Scheduling method and device of mobile equipment, equipment, medium and product

Abstract

The invention discloses a mobile device scheduling method and device, equipment, a medium and a product, and the method comprises the steps: constructing a multi-federated learning task scene which comprises a plurality of mobile devices, edge computing nodes and federated learning tasks; each mobile device participates in at least one federated learning task, and each edge computing node is responsible for one federated learning task; according to the computing power and the communication link quality of the mobile device, constructing a target function which minimizes the total training time of model training of a multi-federated learning task under the condition of ensuring the global model precision and minimizes the total consumed energy of the mobile device under the condition of ensuring the local model precision; and calculating a mobile device scheduling scheme of each federated learning task according to the objective function. By adopting the method and the device, the problem of mobile equipment scheduling in a multi-federated task scene can be solved, the single mobile equipment can participate in multiple federated tasks at the same time, and the task training time and the energy consumption of the mobile equipment are minimized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a scheduling method, apparatus, device, medium, and product for mobile devices. Background Technology

[0002] Currently, federated learning, as a novel distributed computing architecture where data remains local, has attracted significant attention due to its excellent user privacy protection and low data transmission volume. In mobile edge network environments, a large number of mobile devices provide abundant data resources for federated learning, and the concurrent execution of multiple federated learning tasks by multiple edge computing nodes has become the new norm for network operation. Therefore, it is necessary to select or schedule appropriate mobile devices for each federated learning task, allocate transmission and computing resources, and thereby optimize the model training performance of multiple learning tasks.

[0003] While existing technical solutions address scenarios involving multiple concurrent federated tasks, these solutions, in order to simplify the model, limit each user to only one task. In reality, existing mobile devices (such as smartphones and vehicles) are typically equipped with multiple sensors, capable of collecting datasets for various tasks and participating in multiple federated learning tasks. Therefore, there is an urgent need to address the requirement of enabling multiple mobile devices to participate in multiple tasks simultaneously in multi-task federated learning scenarios. Furthermore, current multi-task federated learning technologies do not consider the impact of differences in link quality. In fact, mobile edge network environments are complex, and the link quality in wireless networks is not stable, with data transmission speed fluctuations and interruptions frequently occurring. Therefore, considering link quality factors in the system model of multi-task federated learning scenarios is particularly important. Summary of the Invention

[0004] The purpose of this invention is to provide a scheduling method, apparatus, device, medium, and product for mobile devices, which can solve the scheduling problem of mobile devices in multi-federated task scenarios, enable a single mobile device to participate in multiple federated tasks simultaneously, and minimize task training time and mobile device power consumption.

[0005] To achieve the above objectives, embodiments of the present invention provide a scheduling method for mobile devices, comprising: An application scenario for constructing multiple federated learning tasks is provided. This application scenario includes several mobile devices, edge computing nodes, and federated learning tasks. The mobile devices are used to collect sample data, use the sample data to complete local model training for the federated learning task, and transmit the local model to the edge computing node. Each mobile device participates in at least one federated learning task. The edge computing node is used to complete global model updates for the federated learning task based on the local model. Each edge computing node is responsible for one federated learning task. Based on the computing power of the mobile device and the quality of the communication link between the mobile device and the edge computing node, an objective function is constructed that minimizes the total training time of the model training for multiple federated learning tasks while ensuring the accuracy of the global model, and minimizes the total energy consumption of the mobile device while ensuring the accuracy of the local model. Based on the objective function, calculate the mobile device scheduling scheme for each of the federated learning tasks.

[0006] As an improvement to the above scheme, the objective function, based on the computing power of the mobile device and the communication link quality between the mobile device and the edge computing node, minimizes the total training time of the model training for multiple federated learning tasks while ensuring global model accuracy, and minimizes the total energy consumption of the mobile device while ensuring local model accuracy, includes: Based on the computing power of the mobile device, a mathematical model is constructed to calculate the training time and energy consumed by the mobile device in local model training for the federated learning task, which are respectively denoted as the first time consumed and the first energy consumed. Based on the communication link quality, a mathematical model is constructed of the transmission time and energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, which are respectively denoted as the second time consumption and the second energy consumption. Based on the first consumption time and the second consumption time, a mathematical model is constructed to represent the total consumption time of the mobile device participating in the federated learning task; Based on the first energy consumption and the second energy consumption, a mathematical model is constructed to represent the total energy consumption of the mobile device participating in the federated learning task. A mathematical model for determining the total training time of the model training for the federated learning task is based on the total time consumed by at least one mobile device participating in the model training of the federated learning task in parallel. An objective function is constructed with the goal of minimizing the total training time of all the federated learning tasks and minimizing the total energy consumption of all the mobile devices.

[0007] As an improvement to the above scheme, the step of constructing a mathematical model of the training time and energy consumed by the mobile device in local model training for the federated learning task, based on the computing power of the mobile device, and denoted as the first consumption time and the first consumption energy, includes: Based on the number of sample data collected by the mobile device participating in the federated learning task, the allocated computing resources, and the number of CPU cycles required to process a unit of sample data, a mathematical model is constructed to determine the training time consumed by the mobile device in a single local model training session for the federated learning task, which is denoted as the single training time. Based on a preset local model accuracy threshold, determine the number of local iterations required for the mobile device to train the local model for the federated learning task. Based on the single training time and the number of local iterations, a mathematical model is constructed for the training time consumed by the mobile device in participating in the local model training of the federated learning task, which is denoted as the first consumption time. Based on the number of local iterations, the number of sample data, the computing resources, and the number of CPU cycles, a mathematical model is constructed of the energy consumed by the mobile device in local model training for the federated learning task, denoted as the first energy consumption.

[0008] As an improvement to the above scheme, the step of constructing a mathematical model based on the communication link quality, representing the transmission time and energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, denoted as the second time consumption and the second energy consumption respectively, includes: A mathematical model of the data transmission rate between the mobile device and the edge computing node is constructed based on the sub-channel bandwidth allocated to the mobile device, the data transmission power from the mobile device to the edge computing node, and the wireless communication channel gain between the mobile device and the edge computing node. Based on the data transmission volume and data transmission rate of the mobile device and the edge computing node, a mathematical model is constructed to determine the transmission time consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, which is denoted as the second transmission time. Based on the second consumption time, a mathematical model is constructed of the energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, denoted as the second energy consumption.

[0009] As an improvement to the above scheme, the formula for the objective function is:

[0010] in, The total training time for training the model for the j-th federated learning task. The total energy consumed by the i-th mobile device participating in the j-th federated learning task; m is the number of federated learning tasks, and n is the number of mobile devices. , This indicates that the i-th mobile device participates in the j-th federated learning task. This means that the i-th mobile device does not participate in the j-th federated learning task. As preset time weights, This is the preset energy consumption weight.

[0011] As an improvement to the above scheme, the step of calculating the mobile device scheduling scheme for each federated learning task based on the objective function includes: Using the objective function as the fitness function, a genetic algorithm is used to iteratively generate the optimal individual to obtain the mobile device scheduling scheme for each federated learning task.

[0012] This invention also provides a scheduling device for mobile devices, comprising: An application scenario construction module is used to construct application scenarios for multiple federated learning tasks. Each application scenario includes several mobile devices, edge computing nodes, and federated learning tasks. The mobile devices are used to collect sample data, use the sample data to complete local model training for the federated learning task, and transmit the local model to the edge computing node. Each mobile device participates in at least one federated learning task. The edge computing node is used to complete global model updates for the federated learning task based on the local model. Each edge computing node is responsible for one federated learning task. The objective function construction module is used to construct an objective function based on the computing power of the mobile device and the quality of the communication link between the mobile device and the edge computing node, which minimizes the total training time of the model training for the multi-federated learning task while ensuring the accuracy of the global model, and minimizes the total energy consumption of the mobile device while ensuring the accuracy of the local model. The mobile device scheduling module is used to calculate the mobile device scheduling scheme for each of the federated learning tasks based on the objective function.

[0013] This invention also provides a scheduling device for a mobile device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the scheduling method for the mobile device as described in any of the preceding embodiments.

[0014] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the scheduling method of the mobile device as described in any of the preceding embodiments.

[0015] This invention also provides a computer program product, which includes a computer program or computer instructions. When the computer program or computer instructions are executed by a processor, they implement the scheduling method of the mobile device as described in any of the above embodiments.

[0016] Compared with existing technologies, the scheduling method, apparatus, device, medium, and product for mobile devices disclosed in this invention, combined with the characteristics of mobile edge networks, enables a single mobile device to participate in multiple federated tasks simultaneously. It also comprehensively considers factors such as the computing power of mobile devices, the transmission quality of communication links, and model accuracy. Through a system model in a multi-federated learning task scenario, the scheduling problem of mobile devices in a multi-federated learning task scenario is modeled as a multi-objective optimization problem. Under the constraints of model convergence, transmission reliability, and limited computing resources, it aims to reduce the training time of multiple federated learning tasks while minimizing the energy consumption of each mobile device, thereby obtaining a better scheduling scheme for mobile devices. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a scheduling method for a mobile device provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the application scenario of multiple federated learning tasks in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a scheduling device for a mobile device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0020] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0021] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0022] See Figure 1 This is a flowchart illustrating a scheduling method for a mobile device provided in an embodiment of the present invention. The embodiment of the present invention provides a scheduling method for a mobile device, including steps S11 to S13: S11. Constructing an application scenario for multiple federated learning tasks; wherein, the application scenario includes several mobile devices, edge computing nodes, and federated learning tasks; the mobile devices are used to collect sample data, use the sample data to complete local model training for the federated learning task, and transmit the local model to the edge computing node, each of the mobile devices participating in at least one federated learning task; the edge computing node is used to complete global model updates for the federated learning task based on the local model, each of the edge computing nodes being responsible for one federated learning task; S12. Based on the computing power of the mobile device and the quality of the communication link between the mobile device and the edge computing node, construct an objective function that minimizes the total training time of the model training for multiple federated learning tasks while ensuring the accuracy of the global model, and minimizes the total energy consumption of the mobile device while ensuring the accuracy of the local model. S13. Calculate the mobile device scheduling scheme for each federated learning task according to the objective function.

[0023] In an embodiment of the present invention, see Figure 2 This is a schematic diagram illustrating an application scenario of a multi-federated learning task in an embodiment of the present invention. In this scenario, multiple mobile devices and multiple edge computing nodes are involved. The set of mobile devices is represented as... ,in n This refers to the number of mobile devices. The set of edge computing nodes is represented as... ,in m This refers to the number of edge computing nodes. Each edge computing node is responsible for one federated learning task, acting as a central server or coordinator. Assuming the edge computing nodes... Responsible for the j A federal learning task, or simply a task jThe mobile devices are smart electronic devices such as smartphones and vehicles, which are typically equipped with multiple sensors. Each mobile device can collect data from multiple task types to obtain sample sets belonging to different federated learning tasks, and then participate in one or more federated learning tasks.

[0024] In this scenario, edge computing nodes aim to minimize the training time of the federated learning tasks they are responsible for while ensuring the accuracy of the global model. Meanwhile, mobile devices participating in federated learning model training aim to minimize their own energy consumption while maintaining the accuracy of their local model.

[0025] Based on this, this embodiment of the invention constructs a system model for a multi-federated learning task application scenario based on the computing power of the mobile device and the communication link quality between the mobile device and the edge computing node. Then, based on the system model, a multi-objective optimization problem is constructed to minimize the total task latency and mobile device energy consumption, serving as the objective function. Finally, with the objective function as the objective, a mobile device scheduling scheme for the aforementioned m federated learning tasks is designed. This mobile device scheduling scheme refers to scheduling the aforementioned n mobile devices to determine which mobile devices will participate in the aforementioned m federated learning tasks.

[0026] By employing the technical means of this invention and combining the characteristics of mobile edge networks, a single mobile device can participate in multiple federated tasks simultaneously. It comprehensively considers factors such as mobile device computing power, communication link transmission quality, and model accuracy. Through a system model in a multi-federated learning task scenario, the mobile device scheduling problem in this scenario is modeled as a multi-objective optimization problem. Under constraints of model convergence, transmission reliability, and limited computing resources, the aim is to reduce the training time of multiple federated learning tasks while minimizing the energy consumption of each mobile device, thereby obtaining a better mobile device scheduling scheme.

[0027] As a preferred embodiment, the present invention further implements the above embodiments. Step S12, namely, constructing an objective function based on the computing power of the mobile device and the communication link quality between the mobile device and the edge computing node to minimize the total training time of the model training for the multi-federated learning task while ensuring the accuracy of the global model, and to minimize the total energy consumption of the mobile device while ensuring the accuracy of the local model, includes steps S121 to S126: S121. Based on the computing power of the mobile device, construct a mathematical model of the training time and energy consumed by the mobile device in local model training for the federated learning task, and record them as the first consumption time. And the first energy consumption .

[0028] Preferably, step S121 specifically includes: Based on the number of sample data collected by the mobile device participating in the federated learning task The allocated computing resources and the number of CPU cycles required to process a unit sample of data A mathematical model is constructed to represent the training time consumed in a single local model training session for the mobile device participating in the federated learning task, denoted as the single training time. ; Based on the preset local model accuracy threshold Determine the number of local iterations required for the mobile device to participate in the local model training of the federated learning task; Based on the single training time and the number of local iterations, a mathematical model is constructed to represent the training time consumed by the mobile device in participating in the federated learning task's local model training, denoted as the first training time. ; Based on the number of local iterations and the number of sample data The computing resources and the number of CPU cycles A mathematical model is constructed to represent the energy consumed by the mobile device in local model training for the federated learning task, denoted as the first energy consumption. .

[0029] S122. Based on the communication link quality, construct a mathematical model of the transmission time and energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, and record them as the second consumption time. Second energy consumption .

[0030] Preferably, step S122 specifically includes: According to the sub-channel bandwidth allocated to the mobile device Data transmission power from the mobile device to the edge computing node and the wireless communication channel gain between the mobile device and the edge computing node. Construct the data transmission rate between the mobile device and the edge computing node. Mathematical model; Based on the data transmission volume of the mobile device and the edge computing node and the data transmission rate A mathematical model is constructed to represent the transmission time consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, denoted as the second transmission time. ; According to the second consumption time and the data transmission power A mathematical model is constructed to represent the energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, denoted as the second energy consumption. .

[0031] S123, Based on the first consumption time and the second consumption time Construct the total time consumed by the mobile device in the federated learning task. Mathematical model; S124, Based on the first energy consumption and the second energy consumption Construct the total energy consumed by the mobile device in the federated learning task. Mathematical model; S125. Determine the total training time of the model training for the federated learning task based on the total time consumed by at least one mobile device participating in the model training of the federated learning task in parallel. Mathematical model; S126. Construct an objective function with the goal of minimizing the total training time of all the federated learning tasks and minimizing the total energy consumption of all the mobile devices.

[0032] Specifically, in this embodiment of the invention, the objective function is constructed by building a system mathematical model in a multi-federated learning task scenario, including two stages: a computational model and a communication model.

[0033] (1) Computational Model for m A federal learning task, mobile devices have m Each corresponding local dataset. Having a task-oriented j The local dataset is represented as ,in, Indicates the number of samples.

[0034] During local training, mobile devices The number of CPU cycles required to process a unit of data sample is indivual. The value of is determined by the structure of the federated learning model and its training optimization method. Here, it is assumed that all mobile devices use the same training model and optimization method. Mobile devices Assigned to task j The computing resources are Therefore, mobile speed devices For federal learning tasks j The time consumed to perform one local model training (i.e., the training time per session) is:

[0035] In addition, the local model accuracy threshold is defined as ,in, A smaller local model accuracy threshold means that more precise updates are needed. To obtain the required accuracy threshold, the number of local iterations required can be expressed as... .

[0036] Ultimately, the movement speed is not... For federal learning tasks j implement The computation time consumed by the first local iteration (i.e., the first computation time) is:

[0037] Mobile devices during this process The computational energy consumed is:

[0038] in, For effective capacitor switching, this value is determined by the device. i The structure determines this.

[0039] (2) Communication model After each round of local training, the mobile device The local model parameters need to be uploaded to the edge computing node via a wireless access network. Assuming the wireless access protocol is frequency division multiple access (FDMA) or time division multiple access (TDMA), interference between adjacent mobile devices can be ignored. Furthermore, during the global model's transmission phase, considering that the base station's power is much greater than that of the mobile device, and the downlink bandwidth is greater than the uplink bandwidth, the transmission time can be disregarded compared to the upload time; similarly, the energy consumption of the base station during reception can also be ignored. Mobile devices With edge computing nodes The data transfer rate between them is:

[0040] in, To be assigned to mobile devices Sub-channel bandwidth, For mobile devices To edge computing nodes Data transmission power when transmitting signals; For mobile devices With edge computing nodes Wireless communication channel gain between; It is Gaussian white noise.

[0041] Assuming mobile devices Need to provide edge computing nodes The amount of data transmitted is So, from mobile devices To edge computing nodes The time required to transmit parameters (i.e., the second consumption time) is:

[0042] The energy consumed during transmission (i.e., the second energy consumption) is:

[0043] in, This is used to indicate the energy consumption of the transmitter. For power amplifier coefficient, This represents the circuit power.

[0044] Finally, considering the impact of communication link uncertainty on multi-task federated learning scenarios, a scheduling scheme between mobile devices and edge computing nodes is ultimately implemented based on link quality. Assuming the global model deployment phase is ideal, the quality of the communication link primarily affects the reliability of local model uploads. The probability of successful transmission is defined as: To simplify the problem, we assume here that all devices have the same probability of successful transmission. .

[0045] Furthermore, considering the differences in communication link quality, some wireless transmission links may experience communication interruptions. For the first... j Each federated learning task's completion time for each round will depend on the scheduled mobile device that uploads its parameters last.

[0046] Based on the above considerations, the global precision threshold is defined as follows: ( To achieve the required accuracy threshold, the number of global iteration rounds required is... Currently, we only consider minimizing latency and energy consumption in the system model when completing one round of global aggregation. By ensuring that latency and energy consumption are minimized in each round, we can ultimately achieve the minimum total cost for the entire system to complete all federated learning tasks.

[0047] Therefore, the final mobile device Participating in federal learning tasks j The total training time consumed by performing one round of global iteration can be expressed as:

[0048] Considering that various mobile devices can participate in tasks in parallel, the longest time each mobile device spends on the same task is taken as the total time required to execute that task. Therefore, for participating in federated learning tasks... j The total training time required for model training can be expressed as:

[0049] In the first round of global iteration Participate in the task j The total energy consumed can be expressed as:

[0050] in, Represents edge computing nodes For mobile devices The scheduling status. It is a binary variable. , This indicates that the i-th mobile device participates in the j-th federated learning task. This indicates that the i-th mobile device does not participate in the j-th federated learning task.

[0051] Under this consideration, let's define the cost function. While ensuring the training quality of the task model, we can reduce the latency and energy consumption required for training.

[0052] Preferably, the formula for the objective function is:

[0053] The constraints are as follows:

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] in, The total training time for training the model for the j-th federated learning task. The total energy consumed by the i-th mobile device participating in the j-th federated learning task; m is the number of federated learning tasks, and n is the number of mobile devices. , This indicates that the i-th mobile device participates in the j-th federated learning task. This means that the i-th mobile device does not participate in the j-th federated learning task. As preset time weights, This is the preset energy consumption weight.

[0061] Resource constraints and It defines the adjustment range for the computing resources and transmission power allocated to each task for each mobile device. It is the maximum computing resources that each mobile device can allocate to each task. This refers to the maximum transmission power that each mobile device can allocate to each task; task participation constraints. This means that for any edge computing node, at least one mobile device must be connected to it and assist it in completing the assigned tasks; node capacity constraint. express The number of connected participants cannot exceed its capacity. Model accuracy constraints Introducing the concept of generalization error, and controlling the upper bound of the generalization error to a very small number. This ensures the training quality of the model. Simultaneously, it correlates generalization error with the probability of successful transmission. By combining these factors, the number of mobile devices participating in the same task can be jointly constrained.

[0062] As a preferred embodiment, the present invention further implements the above embodiments, and step S13, namely, calculating the mobile device scheduling scheme for each federated learning task according to the objective function, includes: Using the objective function as the fitness function, a genetic algorithm is used to iteratively generate the optimal individual to obtain the mobile device scheduling scheme for each federated learning task.

[0063] In this embodiment of the invention, a differential evolution algorithm is considered to implement a device scheduling scheme for the scheduling problem of mobile devices in a multi-task federated scenario. This algorithm uses binary encoding, representing the participation of all mobile devices in each federated task using 0 and 1 variables. The population is represented as follows: ,in Indicates the first t Individuals in a population during rotation k This corresponds to the k-th scheduling scheme in round t of the scheduling problem. The population contains a total of M Group scheduling schemes. Each group scheduling scheme is represented as follows: ,in (n and m represent the number of mobile devices and edge computing nodes, respectively) indicates the dimension of the independent variables in the scheduling scheme. Indicates the first k Under the group scheduling scheme, the first The first FL task is for the first The scheduling status of each mobile device. Indicates the first Mobile devices participated in the task Training is conducted only if the objective function is correct; otherwise, the algorithm does not participate. The aforementioned objective function is used as the fitness function of this algorithm. f After population initialization, crossover and mutation operations are performed using the crossover and mutation formulas. Then, the fitness values ​​of different individuals in the population are calculated. M Group scheduling scheme The value is used to obtain the best individual in the current population. T After rounds of iteration, the optimal individual is finally obtained. .

[0064] Using the technical means of this invention, a mobile device scheduling strategy based on differential evolution is proposed. Each feasible mobile device scheduling scheme is treated as an individual, and the weighted sum of the total task latency and energy consumption corresponding to each individual is used as its fitness value. Then, through mutation, crossover, and selection operations, the individual with the best fitness value is iteratively determined as the second-best solution to the problem, and finally the optimal mobile device scheduling scheme for each federated learning task is determined.

[0065] In summary, the scheduling method for mobile devices provided by the embodiments of the present invention has the following beneficial effects: 1. Existing solutions typically only minimize latency during multi-federated learning tasks as the optimization objective. However, in this scenario, the available resources of mobile devices are often limited, and the computational resources and energy consumption from local model training are also significant. This invention comprehensively considers the energy and time consumption during training and transmission, constructing a multi-objective optimization problem that minimizes the total training time and energy consumption.

[0066] 2. Existing solutions do not consider the risk of communication interruption caused by poor wireless transmission links. This invention, taking into account differences in link quality, introduces the concept of successful transmission probability. By adding this constraint variable to the optimization objective, it controls the number of mobile devices participating in the same task under different transmission probabilities.

[0067] 3. Existing solutions limit a single device to participating in only one federated task per round. However, in practice, a single mobile device can participate in multiple tasks simultaneously. This invention enriches the multi-task federated learning scenario, allowing each mobile device to participate in multiple tasks in each training round, making the designed scheduling scheme more closely aligned with real-world application scenarios.

[0068] See Figure 3 This is a schematic diagram of a scheduling device for a mobile device provided in an embodiment of the present invention. The embodiment of the present invention provides a scheduling device 10 for a mobile device, comprising: Application scenario construction module 11 is used to construct application scenarios for multiple federated learning tasks; wherein, the application scenario includes several mobile devices, edge computing nodes, and federated learning tasks; the mobile devices are used to collect sample data, use the sample data to complete local model training for the federated learning task, and transmit the local model to the edge computing node, and each mobile device participates in at least one federated learning task; the edge computing node is used to complete global model update for the federated learning task based on the local model, and each edge computing node is responsible for one federated learning task; The objective function construction module 12 is used to construct an objective function based on the computing power of the mobile device and the quality of the communication link between the mobile device and the edge computing node, which minimizes the total training time of the model training for the multi-federated learning task while ensuring the accuracy of the global model, and minimizes the total energy consumption of the mobile device while ensuring the accuracy of the local model. The mobile device scheduling module 13 is used to calculate the mobile device scheduling scheme for each of the federated learning tasks according to the objective function.

[0069] It should be noted that the scheduling device for a mobile device provided in this embodiment of the invention is used to execute all the process steps of the scheduling method for a mobile device in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0070] This invention also provides a scheduling device for a mobile device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the scheduling method for the mobile device as described in any of the above embodiments.

[0071] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the scheduling method of the mobile device as described in any of the above embodiments.

[0072] This invention also provides a computer program product, which includes a computer program or computer instructions. When the computer program or computer instructions are executed by a processor, they implement the scheduling method of the mobile device as described in any of the above embodiments.

[0073] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0074] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A scheduling method of a mobile device, characterized by, include: An application scenario for constructing multiple federated learning tasks is provided. This application scenario includes several mobile devices, edge computing nodes, and federated learning tasks. The mobile devices are used to collect sample data, use the sample data to complete local model training for the federated learning task, and transmit the local model to the edge computing node. Each mobile device participates in at least one federated learning task. The edge computing node is used to complete global model updates for the federated learning task based on the local model. Each edge computing node is responsible for one federated learning task. Based on the computing power of the mobile device and the quality of the communication link between the mobile device and the edge computing node, an objective function is constructed that minimizes the total training time of the model training for multiple federated learning tasks while ensuring the accuracy of the global model, and minimizes the total energy consumption of the mobile device while ensuring the accuracy of the local model. Based on the objective function, calculate the mobile device scheduling scheme for each of the federated learning tasks.

2. The scheduling method of claim 1, wherein, The objective function, based on the computing power of the mobile device and the communication link quality between the mobile device and the edge computing node, aims to minimize the total training time of the model training for multiple federated learning tasks while ensuring global model accuracy, and to minimize the total energy consumption of the mobile device while ensuring local model accuracy. This objective function includes: Based on the computing power of the mobile device, a mathematical model is constructed to calculate the training time and energy consumed by the mobile device in local model training for the federated learning task, which are respectively denoted as the first time consumed and the first energy consumed. Based on the communication link quality, a mathematical model is constructed of the transmission time and energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, which are respectively denoted as the second time consumption and the second energy consumption. Based on the first consumption time and the second consumption time, a mathematical model is constructed to represent the total consumption time of the mobile device participating in the federated learning task; Based on the first energy consumption and the second energy consumption, a mathematical model is constructed to represent the total energy consumption of the mobile device participating in the federated learning task. A mathematical model for determining the total training time of the model training for the federated learning task is based on the total time consumed by at least one mobile device participating in the model training of the federated learning task in parallel. An objective function is constructed with the goal of minimizing the total training time of all the federated learning tasks and minimizing the total energy consumption of all the mobile devices.

3. The scheduling method of claim 2, wherein, The step of constructing a mathematical model based on the computing power of the mobile device, representing the training time and energy consumed by the mobile device in local model training for the federated learning task, denoted as the first consumption time and the first consumption energy, includes: Based on the number of sample data collected by the mobile device participating in the federated learning task, the allocated computing resources, and the number of CPU cycles required to process a unit of sample data, a mathematical model is constructed to determine the training time consumed by the mobile device in a single local model training session for the federated learning task, which is denoted as the single training time. Based on a preset local model accuracy threshold, determine the number of local iterations required for the mobile device to train the local model for the federated learning task. Based on the single training time and the number of local iterations, a mathematical model is constructed for the training time consumed by the mobile device in the local model training of the federated learning task, which is denoted as the first consumption time. Based on the number of local iterations, the number of sample data, the computing resources, and the number of CPU cycles, a mathematical model is constructed of the energy consumed by the mobile device in local model training for the federated learning task, denoted as the first energy consumption.

4. The scheduling method for mobile devices as described in claim 3, characterized in that, Based on the communication link quality, a mathematical model is constructed to determine the transmission time and energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, denoted as the second time consumption and the second energy consumption, respectively. A mathematical model of the data transmission rate between the mobile device and the edge computing node is constructed based on the sub-channel bandwidth allocated to the mobile device, the data transmission power from the mobile device to the edge computing node, and the wireless communication channel gain between the mobile device and the edge computing node. Based on the data transmission volume and data transmission rate of the mobile device and the edge computing node, a mathematical model is constructed to represent the transmission time consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, which is denoted as the second transmission time. Based on the second consumption time, a mathematical model is constructed of the energy consumed by the mobile device in transmitting the local model trained for the federated learning task to the corresponding edge computing node, denoted as the second energy consumption.

5. The scheduling method for mobile devices as described in claim 2, characterized in that, The formula for the objective function is: in, The total training time for training the model for the j-th federated learning task. The total energy consumed by the i-th mobile device participating in the j-th federated learning task; m is the number of federated learning tasks, and n is the number of mobile devices. , This indicates that the i-th mobile device participates in the j-th federated learning task. This means that the i-th mobile device does not participate in the j-th federated learning task. As preset time weights, This is the preset energy consumption weight.

6. The scheduling method for mobile devices as described in any one of claims 1 to 5, characterized in that, The step of calculating the mobile device scheduling scheme for each federated learning task based on the objective function includes: Using the objective function as the fitness function, a genetic algorithm is used to iteratively generate the optimal individual to obtain the mobile device scheduling scheme for each federated learning task.

7. A scheduling device for a mobile device, characterized in that, include: An application scenario construction module is used to construct application scenarios for multiple federated learning tasks. Each application scenario includes several mobile devices, edge computing nodes, and federated learning tasks. The mobile devices are used to collect sample data, use the sample data to complete local model training for the federated learning task, and transmit the local model to the edge computing node. Each mobile device participates in at least one federated learning task. The edge computing node is used to complete global model updates for the federated learning task based on the local model. Each edge computing node is responsible for one federated learning task. The objective function construction module is used to construct an objective function based on the computing power of the mobile device and the quality of the communication link between the mobile device and the edge computing node, which minimizes the total training time of the model training for the multi-federated learning task while ensuring the accuracy of the global model, and minimizes the total energy consumption of the mobile device while ensuring the accuracy of the local model. The mobile device scheduling module is used to calculate the mobile device scheduling scheme for each of the federated learning tasks based on the objective function.

8. A scheduling device for a mobile device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the scheduling method of the mobile device as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the scheduling method of the mobile device as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which, when executed by a processor, implement the scheduling method for a mobile device as described in any one of claims 1 to 6.

Images