Cloud edge collaborative smart power grid operation and maintenance carbon reduction method based on alternating direction multiplier method

Abstract

The invention provides a cloud-side collaborative smart power grid operation and maintenance carbon reduction method based on an alternating direction multiplier method, and the method comprises the steps: designing a cloud-side collaborative power grid operation and maintenance system which comprises terminal equipment, a cloud computing center and an edge node; modeling is carried out on the related time delay of the cloud edge collaborative power grid operation and maintenance system; modeling the comprehensive carbon emission of the cloud edge collaborative power grid operation and maintenance system according to the modeling of the related time delay; and according to the modeling of the comprehensive carbon emission, an objective function is constructed for the carbon emission minimization problem of the cloud edge collaborative power grid operation and maintenance system, and a globally optimal solution is obtained by adopting an alternating direction multiplier method. According to the invention, effective technical support is provided for realizing low-carbon operation and carbon emission management of the power grid, and the comprehensive carbon emission is effectively reduced while the service time delay is reduced.

Description

Technical Field

[0001] The present invention relates to the field of cloud-edge collaboration and intelligent power grid operation and maintenance, and specifically to a cloud-edge collaborative intelligent power grid operation and maintenance carbon reduction method based on an alternating direction multiplier method. Background Art

[0002] In the current smart grid operation and maintenance field, there are already some methods that can reduce carbon emissions by optimizing the power structure and improving the level of intelligence. These methods usually rely on resource allocation or business scheduling to reduce the comprehensive energy consumption and carbon emissions of the power grid. Although the above work can effectively reduce network energy consumption and carbon emissions, however, these works have ignored different business needs. Faced with the huge amount of data generated by the power grid, its optimization and modeling methods have been difficult to meet the needs of large-scale power grid operation and maintenance.

[0003] With the rapid development of information technology and artificial intelligence technology, cloud-edge collaborative technology has gradually been introduced into the field of smart grid operation and maintenance. Cloud-edge collaborative technology can better meet the needs of various application scenarios through the close collaboration of cloud and edge, thereby amplifying the application value of both. Edge computing is close to terminal devices and is a data acquisition and preprocessing unit that can support cloud applications; cloud computing has powerful processing capabilities and can perform big data analysis optimization and model training, and then send the trained model or business rules to the edge side, and edge computing will run based on the new model or business rules. This technology can make up for the shortcomings of traditional methods to reduce carbon emissions. In cloud-edge collaborative technology, the task offloading optimization algorithm based on the Alternative Direction Method of Multipliers (ADMM) is its key algorithm, and its effectiveness directly determines the overall carbon emissions of power grid operation and maintenance.

[0004] However, in the power system, how to apply the cloud-edge collaborative technology based on the alternating direction multiplier method to the operation and maintenance of smart grids to process the huge amount of data generated and how to establish a mathematical model and provide an optimization method is a question that needs to be discussed. Summary of the invention

[0005] In order to solve the above-mentioned deficiencies, the present invention provides a cloud-edge collaborative power grid operation and maintenance carbon reduction method based on the alternating direction multiplier method, analyzes and mathematically models the carbon emissions of power grid operation and maintenance services under the cloud-edge collaborative architecture, and adopts the alternating direction multiplier method to optimize the mathematical model to solve the impact of the huge amount of data generated by the power grid on large-scale power grid operation and maintenance that cannot be solved by existing technologies.

[0006] To achieve the above object, the present invention adopts the following technical solution:

[0007] A cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on an alternating direction multiplier method comprises the following steps:

[0008] S1: Design a cloud-edge collaborative power grid operation and maintenance system, which includes terminal devices, cloud computing centers and edge nodes; the terminal devices are responsible for collecting different sensory data in the industrial Internet, and aggregate the collected data to the corresponding edge nodes for processing through wireless or wired data links; the cloud computing center is used to provide computing services; the edge nodes provide wireless communication infrastructure for wireless communication with terminal devices and data aggregation, as well as small servers for data processing and computing services;

[0009] S2: Modeling the relevant delay of the cloud-edge collaborative power grid operation and maintenance system;

[0010] S3. Modeling the comprehensive carbon emissions of the cloud-edge collaborative power grid operation and maintenance system according to the modeling of the relevant time delay in step S2;

[0011] S4: Based on the modeling of comprehensive carbon emissions in step S3, an objective function is constructed for the carbon emission minimization problem of the cloud-edge collaborative power grid operation and maintenance system, and the alternating direction multiplier method is used to obtain the global optimal solution.

[0012] Furthermore, in step S1, the cloud-edge collaborative power grid operation and maintenance system includes M edge nodes and 1 cloud computing center, where the cloud computing center is represented by C and the edge nodes are represented by a set S = {S1, ...S m ,…,S M}, m represents the index number of the edge node; the number of terminal devices is N, and the set D = {D1,…D n ,…,D N} represents, n represents the index number of the terminal device;

[0013] Use γ m Represents the edge node S m The proportion of the received data task units that are offloaded to the local for processing, where 0≤γ m ≤; when γ m =1, indicating edge node S m It will use its local computing resources to process all data task units; when γ m =0, indicating edge node S m All received data task units are forwarded to the cloud computing center C for processing.

[0014] Furthermore, the step S2 includes:

[0015] S21: Modeling edge node service latency without local offloading

[0016] Edge Node S mWhen all received data task units are directly forwarded to the cloud computing center C through the wired backbone network for processing, the edge computing network is equivalent to the traditional cloud computing network, that is, all computing tasks are completed by the cloud computing center C, γ m =0;

[0017] Assume that there is a linear relationship between the processing delay and the number of data task units processed. The number of data task units processed per unit time by the cloud computing center C is v. When there is no local offloading, the edge node S m The service delay is expressed as

[0018]

[0019] in, Indicates the edge node S associated with the terminal device m The round-trip data transmission delay between is the edge node S m All data task units S m The round-trip transmission delay between the cloud computing center; m Indicates the number of all data task units;

[0020] S22: Modeling edge node service latency when all offloaded locally

[0021] Edge Node S m When all received data task units are offloaded to the local computer for processing, the cloud computing center will not process any data task units from the edge nodes. m =1; the M / M / 1 queuing system is used to model the data task units to be processed at the edge nodes, and μ m Represents the edge node S m The maximum number of data task units that can be processed by the local computing resources per unit time; when all local offloads are completed, the edge node S m The corresponding delay in providing services to its associated user equipment is modeled as:

[0022]

[0023] where μ m >s m ;

[0024] S23: Cloud-edge collaborative offloading service latency modeling

[0025] In the case of cloud-edge collaborative offloading service, each edge node S m Only use its own local computing resources to process part of the data task unit it receives, using γ m s mRepresented; and the remaining data task units received are forwarded to the cloud, using (1-γ m )s m Indicates that when some local unloading is done, the edge node S m The corresponding delay in providing services to its associated user equipment is modeled as:

[0026]

[0027] Where 0≤γ m ≤1, and γ m s m <μ m .

[0028] Further, the step S3 comprises:

[0029] S31: Carbon Emission Modeling of Edge Nodes

[0030] The total power consumed by the edge node is P m It is expressed as:

[0031]

[0032] where η m It represents the power utilization efficiency, which is defined as the total annual power consumption of the node divided by the total annual power consumption of the node's information technology equipment; and β m They represent edge nodes S m The static power consumption and dynamic power consumption generated by the processing unit data task unit;

[0033] Edge Node S m Task units that process data offloaded locally m s m The queuing delay is Edge Node S m Carbon emissions E m (γ m ) is modeled as:

[0034]

[0035] S32: Carbon Emission Modeling of Cloud Computing Centers

[0036] The total amount of data task units received by the cloud computing center per unit time is expressed as:

[0037]

[0038] Where γ=<γ1,γ2,...,γ M >

[0039] The corresponding delay of the cloud computing center service is Total power consumption P C (γ) is expressed as:

[0040]

[0041] Carbon emissions from cloud computing centers C (γ) is modeled as:

[0042]

[0043] S33: Modeling the carbon emissions of backbone network data transmission

[0044] Use β L represents the dynamic power consumption caused by the transmission unit data task unit in the backbone network, η L represents the power utilization efficiency of the backbone network during data transmission, then the edge node S m The received data task unit s m (1-γ m ) is forwarded to the cloud data center, the transmission delay is Total power consumption P L (γ m ) is expressed as:

[0045] P L (γ m ) = η L β L (1-γ m )s m (3-6)

[0046] Use ∈ L represents the carbon intensity of the power resources used for data transmission in the backbone network. When the edge node forwards some data task units to the cloud data center through the backbone network, the carbon emissions generated are expressed as:

[0047]

[0048] Furthermore, in step S4, when the cloud-edge collaborative network consists of a cloud computing center and M edge nodes, the objective function of the carbon emission minimization problem of the cloud-edge collaborative power grid operation and maintenance system is constructed as follows:

[0049]

[0050] Among them, γ includes the proportion of the business volume unloaded by M edge nodes to the total number of business of all industrial Internet terminals; E C (γ), E m (γ m ) and E L (γ m) are the computing carbon emissions of the cloud computing center, the computing carbon emissions of the mth edge node, and the transmission carbon emissions generated by uploading business data from the industrial Internet terminal corresponding to the mth edge node.

[0051] Furthermore, in step S4, when the alternating direction multiplier method is used to obtain the global optimal solution for the objective function established in step S41:

[0052] First, the iterative formula of the alternating direction multiplier method is simplified based on the augmented Lagrangian expression. Specifically,

[0053] Using indicator function I C (.) Add the feasible region to the objective function, and the objective function is transformed into the following problem:

[0054] min[E C (γ)+1 T (E(γ)+E L (γ))]+I C (γ) (4-2)

[0055] Among them I C (γ) is the indicator function of the feasible domain, that is:

[0056]

[0057] Introducing the constraint z = γ, the objective function is converted to:

[0058]

[0059] where f(y) = E C (y)+1 T (E(y)+E L (y)),h(z)=I C (z);

[0060] After the above equivalent transformation, the target function is converted into the standard alternating direction multiplication method form;

[0061] Then establish the augmented Lagrange multiplier formula and proceed to the next step of solving:

[0062]

[0063] Among them, ρ>0 is the augmented Lagrangian coefficient, λ∈R M is the dual variable;

[0064] Based on the above Lagrange multiplier formula, the alternating direction multiplier method consists of the following iterative steps:

[0065] y k+1 =argminL ρ(y,z k ,λ k ) (4-6)

[0066]

[0067] λ k+1 =λ k +ρ(y k+1 -z k+1 ) (4-8)

[0068] The iterative formula is simplified by combining the linear term and the quadratic term in the augmented Lagrangian and scaling the dual variable. The scaled dual variable is defined as The alternating direction multiplication method is equivalent to the following expression:

[0069]

[0070] u k+1 :=u k +ρ(y k+1 -z k+1 ) (4-11)

[0071] Secondly, determine the stopping criterion of the alternating direction multiplier method ADMM. Specifically, define the primal feasibility residual and the dual feasibility residual of the above problem:

[0072] r k =||y k -z k ||2 (4-12)

[0073] s k =||z k -z k-1 ||2 (4-13)

[0074] where r k is the original feasibility residual, s k It is the dual feasibility residual, and it is used as the stopping criterion to determine whether the two residuals are small enough.

[0075] Furthermore, the determination of whether the two residuals are sufficiently small as a stopping criterion specifically includes:

[0076]

[0077] Among them, ∈ pri >0,∈ dual >0 indicates the feasibility error of the suboptimal solution of the objective function, and the suboptimal solution of the alternating direction multiplier method is obtained as the final solution of the objective function.

[0078] Furthermore, ∈ abs and ∈rel 10 respectively -4 and 10 -2 .

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] 1. The cloud-edge collaborative optimization task offloading method proposed in the present invention can take into account the advantages of both cloud computing centers and edge nodes. The application of cloud-edge collaborative technology in power grid carbon emission management can improve data processing efficiency, reduce latency, enhance data security, and support precise monitoring and metering. It provides effective technical support for the low-carbon operation and carbon emission management of power grids and effectively reduces comprehensive carbon emissions while reducing business delays.

[0081] 2. The mathematical analysis and simulation results of the present invention confirm that the cloud-edge collaborative architecture and collaborative task optimization offloading method can effectively reduce the total carbon emissions of the computing network and transmission system while providing more flexible business performance guarantees. Therefore, it is expected to become an important enabling technology for green and low-carbon power grid operation and maintenance systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a flow chart of the cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method of the present invention.

[0083] Figure 2 It is the flow chart of ADMM algorithm.

[0084] Figure 3 is a comparison of actual carbon emissions and energy consumption when the number of edge node tasks is different and the minimum total energy consumption is used as the objective function, where (a) is the comparison of actual carbon emissions when the number of edge node tasks is different and the minimum total energy consumption is used as the objective function, and (b) is the comparison of actual energy consumption when the number of edge node tasks is different and the minimum total energy consumption is used as the objective function. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0086] See also Figure 1 The embodiment of the present invention provides a cloud-edge collaborative power grid operation and maintenance carbon reduction method based on the alternating direction multiplier method, comprising the following steps:

[0087] S1: Design a cloud-edge collaborative power grid operation and maintenance system, which includes terminal devices, cloud computing centers and edge nodes. The terminal devices are responsible for collecting different sensory data in the industrial Internet, and aggregate the collected data to the corresponding edge nodes for processing through wireless or wired data links; the cloud computing center can provide high-quality and high-performance computing services, with abundant available computing resources and power resources; the edge nodes provide wireless communication infrastructure that can communicate wirelessly with terminal devices and aggregate data, such as wireless communication base stations, access points or aggregation gateways, etc., and also include small servers that can provide data processing and computing services.

[0088] Specifically, consider a cloud-edge collaborative power grid operation and maintenance system consisting of M edge nodes and 1 cloud computing center, where the cloud computing center is represented by C and the edge nodes are represented by the set S = {S1,…S m ,…,S M}, m represents the index number of the edge node; the number of terminal devices is N, and the set D = {D1,…D n ,…,D N}, n represents the index number of the terminal device. m Represents the edge node S m The proportion of the received data task units that are offloaded to the local for processing, where 0≤γ m ≤; when γ m =1, indicating edge node S m It will use its local computing resources to process all data task units; when γ m =0, indicating edge node S m All received data task units are forwarded to the cloud computing center C for processing.

[0089] S2: Modeling the relevant delays of the cloud-edge collaborative power grid operation and maintenance system.

[0090] The latency of power grid services is one of the key indicators for measuring the response speed, stability and reliability of the power grid. It has a vital impact on the overall service quality of the power grid. More importantly, it is closely related to the carbon emissions of power grid operation and maintenance.

[0091] Considering the modeling of the delay related to the cloud-edge collaborative power grid operation and maintenance system, the embodiment of the present invention uses T m Represents the edge node S m The service response delay of the connected terminal device, which includes the round-trip transmission time of the data task unit and the queuing delay for processing. The terminal device and its associated edge node S m The round-trip data transmission delay between express.

[0092] The step S2 comprises the following sub-steps:

[0093] S21: Modeling edge node service latency when there is no local offloading.

[0094] Edge Node S m You can choose to directly forward all received data task units to the cloud computing center C through the wired backbone network for processing. In this case, the edge computing network can be equivalent to the traditional cloud computing network, that is, all computing tasks are completed by the cloud computing center C, γ m = 0. Since cloud computing centers usually have sufficient computing resources, the queuing delay of data task units is negligible compared to the processing delay. Assuming that there is a linear relationship between the processing delay of data task units and their number, the number of data task units that can be processed per unit time by the cloud computing center C is v. In summary, when there is no local offloading, the edge node S m The service delay can be expressed as:

[0095]

[0096] in, is the edge node S m All data task units m The round-trip transmission delay between the cloud computing center; m Indicates the number of all data task units.

[0097] S22: Modeling of edge node service latency when all offloads are local.

[0098] Edge Node S m Choose to offload all received data task units to the local computer for processing. In this case, the cloud computing center will not process any data task units from the edge nodes. m =1. However, due to physical limitations, cost control and other reasons, the local computing resources of edge nodes are relatively limited, so the queuing delay of processing data task units cannot be ignored. In the present invention, the embodiment of the present invention adopts a universal queuing model, that is, the M / M / 1 queuing system to model the data task units to be processed at the edge nodes, and uses μ m Represents the edge node S m The maximum number of data task units that can be processed by the local computing resources per unit time. In this case, the embodiment of the present invention can offload all local m The corresponding delay in providing services to its associated user equipment is modeled as:

[0099]

[0100] where μ m >sm

[0101] S23: Latency modeling of cloud-edge collaborative offloading services.

[0102] In the case of cloud-edge collaborative offloading service, each edge node S m Only use its own local computing resources to process part of the data task unit it receives, which can be used m s m Represented; and the remaining data task units received are forwarded to the cloud, using (1-γ m )s m In this embodiment of the present invention, when some local unloading is performed, the edge node S m The corresponding delay in providing services to its associated user equipment is modeled as:

[0103]

[0104] Where 0≤γ m ≤1, and γ m s m <μ m .

[0105] S3. Model the comprehensive carbon emissions of the cloud-edge collaborative power grid operation and maintenance system based on the modeling of relevant time delays in step S2.

[0106] The comprehensive carbon emissions of the power grid operation and maintenance system mainly include: carbon emissions generated by terminal devices collecting data and transmitting it to the corresponding base station through wireless channels; carbon emissions generated by edge nodes connected to the base station processing locally unloaded data task units; carbon emissions generated by the cloud computing center processing received data task units; and carbon emissions generated by data transmission between edge nodes and cloud computing centers through the backbone network.

[0107] The step S3 comprises the following sub-steps:

[0108] S31: Carbon emission modeling of edge nodes.

[0109] Edge Node S m After deployment is completed, its carbon intensity is determined by the comprehensive source of electricity supply and can be regarded as a constant, denoted by ∈ m Therefore, its carbon emissions are mainly determined by the total operating power and processing latency. The total power consumed by edge nodes depends on the power usage effectiveness (PUE), static and dynamic power consumption. Among them, power usage effectiveness is a widely used data center energy efficiency indicator, which is defined as the total annual power consumption of the node divided by the total annual power consumption of the node's information technology equipment, denoted by η. mStatic power consumption is also called leakage power consumption, which is mainly caused by leakage current in the system and has nothing to do with the computing resource usage of each edge node; while dynamic power consumption is mainly determined by the activity of computing resources. and β m Represents the edge node S m The static power consumption and dynamic power consumption generated by the processing unit data task unit, the total power P m It can be expressed as:

[0110]

[0111] Furthermore, as mentioned above, the edge node S m Task units that process data offloaded locally m s m The queuing delay is Therefore, its carbon emission E m (γ m ) can be modeled as:

[0112]

[0113] S32: Carbon emissions modeling of cloud computing centers.

[0114] Similar to edge nodes, the total power of cloud computing centers is also determined by power utilization efficiency, static power consumption, and dynamic power consumption. The difference is that compared with edge nodes, their static power consumption is Dynamic power consumption generated by the processing unit data task unit β c And the energy efficiency η c At the same time, due to the huge power consumption of cloud computing centers, they are usually deployed in areas where clean energy accounts for a high proportion of power supply. Compared with the edge nodes that sink to the network terminals, their carbon intensity ∈ m Usually small. Since the cloud computing center can receive and process the data task units forwarded by all edge nodes, the total amount of data task units received per unit time can be expressed as:

[0115]

[0116] Where γ=<γ1,γ2,...,γ M >. Accordingly, as mentioned above, due to the sufficient computing resources in the cloud computing center, the queuing delay of the data task unit is negligible compared with the processing delay, and its service response delay is And its total power consumption P C (γ) can be expressed as:

[0117]

[0118] In summary, the carbon emissions of cloud computing centers are C (γ) can be modeled as:

[0119]

[0120] S33: Modeling carbon emissions of backbone network data transmission.

[0121] In the cloud-edge collaborative power grid operation and maintenance system considered by the present invention, the backbone network mainly provides high-speed wired data transmission services between edge servers and cloud computing centers. The total power consumption in the backbone network can also be divided into static power consumption and dynamic power consumption. Since the static power consumption of the backbone network is usually composed of the networking equipment of the edge nodes or cloud computing centers, the static power consumption of the backbone network is regarded as part of the static power consumption of the edge nodes or cloud computing centers in the embodiments of the present invention, that is, included in the static power consumption of the edge nodes or cloud computing centers. and In the process of data transmission, the dynamic power consumption is mainly considered. Since the dynamic power consumption and transmission delay of data transmission in the backbone network depend on the number of data task units transmitted, this paper takes a single edge node forwarding part of the received data task units to the cloud computing center as an example to analyze the carbon emission model of backbone network data transmission.

[0122] Use β L represents the dynamic power consumption caused by the transmission unit data task unit in the backbone network, η L represents the power utilization efficiency of the backbone network during data transmission, then the edge node S m The received data task unit s m (1-γ m ) is forwarded to the cloud data center, the transmission delay is Total power consumption P L (γ m ) can be expressed as:

[0123] P L (γ m ) = η L β L (1-γ m )s m (3-6)

[0124] Therefore, using ∈ L represents the carbon intensity of the power resources used for data transmission in the backbone network. When the edge node forwards some data task units to the cloud data center through the backbone network, the carbon emissions generated can be expressed as:

[0125]

[0126] S4: Based on the modeling of comprehensive carbon emissions in step S3, an objective function is constructed and optimized for the carbon emission minimization problem of the cloud-edge collaborative power grid operation and maintenance system.

[0127] Specifically, it is assumed that all power grid equipment terminals must first upload the services they generate to the nearest base station through a wireless network, and then decide whether to offload to the edge node connected to the base station or upload to the cloud computing center through a backbone data transmission network.

[0128] The step S4 comprises the following sub-steps:

[0129] S41: Construct an objective function for the carbon emission minimization problem of the cloud-edge collaborative power grid operation and maintenance system.

[0130] When the cloud-edge collaborative network consists of a cloud computing center and M edge nodes, the embodiment of the present invention can model the carbon emission minimization problem of the cloud-edge collaborative power grid operation and maintenance system as follows:

[0131]

[0132] Among them, γ includes the proportion of the business volume unloaded by M edge nodes to the total business volume of all industrial Internet terminals. C (γ), E m (γ m ) and E L (γ m ) are the computing carbon emissions of the cloud computing center, the computing carbon emissions of the mth edge node, and the transmission carbon emissions generated by uploading business data from the industrial Internet terminal corresponding to the mth edge node. The embodiment of the present invention notes that different edge nodes may receive different types of services, so the computing carbon emissions of the cloud computing center need to comprehensively consider the business type of each edge node m and the unloading volume γ corresponding to the business type m .

[0133] S42: Use the alternating direction multiplier method to obtain the global optimal solution for the objective function established in step S41.

[0134] Firstly, the iterative formula of the alternating direction multiplier method is simplified based on the augmented Lagrangian expression.

[0135] The above problem is a convex optimization problem with inequality constraints within the allowed parameter range, which can be solved by traditional optimization methods, such as the interior point method. In particular, if the constraints are converted into part of the objective function by introducing an indicator function, the problem can be solved by the Alternative Direction Method of Multipliers (ADMM) to obtain the global optimal solution. The process is as follows: Figure 2As shown in Figure 2, it can be observed that in the objective function of the above problem, the carbon emission E of the edge node network m (γ m ) Carbon emissions from backbone network data transmission E L (γ m ) is independent of each edge node, that is, there is no coupling relationship between them; at the same time, each constraint condition is also independent and uncoupled with each edge node. Therefore, the carbon emission E of the edge node network m (γ m ) Carbon emissions from backbone network data transmission E L (γ m The minimization part of the problem can be solved independently and distributedly among the edge nodes, and the cloud computing center can be responsible for the coordination of the sub-problem solving process among the edge nodes and the C The above alternating direction multiplier method ADMM can realize the task forwarding strategy of minimizing system-level carbon emissions under the premise of protecting the privacy information such as computing power, electricity, and communication resources among edge nodes.

[0136] Since the constraints of the above model are all convex sets, and the feasible domain is the intersection of the four constraints, it can be shown that the feasible domain is a convex set, so the indicator function I can be used C (.) Adding the feasible region to the objective function, the problem can be transformed into the following problem:

[0137] min[E C (γ)+1 T (E(γ)+E L (γ))]+I C (γ) (4-2)

[0138] Among them I C (γ) is the indicator function of the feasible domain, that is:

[0139]

[0140] In order to convert the problem into a standard ADMM form, the embodiment of the present invention introduces the constraint z=γ, and then the problem is converted into:

[0141]

[0142] where f(y) = E C (y)+1 T (E(y)+E L (y)),h(z)=I C (z). After the above equivalent transformation, the problem is converted into the standard ADMM form, and then the augmented Lagrange multiplier formula is established to proceed to the next step of solving:

[0143]

[0144] Among them, ρ>0 is the augmented Lagrangian coefficient, λ∈R M is the dual variable. Based on the above Lagrange multiplier formula, the ADMM algorithm can be composed of the following iterative steps:

[0145] y k+1 =argminL ρ (y,z k ,λ k ) (4-6)

[0146]

[0147] λ k+1 =λ k +ρ(y k+1 -z k+1 ) (4-8)

[0148] In order to improve computational efficiency and concise expression, the present invention simplifies the iterative formula by combining the linear term and the quadratic term in the augmented Lagrangian and scaling the dual variable, and defines the scaled dual variable as So ADMM can be equivalent to the following expression:

[0149]

[0150] u k+1 :=u k +ρ(y k+1 -z k+1 ) (4-11)

[0151] Secondly, the stopping criteria of the alternating direction multiplier method ADMM and the steps of the ADMM algorithm are determined.

[0152] Since the above problem is a constrained optimization problem, the convergence criterion of the algorithm should use the KKT condition of the constrained optimization problem. Therefore, the present invention first defines the original feasibility residual and the dual feasibility residual of the above problem:

[0153] r k =||y k -z k ||2 (4-12)

[0154] s k =||z k -z k-1 ||2 (4-13)

[0155] where r k is the original feasibility residual, s kIt is the dual feasibility residual. According to the original feasibility condition and the dual feasibility condition of the KKT condition with constrained optimization problem, when ADMM converges, the values ​​of the two residuals should be 0. However, in practical applications, it is often judged whether the two residuals are small enough as the stopping criterion, that is:

[0156]

[0157] Among them, ∈ pri >0,∈ dual >0 indicates the feasibility error of the suboptimal solution to the above problem, and the embodiment of the present invention sets ∈ abs and ∈ rel 10 respectively -4 and 10 -2 According to the above formula, the suboptimal solution of the algorithm can be obtained as the final solution to the problem.

[0158] like Figure 2 As shown, the present invention uses the alternating direction multiplier method to obtain the global optimal solution for the objective function established in step S41. The detailed algorithm steps are as follows:

[0159] First, the edge server m initializes its own decision variables, and the parameter coordinator initializes the auxiliary variables, dual variables, and the maximum number of iterations MAX_ITER. Then, it is determined whether the number of iterations i is less than the maximum number of iterations. If so, all edge servers are updated at the same time, and the decision variables are updated according to formula (4-9). The results are fed back to the parameter coordinator, and the parameter coordinator is waiting for the auxiliary variables and dual variables to be sent. After receiving the parameter coordinator, the auxiliary variables are updated according to formula (4-10) and the dual variables are updated according to formula (4-11). Then it is determined whether the stopping criteria are met. If not, the iteration continues. If it is met, the corresponding auxiliary variables and dual variables are sent to the corresponding edge server m, and finally the process ends.

[0160] As shown in FIG3 , the embodiment of the present invention simulates a cloud-edge computing network consisting of 60 power grid terminals, 6 edge nodes, and 1 cloud computing center. The performance and carbon emissions of three offloading schemes are mainly compared, including (1) full offloading, that is, all tasks uploaded from the industrial Internet terminals are processed by the edge nodes, (2) no task offloading, that is, all tasks are uploaded to the cloud computing center for processing, and (3) cloud-edge collaborative optimization offloading, that is, the cloud-edge collaborative optimal task offloading method solved by the convex optimization method.

[0161] In order to compare the performance of the three schemes in reducing carbon emissions in the power grid, the objective function is to minimize the carbon emissions defined in formula (4-1), and the actual carbon emissions and energy consumption when the total number of tasks received by each edge node increases. As can be seen from Figure 3 (a), the proposed cloud-edge collaborative optimization offloading can minimize the carbon emissions generated by the power grid compared to the cases of no local offloading and all local offloading, and it increases slowly with the increase in the number of tasks. As can be seen from Figure 3 (b), the changes in energy consumption and carbon emissions are very different. Although the system and total energy consumption are small when all tasks are offloaded to the local edge node, the actual carbon emissions generated are huge due to the huge differences in the carbon intensity coefficients between edge nodes.

[0162] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on alternating direction multiplier method, characterized in that: The steps include: S1: Design a cloud-edge collaborative power grid operation and maintenance system, which includes terminal devices, cloud computing centers and edge nodes; the terminal devices are responsible for collecting different sensory data in the industrial Internet, and aggregate the collected data to the corresponding edge nodes for processing through wireless or wired data links; the cloud computing center is used to provide computing services; the edge nodes provide wireless communication infrastructure for wireless communication with terminal devices and data aggregation, as well as small servers for data processing and computing services; S2: Modeling the relevant delay of the cloud-edge collaborative power grid operation and maintenance system; S3. Modeling the comprehensive carbon emissions of the cloud-edge collaborative power grid operation and maintenance system according to the modeling of the relevant time delay in step S2; S4: Based on the modeling of comprehensive carbon emissions in step S3, an objective function is constructed for the carbon emission minimization problem of the cloud-edge collaborative power grid operation and maintenance system, and the alternating direction multiplier method is used to obtain the global optimal solution.

2. The cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method according to claim 1 is characterized by: In step S1, the cloud-edge collaborative power grid operation and maintenance system includes M edge nodes and 1 cloud computing center, where the cloud computing center is represented by C and the edge nodes are represented by a set S = {S1, ...S m ,…,S M }, m represents the index number of the edge node; the number of terminal devices is N, and the set D = {D1,…D n ,…,D N } represents, n represents the index number of the terminal device; Use γ m Represents the edge node S m The proportion of the received data task units that are offloaded to the local for processing, where 0≤γ m ≤; when γ m =1, indicating edge node S m It will use its local computing resources to process all data task units; when γ m =0, indicating edge node S m All received data task units are forwarded to the cloud computing center C for processing.

3. The cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method as claimed in claim 2 is characterized by: The step S2 comprises: S21: Modeling edge node service latency without local offloading Edge Node S m When all received data task units are directly forwarded to the cloud computing center C through the wired backbone network for processing, the edge computing network is equivalent to the traditional cloud computing network, that is, all computing tasks are completed by the cloud computing center C, γ m =0; Assume that there is a linear relationship between the processing delay and the number of data task units processed. The number of data task units processed per unit time by the cloud computing center C is v. When there is no local offloading, the edge node S m The service delay is expressed as in, Indicates the edge node S associated with the terminal device m The round-trip data transmission delay between is the edge node S m All data task units m The round-trip transmission delay between the cloud computing center; m Indicates the number of all data task units; S22: Modeling edge node service latency when all offloaded locally Edge Node S m When all received data task units are offloaded to the local computer for processing, the cloud computing center will not process any data task units from the edge nodes. m =1; the M / M / 1 queuing system is used to model the data task units to be processed at the edge nodes, and μ m Represents the edge node S m The maximum number of data task units that can be processed by the local computing resources per unit time; when all local offloads are completed, the edge node S m The corresponding delay in providing services to its associated user equipment is modeled as: where μ m >s m ; S23: Cloud-edge collaborative offloading service latency modeling In the case of cloud-edge collaborative offloading service, each edge node S m Only use its own local computing resources to process part of the data task unit it receives, using γ m s m Represented; and the remaining data task units received are forwarded to the cloud, using (1-γ m )s m Indicates that when some local unloading is done, the edge node S m The corresponding delay in providing services to its associated user equipment is modeled as: Where 0≤γ m ≤1, and γ m s m <μ m .

4. The cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method as claimed in claim 3 is characterized by: The step S3 comprises: S31: Carbon Emission Modeling of Edge Nodes The total power consumed by the edge node is P m It is expressed as: where η m It represents the power utilization efficiency, which is defined as the total annual power consumption of the node divided by the total annual power consumption of the node's information technology equipment; and β m They represent edge nodes S m The static power consumption and dynamic power consumption generated by the processing unit data task unit; Edge Node S m Task units that process data offloaded locally m s m The queuing delay is Edge Node S m Carbon emissions E m (γ m ) is modeled as: S32: Carbon Emission Modeling of Cloud Computing Centers The total amount of data task units received by the cloud computing center per unit time is expressed as: Where γ=<γ1,γ2,...,γ M ); The corresponding delay of the cloud computing center service is Total power consumption P C (γ) is expressed as: Carbon emissions from cloud computing centers C (γ) is modeled as: S33: Modeling the carbon emissions of backbone network data transmission Use β L represents the dynamic power consumption caused by the transmission unit data task unit in the backbone network, η L represents the power utilization efficiency of the backbone network during data transmission, then the edge node S m The received data task unit s m (1-γ m ) is forwarded to the cloud data center, the transmission delay is Total power consumption P L (γ m ) is expressed as: P L (c m )=η L b L (1-c m )s m (3-6) Use ∈ L represents the carbon intensity of the power resources used for data transmission in the backbone network. When the edge node forwards some data task units to the cloud data center through the backbone network, the carbon emissions generated are expressed as:

5. The cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method as claimed in claim 4 is characterized by: In step S4, when the cloud-edge collaborative network consists of a cloud computing center and M edge nodes, the objective function of the carbon emission minimization problem of the cloud-edge collaborative power grid operation and maintenance system is constructed as follows: Among them, γ includes the proportion of the business volume unloaded by M edge nodes to the total number of business of all industrial Internet terminals; E C (γ), E m (γ m ) and E L (γ m ) are the computing carbon emissions of the cloud computing center, the computing carbon emissions of the mth edge node, and the transmission carbon emissions generated by uploading business data from the industrial Internet terminal corresponding to the mth edge node.

6. The cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method as claimed in claim 5, characterized in that: In step S4, when the alternating direction multiplier method is used to obtain the global optimal solution of the objective function established in step S41: First, the iterative formula of the alternating direction multiplier method is simplified based on the augmented Lagrangian expression. Specifically, Using indicator function I C (.) Add the feasible region to the objective function, and the objective function is transformed into the following problem: min[E C (c)+1 T (E(γ)+E L (c))]+I C (c) (4-2) Among them I C (γ) is the indicator function of the feasible domain, that is: Introducing the constraint z = γ, the objective function is converted to: where f(y) = E C (y) + 1 T (E(y) + E L (y)), h(z) = I C (z); After the above equivalent transformation, the target function is converted into the standard alternating direction multiplication method form; Then establish the augmented Lagrange multiplier formula and proceed to the next step of solving: Among them, ρ>0 is the augmented Lagrangian coefficient, λ∈R M is the dual variable; Based on the above Lagrange multiplier formula, the alternating direction multiplier method consists of the following iterative steps: and k+1 =argminL ρ (y,z k ,λ k ) (4-6) l k+1 =λ k +ρ(y k+1 -z k+1 ) (4-8) The iterative formula is simplified by combining the linear term and the quadratic term in the augmented Lagrangian and scaling the dual variable. The scaled dual variable is defined as The alternating direction multiplication method is equivalent to the following expression: you k+1 :=u k +ρ(y k+1 -z k+1 ) (4-11) Secondly, determine the stopping criteria of the alternating direction multiplier method ADMM. Specifically, Define the primal feasibility residual and dual feasibility residual for the above problem: r k =||y k -z k ||2 (4-12) s k =||z k -z k-1 ||2 (4-13) where r k is the original feasibility residual, s k It is the dual feasibility residual, and it is used as the stopping criterion to determine whether the two residuals are small enough.

7. The cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method according to claim 6 is characterized by: The step of judging whether the two residuals are sufficiently small as a stopping criterion specifically includes: Among them, ∈ pri >0,∈ dual >0 indicates the feasibility error of the suboptimal solution of the objective function, and the suboptimal solution of the alternating direction multiplier method is obtained as the final solution of the objective function.

8. The cloud-edge collaborative smart grid operation and maintenance carbon reduction method based on the alternating direction multiplier method according to claim 7, characterized in that: ∈ abs and ∈ rel 10 respectively -4 and 10 -2 .

Images