relative advantage of selectively choosing different actions given a certain state. Specifically, the formula for calculating the state value function is:

[0258] ;

[0259] here, Indicates the state given The probability of making the next action. Represents network parameters, These represent the parameters of the state-value function. Specifically, the formula for calculating the action advantage function is:

[0260] ;

[0261] in, These are the parameters of the action advantage function.

[0262] Finally, by adding equations (37) and (38), the state value function and action value function are calculated:

[0263] ;

[0264] Specifically, the Duel DDQN mainly includes the following parts:

[0265] (1) Network Structure: The Duel DDQN adopts the network structure of the Duel DQN, consisting of a value function and an advantage function. These two functions share some layers but are split into independent branches before the output layer. This allows the network to learn the value of the state and the relative quality of different actions simultaneously. The value function is used to estimate the value of the state, representing the expected cumulative reward for a given state in Equation (37). The action advantage function is used to estimate the relative advantage or disadvantage of different actions, representing the advantage or disadvantage of taking a specific action compared to the average level in Equation (38).

[0266] (2) Target q-value estimation: When calculating the target q-value, Duel DDQN uses the Double DQN estimation method. Specifically, when selecting the optimal action, the current q-network is used to determine the optimal action, denoted as... Subsequently, the q-value corresponding to this action is estimated using a target q-network, denoted as . The specific formula is shown in equation (36). This method helps to reduce the overestimation of the target q value and improve the stability of the algorithm. The final estimate of the q value is achieved by combining the state value function and the action advantage function, as shown in equation (39).

[0267] To better describe the self-adjusting BECIN framework based on Duel DDQN, we first define the rewards, behaviors, and states of Duel DDQN in the scenario.

[0268] Reward: The system's reward stems from the joint optimization of task latency and power consumption, maximizing the system reward in the long run while minimizing system latency and energy consumption. Therefore, a multiplication operation of -1 is performed on the reward before proceeding. In other words, the reward is set to... To more effectively represent the reward, the problem of minimizing task latency and power consumption is transformed into a QoE optimization problem using an exponential function:

[0269] ;

[0270] in this case, This refers to the upper bound of the weighted sum of system latency and energy consumption. As shown in equation (40), the lower the system latency and energy consumption, the higher the QoE. Therefore, we focus on maximizing the system QoE rather than minimizing latency and energy consumption to better characterize the efficiency of the algorithm. Based on equation (40), the optimization problem in equation (27) is transformed into:

[0271] ;

[0272] Therefore, the reward is set as follows: .

[0273] Actions: By observing the system state, actions need to be taken at each time step to handle the task, including offloading decisions, allocating bandwidth resources to edge devices, and allocating computing resources to edge devices. Based on this, the action space can be represented as:

[0274] ;

[0275] The explanation of each action component is as follows:

[0276] Uninstallation decision : (n∈n, m∈m), It refers to MSD n In the time slot The resulting computational tasks are allocated to BEC according to a certain proportion after a decision is made. m Discretize the values ​​between 0 and 1 and allocate tasks according to a specified ratio.

[0277] System bandwidth allocation scheme : This refers to all BEC m Within the coverage area, by MSD n via sub-channel The set of computational tasks being transmitted. In mobile communication systems, subchannels are used to divide spectrum resources into multiple independent frequency bands or segments. Furthermore, in the frequency domain, different subchannels may be affected by different small-scale fading components, depending on the signal propagation path and environment. Therefore, multiple subchannels with different configurations, including small-scale fading components and interference, are established to meet the needs of mobile communication services. Due to variations in propagation paths and bandwidth resource allocation, significantly different transmission rates are achieved.

[0278] BEC resource allocation : Refers to time slot At that time, BEC m Assigned to MSD n A collection of computing resources. , 0≤G n m,t ≤1. Each edge server allocates a portion of its own computing resources to execute tasks based on the current resource and task status.

[0279] Status: The system status consists of five components, including task status. Channel status Bandwidth status and available computing resource status Therefore, the system state is defined as follows:

[0280] ;

[0281] Task status Includes MSD in time slots Information on all data processing tasks executed within the system. Specifically, The definition is as follows:

[0282] ;

[0283] in, and The coordinates represent the location where the task was initiated. Indicates the deadline for the task. This indicates that each MSD is in a time slot. The amount of computational work generated (in bits) meets the conditions. .

[0284] Channel status This indicates the current network status, i.e., the remaining capacity after allocating network channels, and is defined as follows:

[0285] ;

[0286] Bandwidth status This represents the amount of bandwidth resources remaining in the current system, defined as follows:

[0287] ;

[0288] Available computing resource status The current state of computing resources is represented by:

[0289] ;

[0290] In the equation above, and Indicates the current coordinates (horizontal and vertical) of the BEC service. This indicates the total number of available edge processors. This represents the remaining computing resources in the system, calculated as follows:

[0291] ;

[0292] Example 2

[0293] 1. Data and experimental setup;

[0294] In the architecture, a region (ranging from 0 to 500 on both the x and y axes) consists of multiple BEC servers, representing the number of nodes participating in accessing the blockchain network. Considering node joining and leaving, the number of BEC servers varies from approximately 5 to 20. The number of mobile smart devices ranges from 50 to 100. In the simulation, each mobile device generates one task to be processed per time slot. These tasks have different computational and bandwidth requirements depending on the algorithm complexity and task size. Tasks arrive according to a Poisson distribution with a parameter value of 5. The randomly generated data block size is chosen within the range of [10, 100] MB. The computational power of the MSD ranges from [40, 100] cycles / bit, and the computational resources of each MSD are randomly allocated within the range of {0.5, 0.8, 1.0} GHz. The constant power factor of the MSDs is set to 10.10. −27 Watt·[s 3 / cycles 3 The transmit power of MSDs is set to 100mW. In the BECIN architecture, each BEC has a total bandwidth resource of 40MHz and a total computing resource of 10GHz to facilitate communication and task processing between devices within the area. [Noise spectral density...] σ 2 Set to 10 − 9 W.

[0295] 2. Algorithm comparison;

[0296] To demonstrate the effectiveness of the proposed framework, experiments were designed from the following aspects:

[0297] (1) The solution proposed in this invention is a dynamically adjustable algorithm that can learn and optimize its performance according to changes in environmental conditions. Compared with other algorithms that lack dynamic adjustment capabilities, the solution proposed in this invention is more suitable for adapting to different environmental conditions and provides more flexible and stable performance.

[0298] (2) When both the optimization strategy and the baseline algorithm incorporate a dynamic adaptive adjustment mechanism, Dueling DDQN achieves better results than the baseline method. In the experiment, four classic algorithms were selected as benchmarks: PPO (Proximal Policy Optimization), DDQN (Double Deep Q-Network), DQN (Deep Q-Network), and Random.

[0299] 3. Experimental results;

[0300] (1) Training performance evaluation;

[0301] like Figure 3 The diagram illustrates the impact of different learning rates on the average system reward. Lower learning rates lead to a gradual increase in system reward, relying on past experience and reducing volatility, but exhibiting a slower convergence rate. Conversely, higher learning rates result in a rapid increase in system reward, making it more sensitive to new observations, but may lead to unstable behavior. Choosing an appropriate learning rate is crucial, requiring a trade-off between learning speed and stability. Experimental exploration and adjustments based on specific problems and environments are necessary to achieve optimal performance. According to experimental results, a learning rate of 0.01 yields the best reward performance with a satisfactory convergence rate. Therefore, a learning rate of 0.01 is set for subsequent system simulations and evaluations.

[0302] like Figure 4 As shown, the impact of offloading data to all BEC servers on average expected user cost and satisfaction is illustrated. As data offloading progresses, average expected user cost gradually decreases, while satisfaction increases. The results indicate that after converging to the optimal data offloading point, users achieve a high level of satisfaction and a low level of expected cost.

[0303] (2) DRL performance evaluation;

[0304] The system cost and loss function performance of the proposed adaptive dynamic adjustment DRL method were evaluated and compared with two other related methods. The first method is the traditional DRL method, which typically employs static parameters or policies, meaning they do not adjust during operation based on environmental changes. This static approach may lead to poor algorithm performance or instability under different environments. The second method is a DRL method that does not use neural networks. Generally, compared to DRL policies, these methods rely on tabular representations of the value function or policy, and their modeling capabilities are limited, especially in high-dimensional and continuous scenarios.

[0305] like Figure 5 As shown, the performance comparison of different DRL methods is presented, demonstrating significant improvements in cost and loss compared to the baseline algorithm. (a)-(b) represent scenario 1, and (c)-(d) represent scenario 2. The results show that Dynamic-based DRL consistently outperforms the other algorithms. Furthermore, the cost and loss values ​​of the three algorithms (Dynamic DRL, Traditional DRL (non-dynamic), and Traditional DRL) exhibit similar overall trends, showing a decrease followed by stabilization.

[0306] Dynamic Mechanism: The architecture based on dynamic DRL can adapt to environmental changes. It utilizes a model-based approach to model the environment and optimizes strategies online. By observing and analyzing the environment in real time, it selects appropriate actions under different states to maximize expected returns and minimize unnecessary costs.

[0307] Learning ability and optimization process: By using deep reinforcement learning methods, it continuously learns and optimizes through interaction with the environment, gradually improving and adapting to changes in various scenarios. In contrast, traditional DRL algorithms may perform poorly in nonlinear and highly dynamic environments because they typically rely on manually extracted features and linear models.

[0308] Policy Update and Stability: All three algorithms employ similar policy update mechanisms, such as gradient-based methods, continuously adjusting policy parameters for optimization. As the algorithm iterates and trains, the policy converges and becomes stable, leading to a reduction in cost and loss.

[0309] (3) Evaluation of Task Offloading Utility;

[0310] • Performance comparison during iteration, such as Figure 6As shown, (a) illustrates the performance of different time steps and different system rewards. It can be observed that Duel DDQN consistently outperforms other algorithms in all scenarios. Furthermore, the system rewards of the four algorithms (Duel DDQN, DDQN, PPO, and DQN) show a trend of continuous increase followed by stabilization within the time step, while Random exhibits relatively small fluctuations. It can be seen that when the number of iterations is small (≤15), the average system reward increases with the number of iterations. This is because at the beginning of training, the agent may be in an initial random policy state with limited understanding of the environment, leading to suboptimal action choices. As the number of time steps increases, the agent gradually learns from past experience and optimizes its policy, resulting in improved system rewards. However, beyond a certain threshold (e.g., Iterations=15), the system utility tends to stabilize. This is because, as training progresses, the agent accumulates more experience and knowledge, achieving a better balance between exploration and exploitation, leading to a gradual stabilization of the system reward.

[0311] like Figure 6 Figures (b) and (c) illustrate the performance of different methods in terms of latency and energy consumption within time slots. It can be observed that, except for the Random algorithm, both system latency and energy consumption show a decreasing trend during iteration. In particular, Duel-ing DDQN consistently outperforms other algorithms in all scenarios. In the simulation setting, the proposed method demonstrates that latency and energy consumption converge to a minimum when QoE is maximized. Compared to DDQN and PPO methods, the Dynamic Duel-ing DDQN solution achieves energy savings of 42.7% and 20.9%, respectively, and reduces energy consumption by 38.7% compared to deep q-networks. Furthermore, the Dynamic Duel-ing DDQN solution reduces latency by 10.5% and 19.6% in the following cases, and reduces latency by 29.9% compared to DDQN and PPO methods.

[0312] • Performance comparison based on data size, such as Figure 7 Figures (a), (b), and (c) show a comparison of the total cost, energy consumption, and latency of the five algorithms. Here, the dynamic changes in available resources are consistent across the algorithms. By adjusting the parameters of the Poisson distribution, the average output can be controlled to follow a Poisson distribution with a desired mean, where the random value of the data size ranges from 20MB to 100MB. The results show that Duel DDQN significantly outperforms the other four algorithms in terms of latency, energy consumption, and total cost. Furthermore, all five algorithms show an increasing trend with increasing average data size. Moreover, compared to the other three baselines, the growth rate of Duel DDQN proposed in this invention is slower, further validating the effectiveness of the method.

[0313] • Performance comparison based on user device computing power, such as Figure 8 As shown in (a), (b), and (c), the impact of different user device computing capabilities on the total cost, energy consumption, and latency of the offloading system was evaluated. The results show that Duel DDQN outperforms the other four algorithms in all scenarios. Furthermore, the energy efficiency of all five algorithms exhibits a monotonic increase with user device computing capabilities. Conversely, the latency performance of all five algorithms shows a monotonic decrease with user device computing capabilities. Therefore, the overall cost-effectiveness of the five algorithms initially experienced a sharp decline primarily driven by latency performance, transitioned to a gradual plateau, and ultimately showed an upward trend primarily driven by energy consumption. This is likely because as user device computing capabilities increase, devices can process tasks more efficiently, reducing task execution time and thus lowering overall latency. However, energy consumption during task execution also increases with computing capabilities, leading to an overall increase in system energy consumption. From a total cost perspective, the initial rapid decline mainly stemmed from a significant reduction in latency costs, while energy costs increased moderately. As energy consumption continued to escalate rapidly, the rate of decline gradually slowed. Ultimately, a threshold was reached where the reduction in latency costs equaled the increase in energy costs, reaching a minimum. Subsequently, the reduction in energy consumption costs outpaced the reduction in latency costs, leading to an upward trend in total costs. Simulation experiments show that enhancing the computing power of user equipment can significantly improve the system's latency and energy efficiency. However, it is crucial to consider the potential cost increases associated with higher energy consumption. Therefore, when designing cloud computing systems, a trade-off between computing power, latency, and energy consumption should be considered to optimize system performance.

[0314] • Performance comparison at maximum edge server CPU cycle frequency, considering the impact of the edge processor's maximum CPU cycle frequency on overall offloading cost. For example... Figure 9 As shown in (a), (b), and (c), the total cost latency, energy consumption, and total cost of the edge processor at its maximum CPU cycle frequency are presented, respectively. It is observed that the method of this invention outperforms other methods in almost all scenarios. Furthermore, as the maximum CPU cycle frequency of the edge processor increases, the system latency of several algorithms shows a decreasing trend. Regarding energy consumption, the PPO, DQN, and Random algorithms all show a monotonically increasing trend, while the Duel DDQN and DDQN algorithms show a trend of first increasing, then decreasing, and finally increasing again. Possible reasons are as follows:

[0315] a. As the maximum CPU cycle frequency of the edge processor increases, tasks can be executed faster, thereby reducing task execution time and overall latency;

[0316] b. The system needs to balance the trade-off between latency reduction and energy consumption fluctuations caused by edge processor variations. At lower CPU frequencies, these algorithms may not fully utilize processor performance, but above a certain threshold, they can effectively utilize processor resources, thereby reducing the overall cost of the system.

[0317] It is worth noting that the PPO algorithm has relatively poor performance and incurs higher costs at the maximum CPU cycle frequency of edge processors. Therefore, an effective algorithm must be robust. Possible reasons are as follows:

[0318] The PPO algorithm uses the significance sampling ratio and truncation policy to update the policy network. When the maximum CPU cycle frequency of the edge processor increases, the magnitude of the policy update may become larger, resulting in greater fluctuations during the update process and making it difficult to achieve good convergence.

[0319] • Perform performance comparisons at the maximum bandwidth of each BEC, such as Figure 10 As shown in (a), (b), and (c), the performance of different methods under different maximum bandwidths is demonstrated in terms of overall cost, energy consumption, and latency in different BECs. The results show that the Dueling DDQN algorithm outperforms the other algorithms in all scenarios. Furthermore, all five algorithms exhibit a monotonically decreasing trend. This can be explained by the fact that greater bandwidth allows for the transmission of more data in the edge network, thus reducing the overall system cost. Simultaneously, the performance improvement of the algorithms may be attributed to the shorter transmission latency provided by higher bandwidth, enabling tasks to be offloaded and executed faster on both local and edge devices. In addition, higher bandwidth may also lead to lower energy consumption, as it reduces the duration and power requirements of data transmission, allowing edge devices to complete data transmission faster at lower power levels, thereby reducing energy consumption.

[0320] • Performance comparison of MSD maximum transmission power, and study of the impact of MSD maximum transmission power on offloading costs, such as... Figure 11 Figures (a), (b), and (c) illustrate the impact of maximum transmission power on latency, energy consumption, and total cost. The observed results are as follows:

[0321] a. The method proposed in this invention is consistently superior to other methods in all scenarios.

[0322] b. As the maximum transmit power of the MSD increases, the performance of the five methods improves rapidly in terms of energy consumption, but deteriorates rapidly in terms of latency.

[0323] Overall, this trend indicates that energy consumption has a greater impact than latency, leading to a continuous increase in overall cost. This is likely because as the maximum transmission power of the MSD increases, the MSD device can transmit data at higher power levels, thereby reducing transmission time and increasing transmission rate. However, while increasing transmission power can reduce transmission time and latency, it also increases energy consumption. As transmission power increases, the impact of energy consumption on total cost gradually outweighs the cost savings from reduced latency, resulting in an overall upward trend in total cost. This suggests that selecting the maximum transmission power of the MSD is a necessary condition for optimizing total cost when balancing energy consumption and latency.

[0324] Therefore, this invention employs a dynamic entry and exit mechanism based on blockchain cloud computing services to address the challenges of communication and computing resource demands, as well as data privacy issues in the Internet of Things (IoT). Based on the Dynamic Blockchain Cloud Computing IoT (BECIN) framework with dynamic entry and exit mechanisms, it considers the randomness of task arrival and the potential variability in the number, location, and resource supply of cloud computing service providers. The system latency and energy consumption are minimized through the dynamic Dueling DDQN algorithm, which is capable of sensing dynamically changing environments. A comprehensive analysis of the convergence of the proposed dynamic self-adjusting BECIN framework model is conducted. Experimental results verify the feasibility and superior performance of this framework.

[0325] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A dynamic entry and exit mechanism method based on blockchain cloud computing services, characterized in that, Includes the following steps: S1. Establish a dynamic task unloading and resource allocation system model, including: local execution model, edge server computing model and communication model; S2. Analyze the constraints of dynamic task unloading and resource allocation problems, as well as the description of system problems; S3. Based on a join / leave mechanism combining blockchain and cryptography, design a dynamic task unloading and resource allocation strategy. The specific steps are as follows: S31. Introduce a join / leave mechanism based on a combination of blockchain and cryptography. Design dynamic task unloading and resource allocation strategies based on dynamic task arrival and resource changes, and update service strategies as follows: S311, System Initialization: Let D be the registration authority of the BEC system, responsible for system registration and exit management, including key generation, secret distribution, and data deletion upon exit; BEC m This represents the m participants currently added to the system, with the threshold set to... Shared secret set to S b ; S312, Generation of secret sharing and verification information, as detailed below: (1) The secret sharing is jointly generated by the system registry D and the MBEC node; (2) Each BEC i Calculate your own verification information, denoted as and As shown below: ； ； S313. Verification of secret sharing, as shown below: ； S314, Key Recovery; S32. The Duel-based DDQN algorithm is adopted. In a dynamic environment, the task unloading and resource allocation strategies are dynamically updated based on real-time task arrival and resource changes to achieve the optimal task unloading and resource allocation strategy, as detailed below: S321. The formula for calculating the state value function is: ； in, Indicates the state given The probability of making the next action. Represents network parameters, The parameters of the state-value function; S322, Action Advantage Function The calculation formula is: ； in, These are the parameters of the action advantage function; S323. Based on steps S321 and S322, calculate the state value function and action value function, as shown below: 。 2. The method for a dynamic entry and exit mechanism based on blockchain cloud computing services according to claim 1, characterized in that, In step S1, the system model is established, and the local execution model is as follows: S111, Local Delay The calculation formula is as follows: ； in, Representing the BECs set, ={1,2,…, }; For the device MSD n In the time slot Computational power; CPU cycle frequency; Indicates time slot From MSD n To BEC m The proportion of computing tasks that are unloaded; Each time slot The computational tasks generated in the process; S112, Local Execution Capability The calculation formula is as follows: ； in, It is a constant power coefficient; S113, Local Energy Consumption The calculation formula is as follows: 。 3. The method for a dynamic entry and exit mechanism based on blockchain cloud computing services according to claim 2, characterized in that, In step S1, the computational model of the edge server in establishing the system model is as follows: S121, Time Slot From edge server BEC m To MSD n The allocation of computing resources is as follows: ； in, A server computing resource allocation scheme; The resource allocation coefficient represents the time slot. BEC from BEC server m To MSD device MSD n Resource allocation, subject to ∈ constraint; For time slots BEC m The set of MSDs within the wireless coverage area; S122, Time Slot BEC m Total computational task latency As shown below: ； in, BEC m Computational capabilities when processing incoming tasks; S123, BEC m In the time slot energy consumption As shown below: ； in, BEC power consumption for executing one CPU / GPU cycle.

4. The method for a dynamic entry and exit mechanism based on blockchain cloud computing services according to claim 3, characterized in that, In step S1, the communication model is established as follows: S131, in time slot From the edge server BEC m To MSD n The allocated bandwidth is: ； Overall communication bandwidth resource allocation As shown below: ； in, =0,1 ∈ , ∈M; when =1 indicates MSD n AND sub-channel Related, when =0 indicates no association; S132, from arrive Channel power gain As shown below: ； in, Indicates the path loss coefficient. This represents the path loss index. This represents the channel power gain caused by small-scale fading. express Location and Location In the time slot The distance between them; S133, Sub-channel From MSD on time slot t n To BEC m The signal-to-noise ratio (SINR) is shown below: ； in, MSD n Transmission power, Represents the noise power spectral density. This indicates that in time slot t, when BEC m For MSD n Inter-cell interference when allocating sub-channels for transmission tasks; S134, in time slot BEC m and MSD n The transmission rates between them are as follows: ； MSD n To BEC m Transmission delay between As shown below: ； S135, Time Slot From MSD n Transmitted to BEC m The energy consumption is as follows: 。 5. The method for a dynamic entry and exit mechanism based on blockchain cloud computing services according to claim 4, characterized in that, In step S2, the constraints of the dynamic task unloading and resource allocation problem are as follows: S211, The unloading decision constraint is: ； S212, The local processor CPU cycle frequency constraint is: ； in, Indicates the CPU cycle frequency of the local processor; S213, The computing resource constraints of the edge processor are: ； in, This refers to BEC m MSD allocated within the coverage area n Computing resources; S214, Bandwidth constraint is: ； in, This refers to BEC m The coverage area is allocated to MSD n bandwidth resources; S215, The wireless channel power constraint is: ； in, MSD n exist The wireless channel power of the time slot, MSD n exist The maximum permissible wireless channel power during time-slotted wireless communication; S216. Sub-channel constraints, as detailed below: Each subchannel is assigned to only one MSD, satisfying the following constraints: ； Each MSD is assigned to a single subchannel, which obviously satisfies the following condition: ； in, This indicates the bandwidth of the BEC server in time slot t. m To MSD device MSD n The allocation coefficient.

6. The method for a dynamic entry and exit mechanism based on blockchain cloud computing services according to claim 5, characterized in that, In step S2, based on constraints, the joint task unloading and resource allocation problem is addressed as follows: S221, Given time slot The task is performed on the local processor MSD. n and edge server BEC m The total latency for transmission and computation is as follows: ； S222, Given time slot Local Processor MSD n and edge server BEC m The total energy consumption for task transmission and computation is as follows: ； S223. Based on total delay and total energy consumption, the optimization objectives are as follows: ； S224. The BECIN problem, a blockchain-based cloud computing IoT network, is as follows: 。

Images