Cloud side-end collaboration method and system for multi-service heterogeneous resources of novel power distribution network

Abstract

The invention provides a novel power distribution network multi-service heterogeneous resource-oriented cloud side-end cooperation method and system, and the method comprises the steps: determining a corresponding terminal based on the type of a to-be-processed service, and selecting a consensus node through the reputation value of a terminal node; determining a terminal master node from the consensus nodes by using a verifiable random function; setting an execution mode of the terminal main node and then merging the terminal main node into the power distribution network to realize communication with the power distribution substation; establishing a cloud side end collaborative network of a terminal where the power distribution master station, the power distribution sub-station and the master node are located; establishing a calculation model of the cloud edge-end collaborative network; adopting a deep reinforcement learning algorithm to optimize the calculation model under set optimization conditions to obtain an optimized calculation model; and inputting a to-be-processed service into the optimized calculation model, outputting a corresponding unloading strategy, and executing the unloading strategy. Based on the method, the invention further provides a cloud side-end cooperation system for the multi-service heterogeneous resources of the novel power distribution network. The total cost of the task is reduced, and the success rate of task execution is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of collaborative communication, and particularly relates to a cloud-edge-end collaboration method and system for multi-service heterogeneous resources in a new type of distribution network. Background Art

[0002] With the continuous access of a large number of terminal devices and the continuous expansion of data volume, the "cloud-terminal" architecture of traditional distribution networks can no longer meet the application scenarios with high requirements for real-time performance and reliability of terminal devices. The "cloud-edge-end" architecture based on cloud-edge collaboration has emerged as the times require. This architecture consists of three parts: the cloud, the edge, and the terminal. By introducing edge computing technology, the data processing and computing capabilities are migrated from the cloud (or data center) to a location closer to the terminal device, that is, the edge of the network. The computationally intensive tasks are offloaded from the terminal devices of the distribution network to the computing nodes (edge servers) closer to the data source for execution, which can greatly shorten the distance of data transmission in the network and thus reduce the transmission delay. At the same time, migrating some computing tasks to the edge server for execution can also relieve the computing pressure on the terminal device, save energy consumption, and extend the service life of the device.

[0003] However, due to the wide distribution of terminal devices and complex environmental control, problems such as node replication and attacks occur frequently, so it is particularly important to ensure the secure access authentication of various terminal devices. How to provide a cloud-edge-end collaboration method for multi-service heterogeneous resources in a new type of distribution network to solve the difficulties existing in the prior art is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a cloud-edge-end collaboration method and system for multi-service heterogeneous resources in a new type of distribution network, which can reduce the total task cost and improve the success rate of task execution.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A cloud-edge-end collaboration method for multi-service heterogeneous resources in a new type of distribution network includes the following steps:

[0007] Determine the corresponding terminal based on the type of service to be processed, select consensus nodes using the reputation value of the terminal nodes; determine the master node of the terminal from the consensus nodes using a verifiable random function; set the execution mode of the master node of the terminal and then incorporate the master node of the terminal into the distribution network to achieve real-time communication with the distribution substation;

[0008] Establish a cloud-edge-end collaboration network for the distribution main station, the distribution substation, and the terminal where the master node is located; establish a calculation model for the cloud-edge-end collaboration network; use a deep reinforcement learning algorithm to optimize the calculation model under the set optimization conditions to obtain an optimized calculation model;

[0009] Input the business to be processed into the optimized computing model to output the corresponding offloading strategy, and execute the offloading strategy.

[0010] Further, the process of determining the reputation value of the terminal node includes:

[0011] R i = λA i + μB i + γC i ;

[0012] Wherein, R i represents the reputation value of the i-th terminal node; λ is the weight of node activity in the reputation value; A i is the node activity of the i-th terminal node; μ is the weight of historical influence in the reputation value; B i is the historical influence of the i-th terminal node; C i is the consensus reward and punishment degree of the i-th terminal node; γ is the weight of the consensus reward and punishment degree in the reputation value;

[0013] A i = e aT ;

[0014] Wherein, a is the node activity influence factor; T is the consensus contribution rate;

[0015]

[0016] Wherein, v(Δt) is the time decay degree; w(h) is the block height; t is the total number of consensus rounds; h is the height of the block generated by the node's last participation in consensus for the next block;

[0017] C i = 2H + 2 2 I;

[0018] Wherein, H represents whether the node participates in consensus, and I represents the consensus result given by the node.

[0019] Further, the process of selecting consensus nodes using the reputation value of the terminal node includes:

[0020] Use the weighted random sampling algorithm to traverse the terminal nodes to obtain consensus nodes, and the weighted random sampling algorithm makes the probability of selecting a terminal node with a larger reputation value as a consensus node greater.

[0021] Further, after selecting the consensus nodes, it also includes formulating a consistency protocol to determine whether the nodes reach system consensus. If so, update the node reputation value; if not, retain the failure record.

[0022] Further, the process of determining the master node of the terminal from the consensus nodes using a verifiable random function includes:

[0023] Introduce a verifiable random function into the blockchain to select some consensus nodes, and generate a random number and a proof for any input and private key;

[0024] Randomly select terminal nodes, calculate the nodes whose input random numbers are less than the threshold and verify them. If the verification passes, they are determined as the main nodes of the terminal.

[0025] Furthermore, the execution modes of the main nodes of the terminal include local execution, edge execution, and cloud execution.

[0026] Furthermore, the optimization conditions include: under the premise of meeting the maximum latency tolerance and maximum energy consumption tolerance, the overall latency is minimized and the energy consumption cost is minimized.

[0027] Furthermore, the deep reinforcement learning algorithm adopts the FP-DDPG algorithm; the FP-DDPG algorithm includes an input layer, a task clustering layer, a DDPG layer, a prioritized experience replay mechanism layer, and an output layer;

[0028] The input layer is used to receive a set of computing tasks;

[0029] The task clustering layer is used to classify different computing tasks in the set of computing tasks according to task characteristics;

[0030] The DDPG layer is used to train the deep reinforcement learning algorithm with the classified computing tasks to generate corresponding offloading parameters and determine whether they are the optimal offloading parameters;

[0031] The prioritized experience replay mechanism layer is used to store the experiences of the DDPG layer and set priorities for the experiences;

[0032] The output layer is used to output the optimal offloading parameters.

[0033] Furthermore, during the execution of the task clustering layer: the fuzzy clustering algorithm is used to classify different computing tasks according to task characteristics; the task characteristics include task data volume, task latency requirements, and task energy consumption requirements.

[0034] The present invention also proposes a cloud-edge-end collaborative system for multi-service heterogeneous resources in a new type of distribution network, including a preprocessing module, an optimization module, and an output module;

[0035] The preprocessing module is used to determine the corresponding terminal based on the service type to be processed, select consensus nodes using the reputation values of terminal nodes; determine the main nodes of the terminal from the consensus nodes using a verifiable random function; after setting the execution modes of the main nodes of the terminal, incorporate the main nodes of the terminal into the distribution network to achieve real-time communication with the distribution substation;

[0036] The optimization module is used to establish a cloud-edge-end collaborative network for the distribution main station, the distribution substation, and the terminal where the main node is located; establish a computational model for the cloud-edge-end collaborative network; use a deep reinforcement learning algorithm to optimize the computational model under the set optimization conditions to obtain an optimized computational model;

[0037] The output module is used to input the service to be processed into the optimized computational model to output the corresponding offloading strategy, and execute the offloading strategy.

[0038] The effects provided in the invention content are only the effects of the embodiments, rather than all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

[0039] The present invention proposes a cloud-edge-end collaborative method and system for multi-service heterogeneous resources in a new type of distribution network. The method includes the following steps: determining the corresponding terminal based on the service type to be processed, and selecting consensus nodes using the reputation value of the terminal nodes; determining the main node of the terminal from the consensus nodes using a verifiable random function; after setting the execution mode of the main node of the terminal, incorporating the main node of the terminal into the distribution network to achieve real-time communication with the distribution substation; establishing a cloud-edge-end collaborative network for the distribution main station, the distribution substation, and the terminal where the main node is located; establishing a computational model for the cloud-edge-end collaborative network; using a deep reinforcement learning algorithm to optimize the computational model under the set optimization conditions to obtain an optimized computational model; inputting the service to be processed into the optimized computational model to output the corresponding offloading strategy, and executing the offloading strategy. Based on the cloud-edge-end collaborative method for multi-service heterogeneous resources in a new type of distribution network, a cloud-edge-end collaborative system for multi-service heterogeneous resources in a new type of distribution network is also proposed. The present invention reduces the total task cost and improves the success rate of task execution.

[0040] The present invention uses the terminal reputation value as the probability basis for extracting consensus nodes, accurately reducing the authentication scale, preventing attackers from forging puppet nodes to change the consensus result, and effectively enhancing the anti-attack ability of the system.

[0041] The method of selecting the main node based on the verifiable random function in the present invention ensures the security of the main node, avoids wasting time resources by starting the view change protocol, and effectively improves the authentication efficiency. Brief Description of the Drawings

[0042] Figure 1 It is a flowchart of the cloud-edge-end collaborative method for multi-service heterogeneous resources in a new type of distribution network proposed in Embodiment 1 of the present invention;

[0043] Figure 2 It is a schematic diagram of the cloud-edge-end collaborative system for multi-service heterogeneous resources in a new type of distribution network proposed in Embodiment 2 of the present invention. Detailed Description of the Invention

[0044] To clearly illustrate the technical features of this solution, the present invention will be described in detail below through specific embodiments and in conjunction with its accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, the components and settings of specific examples are described below. In addition, the present invention may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. The present invention omits the description of well-known components and processing technologies and processes to avoid unnecessarily limiting the present invention.

[0045] Embodiment 1

[0046] Embodiment 1 of the present invention proposes a cloud-edge-end collaborative method for heterogeneous resources of multiple services in a new type of distribution network to solve the technical problems existing in the security access authentication of various terminal devices in the prior art. Figure 1 It is a flowchart of the cloud-edge-end collaborative method for heterogeneous resources of multiple services in a new type of distribution network proposed in Embodiment 1 of the present invention;

[0047] In step S1, determine the corresponding terminal based on the service type to be processed, select consensus nodes using the reputation value of the terminal node; use a verifiable random function to determine the primary node of the terminal from the consensus nodes.

[0048] First, obtain the distribution main station, distribution sub-station, and the service to be processed, and determine the corresponding terminal based on the service type to be processed.

[0049] The terminal reputation value consists of node activity, historical influence, and consensus rewards and punishments with different weights; the reputation value represents the credibility of the node. The greater the reputation value, the higher the credibility of the node. Common methods for setting the initial reputation value are: 1) Set a fixed value; 2) Related to specific terminal parameters, such as the number of failures, service type, etc.; 3) Allocate a part of the reputation value from each node in the same proportion as the initial reputation value of the new node; 4) The average value of the reputation values of all current nodes.

[0050] The process of determining the reputation value of the terminal node includes:

[0051] R i = λA i + μB i + γC i ;

[0052] where, R i represents the reputation value of the i-th terminal node; λ is the weight of node activity in the reputation value; A iis the node activity of the i-th terminal node, that is, the frequency of participating in consensus within a unit time, reflecting the activity degree of node i; μ is the weight of the historical influence degree in the reputation value; B i is the historical influence degree of the i-th terminal node, which refers to the influence of the historical behavior of node i on the current reputation value; C i is the consensus reward and punishment degree of the i-th terminal node; γ is the weight of the consensus reward and punishment degree in the reputation value; λ, μ, and γ can adjust the influence degree of the three on the node reputation value according to actual applications.

[0053] The calculation process of the node activity includes:

[0054] A i = e aT ;

[0055] where a is the node activity influence factor, used to adjust the growth rate of A i ; T is the consensus contribution rate.

[0056] The calculation process of the historical influence degree includes:

[0057]

[0058] where v(Δt) is the time decay degree, used to measure the influence of the time required for the consensus process on the reputation value, Δt is the time interval between the current update time and the time when the node last participated in the consensus; λ is the parameter for adjusting the influence degree, as Δt increases, v(Δt) decreases, indicating that the longer the time interval, the smaller the influence on the current reputation value of the node; w(h) is the block height, used to measure the influence of the block height where the node is located on the reputation value; t is the total number of consensus rounds; h is the height of the block generated by the node's last participation in the consensus.

[0059] The calculation process of the consensus reward and punishment degree includes:

[0060] C i = 2H + 2 2 I;

[0061] where H represents whether the node participates in the consensus, I represents the consensus result given by the node. When node i makes a correct consensus, that is, the returned result is the same as that of most nodes, it gets a certain reward and makes a positive contribution to the reputation value; when i makes an incorrect consensus, that is, the returned result is different from that of most nodes, it is punished and makes a negative contribution to the reputation value.

[0062] The weighted random sampling algorithm is used to traverse the terminal nodes to obtain the consensus nodes, and the weighted random sampling algorithm makes the probability that the terminal node with a larger reputation value is selected as the consensus node greater.

[0063] In this application, m samples with the largest sampling values are selected to form a consensus group G. The probability of a node being selected is mainly based on the node's reputation value. Adding a random weight makes the probability of a node with a higher reputation value being selected greater, which can prevent the node with the highest reputation value from always being elected as the consensus node and ensure that the distributed performance of the system is not weakened. The weighted random sampling algorithm A-Res traverses the terminal set, and in each round of the loop, it is necessary to find the minimum value of the sampling values in the updated result set. Finally, the result is the consensus node.

[0064] The process of determining the master node of a terminal from the consensus nodes using a verifiable random function includes: introducing the verifiable random function into the blockchain to select some consensus nodes, generating a random number and a proof for any input and private key; randomly selecting terminal nodes, calculating the nodes whose input random numbers are less than the threshold and verifying. If the verification passes, it is determined as the master node of the terminal.

[0065] The detailed process of selecting the master node through VRF includes: generating a random number, a proof, and a threshold, randomly selecting terminal nodes to calculate that the input random numbers are less than the threshold, determining the terminal nodes as the master nodes and broadcasting. The nodes receiving the broadcast judge whether the random numbers are legal and verify whether the input random numbers are less than the threshold. If so, it is determined that this terminal node is the master node.

[0066] Specifically, introducing VRF into the blockchain to select some consensus nodes, for any input s and private key sk, generating a random number and a proof. Any holder of the public key can verify whether the random number is generated according to the input s. The entire process does not expose the node's private key, and malicious attackers cannot obtain the proof. Therefore, attackers cannot distinguish the difference between the output result of VRF and the random source, ensuring the randomness of the master node selection, and thus guaranteeing the security and stability of the entire system.

[0067] After selecting the consensus nodes, it also includes formulating a consistency protocol to determine whether the nodes reach a system consensus. If so, update the node reputation value; if not, retain the failure record.

[0068] In step S2, after setting the execution method of the master node of the terminal, incorporate the master node of the terminal into the distribution network to achieve real-time communication with the distribution substation.

[0069] Based on the successfully obtained sites, build a cloud-edge-terminal model of multiple terminal devices, edge servers, and a single cloud server. Each intelligent terminal device will generate a separate task. The terminal device communicates with the small base station in its service area through a wireless channel, while the base station communicates directly with the edge server through optical fiber. Therefore, the computing resources of the terminal device are limited; the computing resources of the edge server are relatively sufficient but also limited. Each task has three optional execution methods, namely: local execution, edge execution, and cloud execution.

[0070] Local execution means: when the task has relatively low requirements for computing resources, the task is executed on the terminal device; Edge execution means: when the computing resource requirements of the task are relatively large and the local terminal device cannot complete it in time, the task is offloaded to the local edge server for execution; Cloud execution means: when the computing resource requirements of the task are too large and the edge server is overloaded, the task is transmitted from the edge server to the cloud server for execution.

[0071] In step S3, a cloud-edge-terminal collaborative network of the distribution main station, the distribution substation, and the terminal where the main node is located is established; a computing model of the cloud-edge-terminal collaborative network is established; and a deep reinforcement learning algorithm is used to optimize the computing model under the set optimization conditions to obtain an optimized computing model.

[0072] After establishing the cloud-edge-terminal collaborative network of the distribution main station, the distribution substation, and the terminal where the main node is located in this application, the energy consumption of the internal terminals of the distribution network is determined, and the channel transmission rate during edge-terminal communication is set.

[0073] Since the terminal device tasks can be executed in three places: the cloud server, the edge server, and the terminal device, the task delay model is divided into a local computing model, an edge computing model, and a cloud computing model. The local computing model is used to obtain the local task processing time. In the edge computing model, the delay can be divided into computing delay and transmission delay. Because the execution results of the tasks usually have a small amount of data and the data is sent down at a fast speed, the return delay of the task execution results can usually be ignored. Based on the aforementioned channel transmission rate, the edge computing delay is obtained. The cloud computing model has sufficient computing resources and storage resources, so some waiting delays of the tasks do not need to be considered, and the cloud computing delay is obtained.

[0074] In the computing offloading of the distribution network, the edge server and the cloud server are usually directly connected to the power grid, and their energy supply is relatively stable. Therefore, there is no need to pay attention to the energy consumption of the edge server and the cloud server. However, the energy consumption of the terminal device directly affects the device performance. Therefore, when in local computing, the energy consumption of the device itself is used as the local computing energy consumption. When the edge server is computing, it is divided into the energy consumption spent on offloading tasks and the standby energy consumption of the terminal device, and the edge server computing energy consumption is obtained comprehensively. When the cloud server is computing, it is divided into the transmission energy consumption of the terminal device and the standby energy consumption of the terminal device, and the cloud server computing energy consumption is obtained comprehensively.

[0075] It is judged that the task success rate is to minimize the total cost of the overall delay and energy consumption on the premise of meeting the maximum delay tolerance and the maximum energy consumption tolerance.

[0076] The deep reinforcement learning algorithm adopts the FP-DDPG algorithm; the FP-DDPG algorithm includes an input layer, a task clustering layer, a DDPG layer, a prioritized experience replay mechanism layer, and an output layer;

[0077] The input layer is used to receive the computing task set;

[0078] The task clustering layer is used to classify different computing tasks in the computing task set according to task characteristics;

[0079] The DDPG layer is used to train the deep reinforcement learning algorithm with the classified computing tasks to generate corresponding offloading parameters and determine whether they are optimal offloading parameters;

[0080] The prioritized experience replay mechanism layer is used to store the experiences of the DDPG layer and set priorities for the experiences;

[0081] The output layer is used to output the optimal offloading parameters.

[0082] Since the data volume of each task to be processed is different, and different tasks have different latency requirements and energy consumption requirements, offloading so many extremely different tasks simultaneously will greatly increase the risk of distribution network collapse. At the same time, if all tasks are simply mixed together and offloaded simultaneously, it may cause some edge servers to be occupied for a long time or even overloaded, and some tasks are difficult to process in a timely manner, ultimately affecting the normal operation of the distribution network. Therefore, a task clustering layer is set up to cluster different computing tasks using the fuzzy clustering algorithm and divide them into different clusters according to the characteristics of the tasks. The process is as follows: determine the number of clusters as c, randomly select c tasks as the clustering centers of the clusters, initialize the clustering centers of the tasks, determine the energy consumption and latency centers of the clusters, set the fuzzy parameters and convergence thresholds, calculate the membership degrees of the tasks to the clustering centers and update the membership matrix, calculate the change amount of each clustering center to determine whether it converges, determine the clusters, and obtain the tasks after clustering.

[0083] The DDPG layer includes an approximate policy network and an action value network; the approximate policy network is used to map a specific state to a specific action; the action value network is used to evaluate the expected return of executing an action in a specific state.

[0084] The prioritized experience replay mechanism layer divides the experience replay pool into several fixed-length sizes, independently samples equal experience data from each partition, adds information entropy to the objective function, and assigns higher priority values to more valuable experiences to accelerate the convergence speed and stability of the network.

[0085] In step S4, the service to be processed is input into the optimized computing model to output the corresponding offloading strategy, and the offloading strategy is executed.

[0086] In Embodiment 1 of the present invention, a cloud-edge-terminal collaboration method for multi-service heterogeneous resources in a new type of distribution network reduces the total task cost and improves the success rate of task execution. By using the terminal reputation value as the probability basis for extracting consensus nodes, the authentication scale is accurately reduced, preventing attackers from forging puppet nodes to change the consensus result, and effectively enhancing the anti-attack ability of the system. The method of selecting the primary node based on the verifiable random function ensures the security of the primary node, avoids wasting time resources by starting the view change protocol, and effectively improves the authentication efficiency.

[0087] Embodiment 2

[0088] Based on the cloud-edge-terminal collaboration method for multi-service heterogeneous resources in a new type of distribution network proposed in Embodiment 1 of the present invention, Embodiment 2 of the present invention also proposes a cloud-edge-terminal collaboration system for multi-service heterogeneous resources in a new type of distribution network. Figure 2 As shown in the schematic diagram of the cloud-edge-terminal collaboration system for multi-service heterogeneous resources in a new type of distribution network proposed in Embodiment 2 of the present invention, the system includes: a preprocessing module, an optimization module, and an output module.

[0089] The preprocessing module is used to determine the corresponding terminal based on the service type to be processed, select consensus nodes using the reputation value of the terminal node; determine the primary node of the terminal from the consensus nodes using the verifiable random function; set the execution mode of the primary node of the terminal and then incorporate the primary node of the terminal into the distribution network to achieve real-time communication with the distribution substation.

[0090] The optimization module is used to establish a cloud-edge-terminal collaboration network among the distribution main station, the distribution substation, and the terminal where the primary node is located; establish a calculation model of the cloud-edge-terminal collaboration network; use the deep reinforcement learning algorithm to optimize the calculation model under the set optimization conditions to obtain an optimized calculation model.

[0091] The output module is used to input the service to be processed into the optimized calculation model to output the corresponding offloading strategy and execute the offloading strategy.

[0092] In the preprocessing module of the present application, the process of determining the reputation value of the terminal node includes:

[0093] R i =λA i +μB i +γC i ;

[0094] where, R i represents the reputation value of the i-th terminal node; λ is the weight of the node activity in the reputation value; A i is the node activity of the i-th terminal node; μ is the weight of the historical influence in the reputation value; B i is the historical influence of the i-th terminal node; C iis the consensus reward and punishment degree of the i-th terminal node; γ is the weight of the consensus reward and punishment degree in the credit value;

[0095] A i = e aT ;

[0096] where a is the node activity influence factor; T is the consensus contribution rate;

[0097]

[0098] where v(Δt) is the time decay degree; w(h) is the block height; t is the total number of consensus rounds; h is the height of the block generated by the node's most recent participation in consensus for the next block;

[0099] C i = 2H + 2 2 I;

[0100] where H represents whether the node participates in consensus, and I represents the consensus result given by the node.

[0101] The process of selecting consensus nodes using the credit value of terminal nodes includes: traversing terminal nodes using the weighted random sampling algorithm to obtain consensus nodes, and the weighted random sampling algorithm makes the probability of a terminal node with a larger credit value being selected as a consensus node greater.

[0102] After selecting consensus nodes, it also includes formulating a consistency protocol to determine whether the nodes reach a system consensus. If so, update the node credit value; if not, retain a failure record.

[0103] The process of determining the master node of the terminal from consensus nodes using a verifiable random function includes: introducing the verifiable random function into the blockchain to select some consensus nodes, generating a random number and a proof for any input and private key; randomly selecting terminal nodes, calculating the nodes with an input random number less than the threshold and verifying. If the verification passes, it is determined as the master node of the terminal.

[0104] The execution methods of the master node of the terminal include local execution, edge execution, and cloud execution.

[0105] The optimization conditions in the optimization module include: under the premise of meeting the maximum delay tolerance and maximum energy consumption tolerance, the overall delay is minimized and the energy consumption cost is minimized.

[0106] The deep reinforcement learning algorithm uses the FP-DDPG algorithm; the FP-DDPG algorithm includes an input layer, a task clustering layer, a DDPG layer, a prioritized experience replay mechanism layer, and an output layer;

[0107] The input layer is used to receive the set of computing tasks;

[0108] The task clustering layer is used to classify different computing tasks in the computing task set according to task characteristics; during the execution of the task clustering layer: a fuzzy clustering algorithm is used to classify different computing tasks according to task characteristics; the task characteristics include task data volume, task latency requirements, and task energy consumption requirements;

[0109] The DDPG layer is used to train the deep reinforcement learning algorithm with the classified computing tasks to generate corresponding offloading parameters and determine whether they are the optimal offloading parameters;

[0110] The prioritized experience replay mechanism layer is used to store the experiences of the DDPG layer and set priorities for the experiences;

[0111] The output layer is used to output the optimal offloading parameters.

[0112] In Embodiment 2 of the present invention, a cloud-edge-terminal collaborative system for multi-service heterogeneous resources in a new type of distribution network reduces the total task cost and improves the success rate of task execution. Using the terminal reputation value as the probability basis for extracting consensus nodes, it accurately reduces the authentication scale, prevents attackers from forging puppet nodes to change the consensus result, and effectively improves the anti-attack ability of the system. The method of selecting the primary node based on the verifiable random function ensures the security of the primary node, avoids wasting time resources by starting the view change protocol, and effectively improves the authentication efficiency.

[0113] For the description of the relevant parts in the cloud-edge-terminal collaborative system for multi-service heterogeneous resources in a new type of distribution network provided in Embodiment 2 of this application, reference can be made to the detailed description of the corresponding parts in the cloud-edge-terminal collaborative method for multi-service heterogeneous resources in a new type of distribution network provided in Embodiment 1 of this application, which will not be elaborated here.

[0114] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes the inherent elements thereof. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element. In addition, the parts of the above technical solutions provided in the embodiments of this application that are consistent with the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0115] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. For those skilled in the art, other different forms of modifications or deformations can be made on the basis of the above description. It is not necessary and impossible to enumerate all the implementation manners here. Based on the technical solution of the present invention, various modifications or deformations that can be made by those skilled in the art without creative labor are still within the protection scope of the present invention.

Claims

1. Cloud-edge-end collaborative method for multi-service heterogeneous resources in a new-type distribution network, characterized in that, It includes the following steps: Determine the corresponding terminal based on the business type to be processed, and select consensus nodes by using the reputation value of the terminal nodes; determine the primary node of the terminal from the consensus nodes by using a verifiable random function; After setting the execution mode of the primary node of the terminal, incorporate the primary node of the terminal into the distribution network to achieve real-time communication with the distribution substation; Establish a cloud-edge-terminal collaborative network for the distribution main station, the distribution substation, and the terminal where the primary node is located; Establish a computational model for the cloud-edge-terminal collaborative network; use a deep reinforcement learning algorithm to optimize the computational model under the set optimization conditions to obtain an optimized computational model; Input the business to be processed into the optimized computational model to output the corresponding offloading strategy, and execute the offloading strategy.

2. The cloud-edge-terminal collaborative method for multi-service heterogeneous resources in the new type of distribution network according to claim 1, wherein The process of determining the reputation value of the terminal nodes includes: R i = λA i + μB i + γC i ; Among them, R i represents the reputation value of the i-th terminal node; λ is the weight of the node activity in the reputation value; A i is the node activity of the i-th terminal node; μ is the weight of the historical influence in the reputation value; B i is the historical influence of the i-th terminal node; C i is the consensus reward and punishment degree of the i-th terminal node; γ is the weight of the consensus reward and punishment degree in the reputation value; A i = e aT ; where a is the node activity influence factor; T is the consensus contribution rate; where v(Δt) is the time decay degree; w(h) is the block height; t is the total number of consensus rounds; h is the height of the block generated by the node's most recent participation in consensus; C i = 2H + 2 2 I; where H represents whether the node participates in consensus, and I represents the consensus result given by the node.

3. The cloud-edge-end collaborative method for multi-service heterogeneous resources in the new type of distribution network according to claim 1, characterized in that, The process of selecting consensus nodes by using the reputation value of the terminal nodes includes: Traverse the terminal nodes by using the weighted random sampling algorithm to obtain consensus nodes, and the weighted random sampling algorithm makes the probability of a terminal node with a larger reputation value being selected as a consensus node greater.

4. The cloud-edge-end collaborative method for multi-service heterogeneous resources in the new-type distribution network according to claim 3, wherein After selecting the consensus nodes, it also includes formulating a consistency protocol to determine whether the nodes reach system consensus. If so, update the node reputation value; if not, retain the failure record.

5. The cloud-edge-end collaborative method for multi-service heterogeneous resources in the new-type distribution network according to claim 1, wherein The process of determining the primary node of the terminal from the consensus nodes by using a verifiable random function includes: Introduce a verifiable random function into the blockchain to select some consensus nodes, and generate a random number and a proof for any input and private key; Randomly select terminal nodes, calculate the nodes whose input random numbers are less than the threshold and verify them. If the verification passes, it is determined as the primary node of the terminal.

6. The cloud-edge-terminal collaborative method for multi-service heterogeneous resources in the new-type distribution network according to claim 1, characterized in that The execution modes of the primary node of the terminal include local execution, edge execution, and cloud execution.

7. The cloud-edge-end collaborative method for multi-service heterogeneous resources in the new-type distribution network according to claim 1, wherein The optimization conditions include: under the premise of meeting the maximum delay tolerance and the maximum energy consumption tolerance, the overall delay is minimized and the energy consumption cost is minimized.

8. The cloud-edge-terminal collaborative method for multi-service heterogeneous resources in the new-type distribution network according to claim 1, wherein The deep reinforcement learning algorithm uses the FP-DDPG algorithm; the FP-DDPG algorithm includes an input layer, a task clustering layer, a DDPG layer, a prioritized experience replay mechanism layer, and an output layer; The input layer is used to receive the set of computing tasks; The task clustering layer is used to classify different computing tasks in the set of computing tasks according to task characteristics; The DDPG layer is used to train the deep reinforcement learning algorithm by using the classified computing tasks to generate corresponding offloading parameters and determine whether they are the optimal offloading parameters; The prioritized experience replay mechanism layer is used to store the experiences of the DDPG layer and set priorities for the experiences; The output layer is used to output the optimal offloading parameters.

9. The cloud-edge-terminal collaborative method for multi-service heterogeneous resources for a new type of distribution network according to claim 1, characterized in that During the execution of the task clustering layer: use the fuzzy clustering algorithm to classify different computing tasks according to task characteristics; the task characteristics include task data volume, task delay requirement, and task energy consumption requirement.

10. A cloud-edge-end collaborative system for heterogeneous resources of multiple services in a new-type distribution network, characterized in that, It includes a preprocessing module, an optimization module, and an output module; The preprocessing module is used to determine the corresponding terminal based on the service type to be processed, select consensus nodes using the reputation values of the terminal nodes; determine the primary node of the terminal from the consensus nodes using a verifiable random function; After setting the execution mode of the primary node of the terminal, incorporate the primary node of the terminal into the distribution network to achieve real-time communication with the distribution substation; The optimization module is used to establish a cloud-edge-end collaborative network for the distribution main station, distribution substation, and the terminal where the primary node is located; Establish a computational model for the cloud-edge-end collaborative network; use a deep reinforcement learning algorithm to optimize the computational model under the set optimization conditions to obtain an optimized computational model; The output module is used to input the service to be processed into the optimized computational model to output the corresponding offloading strategy, and execute the offloading strategy.

Images