An energy consumption optimization method and system based on inference request offloading

Abstract

The present application provides an energy consumption optimization method and system based on reasoning request offloading, comprising: constructing a mobile device energy consumption model of local execution of a user reasoning request, a mobile device energy consumption model of sending the user reasoning request to a cooperative mobile device for execution, and a mobile device energy consumption model of sending the user reasoning request to an edge server for execution; solving the device resource configuration scheme corresponding to the offloading scheme in each iteration and the total energy consumption of the mobile device; if the total energy consumption of the mobile device cannot be reduced by continuously adjusting the offloading scheme for multiple times, the offloading scheme and the device resource configuration scheme corresponding to the current lowest total energy consumption are established as the final scheme; and the edge server sends the offloading scheme and the device resource configuration scheme of the reasoning request to each mobile device. The present application uses D2D technology to utilize the idle resources of the mobile device to cooperatively provide multiple types of reasoning services and reduce the total energy consumption of the mobile device.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of edge intelligence, and relates to an energy consumption optimization method and system based on inference request offloading. BACKGROUND

[0002] With the booming development of deep learning technology, various new artificial intelligence applications such as image classification, video analysis, and face recognition have gradually become part of daily life, and users' demand for different intelligent applications has increased. In order to enable mobile devices to independently complete inference tasks, various technologies have emerged to improve the computing power of mobile devices. However, due to the limited storage resources of mobile devices, they cannot deploy all the required models. Mobile edge computing is recognized as a promising solution. However, the resources of independent edge servers are limited, and when more users offload inference requests to edge servers, it may cause fierce resource competition, resulting in the allocated resources being unable to meet the user service experience requirements.

[0003] To solve this problem, D2D technology is considered to be introduced to make full use of the models and idle resources deployed on mobile devices to collaboratively obtain multiple types of inference services. However, due to the limited battery capacity of mobile devices, the energy consumed by D2D collaborative inference cannot be ignored.

[0004] Existing technologies propose a series of offloading decisions and resource allocation schemes for users' inference requests. However, they do not consider the case where users' inference requests may not be compatible with the models deployed on their mobile devices, and ignore the fact that mobile devices can share the inference services and idle resources provided by existing technologies through D2D technology to reduce resource competition on independent servers.

[0005] Therefore, how to find an effective method that can reduce the energy consumption of mobile devices while providing multiple types of inference services for users through D2D collaboration and edge-end collaboration is a technical problem that needs to be solved by those skilled in the art. SUMMARY

[0006] In view of the above-mentioned shortcomings of the prior art, the present application provides an energy consumption optimization method and system based on inference request offloading. Considering the diversity of users' demand for different deep learning applications, D2D technology is used to make full use of the idle resources of mobile devices to collaboratively provide multiple types of inference services, thereby reducing the total energy consumption of mobile devices.

[0007] In a first aspect, the present application provides an energy consumption optimization method based on inference request offloading, comprising the following steps:

[0008] Step 1: sending inference request information, model deployment information and computing resource information of a mobile device to an edge server;

[0009] Step 2: constructing a mobile device energy consumption model of executing the user inference request locally, a mobile device energy consumption model of sending the user inference request to a cooperative mobile device for execution, and a mobile device energy consumption model of sending the user inference request to an edge server for execution;

[0010] Step 3: the edge server iteratively adjusts the offloading scheme of the inference request, and solves the device resource configuration scheme corresponding to the offloading scheme in each iteration and the total energy consumption of the mobile device; if the total energy consumption of the mobile device cannot be reduced by continuously adjusting the offloading scheme for multiple times, the offloading scheme and the device resource configuration scheme corresponding to the current lowest total energy consumption are established as the final scheme;

[0011] Step 4: the edge server sends the offloading scheme and the device resource configuration scheme of the inference request to each mobile device.

[0012] In a second aspect, the present application provides an energy consumption optimization system based on inference request offloading, comprising:

[0013] An information sending module is configured to send inference request information, model deployment information and computing resource information of the mobile device to the edge server;

[0014] An inference request constructing module is configured to construct a mobile device energy consumption model of executing the user inference request locally, a mobile device energy consumption model of sending the user inference request to a cooperative mobile device for execution, and a mobile device energy consumption model of sending the user inference request to an edge server for execution;

[0015] An offloading scheme adjusting module is configured to iteratively adjust the offloading scheme of the inference request by the edge server, and solve the device resource configuration scheme corresponding to the offloading scheme in each iteration and the total energy consumption of the mobile device; if the total energy consumption of the mobile device cannot be reduced by continuously adjusting the offloading scheme for multiple times, the offloading scheme and the device resource configuration scheme corresponding to the current lowest total energy consumption are established as the final scheme;

[0016] An offloading scheme distributing module is configured to send the offloading scheme and the device resource configuration scheme of the inference request to each mobile device by the edge server.

[0017] In a third aspect, the present application provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the method of the first aspect of the present application by using the computer program.

[0018] In a fourth aspect, the present application provides a computer readable storage medium comprising a stored computer program, wherein the computer program is executed to perform the method of the first aspect of the present application.

[0019] Compared with the prior art, the beneficial effects of the technical scheme of the present application are:

[0020] The present application takes into account the diversity of user needs for different deep learning applications, and constructs a mobile device energy consumption model for local execution of user inference requests, a mobile device energy consumption model for execution of user inference requests by a mobile device that can cooperate, and a mobile device energy consumption model for execution of user inference requests by an edge server, which iteratively adjusts the offloading scheme of the inference request and solves the device resource configuration scheme corresponding to the offloading scheme and the total energy consumption of the mobile device in each iteration; uses D2D technology to fully utilize the idle resources of the mobile device to cooperatively provide multiple types of inference services and reduce the total energy consumption of the mobile device. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The method flowchart of the present application.

[0022] Figure 2 The method model diagram of the present application. DETAILED DESCRIPTION

[0023] The embodiments of the present application will be described below with reference to the accompanying drawings and preferred embodiments, and other advantages and effects of the present application can be easily understood by those skilled in the art from the disclosure herein. The present application can also be implemented or applied in other different specific embodiments, and various modifications or changes can be made to the details herein based on different views and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for illustration of the present application, and are not intended to limit the protection scope of the present application.

[0024] It should be noted that the diagrams provided in the following embodiments only schematically illustrate the basic concept of the present application, and only the components related to the present application are shown in the diagrams, not the number, shape and size of the components when actually implemented. The actual implementation of each component may be a random change in shape, number and proportion, and the layout pattern of the components may also be more complex.

[0025] Before introducing the embodiments of the present application, the related terms involved in the embodiments of the present application are first explained as follows:

[0026] D2D: Device to Device, terminal direct communication.

[0027] Embodiment

[0028] The present embodiment proposes an energy consumption optimization method based on inference request offloading. Please refer to Figure 1 , comprising the following steps:

[0029] Step 1: sending inference request information, model deployment information of the mobile device and computing resource information to the edge server;

[0030] Step 2: constructing a mobile device energy consumption model in which the user inference request is executed locally, a mobile device energy consumption model in which the mobile device sends the user inference request to a cooperative mobile device for execution, and a mobile device energy consumption model in which the mobile device sends the user inference request to the edge server for execution;

[0031] Step 3: the edge server iteratively adjusts the offloading scheme of the inference request and solves the device resource configuration scheme corresponding to the offloading scheme and the total energy consumption of the mobile device in each iteration; if the total energy consumption of the mobile device cannot be reduced by continuously adjusting the offloading scheme for multiple times, the offloading scheme and the device resource configuration scheme corresponding to the current lowest total energy consumption are established as the final scheme;

[0032] Step 4: the edge server sends the offloading scheme and the device resource configuration scheme of the inference request to each mobile device.

[0033] In this embodiment, the mobile device sends the user inference request to the cooperative mobile device through the D2D link.

[0034] As a preferred technical solution, in this embodiment, step 1 is specifically:

[0035] Define N as a set of users / mobile devices, M as a total set of mobile devices and edge servers, and K as a set of DNN models; define x i，m ∈{0,1} represents whether the inference request T i of the mobile device i, i∈N is completed on m, m∈M.

[0036] It is assumed that the edge server deploys all DNN models in the set K, and the mobile device deploys one of the DNN models, and each mobile device requests one service in the set K; it is assumed that the mobile device completes one inference request at a time, and the edge server completes multiple inference requests.

[0037] It can be understood that in the communication system of the present application, there are multiple users, and each user carries a mobile device. In addition, there is an edge server in the system. It is assumed that a total of |K| models are involved in the whole system, and the |K| models respectively provide different inference services. In the server, all the |K| models are deployed; on each mobile device, only one of them is deployed. In this embodiment, it is assumed that x i，m =0 represents that the inference request T i of the mobile device i, i∈N is not completed on m, m∈M, and x i，m =1 represents that the inference request T i of the mobile device i, i∈N is completed on m, m∈M.

[0038] Each user will request a model service, and the user request can only be completed on the mobile device or edge server (the edge server is deployed with all |K| models, so it can be offloaded to the edge server; but the mobile device may not) deployed with the corresponding DNN model.

[0039] Collaborative mobile device interpretation: If there are users A, B, C, and D in the system. If user A requests A-type service, and users B and C deploy A-type models, and user D deploys C-type model, then the collaborative mobile device of user A is the mobile device of users B and C. (The collaborative mobile device does not include its own mobile device, even if the corresponding model is deployed). Here, mobile device j is the collaborative mobile device.

[0040] As a preferred technical solution, in the embodiment, the total energy consumption of all mobile devices obtained in step 2 is specifically:

[0041] Step 2.1: For the way of completing the inference request locally, obtain the inference request T of mobile device i, i∈N i The energy consumption generated when executed locally

[0042]

[0043] Where σ i is the capacitance coefficient of mobile device i, ρ i is the workload of inference request T i , f i，i is the computing resource allocated by mobile device i to inference request T i ;

[0044] Step 2.2: For the way of completing the inference request on the collaborative mobile device, obtain the transmission energy consumption generated when mobile device i, i∈N sends inference request T i to collaborative mobile device j, j∈M\{0,i}

[0045]

[0046] Where d i is the input data size of inference request T i , in bits, w i is the bandwidth resource allocated to mobile device i, p i is the transmission power of mobile device i, in w, g i，j is the channel gain between mobile device i and collaborative mobile device j, δ 2 is the noise;

[0047] The computation energy consumption generated by the cooperative mobile device j, j ∈ M \ {0, i} performing the inference request T is:

[0048]

[0049] where σ j is the capacitance coefficient of the cooperative mobile device j, f i，j is the computation resource allocated to the cooperative mobile device j by the inference request T i ;

[0050] Step 2.3: Obtain the transmission energy consumption generated by the mobile device i, i ∈ N sending the inference request T i to the edge server in the manner of completing the inference request by the user to the edge server

[0051]

[0052] where g i，0 is the channel gain between the mobile device i and the edge server;

[0053] Step 2.4: Obtain the total energy consumption E sum of all the mobile devices:

[0054]

[0055] As a preferred technical solution, in the embodiment, the step 3 is specifically:

[0056] Step 3.1: Construct a feasible offloading set u between the inference request and the mobile device and the edge server, and initialize a feasible offloading scheme;

[0057] Step 3.2: Iteratively adjust the offloading scheme of the inference request based on the feasible offloading set u;

[0058] Step 3.3: Solve the resource configuration scheme of the computation resource and the bandwidth resource for the offloading decision X' obtained by the current iteration adjustment;

[0059] Step 3.4: Calculate the current mobile device total energy consumption E' under the current offloading decision X' and the resource configuration scheme thereof;

[0060] Step 3.5: If the mobile device total energy consumption E under the latest scheme is greater than the current mobile device total energy consumption E', set the offloading scheme of this iteration as the latest offloading scheme, set the resource allocation strategy of the latest offloading scheme as the latest resource allocation scheme, and set the current mobile device total energy consumption E' as the latest mobile device total energy consumption; otherwise, return to step 3.2.

[0061] Step 3.6: If all feasible offloading decisions in the feasible offloading set u have been adjusted in step 3.2, and no offloading scheme with lower energy consumption than the latest offloading scheme can be found, then the latest offloading scheme is set as the final offloading scheme, the resource allocation scheme obtained by the final offloading scheme is set as the final resource allocation scheme, and the iteration process ends; otherwise, return to step 3.2.

[0062] As a preferred technical solution, in the embodiment, step 3.1 is specifically:

[0063] Step 3.1.1: Construct a feasible offloading set u between a reasoning request and a mobile device and an edge server:

[0064] u←{x i，m =0|i∈N,m∈M,k∈K,z i，k =y m，n ,k=1}

[0065] wherein z i，k ∈{0,1} represents whether the mobile device i requests the service of type k, y m，k ∈{0,1} represents whether the cooperative mobile device m deploys the DNN model of type k;

[0066] Step 3.1.2: Initialize the offloading scheme: for each reasoning request T i , the user randomly selects to offload the reasoning request T i to the edge server, that is, x i，0 =1; or offload the reasoning request T i to an unexecuted feasible offloading decision x i，j , x i，j =0, x i，j ∈u, j∈M\{0,i} corresponds to the cooperative mobile device j, that is, x j =1, and the cooperative mobile device j is an idle device (that is, no reasoning request to be completed), which satisfies x n，j =0,

[0067] Step 3.1.3: Solve the resource configuration scheme of the computing resource and the bandwidth resource under the initial offloading scheme;

[0068] Step 3.1.4: Calculate the total energy consumption of the mobile device under the initial offloading scheme and the resource configuration scheme thereof;

[0069] Step 3.1.5: Set the initial offloading scheme and the resource configuration scheme thereof as the latest scheme, and set the total energy consumption of the mobile device as the latest total energy consumption.

[0070] As a preferred technical solution, in this embodiment, the specific steps of the adjustment process in step 3.2 are as follows:

[0071] Step 3.2.1: If the latest solution is the initial solution, or if the latest solution is different from the latest solution at the beginning of the previous iteration, then start from the beginning of the feasible unload set u and sequentially select the first unexecuted feasible unload variable; otherwise, start from the unload variable adjusted in the previous iteration in the feasible unload set u and sequentially select the first unexecuted feasible unload variable; assuming the selected variable is x. i，m ;

[0072] Step 3.2.2: Transfer the inference request T i The uninstallation decision is changed to uninstall to m, that is, let x... i，m =1;

[0073] Step 3.2.3: Transfer the inference request T i The original unloading decision value is changed to 0, that is, x is set to 0. i，m′ =0,

[0074] Step 3.2.4: If m is an edge server, then proceed to step 3.2.7;

[0075] Step 3.2.5: If m is an idle mobile device, that is, m has no pending inference requests, then x is satisfied. n，m =0, Then proceed to step 3.2.7;

[0076] Step 3.2.6: Transfer the inference request T to be executed on m. n x n，m =1,x n，m ∈ Unload to one of the cooperating devices j, i.e., let x n，m =0 and x n，j =1, and the collaborating device is an idle mobile device, i.e., satisfying x n′，j =0, Or offload to an edge server, i.e., let x n，m =0 and x n，0 =1.

[0077] Step 3.2.7: The adjustment process is complete.

[0078] As a preferred technical solution, in this embodiment, step 3.3 specifically includes:

[0079] Initialize weight λ i , α, β j η i and θ i，j, i∈N, j∈M are random positive values; based on the calculation formula of computing resources and bandwidth resources, the weight is iteratively updated, and the specific iteration process is as follows:

[0080] Initialize the weight λ i , α, β m , η i and θ i，m , i∈N, m∈M are random positive values; based on the calculation formula of computing resources and bandwidth resources, the weight is iteratively updated, and the specific iteration process is as follows:

[0081] Step 3.3.1: If the inference request T i is offloaded to a cooperative mobile device or an edge server for completion, then based on the weight λ i , α and η i of the current iteration, the bandwidth resource allocated to the mobile device i is calculated The calculation formula is:

[0082]

[0083] Wherein, Otherwise, the mobile device i does not allocate bandwidth resources, that is,

[0084] If the inference request T i is offloaded to an edge server for completion, then based on the weight λ i , β0and θ i，0 of the current iteration, the computing resource allocated to the inference request T i by the edge server is calculated

[0085]

[0086] If the inference request T i is completed locally or offloaded to a cooperative mobile device for completion, then based on the weight λ i , β m and θ i，m of the current iteration, m∈M\{0}, the computing resource allocated to the inference request T i by m is calculated

[0087]

[0088] Wherein,

[0089]

[0090]

[0091] ξ=arccos T,

[0092]

[0093]

[0094]

[0095]

[0096] Step 3.3.2: weight λ i , α, β m , η i and θ i，m , i∈N, m∈M, iteratively update according to the following rules:

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] wherein,

[0103]

[0104]

[0105]

[0106]

[0107]

[0108] wherein, is the delay requirement of user i, w max is the total bandwidth resource of the system, is the total computing resource of m, t is the iteration round; a(t), b(t), c(t), d(t) and e(t) are small positive steps;

[0109] Step 3.3.3: if the iteration number reaches the upper limit, i.e. t=t max , or the change degree of all weights between two iterations is less than the preset value ε, the resource allocation scheme of this iteration is set as the resource allocation scheme based on the current offloading scheme, and the iteration process ends; otherwise, return to step 3.3.1.

[0110] In this embodiment, the random positive value and the preset value ε are small.

[0111] The inference request offloading based energy consumption optimization system provided by the embodiments of the present application is described below, and the inference request offloading based energy consumption optimization system described below can be referred to the inference request offloading based energy consumption optimization method described above.

[0112] The inference request offloading based energy consumption optimization system provided by the embodiments of the present application comprises:

[0113] The information sending module is configured to send the inference request information, the model deployment information of the mobile device and the computing resource information to the edge server.

[0114] The model construction module is configured to construct a mobile device energy consumption model in which the user inference request is executed locally, a mobile device energy consumption model in which the user inference request is sent to a cooperative mobile device for execution, and a mobile device energy consumption model in which the user inference request is sent to the edge server for execution.

[0115] The offloading scheme adjustment module is configured to iteratively adjust the offloading scheme of the inference request by the edge server, and solve the device resource configuration scheme corresponding to the offloading scheme in each iteration and the total energy consumption of the mobile device; if the total energy consumption of the mobile device cannot be reduced by continuously adjusting the offloading scheme for multiple times, the offloading scheme and the device resource configuration scheme corresponding to the current lowest total energy consumption are established as the final scheme.

[0116] The offloading scheme allocation module is configured to send the offloading scheme of the inference request and the device resource configuration scheme result to each mobile device by the edge server.

[0117] In another embodiment of the present application, an electronic device is also provided, comprising: one or more processors; a storage device for storing one or more programs, which, when executed by the one or more processors, cause the present application to implement the method in any of the above embodiments.

[0118] In the present embodiment, the computer system suitable for the electronic device for implementing the embodiments of the present application comprises a central processing unit (CPU) which can perform various appropriate actions and processes according to the programs stored in the read-only memory (ROM) or loaded from the storage part to the random access memory (RAM), such as performing the method described in the above embodiments. In the RAM, various programs and data required for system operation are also stored. The CPU, the ROM and the RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0119] The following components are connected to the I / O interface: an input section including a keyboard, a mouse, etc.; an output section including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), etc., and a speaker, etc.; a storage section including a hard disk, etc.; and a communication section including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as necessary. A removable media such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive as necessary, so that a computer program read out from it is installed into the storage section as necessary.

[0120] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present application. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing a computer program for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section, and / or installed from a removable media. When the computer program is executed by a central processing unit (CPU), various functions defined in the system of the present application are executed.

[0121] It should be noted that, although several modules or units of the apparatus for action execution are mentioned in the foregoing detailed description, such division into modules or units is not mandatory. Indeed, according to embodiments of the present application, features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, features and functions of one module or unit described above can be further divided into several modules or units.

[0122] In the present embodiments, the present application also provides a computer readable storage medium having stored thereon a computer program, which, when executed by a processor of a computer, causes the computer to perform the method provided by any of the preceding embodiments. The computer readable storage medium can be included in the electronic device described in the above embodiments, or can exist separately from the electronic device without being assembled into the electronic device.

[0123] Those skilled in the art can clearly understand the example embodiments described herein through the above description of the example embodiments that the example embodiments described herein can be implemented by software or by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.) or on a network, and includes a plurality of instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0124] Obviously, the above embodiments of the present application are merely examples for clearly illustrating the present application, and are not intended to limit the embodiments of the present application. Based on the above description, other different forms of changes or variations can also be made by those of ordinary skill in the art. Here, it is not necessary and impossible to enumerate all the embodiments. Any modification, equivalent replacement, and improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. An energy consumption optimization method based on inference request offloading, characterized in that, The method comprises the following steps: Step 1: the user sends inference request information, model deployment information of the mobile device and computing resource information to the edge server; Let N be the set of users or mobile devices, M be the total set of mobile devices and edge servers, and K be the set of DNN models; let x i,m ∈{0,1} represent the inference request T of mobile device i, i∈N i whether it is completed on m, m∈M It is assumed that the edge server deploys all DNN models in the set K, and the mobile device deploys one of the DNN models, and each mobile device requests a service in the set K; it is assumed that the mobile device completes one inference request at a time, and the edge server completes multiple inference requests; Step 2: a mobile device energy consumption model in which a user inference request is executed locally by a mobile device, a mobile device energy consumption model in which a user inference request is sent to a mobile device for execution by a cooperative mobile device, and a mobile device energy consumption model in which a user inference request is sent to an edge server for execution are constructed; Step 3: the edge server iteratively adjusts the offloading scheme of the inference request, and solves the resource configuration scheme corresponding to the offloading scheme in each iteration and the total energy consumption of the mobile device; if the total energy consumption of the mobile device cannot be reduced by continuously adjusting the offloading scheme for multiple times, the offloading scheme and the resource configuration scheme corresponding to the current lowest total energy consumption are established as the final scheme; The total energy consumption of the mobile device in step 3 is specifically: Obtaining a reasoning request T of a mobile device i, i e N i Energy consumption generated when executed locally where σ i is the capacitance coefficient of mobile device i, p i is the workload of inference request T i , f i,i is the computing resource allocated by mobile device i to inference request T i ; Obtaining a mobile device i, i e N sends an inference request T i Transmission energy consumption generated by sending to a cooperative mobile device j, j e M \ {0, i} where d i is the input data size of the inference request T i , w i is the bandwidth resource allocated to the mobile device i, p i is the transmission power of the mobile device i, g i,j is the channel gain between the mobile device i and the cooperative mobile device j, and δ 2 is the noise. Collaborating mobile devices j, j e M \ {0, i} perform inference request T i The resulting computational energy consumption is: where σ j is the capacitance coefficient of the cooperative mobile device j, f i,j is the computing resource allocated to the inference request T i by the cooperative mobile device j; Obtaining mobile device i, i e N will inference request T i Transmission energy consumption generated by sending to edge server where g i,0 is the channel gain between mobile device i and the edge server; Obtaining total energy consumption E of all mobile devices sum : Step 4: the edge server sends the offloading scheme and the resource configuration scheme result to each mobile device.

2. The method of claim 1, wherein, The step 3 is specifically: Step 3.1: a feasible offloading set u between a mobile device and an edge server for one inference request is constructed, and a feasible offloading scheme is initialized; Step 3.2: the offloading scheme of the inference request is iteratively adjusted based on the feasible offloading set u; Step 3.3: for the offloading decision X' obtained by the current iterative adjustment, the resource configuration scheme of the computing resource and the bandwidth resource is solved; Step 3.4: the current total energy consumption E' of the mobile device under the current offloading decision X' and the resource configuration scheme thereof is calculated; Step 3.5: if the total energy consumption E of the mobile device under the latest scheme is greater than the current total energy consumption E' of the mobile device, the offloading scheme of this iteration is set as the latest offloading scheme, the resource allocation strategy of the latest offloading scheme is set as the latest resource allocation scheme, and the current total energy consumption E' of the mobile device is set as the latest total energy consumption of the mobile device; otherwise, return to step 3.2; Step 3.6: if all feasible offloading decisions in the feasible offloading set u have been adjusted in step 3.2, and no offloading scheme with lower energy consumption than the latest offloading scheme has been found, the latest offloading scheme is set as the final offloading scheme, the resource allocation scheme obtained by the final offloading scheme is set as the final resource allocation scheme, and the iteration process ends; Otherwise, return to step 3.

2.

3. The method of claim 2, wherein, Step 3.1 is specifically: Step 3.1.1: a feasible offloading set u between a mobile device and an edge server for one inference request is constructed: u <- {x i,m = 0 | i e N, m e M, k e K, z i,k = y m,k = 1} wherein z i,k ∈{0,1} indicates whether the mobile device i requests the service of type k, y m,k ∈{0,1} indicates whether the m has deployed the DNN model of type k; Step 3.1.2: Initialization of offloading scheme: for each inference request T i , the user randomly selects to offload the inference request T i to the edge server, i.e., let x i,0 = 1; or offload the inference request T i to one unexecuted feasible offloading decision x i,j , x i,j = 0, x i,j ∈ u, j ∈ M \ {0, i} on the corresponding cooperative mobile device j, i.e., let x i,j = 1, and the cooperative mobile device j is an idle device, satisfying Step 3.1.3: the resource configuration scheme of the computing resource and the bandwidth resource under the initial offloading scheme is solved; Step 3.1.4: the total energy consumption of the mobile device under the initial offloading scheme and the resource configuration scheme thereof is calculated; Step 3.1.5: the initial offloading scheme and the resource configuration scheme thereof are set as the latest scheme, and the total energy consumption of the mobile device is set as the latest total energy consumption.

4. The method of claim 3, wherein, The step 3.2 is specifically: Step 3.2.1: if the current latest scheme is the initial scheme, or the current latest scheme is different from the latest scheme at the beginning of the last iteration, then starting from the beginning in the feasible offloading set u, sequentially select the first unexecuted feasible offloading variable; otherwise, starting from the offloading variable adjusted in the last iteration in the feasible offloading set u, sequentially select the first unexecuted feasible offloading variable; Assume the selected variable is x i,m ; Step 3.2.2: Change the offload decision of the inference request T i to offload to m, i.e. let x i,m = 1; Step 3.2.3: Change the value of the original offload decision of the inference request T i to 0, i.e. let Step 3.2.4: if m is an edge server, then jump to step 3.2.7; Step 3.2.5: If m is a free mobile device, i.e. m has no inference requests to be executed, then jump to step 3.2.

7. Step 3.2.6: The inference request T to be executed on m n x n,m = 1 n,m ∈ u is offloaded to one of the cooperating devices j, i.e. x n,m = 0 and x n,j = 1 and this cooperating device is the idle mobile device, i.e. it satisfies or offloaded to the edge server, i.e. x n,m = 0 and x n,0 = 1 Step 3.2.7: the adjustment process ends.

5. An energy consumption optimization system based on inference request offloading using the energy consumption optimization method based on inference request offloading according to any one of claims 1 to 4. Comprise: An information sending module configured to send the inference request information of a user, model deployment information of a mobile device, and computing resource information to an edge server; A model construction module configured to construct a mobile device energy consumption model in which a user inference request is executed locally on a mobile device, a mobile device energy consumption model in which a user inference request is sent to a cooperative mobile device for execution, and a mobile device energy consumption model in which a user inference request is sent to an edge server for execution; An offloading scheme adjustment module configured to iteratively adjust an offloading scheme for an inference request by an edge server, and to solve a resource configuration scheme corresponding to the offloading scheme and total mobile device energy consumption in each iteration; if the offloading scheme cannot reduce the total mobile device energy consumption after being adjusted for a plurality of times, then the offloading scheme and the resource configuration scheme corresponding to the lowest total energy consumption are established as a final scheme; An offloading scheme allocation module configured to send the offloading scheme and the resource configuration scheme for an inference request to each mobile device by an edge server. 6.An electronic device, comprising a memory and a processor, wherein, The memory stores a computer program, and the processor is configured to execute the method in any one of claims 1 to 4 by the computer program.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer program executes the method in any one of claims 1 to 4 when running. The computer readable storage medium comprises a stored computer program, wherein the computer program executes the method in any one of claims 1 to 4 when running.

Images