A method and system for edge computing optical network collaboration in distribution network

Abstract

The present invention discloses an edge computing optical network collaboration method and system for a distribution network, specifically comprising: constructing an edge computing optical network system architecture for the distribution network, constructing a collaborative model of edge computing of the distribution network based on the network architecture, proposing a computing power collaboration and communication collaboration method between multiple edge computing nodes in the edge computing optical network, designing an edge computing optical network node collaborative training process based on the Q-Learning algorithm model, and improving the efficient processing of distributed services and flexible deployment of applications of the edge computing system of the distribution Internet of Things.

Description

Technical Field

[0001] The patent of this invention relates to the field of power Internet of Things, specifically a method and system for edge computing optical network collaboration in a distribution network. Background Art

[0002] With the rapid development of the power Internet of Things, the explosive growth of data volume far exceeds the bearing capacity of network bandwidth, and the emergence of a large number of new intelligent applications has put forward higher requirements for system latency performance. Edge computing, as a new computing mode, can enable data to be processed promptly and effectively in the edge network near the source. In the edge computing mode, the terminal offloads tasks to nearby edge computing devices for execution, effectively alleviating the network congestion and high latency problems caused by transmitting all data to the data center in traditional cloud computing. The bottleneck of edge computing lies mainly in its limited computing, communication, storage and other resources. When the number of terminals and the scale of data increase further, the edge nodes tend to be saturated, resulting in a serious decline in service quality.

[0003] The access scale of terminals, data and services in the current power Internet of Things is constantly growing, and massive power data needs to be analyzed and calculated on edge computing devices to achieve efficient and flexible business processing and decision-making. Due to the distributed nature of new power system services, the working mode of edge computing node devices has also changed from a single-machine independent operation mode to a distributed collaborative mode of multiple edge computing nodes. Therefore, the growing distributed business deployment urgently needs to solve the problem of collaborative operation between multiple edge computing nodes, and the computing power collaboration between "edge-to-edge" is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The existing technology has an urgent need to solve the problem of coordinated operation between multiple edge computing nodes and computing power coordination between "edge-to-edge" in the face of the growing distributed business deployment.

[0005] In order to solve the above problems, the present invention proposes an edge computing optical network collaboration method for a distribution network, comprising:

[0006] S1: Build an edge computing optical network system architecture for distribution networks;

[0007] The edge computing optical network system architecture for distribution networks is mainly composed of multiple distributed edge computing nodes, multiple ONU nodes and one OLT device; Node represents the edge computing terminal, which performs distributed collaborative computing on the services; ONU represents the ONU node of the passive optical network, and each ONU is equipped with an edge computing terminal; OLT represents the OLT device of the passive optical network, which has an embedded communication and computing power collaboration module of the edge computing optical network, and is mainly responsible for the collection of computing power and communication requirements, the training of computing power and communication collaboration algorithms, and the dynamic scheduling and allocation of computing power and communication resources;

[0008] Based on the above architecture, the specific steps of this method are as follows:

[0009] S1-1: Communication and computing power coordination model of edge computing optical network;

[0010] S1-1-1: The computing power requirement of edge computing Node_i is:

[0011]

[0012] Among them, Zp is the data volume of power distribution service p, λp is the arrival rate of power distribution service p, that is, the number of services arriving per unit time, P APP It is the set of all power distribution services connected to edge computing Node_i;

[0013] S1-1-2: Define the communication and computing power coordination model between edge computing nodes i and j as:

[0014]

[0015] Among them, T total is the total delay of communication and computation, T tr is the transmission delay between edge computing node i and edge computing node j in the optical network, Z i is the computing power requirement of the computing task of edge computing node i, C edge is the collaborative computing resource allocated to node i by edge computing node j, ω j is the ratio of collaborative computing resources allocated from computing node j to computing node i, C j is the available computing power resource of edge computing node j, and t is the tth time slot;

[0016] S2: Define the edge computing optical network collaborative working mode of the distribution network;

[0017] According to the power distribution service collaboration requirements of edge computing optical networks, three collaborative working modes of edge computing optical networks are defined, namely, collaboration between edge computing master nodes and slave nodes: 1:1 collaboration, 1:M collaboration, and N:M collaboration;

[0018] When edge computing node i accesses a single type of service and the service computing volume is less than 2C j When edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C, a 1:1 collaborative working mode can be adopted; when edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C j When multiple edge computing nodes are connected to multiple types of concurrent services, the 1:M collaborative working mode can be adopted; when multiple edge computing nodes are connected to multiple types of concurrent services, the N:M collaborative working mode can be adopted;

[0019] S2-1: Define 1:1 collaboration mode: The edge computing master node i selects one edge computing node j as a slave node to perform computing power collaboration:

[0020]

[0021] Among them, Ttask is the task delay of communication and computing, I=1 indicates that the edge computing task is processed successfully, and I=0 indicates that the edge computing task fails to be processed. If it is 0, the computing task fails to be processed, and the next edge computing task is processed without changing to other collaborative modes;

[0022] S2-2: Definition 1: M collaboration mode: The edge computing node selects M edge computing nodes as slave nodes to perform computing power collaboration:

[0023]

[0024] S2-3: Define N:M collaboration mode: N edge computing nodes are used as master nodes, and M edge computing nodes are selected as slave nodes to coordinate computing power:

[0025]

[0026] S3: Edge computing optical network collaboration method;

[0027] S3-1: Edge computing optical network collaboration based on Q-Learning algorithm. First, set the triple {S, A, R}, where S is the state space, A is the action space, and R is the reward function;

[0028] S3-1-1: State space S: Assume the system state at time t is:

[0029]

[0030] Among them, Zn is the computing power requirement of the Nth business;

[0031] S3-1-2: Action space A: Set the action at time t to:

[0032]

[0033] S3-1-3: Reward function R: Set the reward as the number of tasks successfully processed at a time:

[0034]

[0035] S3-1-4: The value of taking action a in state s is recorded as Q{s,a}, which is initialized to 0. During operation, the value of each state-action pair is updated according to formula 9:

[0036]

[0037] Set the cycle time T, count the number of tasks successfully processed in all time slots of this cycle, and set the MDP goal to the strategy of finding the maximum number of successful tasks processed in this cycle according to the total number of tasks in this cycle, that is, A range of Q values ​​corresponding to the maximum value.

[0038] In the preferred mode, edge computing edge-edge collaborative Q-Learning model training:

[0039] S1: Initialization: Initialize the learning rate α, which is used to control the amplitude of the algorithm update, and the discount rate β, which is a parameter used to measure the relative importance of future rewards. The learning rate and discount rate remain unchanged during algorithm training. The initial value of the Q table is 0, that is, the initial value of all state-action pairs is set to 0;

[0040] S2: At each time slot t within the training period Ts, obtain the current system state s(t);

[0041] S3: Randomly select an action a(t);

[0042] S4: Calculate the reward R(t+1) and observe the next system state s(t+1);

[0043] S5: Update Q(s(t), a(t)) according to formula 9;

[0044] S6: The training of this cycle Ts ends.

[0045] An edge computing optical network collaborative system for a distribution network includes: a distributed edge computing node module, an ONU module, an OLT device module, and a communication and computing power collaborative module, specifically including:

[0046] M1: Building edge computing optical network system architecture for distribution networks;

[0047] The edge computing optical network system architecture for distribution networks mainly consists of multiple distributed edge computing node modules, multiple ONU modules and one OLT device module; Node represents the edge computing terminal, which performs distributed collaborative computing on services; ONU represents the ONU node of the passive optical network, and each ONU is equipped with an edge computing terminal; OLT represents the OLT device of the passive optical network, which has an embedded communication and computing power collaboration module of the edge computing optical network, and is mainly responsible for the collection of computing power and communication requirements, the training of computing power and communication collaboration algorithms, and the dynamic scheduling and allocation of computing power and communication resources;

[0048] M1-1: Communication and computing power coordination model of edge computing optical network;

[0049] M1-1-1: The computing power requirement of edge computing Node_i is:

[0050]

[0051] Among them, Zp is the data volume of power distribution service p, λp is the arrival rate of power distribution service p, that is, the number of services arriving per unit time, P APP It is the set of all power distribution services connected to edge computing Node_i;

[0052] M1-1-2: Define the communication and computing power coordination model between edge computing nodes i and j as:

[0053]

[0054] Among them, T total is the total delay of communication and computation, T tr is the transmission delay between edge computing node i and edge computing node j in the optical network, Z i is the computing power requirement of the computing task of edge computing node i, C edge is the collaborative computing resource allocated to node i by edge computing node j, ω j is the ratio of collaborative computing resources allocated from computing node j to computing node i, C j is the available computing power resource of edge computing node j, and t is the tth time slot;

[0055] M2: Defines the edge computing optical network collaborative working mode of the distribution network;

[0056] According to the power distribution service collaboration requirements of edge computing optical networks, three collaborative working modes of edge computing optical networks are defined, namely, collaboration between edge computing master nodes and slave nodes: 1:1 collaboration, 1:M collaboration, and N:M collaboration;

[0057] When edge computing node i accesses a single type of service and the service computing volume is less than 2C j When edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C, a 1:1 collaborative working mode can be adopted; when edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C j When multiple edge computing nodes are connected to multiple types of concurrent services, the 1:M collaborative working mode can be adopted; when multiple edge computing nodes are connected to multiple types of concurrent services, the N:M collaborative working mode can be adopted;

[0058] M2-1: Define 1:1 collaboration mode: The edge computing master node i selects one edge computing node j as a slave node to perform computing power collaboration:

[0059]

[0060] Among them, Ttask is the task delay of communication and computing, I=1 indicates that the edge computing task is processed successfully, and I=0 indicates that the edge computing task fails to be processed. If it is 0, the computing task fails to be processed, and the next edge computing task is processed without changing to other collaborative modes;

[0061] M2-2: Definition 1: M collaboration mode: The edge computing node selects M edge computing nodes as slave nodes to perform computing power collaboration:

[0062]

[0063] M2-3: Define N:M collaboration mode: N edge computing nodes are used as master nodes, and M edge computing nodes are selected as slave nodes to coordinate computing power:

[0064]

[0065] M3: Edge computing optical network collaboration method;

[0066] M3-1: Edge computing optical network collaboration based on Q-Learning algorithm. First, set the triple {S, A, R}, where S is the state space, A is the action space, and R is the reward function;

[0067] M3-1-1: State space S: Assume the system state at time t is:

[0068]

[0069] Among them, Zn is the computing power requirement of the Nth business;

[0070] M3-1-2: Action space A: Set the action at time t to:

[0071]

[0072] M3-1-3: Reward function R: Set the reward as the number of tasks successfully processed at a time:

[0073]

[0074] M3-1-4: The value of taking action a in state s is recorded as Q{s,a}, which is initialized to 0. During operation, the value of each state-action pair is updated according to formula 9:

[0075]

[0076] Set the cycle time T, count the number of tasks successfully processed in all time slots of this cycle, and set the MDP goal to the strategy of finding the maximum number of successful tasks processed in this cycle according to the total number of tasks in this cycle, that is, A series of Q values ​​corresponding to the maximum value, among which MPD is the abbreviation of Markov decision process. Q-Learing is a reinforcement learning algorithm that sets the maximum value strategy to improve the overall performance.

[0077] The present invention can achieve the following beneficial effects: solving the interdependence problem between high availability confirmation and business processing, reducing the internal processing delay of the cluster, business processing state independence, cluster state management seamless switching, and improving overall performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is the edge-to-edge collaborative system architecture of edge computing. DETAILED DESCRIPTION

[0079] S1: Build an edge computing optical network system architecture for distribution networks;

[0080] The edge computing optical network system architecture for distribution networks consists of multiple distributed edge computing Node nodes, multiple ONU nodes, and one OLT device; Node represents the edge computing terminal, which performs distributed collaborative computing on services; ONU represents the ONU node of the passive optical network, and each ONU is equipped with an edge computing terminal; OLT represents the OLT device of the passive optical network, which has an embedded communication and computing power collaboration module of the edge computing optical network, responsible for the collection of computing power and communication requirements, the training of computing power and communication collaboration algorithms, and the dynamic scheduling and allocation of computing power and communication resources;

[0081] Based on the above architecture, the specific steps of this method are as follows:

[0082] S1-1: Constructing a communication and computing power coordination model for edge computing optical networks;

[0083] S1-1-1: The computing power requirement of edge computing Node_i is:

[0084]

[0085] Among them, Zp is the data volume of power distribution service p, λp is the arrival rate of power distribution service p, that is, the number of services arriving per unit time, and PAPP is the set of all power distribution services connected to edge computing Node_i;

[0086] S1-1-2: Define the communication and computing power coordination model between edge computing nodes i and j as:

[0087]

[0088] Among them, T total is the total delay of communication and computation, T tris the transmission delay between edge computing node i and edge computing node j in the optical network, Z i is the computing power requirement of the computing task of edge computing node i, C edge is the collaborative computing resource allocated to node i by edge computing node j, ω j is the ratio of collaborative computing resources allocated from computing node j to computing node i, C j is the available computing power resource of edge computing node j, and t is the tth time slot;

[0089] S2: Define the edge computing optical network collaborative working mode of the distribution network;

[0090] According to the power distribution service collaboration requirements of edge computing optical networks, three collaborative working modes of edge computing optical networks are defined, namely, collaboration between edge computing master nodes and slave nodes: 1:1 collaboration, 1:M collaboration, and N:M collaboration;

[0091] When edge computing node i accesses a single type of service and the service computing volume is less than 2C j When edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C, a 1:1 collaborative working mode can be adopted; when edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C j When multiple edge computing nodes are connected to multiple types of concurrent services, the 1:M collaborative working mode can be adopted; when multiple edge computing nodes are connected to multiple types of concurrent services, the N:M collaborative working mode can be adopted;

[0092] S2-1: Define 1:1 collaboration mode: The edge computing master node i selects one edge computing node j as a slave node to perform computing power collaboration:

[0093]

[0094] Among them, Ttask is the task delay of communication and computing, I=1 indicates that the edge computing task is processed successfully, and I=0 indicates that the edge computing task fails to be processed. If it is 0, the computing task fails to be processed, and the next edge computing task is processed without changing to other collaborative modes;

[0095] S2-2: Definition 1: M collaboration mode: The edge computing node selects M edge computing nodes as slave nodes to perform computing power collaboration:

[0096] S2-3: Define N:M collaboration mode: N edge computing nodes are used as master nodes, and M edge computing nodes are selected as slave nodes to coordinate computing power:

[0097]

[0098] S3: Edge computing optical network collaboration method;

[0099] S3-1: Edge computing optical network collaboration based on Q-Learning algorithm. First, set the triple {S, A, R}, where S is the state space, A is the action space, and R is the reward function;

[0100] S3-1-1: State space S: Assume the system state at time t is:

[0101]

[0102] Among them, Zn is the computing power requirement of the Nth business;

[0103] S3-1-2: Action space A: Set the action at time t to:

[0104]

[0105] S3-1-3: Reward function R: Set the reward as the number of tasks successfully processed at a time:

[0106]

[0107] S3-1-4: The value of taking action a in state s is recorded as Q{s,a}, which is initialized to 0. During operation, the value of each state-action pair is updated according to formula 9:

[0108]

[0109] Set the cycle time T, count the number of tasks successfully processed in all time slots of this cycle, and set the MDP goal to the strategy of finding the maximum number of successful tasks processed in this cycle according to the total number of tasks in this cycle, that is, A range of Q values ​​corresponding to the maximum value.

[0110] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An edge computing optical network collaboration method for a distribution network, characterized in that: include: S1: Build an edge computing optical network system architecture for distribution networks; The edge computing optical network system architecture for distribution networks consists of multiple distributed edge computing Node nodes, multiple ONU nodes, and one OLT device; Node represents the edge computing terminal, which performs distributed collaborative computing on services; ONU represents the ONU node of the passive optical network, and each ONU is equipped with an edge computing terminal; OLT represents the OLT device of the passive optical network, which has an embedded communication and computing power collaboration module of the edge computing optical network, responsible for the collection of computing power and communication requirements, the training of computing power and communication collaboration algorithms, and the dynamic scheduling and allocation of computing power and communication resources; Based on the above architecture, the specific steps of this method are as follows: S1-1: Constructing a communication and computing power coordination model for edge computing optical networks; S1-1-1: The computing power requirement of edge computing Node_i is: Among them, Zp is the data volume of power distribution service p, λp is the arrival rate of power distribution service p, that is, the number of services arriving per unit time, P APP It is the set of all power distribution services connected to edge computing Node_i; S1-1-2: Define the communication and computing power coordination model between edge computing nodes i and j as: Among them, T total is the total delay of communication and computation, T tr is the transmission delay between edge computing node i and edge computing node j in the optical network, Z i is the computing power requirement of the computing task of edge computing node i, C edge is the collaborative computing resource allocated to node i by edge computing node j, ω j is the ratio of collaborative computing resources allocated from computing node j to computing node i, C j is the available computing power resource of edge computing node j, and t is the tth time slot; S2: Define the edge computing optical network collaborative working mode of the distribution network; According to the power distribution service collaboration requirements of edge computing optical networks, three collaborative working modes of edge computing optical networks are defined, namely, collaboration between edge computing master nodes and slave nodes: 1:1 collaboration, 1:M collaboration, and N:M collaboration; When edge computing node i accesses a single type of service and the service computing volume is less than 2C j When edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C, a 1:1 collaborative working mode is adopted; when edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C j When multiple edge computing nodes are connected to multiple types of concurrent services, the 1:M collaborative working mode is adopted; when multiple edge computing nodes are connected to multiple types of concurrent services, the N:M collaborative working mode is adopted; S2-1: Define 1:1 collaboration mode: The edge computing master node i selects one edge computing node j as a slave node to perform computing power collaboration: Among them, Ttask is the task delay of communication and computing, I=1 indicates that the edge computing task is successfully processed, and I=0 indicates that the edge computing task fails to be processed. If it is 0, the computing task fails to be processed, and the next edge computing task is processed without changing to other collaborative modes; S2-2: Definition 1: M collaboration mode: The edge computing node selects M edge computing nodes as slave nodes to perform computing power collaboration: S2-3: Define N:M collaboration mode: N edge computing nodes are used as master nodes, and M edge computing nodes are selected as slave nodes to coordinate computing power: S3: Edge computing optical network collaboration method; S3-1: Edge computing optical network collaboration based on Q-Learning algorithm. First, set the triple {S, A, R}, where S is the state space, A is the action space, and R is the reward function; S3-1-1: State space S: Assume the system state at time t is: Among them, Zn is the computing power requirement of the Nth business; S3-1-2: Action space A: Set the action at time t to: S3-1-3: Reward function R: Set the reward as the number of tasks successfully processed at a time: S3-1-4: The value of taking action a in state s is recorded as Q{s,a}, which is initialized to 0. During operation, the value of each state-action pair is updated according to formula 9: Formula 9: In the formula, α represents the learning rate, β represents the discount rate, and the cycle time T is set. The number of tasks successfully processed in all time slots of this cycle is counted. According to the total number of tasks in this cycle, the goal of MDP is set as the strategy to maximize the number of successful tasks processed in this cycle, that is, Q(s (t+1) ,a (t+1) )A series of Q values ​​corresponding to the maximum value.

2. The edge computing optical network collaboration method for distribution network according to claim 1 is characterized in that: Edge computing edge-to-edge collaborative Q-Learning model training: S1: Initialization: Initialize the learning rate α, which is used to control the amplitude of the algorithm update, and the discount rate β, which is a parameter used to measure the relative importance of future rewards. The learning rate and discount rate remain unchanged during algorithm training. The initial value of the Q table is 0, that is, the initial value of all state-action pairs is set to 0; S2: At each time slot t within the training period Ts, obtain the current system state s(t); S3: Randomly select an action a(t); S4: Calculate the reward R(t+1) and observe the next system state s(t+1); S5: Update Q(s(t), a(t)) according to formula 9; S6: The training of this cycle Ts is completed.

3. An edge computing optical network collaborative system for a distribution network, characterized in that: include: Distributed edge computing node module, ONU module, OLT device module, and communication and computing power coordination module, specifically including: M1: Building edge computing optical network system architecture for distribution networks; The edge computing optical network system architecture for distribution networks includes: multiple distributed edge computing Node modules, multiple ONU modules, and one OLT device module; Node represents an edge computing terminal that performs distributed collaborative computing on services; ONU represents an ONU node of a passive optical network, and each ONU is equipped with an edge computing terminal; OLT represents an OLT device of a passive optical network, with an embedded communication and computing power collaboration module of an edge computing optical network, responsible for the collection of computing power and communication requirements, the training of computing power and communication collaboration algorithms, and the dynamic scheduling and allocation of computing power and communication resources; M1-1: Communication and computing power coordination model of edge computing optical network; M1-1-1: The computing power requirement of edge computing Node_i is: Among them, Zp is the data volume of power distribution service p, λp is the arrival rate of power distribution service p, that is, the number of services arriving per unit time, P APP It is the set of all power distribution services connected to edge computing Node_i; M1-1-2: Define the communication and computing power coordination model between edge computing nodes i and j as: Among them, T total is the total delay of communication and computation, T tr is the transmission delay between edge computing node i and edge computing node j in the optical network, Z i is the computing power requirement of the computing task of edge computing node i, C edge is the collaborative computing resource allocated to node i by edge computing node j, ω j is the ratio of collaborative computing resources allocated from computing node j to computing node i, C j is the available computing power resource of edge computing node j, and t is the tth time slot; M2: Defines the edge computing optical network collaborative working mode of the distribution network; According to the power distribution service collaboration requirements of edge computing optical networks, three collaborative working modes of edge computing optical networks are defined, namely, collaboration between edge computing master nodes and slave nodes: 1:1 collaboration, 1:M collaboration, and N:M collaboration; When edge computing node i accesses a single type of service and the service computing volume is less than 2C j When edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C, a 1:1 collaborative working mode is adopted; when edge computing node i is connected to multiple types of services and the business computing volume is greater than 2C j When multiple edge computing nodes are connected to multiple types of concurrent services, the 1:M collaborative working mode is adopted; when multiple edge computing nodes are connected to multiple types of concurrent services, the N:M collaborative working mode is adopted; M2-1: Define 1:1 collaboration mode: The edge computing master node i selects one edge computing node j as a slave node to perform computing power collaboration: Among them, Ttask is the task delay of communication and computing, I=1 indicates that the edge computing task is successfully processed, and I=0 indicates that the edge computing task fails to be processed. If it is 0, the computing task fails to be processed, and the next edge computing task is processed without changing to other collaborative modes; M2-2: Definition 1: M collaboration mode: The edge computing node selects M edge computing nodes as slave nodes to perform computing power collaboration: M2-3: Define N:M collaboration mode: N edge computing nodes are used as master nodes, and M edge computing nodes are selected as slave nodes to coordinate computing power: M3: Edge computing optical network collaboration method; M3-1: Edge computing optical network collaboration based on Q-Learning algorithm. First, set the triple {S, A, R}, where S is the state space, A is the action space, and R is the reward function; M3-1-1: State space S: Assume the system state at time t is: Among them, Zn is the computing power requirement of the Nth business; M3-1-2: Action space A: Set the action at time t to: M3-1-3: Reward function R: Set the reward as the number of tasks successfully processed at a time: M3-1-4: The value of taking action a in state s is recorded as Q{s,a}, which is initialized to 0. During operation, the value of each state-action pair is updated according to formula 9: Formula 9: In the formula, α represents the learning rate, β represents the discount rate, and the cycle time T is set. The number of tasks successfully processed in all time slots of this cycle is counted. According to the total number of tasks in this cycle, the goal of MDP is set as the strategy to maximize the number of successful tasks processed in this cycle, that is, Q((s (t+1) ,a (t+1) )) A series of Q values ​​corresponding to the maximum value, setting the maximum value strategy to improve the overall performance.

Images