Decentralization convex optimization method and system for enhancing AGV edge calculation energy efficiency

Abstract

The invention provides a decentration convex optimization method and system for enhancing AGV edge computing energy efficiency, and the method comprises the steps: building a task unloading model based on an edge computing network; based on a preset base station index set, an AGV index set and the calculation task of the AGV, obtaining a state space and an action space of the task unloading model; establishing a flow model, a calculation model and a communication model; integrating the models to obtain a joint optimization problem of task unloading and resource allocation; decomposing the joint optimization problem into a task unloading problem and a resource allocation problem by adopting a hierarchical decomposition method; a probability task unloading strategy is adopted, load prediction is carried out on a task unloading problem, a resource allocation problem is decomposed and convexly optimized, and a joint optimization problem is solved in combination with a load prediction result. The invention aims to realize calculation unloading satisfying energy constraint in the AGV edge calculation network so as to reduce the time delay of the edge calculation network and promote the development of the edge calculation technology.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle edge computing technology, and specifically relates to a decentralized convex optimization method and system for enhancing the energy efficiency of AGV edge computing. Background Art

[0002] In the field of high-end equipment cable manufacturing, logistics automation and intelligence are the key to improving production efficiency and reducing costs. In response to the country's call to promote the digital transformation of the manufacturing industry, it is now necessary to digitally transform and upgrade old factories. This transformation is not only the key to improving production efficiency and reducing costs, but also the only way to achieve modernization in the field of high-end equipment cable manufacturing. In order to ensure real-time production process control in the complex manufacturing logistics scenarios of high-end equipment cables, it is necessary to provide ultra-high reliability and millimeter-level latency for the automatic guided vehicles (AGVs) performing tasks. High-complexity real-time computing often requires explosive computing resources and strict time delays, but the computing power of AGVs is limited and insufficient to meet these requirements. Therefore, vehicle edge computing (VEC) is proposed to solve the above problems, and latency is greatly reduced by deploying computing services near AGVs.

[0003] However, VEC systems face the non-convexity and discreteness of optimization problems when deciding task offloading and resource allocation, as well as the additional communication delays and security issues brought by centralized solutions. To overcome these challenges, a decentralized convex optimization method and system for enhancing the energy efficiency of AGV edge computing is needed. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a decentralized convex optimization method and system for enhancing the energy efficiency of AGV edge computing. By selecting offloading targets and allocating resources, computing offloading that meets energy constraints in the AGV edge computing network is achieved, so as to reduce the latency of the edge computing network and promote the development of edge computing technology.

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

[0006] A decentralized convex optimization method for enhancing the energy efficiency of AGV edge computing includes the following steps:

[0007] Based on the edge computing network, a task offloading model of AGV is constructed; wherein the edge computing network includes several base stations, each of which is equipped with a VEC server that provides wireless access and computing resources for AGV;

[0008] Based on the preset base station index set, AGV index set and AGV computing tasks, the state space and action space of the task offloading model are obtained;

[0009] Based on the state space and action space of the task offloading model, the AGV traffic model, the calculation model of the AGV calculation task, and the communication model between the AGV and the base station are established;

[0010] Integrate the AGV traffic model, the computation model, and the communication model to obtain a joint optimization problem of task offloading and resource allocation;

[0011] Using a hierarchical decomposition method, the joint optimization problem is decomposed into a task offloading problem and a resource allocation problem;

[0012] Using a probabilistic task offloading strategy, load prediction is performed on the task offloading problem to obtain a load prediction result;

[0013] The resource allocation problem is decomposed and convexly optimized, and the joint optimization problem is solved in combination with the load prediction result to complete the decentralized convex optimization for enhancing the energy efficiency of AGV edge computing.

[0014] Preferably, the method for establishing the AGV flow model includes:

[0015] Establishing a two-dimensional geographical location set of passable locations in the task offloading model, and obtaining the location of the AGV and the location of the base station based on the two-dimensional geographical location set;

[0016] Calculate the Euclidean distance between the position of the AGV and the position of the base station based on the square norm;

[0017] When the Euclidean distance is within the preset base station coverage radius, the base station receives the task from the AGV and completes the construction of the AGV traffic model.

[0018] Preferably, the method for establishing a computing model for an AGV computing task includes:

[0019] Based on DVFS technology, the maximum nominal CPU frequency of the base station is obtained;

[0020] The CPU frequency scaling ratio determined by the preset AGV to complete the computing task is combined with the maximum nominal CPU frequency of the base station and the computing resources required by the AGV to complete the computing task to obtain the execution time and energy consumption of the computing task;

[0021] The computing model is obtained based on the execution time and energy consumption of the computing task.

[0022] Preferably, the method for establishing a communication model between the AGV and the base station includes:

[0023] Based on the maximum transmission power of the AGV and the decision variables, the actual transmission power of the AGV is obtained;

[0024] Based on the AGV position and the base station position, the channel gain is obtained;

[0025] Based on the actual transmission power, the channel gain, the fixed bandwidth occupied by the AGV when communicating with the base station, and the noise power, the transmission power of the data transmission between the AGV and the base station is obtained;

[0026] Obtaining a transmission delay based on the transmission power and the length of the data transmitted between the AGV and the base station;

[0027] Based on the actual transmission power and the transmission delay, the energy consumption of data transmission between the AGV and the base station is obtained, and the construction of the communication model is completed.

[0028] Preferably, the goal of the joint optimization problem of task offloading and resource allocation is to minimize the total task response time on the AGV and the energy consumption of data transmission between the AGV and the base station, while satisfying the base station constraints; wherein the total task response time includes the execution time of the computing task, the transmission delay, and the context transmission delay when context transmission occurs; and the base station constraints include computing utilization constraints, energy consumption constraints, and time constraints.

[0029] Preferably, a probabilistic task offloading strategy is adopted, and load prediction for the task offloading problem includes two scenarios: a non-overlapping base station scenario and an overlapping base station scenario, wherein the overlapping base station scenario is an overlapping area at the boundary of the coverage areas of two base stations; wherein, in the overlapping base station scenario, the method for obtaining the load prediction result includes:

[0030] Based on the probability of AGV unloading tasks to the preset base station, an unloading probability matrix is ​​established;

[0031] Based on the adjacency matrix and the offloading probability matrix, obtaining the offloading workload on the base station and the total estimation of the context transmission delay of the base station;

[0032] Based on the offloading workload and the total estimation of context transmission delay, obtaining the optimal offloading delay of all AGVs of the corresponding base station;

[0033] Based on the optimal delay of all AGV unloading, the computing time of all AGV computing tasks is optimized to obtain the load prediction results.

[0034] Preferably, the method for solving the joint optimization problem includes:

[0035] Decomposing the resource allocation problem to obtain a decentralized resource allocation problem for a single base station, and rewriting the decentralized resource allocation problem as a convex optimization problem;

[0036] Introducing Lagrangian variables into the convex optimization problem for dual decomposition to obtain a computation subproblem and a communication subproblem;

[0037] Based on the computational subproblem, the communication subproblem and the load prediction result, the joint optimization problem is solved to achieve optimal allocation of resources.

[0038] The present invention also provides a decentralized convex optimization system for enhancing the energy efficiency of AGV edge computing, which is used to implement the method, including:

[0039] A task offloading model building module is used to build a task offloading model for AGV based on an edge computing network; wherein the edge computing network includes a plurality of base stations, each of which is equipped with a VEC server that provides wireless access and computing resources for the AGV;

[0040] A space partitioning module is used to obtain the state space and action space of the task offloading model based on a preset base station index set, an AGV index set, and the computing tasks of the AGV;

[0041] A model building module is used to build an AGV traffic model, a calculation model of AGV calculation tasks, and a communication model between AGV and base station based on the state space and action space of the task offloading model;

[0042] A model integration module, used to integrate the AGV traffic model, the calculation model and the communication model to obtain a joint optimization problem of task offloading and resource allocation;

[0043] A problem decomposition module, used to decompose the joint optimization problem into a task offloading problem and a resource allocation problem by adopting a hierarchical decomposition method;

[0044] A load prediction module is used to adopt a probabilistic task offloading strategy to perform load prediction on the task offloading problem and obtain a load prediction result;

[0045] The problem-solving module is used to decompose and convexly optimize the resource allocation problem, combine the load prediction results, solve the joint optimization problem, and complete the decentralized convex optimization for enhancing the energy efficiency of AGV edge computing.

[0046] Compared with the prior art, the invention has the following beneficial effects: the invention converts complex optimization problems into easily solvable convex subproblems by means of hierarchical decomposition, converts task offloading problems into continuous optimization problems by means of probability-based strategies, and reduces the communication between AGVs and base stations by means of decentralized convex optimization, thereby achieving efficient computation offloading and resource allocation while ensuring task execution time constraints and energy consumption constraints, reducing system latency, and improving overall production efficiency. It promotes the development of edge computing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0048] Figure 1 This is a flow chart of a decentralized convex optimization method for enhancing the energy efficiency of AGV edge computing according to Embodiment 1 of the present invention;

[0049] Figure 2 This is a diagram of a task offloading model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in 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. 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.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Embodiment 1

[0053] like Figure 1 As shown, a decentralized convex optimization method for enhancing the energy efficiency of AGV edge computing includes the following steps:

[0054] S1: Based on the edge computing network, a task offloading model of AGV is constructed; wherein the edge computing network includes several base stations, each of which is equipped with a VEC server that provides wireless access and computing resources for AGV; specifically, Figure 2 As shown, the VEC system where the mobile AGV is located includes N base stations. Each base station is equipped with a VEC server, which can provide edge computing. The AGV can only be connected to one base station during movement.

[0055] S2: Based on the preset base station index set, AGV index set and AGV computing tasks, obtain the state space and action space of the task offloading model; in this embodiment, the number of base stations is defined as m, and the number of AGVs is defined as n. M = {1, 2, ..., m} is defined to represent the index set of the base station, and N = {1, 2, ..., n} is defined to represent the index set of the AGV.

[0056] Among them, j∈M is the index of the base station, and i∈N is the index of the AGV. The edge computing network consists of base stations, each of which contains a MEC server.

[0057] Among them, the server provides wireless access and computing resources for AGV.

[0058] Among them, AGV is defined to solve periodic tasks with a period of ΔT. At each time step t, each vehicle i∈N will generate a task.

[0059] In the task offloading model, the period ΔT is the longest acceptable delay to complete the task. The computational task of AGV-i is defined as [D i ,C i ].

[0060] Among them, D i is the length of the data transmitted between the AGV and the base station (in bits). The transmission data is mainly composed of calculation instructions, so D i Considered as a constant.

[0061] Among them, C i It represents the computing resources required to complete the task (in CPU cycles), and its average value is C i .

[0062] In the present invention, tasks are considered as atoms and cannot be split. Only the pure offloading case where the tasks are completely executed on a single base station is considered.

[0063] S3: Based on the state space and action space of the task offloading model, establish the AGV traffic model, the calculation model of the AGV calculation task, and the communication model between the AGV and the base station;

[0064] S31: In this embodiment, a further implementation method is that the method for establishing the AGV flow model includes:

[0065] A two-dimensional geographic location set of passable locations in the task offloading model is established, and based on the two-dimensional geographic location set, the location of the AGV and the location of the base station are obtained; specifically, the two-dimensional geographic location set of passable locations in the system (task offloading model) is defined as Define the position of AGV-i as P i , the base station location is r j .

[0066] The Euclidean distance between the AGV position and the base station position is calculated based on the square norm; the base station coverage radius is defined as L j .

[0067] When the Euclidean distance is within the preset base station coverage radius, the base station receives the task from the AGV and completes the construction of the AGV flow model. j ,L j ) before it can receive tasks from AGV.

[0068] Furthermore, the proactive migration service can better manage the context transfer between base stations, and the context transmission delay across base stations is assumed to be known and proportional to the workload.

[0069] S32: A further embodiment is that the computing power of the base station is provided by a CPU supporting dynamic voltage and frequency scaling (DVFS). DVFS is a technique for adjusting CPU frequency to balance energy consumption and task execution time, and its available frequencies are limited to a limited set of values, i.e., the available clock frequency vector F of base station-j = {f j1 ,f j2 ,...,f jN}. The difference between consecutive frequencies is so small that the frequency can be approximately treated as a continuous variable. Define the maximum nominal CPU frequency of base station-j as f j .

[0070] Specifically, the method of establishing the calculation model of the AGV calculation task includes:

[0071] Based on DVFS technology, the maximum nominal CPU frequency of the base station is obtained;

[0072] The CPU frequency scaling ratio determined by the preset AGV to complete the computing task is combined with the maximum nominal CPU frequency of the base station and the computing resources required by the AGV to complete the computing task to obtain the execution time and energy consumption of the computing task; specifically, define ω ij ∈(0,1) is the CPU frequency scaling ratio determined by AGV-i to complete the task. The execution time can be expressed as:

[0073]

[0074] The energy consumption is:

[0075]

[0076] Among them, θ j It depends on the energy consumption coefficient of the chip architecture.

[0077] Based on the execution time and energy consumption of the computing tasks, a computing model is obtained.

[0078] A further implementation method is that the method for establishing a communication model between the AGV and the base station includes:

[0079] Based on the maximum transmission power p of AGV i and the decision variable β ij , get the actual transmission power β of AGV ij p i ; Among them, β ij ∈(0,1) is a decision variable that allows AGV to make a trade-off between energy consumption and transmission delay.

[0080] Based on the AGV position and the base station position, the channel gain is obtained;

[0081] Based on the actual transmission power, channel gain, fixed bandwidth occupied by the AGV and the base station when communicating, and noise power, the transmission power of the data transmission between the AGV and the base station is obtained;

[0082] Specifically, define the channel gain as h ij =g(p i -r j ), including path loss and fading. When AGV communicates with the base station, it occupies a fixed bandwidth B p , the noise power is defined as δ j , g represents the corresponding base station. The transmission power is expressed as:

[0083]

[0084] Based on the transmission power and the length of the data exchanged between the AGV and the base station, the transmission delay is obtained; the transmission delay is expressed as:

[0085]

[0086] If context transfer occurs, the additional delay is C i ε.

[0087] Based on the actual transmission power and transmission delay, the energy consumption of data transmission between AGV and base station is obtained, and the construction of the communication model is completed.

[0088] Specifically, the energy consumption of data transmission can be expressed by transmission power and transmission delay as:

[0089]

[0090] S4: Integrate the AGV traffic model, computing model and communication model to obtain the joint optimization problem of task offloading and resource allocation; a further implementation method is that the goal of the joint optimization problem of task offloading and resource allocation is to minimize the total task response time on the AGV and the energy consumption of data transmission between the AGV and the base station, while satisfying the base station constraints; wherein, the present invention focuses on a single time step, and the total task response time includes the execution time of the computing task, the transmission delay and the context transmission delay when the context transmission occurs; the base station constraints include computing utilization constraints, energy consumption constraints and time constraints.

[0091] In this embodiment, the offloading decision variable Z and the decision matrix Z' of the next time step are defined. If AGV-i offloads a task to base station-j in the current time step and is still within the coverage of base station-j in the next time step, the AGV maintains the same offloading target, otherwise, it must communicate with another base station immediately. In this case, context transfer occurs and additional communication delay is calculated.

[0092] Among them, the context transmission delay is expressed as:

[0093]

[0094] The joint optimization problem is defined as:

[0095]

[0096] Among them, η i is a weighting parameter used to strike a balance between latency and energy saving.

[0097] Among them, the base station calculation utilization constraint is expressed as:

[0098]

[0099] Among them, U j Indicates the upper limit of CPU usage.

[0100] The energy consumption constraint is expressed as:

[0101]

[0102] in, Represents the energy consumption limit at the base station.

[0103] The time constraint is expressed as:

[0104]

[0105] Z ij The sum for all base stations should be 1, expressed as a constraint:

[0106]

[0107] S5: Using the hierarchical decomposition method, the joint optimization problem is decomposed into the task offloading problem and the resource allocation problem; specifically, the joint optimization objective is nonlinear, with a binary variable φ ij , so it is a MINLP problem and is difficult to solve. In order to solve this problem, the present invention proposes a hierarchical decomposition method that can perform calculations in a decentralized manner.

[0108] Among them, the original problem is decomposed into several decoupled sub-problems, which have lower complexity and can be solved efficiently.

[0109] The Tammer decomposition method is used to search for the optimal solution of the three groups of variables in turn.

[0110] In step S5, we first fix an arbitrary task offloading decision and transform the problem into an optimization problem of the remaining two free variables. Then the optimization becomes a resource allocation problem:

[0111]

[0112] If the given task offloading decision is not feasible, then the solution cannot possibly satisfy all constraints in step S4.

[0113] The task offloading problem is expressed as:

[0114]

[0115] S6: adopting a probabilistic task offloading strategy to perform load prediction on the task offloading problem and obtain a load prediction result; a further implementation method is that adopting the probabilistic task offloading strategy to perform load prediction on the task offloading problem includes two scenarios: a non-overlapping base station scenario and an overlapping base station scenario, and the overlapping base station scenario is that there is an overlapping area at the boundary of the coverage areas of two base stations;

[0116] Load prediction is crucial for vehicles when making unloading decisions. The present invention considers two scenarios with different base station deployment architectures and classifies them according to whether the coverage areas overlap or not.

[0117] S61: In the non-overlapping base station scenario, the AGV has only one unloading target, and the original problem degenerates into m independent problems, which can be decoupled. The task offloading problem can be ignored and the load within the base station coverage area can be estimated:

[0118]

[0119] Where S = {i∈N|P i ∈U(r j ,L j )}

[0120] The Lagrangian variables can be estimated with a high-precision load estimation system and the proposed decentralized offloading strategy can be implemented by decomposition. Therefore, the optimal offloading solution can be achieved through a non-overlapping deployment architecture.

[0121] S62: In the overlapping base station scenario, the present invention assumes that there is an overlapping area at the boundary of the coverage area of ​​two base stations. Context transmission delay only occurs when unloading to the backward base station. The present invention proposes a probabilistic task offloading strategy. This strategy determines the probability of assigning tasks to adjacent base stations (two adjacent base stations with overlapping coverage areas), which helps to calculate E(C j ) to solve the optimization problem. In the task offloading prediction process, the updated adjacency matrix A information is collected, the offloading probability matrix P is calculated, and E(C j ) and λ j , predicting Ρ(:,j) and Ρ(j,:). For all AGVs in the same area, the optimal computation time of the task increases monotonically with the total workload on the connected base stations. The unloading probability is also related to the driving direction of the AGV.

[0122] Methods for obtaining load forecast results include:

[0123] Based on the probability of AGV unloading tasks to the preset base station, an unloading probability matrix is ​​established; specifically, define P = [P jk ]∈[0,1] m×m represents the unloading probability matrix, and the element P jk represents the probability that the vehicle offloads the task to base station-k (jk represents the AGV from base station j to base station k).

[0124] Based on the adjacency matrix and the offloading probability matrix, the offloading workload on the base station and the total estimation of the context transmission delay of the base station are obtained; specifically, from the adjacency matrix A and the offloading probability matrix P, it can be known that the offloading workload on the base station-j is:

[0125] E(C j )=Α(:,j) Τ Ρ(:,j)+Α(j,:)(1 m -Ρ(j,:)) Τ

[0126] Among them, (:,j) and (j,:) are the j-th column vector and j-th row vector of the matrix respectively.

[0127] The total estimated context transmission delay for base station-j is:

[0128]

[0129] Based on the offloading workload and the total estimation of context transmission delay, the optimal offloading delay of all AGVs of the corresponding base station is obtained; the optimal offloading delay of all AGVs of base station-j is expressed as:

[0130]

[0131] Based on the optimal delay of all AGV unloading, the calculation time of all AGV computing tasks is optimized to obtain the load prediction results. The calculation time of all AGVs in the optimized system is:

[0132]

[0133] S7: Decompose and convexly optimize the resource allocation problem, combine the load prediction results, solve the joint optimization problem, and complete the decentralized convex optimization for enhancing the energy efficiency of AGV edge computing. Collect Ρ(:,j), λ from the base station j and Ρ(j,:), select the unloading action Z based on probability ij , calculate ω ij , and α ij Finally, we get the control variable Z ij , α ij And ω ij .

[0134] A further implementation method is that the method for solving the joint optimization problem includes:

[0135] S71: Decompose the resource allocation problem, obtain the decentralized resource allocation problem of a single base station, and rewrite the decentralized resource allocation problem as a convex optimization problem; specifically, according to the definition of the resource allocation problem, since the value of Z is known, the unloading AGV set of each base station-j can be determined as V j ={i∈N|Z ij =1}.

[0136] The resource allocation problem can be expressed as:

[0137]

[0138] Since only one base station is considered, the dimension of the variables is reduced and the time constraint is replaced by:

[0139]

[0140] Where, ΔT i '=ΔT-Δd i , decision variable Z is given, ΔT i ' is known.

[0141] The resource allocation problem in the system is decomposed into the decentralized resource allocation problem of a single base station.ij Convexify the resource allocation problem. The resource allocation problem is rewritten as a convex optimization problem:

[0142]

[0143] S72: Introduce Lagrangian variables into the convex optimization problem for dual decomposition to obtain computational subproblems and communication subproblems; specifically, introduce Lagrangian variables To form the dual problem:

[0144]

[0145] The optimization objectives include computation and communication. They have independent structures of objective and constraint functions, and can separate the computation and communication parts of each AGV to decompose the problem.

[0146] The dual function is:

[0147]

[0148] in:

[0149]

[0150] and Represents the dual function of the computation subproblem and the communication subproblem at AGV-i.

[0151] Furthermore, from the definition, the corresponding computation and communication sub-problems are:

[0152]

[0153] Both subproblems are convex and have zero duality gap.

[0154] S73: Based on the computational subproblem, the communication subproblem, and the load prediction results, solve the joint optimization problem to achieve optimal resource allocation.

[0155] Combined with the task offloading decision, the optimal delay of all AGVs offloading at base station-j is expressed as:

[0156]

[0157] Furthermore, the computation time of all AGVs in the optimization system is:

[0158]

[0159] Embodiment 2

[0160] like Figure 2As shown, this example takes n AGVs in an industrial park as an example. M base stations are evenly deployed along the AGV movement road, and each base station is equipped with a CPU computing module as an edge server.

[0161] Specifically, it is achieved through the following steps:

[0162] Step S1: Define the system state space and action space, define the number of base stations as m, the number of AGVs as n, and the calculation task of AGV-i as [D i ,C i ];

[0163] Step S2: Calculate model execution time Energy consumption Transmission delay Energy consumption for data transmission Context transfer delay Δd i .

[0164] Step S3: transform the joint problem into a task offloading problem and a resource allocation problem through a hierarchical decomposition method;

[0165] Step S4: Perform load prediction for the task offloading problem through a probability strategy-based method;

[0166] Specifically, in step S4, at the base station, the updated adjacency matrix A information is collected during the task offloading prediction process, the offloading probability matrix P is calculated, and E(C j ) and λ j , predict Ρ(:,j) and Ρ(j,:);

[0167] Step S5: decompose the resource allocation problem, obtain the computation subproblem and the communication subproblem, and solve the joint optimization problem;

[0168] Specifically, in step S5, at the AGV end, P(:, j), λ j and Ρ(j,:), select the unloading action Z based on probability ij , calculate ω ij , and α ij Finally, we get the control variable Z ij , α ij And ω ij .

[0169] Implementation Column 3

[0170] The present invention also provides a decentralized convex optimization system for enhancing the energy efficiency of AGV edge computing, which is used to implement the method, including:

[0171] A task offloading model building module is used to build a task offloading model for AGV based on an edge computing network; wherein the edge computing network includes several base stations, each of which is equipped with a VEC server that provides wireless access and computing resources for AGV;

[0172] A space partitioning module is used to obtain the state space and action space of the task offloading model based on a preset base station index set, an AGV index set, and the computing tasks of the AGV;

[0173] A model building module is used to build an AGV traffic model, a calculation model of AGV calculation tasks, and a communication model between AGV and base station based on the state space and action space of the task offloading model;

[0174] The model integration module is used to integrate the AGV traffic model, computation model, and communication model to obtain the joint optimization problem of task offloading and resource allocation;

[0175] A problem decomposition module is used to decompose the joint optimization problem into a task offloading problem and a resource allocation problem by using a hierarchical decomposition method;

[0176] The load prediction module is used to adopt a probabilistic task offloading strategy to perform load prediction on the task offloading problem and obtain load prediction results;

[0177] The problem-solving module is used to decompose and convexly optimize the resource allocation problem, combine the load prediction results, solve the joint optimization problem, and complete the decentralized convex optimization for enhancing the energy efficiency of AGV edge computing.

[0178] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A decentralized convex optimization method for enhancing the energy efficiency of AGV edge computing, characterized in that: The following steps are involved: Based on the edge computing network, a task offloading model of AGV is constructed; wherein the edge computing network includes several base stations, each of which is equipped with a VEC server that provides wireless access and computing resources for AGV; Based on the preset base station index set, AGV index set and AGV computing tasks, the state space and action space of the task offloading model are obtained; Based on the state space and action space of the task offloading model, the AGV traffic model, the calculation model of the AGV calculation task, and the communication model between the AGV and the base station are established; Integrate the AGV traffic model, the computation model, and the communication model to obtain a joint optimization problem of task offloading and resource allocation; Using a hierarchical decomposition method, the joint optimization problem is decomposed into a task offloading problem and a resource allocation problem; Using a probabilistic task offloading strategy, load prediction is performed on the task offloading problem to obtain a load prediction result; The resource allocation problem is decomposed and convexly optimized, and the joint optimization problem is solved in combination with the load prediction result to complete the decentralized convex optimization for enhancing the energy efficiency of AGV edge computing.

2. The decentralized convex optimization method for enhancing AGV edge computing energy efficiency according to claim 1, characterized in that: Methods for building AGV traffic models include: Establishing a two-dimensional geographical location set of passable locations in the task offloading model, and obtaining the location of the AGV and the location of the base station based on the two-dimensional geographical location set; Calculate the Euclidean distance between the position of the AGV and the position of the base station based on the square norm; When the Euclidean distance is within the preset base station coverage radius, the base station receives the task from the AGV and completes the construction of the AGV traffic model.

3. The decentralized convex optimization method for enhancing the energy efficiency of AGV edge computing according to claim 1 is characterized in that: The method of establishing the computational model of the AGV computing task includes: Based on DVFS technology, the maximum nominal CPU frequency of the base station is obtained; The CPU frequency scaling ratio determined by the preset AGV to complete the computing task is combined with the maximum nominal CPU frequency of the base station and the computing resources required by the AGV to complete the computing task to obtain the execution time and energy consumption of the computing task; The computing model is obtained based on the execution time and energy consumption of the computing task.

4. The decentralized convex optimization method for enhancing AGV edge computing energy efficiency according to claim 1, characterized in that: The method of establishing a communication model between the AGV and the base station includes: Based on the maximum transmission power of the AGV and the decision variables, the actual transmission power of the AGV is obtained; Based on the AGV position and the base station position, the channel gain is obtained; Based on the actual transmission power, the channel gain, the fixed bandwidth occupied by the AGV when communicating with the base station, and the noise power, the transmission power of the data transmission between the AGV and the base station is obtained; Obtaining a transmission delay based on the transmission power and the length of the data transmitted between the AGV and the base station; Based on the actual transmission power and the transmission delay, the energy consumption of data transmission between the AGV and the base station is obtained, and the construction of the communication model is completed.

5. The decentralized convex optimization method for enhancing AGV edge computing energy efficiency according to claim 4, characterized in that: The goal of the joint optimization problem of task offloading and resource allocation is to minimize the total task response time on the AGV and the energy consumption of data transmission between the AGV and the base station, while satisfying the base station constraints; wherein the total task response time includes the execution time of the computing task, the transmission delay, and the context transmission delay when context transmission occurs; the base station constraints include computing utilization constraints, energy consumption constraints, and time constraints.

6. The decentralized convex optimization method for enhancing AGV edge computing energy efficiency according to claim 4, characterized in that: Using a probabilistic task offloading strategy, load prediction for the task offloading problem includes two scenarios: a non-overlapping base station scenario and an overlapping base station scenario. The overlapping base station scenario is an overlapping area at the boundary of the coverage areas of two base stations. In the overlapping base station scenario, the method for obtaining the load prediction result includes: Based on the probability of AGV unloading tasks to the preset base station, an unloading probability matrix is ​​established; Based on the adjacency matrix and the offloading probability matrix, obtaining the offloading workload on the base station and the total estimation of the context transmission delay of the base station; Based on the offloading workload and the total estimation of context transmission delay, obtaining the optimal offloading delay of all AGVs of the corresponding base station; Based on the optimal delay of all AGV unloading, the computing time of all AGV computing tasks is optimized to obtain the load prediction results.

7. The decentralized convex optimization method for enhancing AGV edge computing energy efficiency according to claim 4, characterized in that: The method for solving the joint optimization problem includes: Decomposing the resource allocation problem to obtain a decentralized resource allocation problem for a single base station, and rewriting the decentralized resource allocation problem as a convex optimization problem; Introducing Lagrangian variables into the convex optimization problem for dual decomposition to obtain a computation subproblem and a communication subproblem; Based on the computational subproblem, the communication subproblem and the load prediction result, the joint optimization problem is solved to achieve optimal allocation of resources.

8. A decentralized convex optimization system for enhancing the energy efficiency of AGV edge computing, used to implement the method described in any one of claims 1 to 7, characterized in that: include: A task offloading model building module is used to build a task offloading model for AGV based on an edge computing network; wherein the edge computing network includes a plurality of base stations, each of which is equipped with a VEC server that provides wireless access and computing resources for the AGV; A space partitioning module is used to obtain the state space and action space of the task offloading model based on a preset base station index set, an AGV index set, and the computing tasks of the AGV; A model building module is used to build an AGV traffic model, a calculation model of AGV calculation tasks, and a communication model between AGV and base station based on the state space and action space of the task offloading model; A model integration module, used to integrate the AGV traffic model, the calculation model and the communication model to obtain a joint optimization problem of task offloading and resource allocation; A problem decomposition module, used to decompose the joint optimization problem into a task offloading problem and a resource allocation problem by adopting a hierarchical decomposition method; A load prediction module is used to adopt a probabilistic task offloading strategy to perform load prediction on the task offloading problem and obtain a load prediction result; The problem-solving module is used to decompose and convexly optimize the resource allocation problem, combine the load prediction results, solve the joint optimization problem, and complete the decentralized convex optimization for enhancing the energy efficiency of AGV edge computing.

Images