Alternating current and direct current power supply system cloud side end cooperative task allocation method and system

Abstract

The invention provides an AC / DC power supply system cloud edge end cooperative task allocation method and system, and the method comprises the steps: determining the number of tasks, needing to be processed at a cloud end and an edge end, of an AC / DC power supply system, and building a multi-objective function of time delay and energy consumption joint optimization according to the time and energy consumed when each task is processed at the cloud end and the edge end; and optimizing the established multi-objective function through an adaptive reservation simulated annealing algorithm. The technical problem that a global optimal solution is difficult to solve due to the fact that a corresponding intelligent optimization algorithm is not suitable for optimization of a multi-objective function and depends on a large amount of operation is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud-edge-end collaboration, edge computing and intelligent optimization algorithms, and specifically to a method and system for cloud-edge-end collaborative task allocation in an AC / DC power supply system. Background Art

[0002] In order to ensure the stability of the AC / DC power supply system, many tasks often need to be processed during operation. The complexity and real-time requirements of these tasks are usually quite different. The traditional cloud-based collaborative architecture uploads all pending tasks to the cloud server for centralized processing. This not only increases the computing burden and storage pressure of the cloud server and consumes a lot of energy, but also the transmission delay caused by communication makes it difficult to achieve tasks with high real-time requirements. The currently more widely used cloud-edge collaborative architecture delegates part of the computing resources of the cloud server to the edge server closer to the AC / DC power supply system, which can effectively reduce the transmission delay caused by communication and meet the real-time requirements of the AC / DC power supply system.

[0003] In a cloud-edge-device collaborative architecture, the primary challenge is how to rationally allocate pending tasks in the AC / DC power system between cloud servers and edge servers to achieve low latency and low energy consumption. Existing technologies primarily employ various intelligent optimization algorithms, such as greedy algorithms and simulated annealing algorithms, to find the optimal solution to the cloud-edge-device task allocation problem. For example:

[0004] In the existing open literature, "Research on Intelligent Diagnosis System for Machine Tool Faults Based on Cloud-Edge-End Collaboration [J]," China Mechanical Engineering, 2023, 34(05): 584-594, a time-delay optimization objective function for the cloud-edge-end collaborative task allocation problem for machine tool fault diagnosis is first established. Then, a greedy algorithm is used to solve the objective function and obtain the task allocation strategy. However, this method only considers the time-delay optimization objective function. For multi-objective optimization functions such as time delay and energy consumption, the greedy algorithm has difficulty in obtaining the optimal solution.

[0005] The existing published literature, "Data Management Framework and Optimized Operation Method for Regional Wind Farm Big Data Centers Based on Cloud-Edge-End Collaboration [J / OL]. High Voltage Technology, 1-11 [2024-06-03]," first establishes a multi-objective function for jointly optimizing latency, energy consumption, and transmitted data volume. It then uses a traditional simulated annealing algorithm to find the optimal solution and determine the task allocation strategy. However, this traditional simulated annealing algorithm not only has poor optimization results but also is slow, making it difficult to achieve satisfactory results.

[0006] In summary, the prior art has the following deficiencies:

[0007] 1) Only the delay optimization objective function is considered, and the corresponding intelligent optimization algorithm applied is not suitable for optimizing multiple objective functions;

[0008] 2) Traditional simulated annealing algorithms require a large amount of computation to obtain the global optimal solution, which takes a lot of time, making it impossible to obtain the global optimal solution under normal circumstances.

[0009] The existing invention patent application document "Cloud-edge joint data center balanced optimization scheduling method and joint data center system" with publication number CN117742916A, the existing method includes: S1, constructing a cloud-edge collaborative computing model that integrates deep reinforcement learning and hybrid particle swarm algorithm. The cloud-edge collaborative computing model includes: a task listener, an IPSO module, a cloud-edge joint data center, and a convolution cloud neural network model; the cloud-edge collaborative computing model uses a method that combines a hybrid particle swarm algorithm with deep reinforcement learning to achieve balanced scheduling of the cloud-edge joint data center; S2, using the hybrid particle swarm algorithm to solve the optimal remaining resources. The particle swarm algorithm in the cloud-edge collaborative computing model uses the global optimal solution capability and the crossover, selection, and mutation capabilities of the genetic algorithm, and combines the Metropolis acceptance criterion verification in the simulated annealing algorithm to establish the initial state space of deep reinforcement learning; S3, using deep reinforcement learning to achieve load balancing. The policy gradient-based RL model in the cloud-edge collaborative computing model is used to perform resource scheduling of the cloud-edge data center, avoid the calculation of the reward value, determine the next action by probability, and then continuously iterate and improve the strategy until the strategy converges.

[0010] However, the aforementioned existing solutions only consider the latency optimization goal without considering energy consumption. In addition, the existing technology mainly uses deep reinforcement learning methods to solve the optimal offloading strategy, which requires a lot of training time and computing resources.

[0011] The existing invention patent application document with publication number CN117632488A is "Multi-user fine-grained task offloading scheduling method and device based on cloud-edge-end collaboration". The existing method includes: first, constructing a multi-user fine-grained task offloading network model in a cloud-edge-end collaboration scenario, and then modeling the applications with dependent tasks on the user device as a directed acyclic graph, establishing a system delay, energy consumption and server rental cost model, and constructing a task offloading decision model by minimizing the weighted sum of system delay, energy consumption and server rental cost, calculating task priority, and designing a multi-user subtask scheduling scheme based on task priority, and solving the optimal offloading decision based on the improved simulated annealing particle swarm algorithm.

[0012] However, the aforementioned existing solutions are mainly based on a simulated annealing particle swarm optimization hybrid algorithm to solve the optimal unloading strategy. Compared with a single heuristic optimization algorithm, this type of hybrid heuristic optimization algorithm usually has higher computational complexity and requires adjusting more hyperparameters. When solving high-dimensional 0-1 planning problems, it is easily affected by particle aggregation and has difficulty escaping the local optimum.

[0013] In summary, the existing technologies have technical problems such as the corresponding intelligent optimization algorithms are not suitable for optimizing multi-objective functions, and rely on a large number of calculations, which makes it difficult to find the global optimal solution. Summary of the Invention

[0014] The technical problem to be solved by the present invention is: how to solve the technical problem that the corresponding intelligent optimization algorithm in the existing technology is not suitable for optimizing multiple objective functions, and relies on a large number of calculations, which makes it difficult to solve the global optimal solution.

[0015] The present invention solves the above technical problems by adopting the following technical solutions: A method for allocating tasks between the cloud and the edge of an AC / DC power supply system includes:

[0016] S1. Determine the number of tasks to be processed on the cloud and the number of tasks to be processed on the edge.

[0017] S2. Execute the pending tasks on the cloud and edge, measure and record the execution time consumption data and the execution energy consumption data, and process them to obtain the edge time matrix, edge energy matrix, cloud time matrix, and cloud energy matrix;

[0018] S3, obtains the edge allocation strategy and the cloud allocation strategy to construct the edge task allocation matrix and the cloud task allocation matrix for the edge and the cloud, and establishes and solves the multi-objective function of joint optimization of delay and energy consumption to obtain the edge task allocation solution matrix;

[0019] S4. Based on the edge task allocation solution matrix, the adaptive retained simulated annealing algorithm is used to optimize the multi-objective function of delay and energy consumption, and the optimal edge task allocation matrix is ​​obtained: α(L), α(L) = [α1(L), ..., α i (L),...,α n (L)], where α i (L) is the final allocation strategy of the i-th task to be executed at the edge.

[0020] The present invention not only considers the time delay optimization objective function, but also considers the multi-objective function of energy consumption optimization. The corresponding intelligent optimization algorithm applied by the present invention is more suitable for optimizing multi-objective functions; the simulated annealing algorithm of the present invention does not rely on a large number of calculations to obtain the global optimal solution, which improves the computing efficiency and can obtain the global optimal solution in more application scenarios.

[0021] In a more specific technical solution, S2 includes:

[0022] S21, execute n pending tasks in sequence on the edge, measure the execution time consumption data and execution energy consumption data of each pending task on the edge, and obtain the edge time matrix T e , edge energy matrix E e :

[0023] T e =[T e1 ,K,T ei ,K,T en ] T

[0024] E e =[E e1 ,K,E ei ,K,E en ] T

[0025] Where, T ei is the time it takes for the i-th task to be executed once at the edge, E ei The energy consumed by executing the i-th task on the edge.

[0026] S22, execute n tasks to be executed in the cloud in sequence, measure the execution time consumption data and execution energy consumption data of each task to be executed in the cloud, and obtain the cloud time matrix T c , Cloud Energy Matrix E c :

[0027] T c =[T c1 ,K,T ci ,K,T cn ] T

[0028] E c =[E c1 ,K,E ci ,K,E cn ] T

[0029] Where, T ci is the time consumed by the execution of the i-th task in the cloud, E ci The energy consumed by executing the i-th task in the cloud.

[0030] In more specific technical solutions, S3 includes:

[0031] S31. Construct an edge task allocation matrix using the following logic:

[0032] α, α=[α1,...,α i,...,α n ]

[0033] Where, α i Assign a strategy to the edge for the i-th task to be executed, and α i ∈{0,1};

[0034] Use the following logic to construct the cloud task allocation matrix:

[0035] β, β=[β1,...,β i ,...,β n ]

[0036] Where, β i Assign a strategy to the i-th task to be executed in the cloud, and β i ∈{0,1};

[0037] S32. Construct a time delay optimization objective function and an energy consumption optimization objective function, and obtain a multi-objective function for joint optimization of time delay and energy consumption based on the obtained function.

[0038] Compared with the traditional simulated annealing algorithm, the adaptive retention simulated annealing algorithm proposed in the present invention adaptively retains some features of the current solution according to the current temperature, overcomes the disadvantage that the traditional simulated annealing algorithm requires a large amount of calculations to obtain the global optimal solution, and significantly improves the optimization speed.

[0039] In a more specific technical solution, in S31, the edge allocation strategy α of the i-th task to be executed is set i , and the cloud allocation strategy β of the i-th task to be executed i It is an exclusive OR relationship.

[0040] In a more specific technical solution, S32 also includes:

[0041] S321. Construct a latency optimization objective function using the following logic:

[0042] f T ,f T =αT e +βT c ;

[0043] S322. Use the following logic to construct an energy consumption optimization objective function:

[0044] f E , f E =αE e +βE c ;

[0045] S323. Combine the delay optimization objective function and the energy consumption optimization objective function to establish a multi-objective function for joint optimization of delay and energy consumption:

[0046] A=min(xf T +yf E ).

[0047] Where A is the multi-objective function value, x is the weight parameter of the low latency objective, and y is the weight parameter of the low energy consumption objective, where: x + y = 1.

[0048] The present invention comprehensively considers the latency and energy consumption of cloud-edge collaborative task allocation by establishing a multi-objective function for joint optimization of latency and energy consumption. At the same time, it uses the heuristic algorithm of adaptive retained simulated annealing algorithm as the solution method, which saves the training time of deep learning models, saves computing resources, and improves the efficiency and convenience of solution.

[0049] In a more specific technical solution, S4 includes:

[0050] S41, given initial temperature H0, end temperature H f 、Current update temperature H c (k), where k is the number of temperature updates. Assume that the number of temperature updates k=0 under the initial condition, and let H c (0) = H0;

[0051] S42, calculating the element retention probability P in the current edge task allocation solution matrix α, and setting the current update temperature optimization number;

[0052] S43. Perform an element retention determination operation on the edge task allocation solution matrix α according to the element retention probability P, and obtain a new edge task allocation solution matrix α(new);

[0053] S44. Calculate the difference d between the objective functions of the new edge-end task allocation solution matrix α(new) and the edge-end task allocation solution matrix α(j) using the following logic:

[0054] d=[xf T (α(new))+yf E (α(new))]-[xf T (α(j))+yf E (α(j))]

[0055] S45. Determine whether the difference d between the objective functions satisfies: d<0; calculate the current updated temperature H. c (k), the probability P1 of accepting a new solution in the j-th optimization search is calculated as follows:

[0056]

[0057] S46. If yes, let the edge task allocation solution matrix α(j+1) of the j+1th optimization be equal to the new edge task allocation solution matrix α(new);

[0058] S47. If not, accept the new edge task assignment solution matrix α(new) using the Metropolis criterion:

[0059] S48. After L iterations, perform a temperature update operation to obtain an updated temperature value H. c (k+1); if j=L, then jump to step S410 and record the edge task allocation solution matrix α(j+1) of the j+1th optimization as the edge task allocation solution matrix α(L) of the Lth optimization; if j=L, then return to step S42 and perform the j+1th optimization operation;

[0060] S49, updating the temperature value H c When (k+1) satisfies the preset conditions, the optimal edge task allocation solution matrix α(L) is determined and output.

[0061] Compared with the simulated annealing hybrid algorithm, the adaptive retention simulated annealing algorithm proposed in the present invention does not incorporate other algorithms, but only improves the search strategy based on the traditional simulated annealing algorithm. It is easier to implement and has better optimization effect.

[0062] Specifically, the present invention does not mix other heuristic algorithms, but only improves the traditional simulated annealing algorithm and proposes an adaptive retention simulated annealing algorithm. As the algorithm progresses, the retention probability of each element in the current solution is continuously improved, the optimization range is dynamically changed, and the ability to search for the global optimal solution is enhanced.

[0063] In a more specific technical solution, in S42, the current update temperature H is set c The number of optimization searches under (k) is L, and any optimization search in L is recorded as j, where j = 0, 1, 2, K, L;

[0064] The following logic is used to express the edge task allocation solution matrix for the jth optimization of the current temperature update:

[0065] α(j),α(j)=[α1(j),...,α i (j),...,α n (j)]

[0066] Where, α i (j) edge allocation strategy for the ith task of the edge task allocation solution matrix α(j);

[0067] Use the following logic to calculate the current updated temperature H c(k), the element retention probability P in the edge task allocation solution matrix α(j) of the jth optimization is:

[0068]

[0069] Where λ is the tuning parameter.

[0070] In a more specific technical solution, in S43, a new edge-end task allocation solution matrix having the same dimension as the edge-end task allocation solution matrix α(j) obtained by the j-th optimization is given:

[0071] α(new)=[α1(new),...,α i (new),...,α n (new)]

[0072] Where, α i (new) is the allocation strategy for the i-th task to be executed in the new edge task allocation solution matrix α(new);

[0073] Using the following logic, each element α in the edge task allocation matrix α(j) is i (j) Perform element retention determination operation:

[0074] Given a random number ε∈(0,1), if ε≤P, then retain the allocation strategy α of the i-th task to be executed i (j), let α i (new) = α i (j);

[0075] If not, the allocation strategy α of the i-th task to be executed is not retained i (j) Randomly select a number from 0 and 1 and assign it to the allocation strategy α of the i-th task to be executed i (new);

[0076] When all elements in the edge task allocation solution matrix α are judged to be complete, a new edge task allocation matrix α(new) is obtained;

[0077] In more specific technical solutions, S49 also includes:

[0078] The edge task allocation matrix α(L) optimized for the Lth time is cooled according to the following linear cooling strategy:

[0079] H c (k+1)=rH c (k)

[0080] In the formula, r is the cooling coefficient;.

[0081] If the updated temperature value H c(k+1) satisfies: H c (k+1)≥H f , then return to step S42;

[0082] If the updated temperature value H c (k+1) satisfies: H c (k+1)<H f , then the edge task allocation matrix α(L) obtained by the L-th optimization is used as the optimal solution for the optimization of the multi-objective function A for the joint optimization of delay and energy consumption, and is recorded as the optimal edge task allocation solution matrix α(L).

[0083] In a more specific technical solution, the AC / DC power system cloud-edge-device collaborative task allocation system includes:

[0084] The pending task determination module is used to determine the number of pending tasks on the cloud and the number of pending tasks on the edge;

[0085] A time and energy matrix construction module is used to execute pending tasks on the cloud and edge, measure and record execution time consumption data and execution energy consumption data, and process them to obtain the edge time matrix, edge energy matrix, cloud time matrix and cloud energy matrix. The time and energy matrix construction module is connected to the pending task determination module;

[0086] A multi-objective function construction and solution module is used to obtain edge allocation strategies and cloud allocation strategies, construct edge task allocation matrices and cloud task allocation matrices for the edge and cloud, and establish and solve a multi-objective function for joint optimization of latency and energy consumption to obtain the edge task allocation solution matrix. The multi-objective function construction and solution module is connected to the time and energy matrix construction module.

[0087] The final allocation strategy determination module is used to solve the edge task allocation matrix and optimize the multi-objective function of delay and energy consumption through the adaptive retention simulated annealing algorithm to obtain the optimal edge task allocation matrix: α(L), α(L) = [α1(L), ..., α i (L),...,α n (L)], where α i (L) is the final allocation strategy of the i-th task to be executed at the edge end. The final allocation strategy determination module is connected with the multi-objective function construction and solution module.

[0088] Compared with the prior art, the present invention has the following advantages:

[0089] The present invention not only considers the time delay optimization objective function, but also considers the multi-objective function of energy consumption optimization. The corresponding intelligent optimization algorithm applied by the present invention is more suitable for optimizing multi-objective functions; the simulated annealing algorithm of the present invention does not rely on a large number of calculations to obtain the global optimal solution, which improves the computing efficiency and can obtain the global optimal solution in more application scenarios.

[0090] Compared with the traditional simulated annealing algorithm, the adaptive retention simulated annealing algorithm proposed in the present invention adaptively retains some features of the current solution according to the current temperature, overcomes the disadvantage that the traditional simulated annealing algorithm requires a large amount of calculations to obtain the global optimal solution, and significantly improves the optimization speed.

[0091] The present invention comprehensively considers the latency and energy consumption of cloud-edge collaborative task allocation by establishing a multi-objective function for joint optimization of latency and energy consumption. At the same time, it uses the heuristic algorithm of adaptive retained simulated annealing algorithm as the solution method, which saves the training time of deep learning models, saves computing resources, and improves the efficiency and convenience of solution.

[0092] Compared with the simulated annealing hybrid algorithm, the adaptive retention simulated annealing algorithm proposed in the present invention does not incorporate other algorithms, but only improves the search strategy based on the traditional simulated annealing algorithm. It is easier to implement and has better optimization effect.

[0093] The present invention does not mix other heuristic algorithms, but only improves the traditional simulated annealing algorithm and proposes an adaptive retention simulated annealing algorithm. As the algorithm progresses, the retention probability of each element in the current solution is continuously improved, the optimization range is dynamically changed, and the ability to search for the global optimal solution is enhanced.

[0094] The present invention solves the technical problems in the prior art that the corresponding intelligent optimization algorithm is not suitable for optimizing multiple objective functions and relies on a large number of calculations, which makes it difficult to find the global optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 This is a schematic diagram of the basic steps of the method for allocating tasks between the cloud and the edge in an AC / DC power system according to Example 1 of the present invention;

[0096] Figure 2 Schematic diagram of the specific steps of establishing a multi-objective function in Example 1 of the present invention;

[0097] Figure 3 Schematic diagram of specific steps for optimizing a multi-objective function according to embodiment 1 of the present invention;

[0098] Figure 4 This is a comparison chart of simulation results between the adaptive retention simulated annealing algorithm of Example 2 of the present invention and the traditional simulated annealing algorithm;

[0099] Figure 5 This is a comparison chart of simulation results of the adaptive retention simulated annealing algorithm and the simulated annealing hybrid algorithm in Example 2 of the present invention. DETAILED DESCRIPTION

[0100] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0101] Example 1

[0102] like Figure 1 As shown, the AC / DC power system cloud-edge collaborative task allocation method provided by the present invention includes the following basic steps:

[0103] S1. Determine the number of tasks that the system needs to process in the cloud and at the edge.

[0104] In this embodiment, the AC and DC power supplies are denoted as the system, and the number of tasks that the system needs to process at the cloud and edge is n, and any one of the n tasks is denoted as the i-th task, i = 1, 2, K, n;

[0105] S2, measure the time and energy corresponding to executing each task;

[0106] In this embodiment, n tasks are executed simultaneously on the edge and the cloud, and the execution time and energy consumption data are recorded. Specifically, n tasks are executed on the edge in sequence, and the time and energy consumed by each task executed once on the edge are measured to obtain the edge time matrix T e and the edge energy matrix E e ,T e =[T e1 ,K,T ei ,K,T en ] T , E e =[E e1 ,K,E ei ,K,E en ] T , where T ei is the time consumed by the execution of the i-th task at the edge, E ei The energy consumed by executing the i-th task once at the edge;

[0107] Execute n tasks in the cloud in sequence, measure the time and energy consumed by each task in the cloud, and obtain the cloud time matrix T c and cloud energy matrix E c , T c =[T c1 ,K,T ci ,K,T cn ] T , E c =[E c1 ,K,E ci ,K,E cn ] T , where T ci is the time consumed by the execution of the i-th task in the cloud, E ci The energy consumed by executing the i-th task in the cloud once;

[0108] S3, establish a multi-objective function for joint optimization of delay and energy consumption;

[0109] like Figure 2 As shown, in this embodiment, step S3 of establishing a multi-objective function further includes the following specific steps:

[0110] S31. Construct an edge task allocation matrix and a cloud task allocation matrix;

[0111] In this embodiment, the edge task allocation matrix is ​​expressed as follows:

[0112] α, α=[α1,...,α i ,...,α n ]

[0113] Among them, α i is the allocation strategy of the i-th task at the edge, and α i ∈{0,1}, if α i =1, indicating that the i-th task is assigned to the edge. If α i = 0, indicating that the i-th task is assigned to the cloud; the cloud task assignment matrix is ​​β, β = [β1, ..., β i ,...,β n ], where β i is the allocation strategy of the i-th task in the cloud, and β i ∈{0,1}, if β i =1, indicating that the i-th task is assigned to the cloud. If β i =0, indicating that the i-th task is assigned to the edge;

[0114] In this embodiment, the allocation strategy α of the i-th task at the edge is i and the allocation strategy β of the i-th task in the cloudi It is an exclusive OR relationship, that is, if one of the parameters is determined, the other parameter can be determined;

[0115] S32. Construct a multi-objective function with low latency and low energy consumption as optimization goals;

[0116] In this embodiment, the latency optimization objective function is as follows:

[0117] f T ,f T =αT e +βT c ;

[0118] The energy consumption optimization objective function is recorded as follows:

[0119] f E , f E =αE e +βE c ;

[0120] In this embodiment, a multi-objective function with low latency and low energy consumption as optimization goals is established, and its expression is:

[0121] A=min(xf T +yf E )

[0122] Where A is the multi-objective function value, x is the weight parameter of the low latency objective, and y is the weight parameter of the low energy consumption objective, and the condition is satisfied: x + y = 1;

[0123] The solution of the multi-objective function A is the task allocation matrix α at the edge;

[0124] S4, optimizing the multi-objective function through adaptive retention simulated annealing algorithm;

[0125] In this embodiment, the multi-objective function A is optimized by the adaptive retention simulated annealing algorithm to obtain the final solution of the multi-objective function A, which is recorded as the optimal edge task allocation matrix:

[0126] α(L), α(L)=[α1(L),...,α i (L),...,α n (L)]

[0127] Among them, α i (L) is the final allocation strategy of the i-th task at the edge.

[0128] like Figure 3 As shown, in this embodiment, step S4 of optimizing the multi-objective function further includes the following specific steps:

[0129] S41. Given the initial temperature, final temperature, and current updated temperature;

[0130] In this embodiment, the initial temperature is set to H0 and the final temperature is set to H f , the current updated temperature is H c (k), k is the number of temperature updates;

[0131] In this embodiment, it is assumed that the temperature update times k=0 under the initial condition, and H c (0) = H0;

[0132] S42, calculate the retention probability P of the elements in the current solution α, and set the number of optimization times for the current updated temperature;

[0133] S43, determining whether to retain the elements of the current solution with probability P;

[0134] In this embodiment, the current update temperature H is set c The number of optimization searches under (k) is L, and any optimization search in L is recorded as j, j = 0, 1, 2, K, L, and the edge task allocation matrix of the jth optimization search of the current updated temperature is recorded as:

[0135] α(j),α(j)=[α1(j),...,α i (j),...,α n (j)]

[0136] Among them, α i (j) is the allocation strategy of the i-th task in the edge task allocation matrix α(j);

[0137] Calculate the current updated temperature H c (k), the retention probability P of the elements in the edge task allocation matrix α(j) of the jth optimization is calculated as follows:

[0138]

[0139] Among them, λ is the tuning parameter;

[0140] S44, processing to obtain a new solution α(new);

[0141] In this embodiment, a new edge task allocation matrix with the same dimension as the edge task allocation matrix α(j) optimized for the jth time is given:

[0142] α(new)=[α1(new),...,α i (new),...,α n (new)]

[0143] Among them, αi (new) is the allocation strategy of the i-th task in the new edge task allocation matrix α(new);

[0144] For each element α in the edge task assignment matrix α(j), i (j) Make the following judgment:

[0145] Given a random number ε∈(0,1), if ε≤P, then keep α i (j), let α i (new) = α i (j); otherwise, α is not retained i (j) Randomly select a number from 0 and 1 and assign it to α i (new);

[0146] After all elements are judged, the new edge task allocation matrix α(new) is obtained after the assignment is completed;

[0147] S45, calculating the difference d;

[0148] In this embodiment, the difference d between the objective function of the new edge task allocation matrix α(new) and the edge task allocation matrix α(j) is calculated using the following formula:

[0149] d=[xf T (α(new))+yf E (α(new))]-[xf T (α(j))+yf E (α(j))]

[0150] S46, the difference d between the objective function of the new edge-end task allocation matrix α(new) and the edge-end task allocation matrix α(j) is less than 0;

[0151] Calculate the current updated temperature H c (k), the probability P1 of accepting a new solution in the j-th optimization search is calculated as follows:

[0152]

[0153] S47. If yes, then α=α(new);

[0154] In this embodiment, if the difference d between the objective function of the new edge task allocation matrix α(new) and the edge task allocation matrix α(j) is less than 0, then the edge task allocation matrix α(j+1) of the j+1th optimization is set to be equal to the new edge task allocation matrix α(new);

[0155] S48. Otherwise, accept the new solution using the Metropolis criterion:

[0156] In this embodiment, a random number ε∈(0,1) is given. If ε≤P1, the edge task allocation matrix α(j+1) of the j+1th optimization is set to be equal to the new edge task allocation matrix α(new); otherwise, the edge task allocation matrix α(j+1) of the j+1th optimization is set to be equal to the edge task allocation matrix α(j) of the jth optimization.

[0157] After the judgment is completed, the edge task allocation matrix α(j+1) of the j+1th optimization is obtained;

[0158] S49, update the temperature after L iterations;

[0159] In this embodiment, if j=L, the process proceeds to step S410, and the edge task allocation matrix α(j+1) obtained in step S46 for the j+1th optimization is recorded as the edge task allocation matrix α(L) for the Lth optimization. Otherwise, the process returns to step S42 and performs the j+1th optimization.

[0160] S410, determine whether H is satisfied c (k+1)≧H f ;

[0161] In this embodiment, the edge task allocation matrix α(L) optimized for the Lth time is cooled according to the linear cooling strategy, which is expressed as follows:

[0162] H c (k+1)=rH c (k)

[0163] Among them, r is the temperature reduction coefficient, and its value range is (0.8, 1);

[0164] The updated temperature value H after the k+1th temperature update c (k+1) makes the following judgment:

[0165] If H c (k+1)≥H f , then return to step S42;

[0166] If H c (k+1)<H f , then the edge-end task allocation matrix α(L) obtained in the L-th optimization in step S410 is used as the optimal solution for the optimization of the multi-objective function A, and is recorded as the optimal edge-end task allocation matrix α(L). The optimal edge-end task allocation matrix α(L) is output, and the optimization ends.

[0167] S411, output the optimal solution;

[0168] In this embodiment, when the number of optimization searches j=0 in step S4, if the number of temperature updates k=0, the initial edge task allocation matrix α(0) is randomly given, otherwise the initial edge task allocation matrix α(0) is equal to the edge task allocation matrix α(L) of the Lth optimization search in step S48.

[0169] Example 2

[0170] In this embodiment, a simulation is performed to illustrate the effectiveness of the present invention.

[0171] The simulation parameter settings of the adaptive retention simulated annealing algorithm are shown in Table 1:

[0172] Table 1

[0173]

[0174]

[0175] During the simulation, it is assumed that the number of tasks that the AC / DC power supply system needs to process is 20. The optimal solution is obtained according to the parameters in Table 1. The optimization results are compared with other algorithms, and a total of 30 comparative experiments are conducted.

[0176] See also Figure 4 Comparison of simulation results between the adaptive retention simulated annealing algorithm and the traditional simulated annealing algorithm, and Figure 5 Comparison of simulation results between the adaptive preserved simulated annealing algorithm and the simulated annealing-particle swarm optimization hybrid algorithm. This comparison shows that the adaptive preserved simulated annealing algorithm proposed in this invention achieves better optimization results, and the fluctuations in the optimal objective function value over multiple experiments are far less than those of the traditional simulated annealing algorithm, demonstrating greater stability.

[0177] In summary, the present invention not only considers the time delay optimization objective function, but also considers the energy consumption optimization multi-objective function. The corresponding intelligent optimization algorithm applied by the present invention is more suitable for optimizing multi-objective functions; the simulated annealing algorithm of the present invention does not rely on a large number of calculations to obtain the global optimal solution, which improves the computing efficiency and can obtain the global optimal solution in more application scenarios.

[0178] Compared with the traditional simulated annealing algorithm, the adaptive retention simulated annealing algorithm proposed in the present invention adaptively retains some features of the current solution according to the current temperature, overcomes the disadvantage that the traditional simulated annealing algorithm requires a large amount of calculations to obtain the global optimal solution, and significantly improves the optimization speed.

[0179] The present invention comprehensively considers the latency and energy consumption of cloud-edge collaborative task allocation by establishing a multi-objective function for joint optimization of latency and energy consumption. At the same time, it uses the heuristic algorithm of adaptive retained simulated annealing algorithm as the solution method, which saves the training time of deep learning models, saves computing resources, and improves the efficiency and convenience of solution.

[0180] Compared with the simulated annealing hybrid algorithm, the adaptive retention simulated annealing algorithm proposed in the present invention does not incorporate other algorithms, but only improves the search strategy based on the traditional simulated annealing algorithm. It is easier to implement and has better optimization effect.

[0181] The present invention solves the technical problems in the prior art that the corresponding intelligent optimization algorithm is not suitable for optimizing multiple objective functions and relies on a large number of calculations, which makes it difficult to find the global optimal solution.

[0182] The present invention does not mix other heuristic algorithms, but only improves the traditional simulated annealing algorithm and proposes an adaptive retention simulated annealing algorithm. As the algorithm progresses, the retention probability of each element in the current solution is continuously improved, the optimization range is dynamically changed, and the ability to search for the global optimal solution is enhanced.

[0183] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for allocating tasks between the cloud and the edge in an AC / DC power supply system, characterized in that: The method comprises: S1. Determine the number of tasks to be processed on the cloud and the number of tasks to be processed on the edge. S2. Execute the task to be processed on the cloud and edge, measure and record execution time consumption data and execution energy consumption data, and process the data to obtain an edge time matrix, an edge energy matrix, a cloud time matrix, and a cloud energy matrix; S3, obtaining an edge allocation strategy and a cloud allocation strategy to construct an edge task allocation matrix and a cloud task allocation matrix for the edge and the cloud, thereby establishing and solving a multi-objective function for joint optimization of latency and energy consumption to obtain an edge task allocation solution matrix; S4. According to the edge task allocation solution matrix, an adaptive retained simulated annealing algorithm is used to optimize the multi-objective function of delay and energy consumption joint optimization to obtain the optimal edge task allocation matrix: α(L), α(L) = [α1(L), ..., α i (L),...,α n (L)], where α i (L) is the final allocation strategy of the i-th task to be executed at the edge end.

2. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 1, characterized in that: The S2 includes: S21, execute n tasks to be processed in sequence on the edge end, measure the execution time consumption data and the execution energy consumption data of each task to be processed in a single execution on the edge end, and obtain the edge end time matrix T e , the edge energy matrix E e : T e =[T e1 ,K,T ei ,K,T en ] T E e =[E e1 ,K,E ei ,K,E en ] T Where, T ei is the time consumed by the i-th task to be executed once at the edge end, E ei The energy consumed by executing the i-th task to be executed once at the edge end; S22, executing n tasks to be executed in sequence on the cloud, measuring the execution time consumption data and the execution energy consumption data of each task to be executed in the cloud for a single execution, and obtaining the cloud time matrix T c 、The cloud energy matrix E c : T c =[T c1 ,K,T ci ,K,T cn ] T E c =[E c1 ,K,E ci ,K,E cn ] T Where, T ci is the time consumed by the i-th task to be executed once in the cloud, E ci The energy consumed when the i-th task to be executed is executed once in the cloud.

3. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 1, characterized in that: The S3 includes: S31. Construct the edge task allocation matrix using the following logic: α，α＝[α1,...,α i ,...,α n ] Where, α i Assign a strategy to the edge end for the i-th task to be executed, and α i ∈{0,1}; The cloud task allocation matrix is ​​constructed using the following logic: β，β＝[β1,...,β i ,...,β n ] Where, β i Assign a strategy to the i-th task to be executed in the cloud, and β i ∈{0,1}; S32: Construct a time delay optimization objective function and an energy consumption optimization objective function, and obtain the multi-objective function for joint optimization of time delay and energy consumption based on the obtained objective function.

4. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 3, characterized in that: In the step S31, the edge allocation strategy α of the i-th task to be executed is set. i , and the cloud allocation strategy β of the i-th task to be executed i It is an exclusive OR relationship.

5. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 3, characterized in that: The S32 further includes: S321. Construct the delay optimization objective function using the following logic: f T ,f T =αT e +βT c ; S322: Construct the energy consumption optimization objective function using the following logic: f E ,f E =αE e +βE c ; S323: Combine the delay optimization objective function and the energy consumption optimization objective function to establish the delay and energy consumption joint optimization multi-objective function: A=min(xf T +yf E )。 Where A is the multi-objective function value, x is the weight parameter of the low latency objective, and y is the weight parameter of the low energy consumption objective, where: x + y = 1.

6. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 1, characterized in that: The S4 includes: S41, given initial temperature H0, end temperature H f 、Current update temperature H c (k), where k is the number of temperature updates. Assume that the number of temperature updates k=0 under the initial condition, and let H c (0) = H0; S42, calculating the element retention probability P in the current edge task allocation solution matrix α, and setting the current update temperature optimization number; S43. Perform an element retention determination operation on the edge task allocation solution matrix α according to the element retention probability P, and obtain a new edge task allocation solution matrix α(new); S44. Calculate the difference d between the objective functions of the new edge-end task allocation solution matrix α(new) and the edge-end task allocation solution matrix α(j) using the following logic: d=[xf T (α(new))+yf E (α(new))]-[xf T (α(j))+yf E (α(j))] S45, determine whether the difference d between the objective functions satisfies: d<0; calculate the current updated temperature H c (k), the probability P1 of accepting a new solution in the j-th optimization search is calculated as follows: S46. If yes, set the edge task allocation solution matrix α(j+1) obtained by the j+1th optimization to be equal to the new edge task allocation solution matrix α(new); S47. If not, accept the new edge task allocation solution matrix α(new) using the Metropolis criterion: S48. After L iterations, perform a temperature update operation to obtain an updated temperature value H. c (k+1); if j=L, then jump to step S410 and record the edge task allocation solution matrix α(j+1) of the j+1th optimization as the edge task allocation solution matrix α(L) of the Lth optimization; if j=L, then return to step S42 and perform the j+1th optimization operation; S49, in the updating temperature value H c When (k+1) satisfies the preset conditions, the optimal edge-end task allocation solution matrix α(L) is determined and output.

7. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 6, characterized in that: In the step S42, the current update temperature H is set c The number of optimization searches under (k) is L, and any optimization search in the number of optimization searches L is recorded as j, where j = 0, 1, 2, K, L; The following logic is used to express the edge task allocation solution matrix of the j-th optimization of the current updated temperature: α(j),α(j)=[α1(j),...,α i (j),...,α n (j)] Where, α i (j) solving the edge task allocation strategy for the i-th task in the edge task allocation matrix α(j); Calculate the current update temperature H using the following logic c (k), the element retention probability P in the edge task allocation solution matrix α(j) of the j-th optimization is: Where λ is the tuning parameter.

8. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 6, wherein: In S43, a new edge-end task allocation solution matrix having the same dimension as the edge-end task allocation solution matrix α(j) obtained by the j-th optimization is given: α(new)=[α1(new),...,α i (new),...,a n (new)] Where, α i (new) is the allocation strategy of the i-th task to be executed in the new edge task allocation solution matrix α(new); Using the following logic, each element α in the edge task allocation matrix α(j) is i (j) performing the element retention determination operation: Given a random number ε∈(0,1), if ε≤P, then retain the allocation strategy α of the i-th task to be executed i (j), let α i (new) = α i (j); If not, the allocation strategy α of the i-th task to be executed is not retained i (j) Randomly select a number from 0 and 1 and assign it to the allocation strategy α of the i-th task to be executed i (new); When all elements in the edge-end task allocation solution matrix α are judged to be complete, the new edge-end task allocation matrix α(new) is obtained.

9. The method for allocating cloud-edge-device collaborative tasks in an AC / DC power supply system according to claim 6, wherein: The S49 further includes: The edge task allocation matrix α(L) obtained by the L-th optimization is cooled according to the following linear cooling strategy: H c (k+1)=rH c (k) In the formula, r is the cooling coefficient;. If the updated temperature value H c (k+1) satisfies: H c (k+1)≥H f , then return to step S42; If the updated temperature value H c (k+1) satisfies: H c (k+1)<H f , then the edge task allocation matrix α(L) obtained by the L-th optimization is used as the optimal solution for the optimization of the multi-objective function A for the joint optimization of time delay and energy consumption, and is recorded as the optimal edge task allocation solution matrix α(L).

10. AC / DC power supply system cloud-edge collaborative task allocation system, characterized by: The system comprises: The pending task determination module is used to determine the number of pending tasks on the cloud and the number of pending tasks on the edge; A time and energy matrix construction module is used to execute the pending tasks on the cloud and edge, measure and record the execution time consumption data and the execution energy consumption data, and process the data to obtain the edge time matrix, edge energy matrix, cloud time matrix and cloud energy matrix. The time and energy matrix construction module is connected to the pending task determination module; A multi-objective function construction and solution module is used to obtain edge allocation strategies and cloud allocation strategies, so as to construct edge task allocation matrices and cloud task allocation matrices for the edge and the cloud, and to establish and solve a multi-objective function for joint optimization of delay and energy consumption to obtain an edge task allocation solution matrix. The multi-objective function construction and solution module is connected to the time and energy matrix construction module; The final allocation strategy determination module is used to perform an optimization operation on the multi-objective function of joint optimization of delay and energy consumption according to the edge task allocation solution matrix through the adaptive retention simulated annealing algorithm to obtain the optimal edge task allocation matrix: α(L), α(L) = [α1(L), ..., α i (L),...,α n (L)], where α i (L) is the final allocation strategy of the i-th task to be executed at the edge end, and the final allocation strategy determination module is connected to the multi-objective function construction and solution module.

Images