A network resource scheduling system and method for federated learning service flow

Abstract

The application discloses a network resource scheduling system and method for a federal learning service flow. The system effectively alleviates the problem of too frequent service flow transmission and high bandwidth requirement of the existing two-layer architecture of the federal learning mechanism through a three-layer architecture of a factory layer, a field layer and a device layer. Meanwhile, the application generates a scheduling plan and provides a deterministic transmission service for industrial Internet of Things devices through the field layer, solves the problems of low resource scheduling efficiency and high transmission delay in a high-concurrency federal learning service flow and a centralized interaction scene, and greatly improves the learning accuracy and convergence speed of the federal learning in the industrial Internet of Things.

Description

Technical Field

[0001] This invention relates to the field of data transmission technology, and specifically to a network resource scheduling system and method for federated learning service flows. Background Technology

[0002] By integrating emerging artificial intelligence technologies, the next generation of Industrial Internet of Things (IIoT) is expected to accelerate industrial automation. In this context, distributed IIoT devices will generate massive amounts of data. Due to stringent latency requirements and privacy regulations, industrial data needs to be processed in a distributed manner at the network edge. In recent years, a significant amount of research has focused on this area, resulting in many new distributed learning paradigms that can effectively utilize massive amounts of dispersed data and distributed computing resources without requiring raw data transmission over the network. In particular, federated learning, as one of the most widely studied learning paradigms, enables distributed devices to collaboratively train deep neural network models by periodically exchanging model parameters between the devices and a central server (i.e., federated learning workflows). This paradigm effectively protects the privacy of data providers while reducing communication latency and overhead.

[0003] However, running federated learning in resource-constrained and latency-sensitive Industrial IoT scenarios still faces significant challenges, namely the coexistence of multiple forms of federated learning traffic flows (such as time-triggered federated learning flows, best-effort federated learning flows, and federated learning flows). Therefore, the ability to efficiently aggregate model parameters from the devices on a central server is crucial for accelerating federated learning convergence. Time-Sensitive Networking (TSN) technology, evolved from traditional industrial Ethernet, can provide bounded end-to-end latency, jitter, and packet loss rate for multiple federated learning traffic flows within a local area network. Federated learning requires all devices to interact with the central server in parallel, which can lead to severe resource contention and transmission latency issues, especially when the number of devices participating in training is large. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a network resource scheduling system for federated learning service flows to solve the problems of inefficient resource scheduling and high transmission latency in high-concurrency federated learning service flows and centralized interaction scenarios in local area networks.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention provides a network resource scheduling system for federated learning service flows, comprising:

[0007] The factory level, the site level, and the equipment level, among which,

[0008] The factory layer includes a factory server and a central controller; the field layer includes wireless access switches and wired access switches; the equipment layer includes industrial gateways.

[0009] The factory server is used to: obtain instantiation parameters; construct a federated learning model based on the instantiation parameters; receive trained model parameters sent by the industrial gateway, and update the federated learning model based on the trained model parameters; aggregate the trained model parameters, and distribute the aggregated model parameters to the industrial gateway.

[0010] The central controller is used to generate a resource scheduling plan based on a federated learning model; and to issue resource scheduling instructions to the wireless access switch and the wired access switch based on the resource scheduling plan.

[0011] The industrial gateway is used to perform local training based on local data and model parameters sent by the factory server to obtain trained model parameters, and then send the trained model parameters to the factory server.

[0012] The wireless access switch and the wired access switch are used to establish a network connection between the factory layer, the field layer and the equipment layer, and to receive resource scheduling instructions to execute transmission scheduling.

[0013] Optionally, the wireless access switch and the wired switch are used to establish a network connection between the factory layer, the field layer, and the equipment layer, including:

[0014] The wireless access switch is used to establish a connection between the industrial gateway and the factory server;

[0015] The wired access switch is used to create a network between the field layer and the factory layer.

[0016] Optionally, the system further includes: a shaper, which is disposed in each of the wireless access switches and the wired access switches; the shaper is used to construct a queuing mechanism for federated learning traffic flows.

[0017] Optionally, the system further includes a field server, which is set up on each production line on-site for aggregating and distributing federated learning model parameters.

[0018] This invention also provides a network resource scheduling method for federated learning service flows, applied to the central controller of the aforementioned network resource scheduling system for federated learning service flows, comprising:

[0019] Obtain business flow information and preset federated learning models for federated learning business processes;

[0020] A resource scheduling model is established based on the business flow information and the federated learning model;

[0021] The business flow information is input into the resource scheduling model to obtain a resource scheduling plan.

[0022] Optionally, establishing a resource scheduling model based on the business flow information and the federated learning model includes:

[0023] The federated learning model is optimized based on the preset federated learning business flow parameter constraints to obtain a multi-objective resource scheduling constraint optimization model;

[0024] Extract status information from the business flow information;

[0025] The remaining time slot capacity is obtained based on the service flow information and the multi-objective resource scheduling constraint optimization model.

[0026] A resource scheduling model is established based on status information and remaining time slot capacity.

[0027] Optionally, the preset federated learning business flow parameter constraints include:

[0028] The scheduling cycle, time slot length, transmission time, transmission time slot capacity, and deterministic requirements of the federated learning service flow.

[0029] Optionally, establishing a resource scheduling model based on state information and remaining time slot capacity includes:

[0030] The state information is subjected to feature extraction to obtain state features;

[0031] A state space is established based on the state characteristics and the remaining time slot capacity;

[0032] An action space is established based on the state information and the preset time slot allocation decision;

[0033] Decision analysis is performed based on the state space and the action space, and a resource scheduling model is established based on a preset network model.

[0034] This invention also provides an electronic device, comprising:

[0035] The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes these computer instructions to perform the network resource scheduling method for federated learning service flows provided in this embodiment of the invention.

[0036] This invention also provides a computer-readable storage medium storing computer instructions for causing a computer to execute the network resource scheduling method for federated learning service flows provided in this invention.

[0037] The technical solution of this invention has the following advantages:

[0038] 1. This invention provides a network resource scheduling system for federated learning service flows. Through a three-layer architecture (factory layer, field layer, and device layer), it effectively alleviates the problems of excessively frequent service flow transmission and high bandwidth requirements in existing two-layer federated learning mechanisms. Simultaneously, this invention generates scheduling plans and provides deterministic transmission services to industrial IoT devices through the field layer, solving the problems of inefficient resource scheduling and high transmission latency in high-concurrency federated learning service flows and centralized interaction scenarios. This significantly improves the learning accuracy and convergence speed of federated learning in the industrial IoT.

[0039] 2. This invention provides a network resource scheduling method for federated learning service flows. By establishing a resource scheduling model and generating a resource scheduling plan, and comprehensively considering multiple factors, the overall system performance is effectively improved. At the same time, by comprehensively considering the accuracy of federated learning and the scheduling success rate of the TSN network, the deterministic transmission of federated learning model parameters is guaranteed. Attached Figure Description

[0040] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a schematic diagram of a network resource scheduling system for federated learning service flows in an embodiment of the present invention;

[0042] Figure 2 This is a flowchart of a network resource scheduling method for federated learning service flows in an embodiment of the present invention.

[0043] Figure 3 A flowchart illustrating the establishment of a resource scheduling model according to an embodiment of the present invention;

[0044] Figure 4 A flowchart illustrating the establishment of a resource scheduling model according to an embodiment of the present invention;

[0045] Figure 5 This is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0047] According to an embodiment of the present invention, a network resource scheduling system embodiment for federated learning service flows is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0048] This embodiment provides a network resource scheduling system for federated learning service flows, which can be used in scenarios such as the Industrial Internet of Things (IIoT). Figure 1 As shown, this network resource scheduling system for federated learning business flows includes: a factory layer, a field layer, and a device layer, wherein...

[0049] The factory layer includes factory servers and central controllers; the field layer includes wireless access switches and wired access switches; and the equipment layer includes industrial gateways.

[0050] The factory server is used for: obtaining instantiation parameters; building a federated learning model based on the instantiation parameters; receiving the trained model parameters sent by the industrial gateway and updating the federated learning model based on the trained model parameters; aggregating the trained model parameters and distributing the aggregated model parameters to the industrial gateway.

[0051] The central controller is used to generate resource scheduling plans based on the federated learning model; and to issue resource scheduling instructions to wireless access switches and wired access switches based on the resource scheduling plans.

[0052] The industrial gateway is used to perform local training based on local data and model parameters sent by the factory server to obtain trained model parameters, and then send the trained model parameters to the factory server.

[0053] Wireless access switches and wired access switches are used to establish network connections between the factory layer, field layer, and equipment layer, and to receive resource scheduling instructions to execute transmission scheduling.

[0054] Specifically, the network resource scheduling system for federated learning service flows provided in this embodiment of the invention, through a three-layer architecture of factory layer, field layer, and device layer, effectively alleviates the problems of excessively frequent service flow transmission and high bandwidth requirements in the existing two-layer architecture of federated learning mechanisms. Simultaneously, this invention generates scheduling plans and provides deterministic transmission services to industrial IoT devices through the field layer, solving the problems of inefficient resource scheduling and high transmission latency in high-concurrency federated learning service flows and centralized interaction scenarios, greatly improving the learning accuracy and convergence speed of federated learning in the industrial IoT.

[0055] Specifically, traditional federated learning scheduling systems have only a simple two-layer architecture: a factory layer and a device layer. This two-layer structure leads to excessively frequent federated learning service flow transmission and high bandwidth requirements. Therefore, the resource scheduling system for service flow transmission proposed in this application consists of three layers: a factory layer, a field layer, and a device layer. By adding an intermediate field layer, the pressure on service flow transmission is alleviated, and transmission efficiency is improved. Through precise federated learning service flow scheduling, the parameter aggregation and distribution between the three layers are guaranteed deterministically. The amount of model parameters transmitted between the field layer and the factory center is very small, effectively avoiding bandwidth resource contention, thereby greatly saving bandwidth resource consumption for service flow transmission.

[0056] In the following embodiment, taking the deployment of 30 terminal nodes, 3 field-level servers, and 1 factory-level server in the considered network scenario as an example, the system supports the operation of a network resource scheduling system for federated learning service flows. These network nodes are interconnected through 10 TSN switches to form a tree topology. 66 federated learning service flows need to be scheduled within one scheduling cycle (i.e., one federated learning round). Furthermore, the link bandwidth of the converged wireless and wired TSN network is set to 1 Gbit / s. Of this, 90% of the bandwidth is used to support the operation of the network resource scheduling system for federated learning service flows, and 10% of the bandwidth is used to transmit other industrial federated learning service flows, such as control and / or safety-related federated learning service flows. The propagation delay τproc, processing delay τprop, queuing delay τqueue, and synchronization delay τsync are set to 10 μs. The time slot is set to 40ms, therefore the scheduling period is 40ms, and the time slot size within each scheduling period is set to 10ms. For the DynamicSched algorithm, the online and target networks respectively have [S dim 1000, 500, A dim Neuron; here, S dim A represents the dimension by which an agent observes the state of its environment. dim The number of time slots in the network is denoted as τ. Additionally, the update factor τ of the target network is set to 0.005, a value that depends on the scheduling period and the time slot size.

[0057] For the federated learning model, this invention utilizes a convolutional recurrent neural network based on an attention mechanism. This model comprises 10 layers: two convolutional layers, two max-pooling layers, one flattened layer, one fully connected layer, one forward-gated recurrent unit layer, one backward-gated recurrent unit layer, one attention layer, and one softmax output layer. Based on this, the total number of model parameters required for training is approximately 15099. Since each parameter can be quantized to 4 bits, the data size of the federated learning model parameters is approximately 60395 bits.

[0058] Specifically, in one embodiment, the aforementioned wireless access switch and wired switch are used to establish a network connection between the factory layer, the field layer, and the equipment layer, including:

[0059] Wireless access switches are used to establish connections between industrial gateways and factory servers;

[0060] Wired access switches are used to create networks between the field layer and the factory layer.

[0061] Specifically, the switch is a TSN (Time-Sensitive Networking) switch. By deploying multiple wireless and wired access switches at the field layer, it uses time-sensitive networking technology that combines wireless and wired connections to connect the three layers, providing more efficient deterministic transmission services for industrial IoT devices and supporting efficient federated learning with deterministic requirements.

[0062] Specifically, in one embodiment, the system further includes a shaper, which is disposed in each wireless access switch and wired access switch; the shaper is used to construct a queuing mechanism for federated learning service flows. Specifically, the shaper is a CQF-based shaper, which can better guarantee deterministic transmission requirements, and by constructing a queuing mechanism for federated learning service flows through the shaper, it can provide service guarantees with bounded transmission delays for each federated learning service flow.

[0063] Specifically, in one embodiment, the system further includes a field server, which is set up on each production line on-site and is used to aggregate and distribute parameters of the federated learning model.

[0064] This embodiment also provides a network resource scheduling method for federated learning service flows, applicable to, for example... Figure 1 The central controller in the network resource scheduling system for federated learning traffic flow shown is, for example... Figure 2 As shown, the method specifically includes the following steps:

[0065] Step S1: Obtain the business flow information of the federated learning business flow and the preset federated learning model.

[0066] Step S2: Establish a resource scheduling model based on business flow information and federated learning model.

[0067] Step S3: Input the business flow information into the resource scheduling model to obtain the resource scheduling plan.

[0068] Through steps S1 to S3 above, the network resource scheduling method for federated learning service flows provided in this embodiment of the invention effectively improves the overall system performance by establishing a resource scheduling model and generating a resource scheduling plan, taking into account multiple factors. At the same time, it ensures the deterministic transmission of federated learning model parameters by comprehensively considering the accuracy of federated learning and the scheduling success rate of the TSN network.

[0069] Specifically, in one embodiment, step S2 described above is as follows: Figure 3 As shown, the specific steps include the following:

[0070] Step S21: Optimize the federated learning model according to the preset federated learning service flow parameter constraints to obtain a multi-objective resource scheduling constraint optimization model. Specifically, based on the preset federated learning service flow parameter constraints, and comprehensively considering the model's learning accuracy and scheduling capability, the overall system performance can be maximized, and the optimal time interval allocation for parameter transmission can be made.

[0071] Step S22: Extract status information from business flow information.

[0072] Step S23: Obtain the remaining time slot capacity based on the service flow information and the multi-objective resource scheduling constraint optimization model.

[0073] Step S24: Establish a resource scheduling model based on the status information and remaining time slot capacity.

[0074] Specifically, based on constraints and the federated learning model, a multi-objective resource scheduling constraint optimization model is established to formulate time slot resource allocation decisions that maximize overall system performance for deterministic model parameter transmission in a combined wireless and wired scheduling model. Since the learning accuracy of the server-side model and the scheduling capability of the fusion network are interrelated, both need to be considered simultaneously to maximize overall system performance. The hierarchical federated learning framework in the network resource scheduling system for federated learning traffic flows needs to calculate a time slot allocation method that maximizes overall system performance for deterministic model parameter transmission on the fused TSN network. Therefore, the multi-objective resource scheduling problem is constructed as follows:

[0075] Maximizgξ|A(θ r )|+(1-ξ)|S r |,

[0076]

[0077]

[0078]

[0079] T≤[T min T max ].

[0080] The above constraints ensure that the needs of federated learning traffic can be met, thus providing deterministic transmission guarantees for federated learning traffic.

[0081] Specifically, in one embodiment, the preset federated learning service flow parameter constraints in step S21 above include the federated learning service flow scheduling period, time slot length, federated learning service flow transmission time slot capacity, and deterministic requirements of the federated learning service flow.

[0082] Specifically, define parameter constraints for the model federated learning service flow to ensure deterministic end-to-end transmission of the model federated learning service flow, wherein:

[0083] Federated learning workflow scheduling period: Since the gating list for all federated learning workflows needs to be calculated periodically, the scheduling period SC should be equal to the least common multiple of the periods of all federated learning workflows, i.e.:

[0084] SC = LCM(K|cycles|).

[0085] Where SC is a constant, and LCM(·) is a function used to return the least common multiple of all federated learning business flow cycles;

[0086] Time slot length: The value of time slot length T should be within [T min ,T max Within the range:

[0087] T min =τ send +τ prop +τ proc +τ sync +τ queue ,

[0088] T max =GCD(K[cycles]).

[0089] Where, τ send τ prop τ proc τ queue τ syncThese represent the packet transmission, propagation, processing, queuing, and synchronization delays between any two adjacent connected nodes, respectively; GCD(·) is a function used to return the greatest common divisor of all federated learning traffic cycles;

[0090] Federated learning traffic transmission time: Under any circumstances, the transmission time of each federated learning traffic should be less than its federated learning traffic cycle. This prevents data packets from different federated learning traffic cycles from simultaneously consuming limited network interface card resources; therefore:

[0091]

[0092] The cycle time of the federated learning workflow should be divisible by the time slot length T:

[0093]

[0094] Federated learning service flow transmission slot capacity: By determining the transmission time of the federated learning service flow, the time resources already occupied for each slot in the current scheduling period can be calculated, i.e.

[0095]

[0096] Each data packet can only be inserted into a single time slot for transmission, that is:

[0097]

[0098] The total size of the data packets sent in each time slot cannot exceed its maximum capacity:

[0099]

[0100] Among them, Ω s (·) represents an indicator function. Indicates the federated learning business flow t k The xth data packet, This is the maximum capacity of time slot s;

[0101] Deterministic requirements for federated learning workflows (latency, jitter, packet loss rate): Based on the above definitions, the end-to-end latency of each federated learning workflow can be obtained. Shaking and packet loss rate Right now

[0102]

[0103]

[0104]

[0105] Among them, Hk Indicates the federated learning business flow t k The number of switch nodes traversed in an end-to-end transmission.

[0106] Specifically, in one embodiment, step S24 described above is as follows: Figure 4 As shown, the specific steps include the following:

[0107] Step S241: Extract features from the state information to obtain state features. Specifically, the features of the service flow parameters include: the features of the federated learning service flow parameters are: the federated learning service flow scheduling period, slot length, transmission time, transmission slot capacity, deterministic requirements (latency, jitter, packet loss rate), etc.

[0108] Step S242: Establish the state space based on state characteristics and remaining time slot capacity. Specifically, in each scheduling cycle, federated learning traffic flow information Φ is collected from industrial IoT devices and TSN switches. t and remaining time slot capacity The state space can be defined as:

[0109]

[0110] in, This represents the number of federated learning business flows t received in the r-th round of federated learning. k State characteristics, This indicates the remaining capacity of each receive queue on switch port p.

[0111] Step S243: Establish the action space based on state information and preset time slot allocation decisions. Specifically, based on state information, the agent can formulate time slot resource allocation decisions to determine the federated learning business flow t. k The scheduling of time slots enables deterministic transmission of federated learning traffic flows, i.e.:

[0112]

[0113] A reward function can also be established: once the agent executes the predetermined decision, it will receive a reward to evaluate its performance on the given decision. t The effectiveness of decision-making under certain conditions is assessed to maximize learning accuracy and scheduling success rate.

[0114]

[0115] Step S244: Perform decision analysis based on the state space and action space, and establish a resource scheduling model based on a preset network model. Specifically, a resource scheduling model, called DynamicSched, is established using a resource dynamic scheduling algorithm based on dual deep Q-networks (D3QN). By employing a dual deep Q-network (DQN) module and a dual DQN module, the decision overestimation problem can be overcome, effectively improving decision performance. The dual DQN module uses an online network to evaluate the ε-greedy policy and a target network to estimate its value. In this way, the operation of finding the maximum Q value in the target Q-value estimation of the DQN algorithm can be decomposed into decision selection and decision evaluation, thereby effectively alleviating the decision overestimation problem. The lower layers of the online network and the target network in the dual DQN module are respectively constructed using... and In this parameterized fully connected network structure, the deeper layers of the two networks are each assigned a separate fully connected layer, instead of the original single fully connected layer. Here, these two fully connected layers represent value functions. and dominance function In the output aggregation layer, the outputs of the two fully connected layers share a common feature learning layer, thereby combining the two fully connected layers to produce a single state decision Q-value function, i.e.,

[0116]

[0117] Where V(·) represents a one-dimensional scalar and A(·) represents a |A|-dimensional vector.

[0118] This invention also provides an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 901 and a memory 902, wherein the processor 901 and the memory 902 may be connected via a bus or other means. Figure 5 Taking the example of a connection between China and Israel via a bus.

[0119] Processor 901 can be a Central Processing Unit (CPU). Processor 901 can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0120] The memory 902, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the method embodiments of the present invention. The processor 901 executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory 902, thereby implementing the methods in the above method embodiments.

[0121] The memory 902 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor 901, etc. Furthermore, the memory 902 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 902 may optionally include memory remotely located relative to the processor 901, and these remote memories may be connected to the processor 901 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0122] One or more modules are stored in memory 902, and when executed by processor 901, they perform the methods described in the above method embodiments.

[0123] The specific details of the aforementioned electronic device can be understood by referring to the relevant descriptions and effects in the above method embodiments, and will not be repeated here.

[0124] Those skilled in the art will understand that all or part of the processes in the methods of 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 methods described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0125] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A network resource scheduling system for federated learning service flows, characterized in that, include: The factory level, the site level, and the equipment level, among which, The factory layer includes a factory server and a central controller; the field layer includes wireless access switches and wired access switches; the equipment layer includes industrial gateways. The factory server is used to: obtain instantiation parameters; construct a federated learning model based on the instantiation parameters; receive trained model parameters sent by the industrial gateway, and update the federated learning model based on the trained model parameters; aggregate the trained model parameters, and distribute the aggregated model parameters to the industrial gateway. The central controller is used to generate a resource scheduling plan based on a federated learning model; and to issue resource scheduling instructions to the wireless access switch and the wired access switch based on the resource scheduling plan. The industrial gateway is used to perform local training based on local data and model parameters sent by the factory server to obtain trained model parameters, and then send the trained model parameters to the factory server. The wireless access switch and the wired access switch are used to establish a network connection between the factory layer, the field layer and the equipment layer, and to receive resource scheduling instructions to execute transmission scheduling. The system further includes: a shaper, which is disposed in each of the wireless access switches and the wired access switches; the shaper is used to construct a queuing mechanism for federated learning traffic flows; A field server is set up on each production line on-site and is used to aggregate and distribute parameters of the federated learning model.

2. The network resource scheduling system for federated learning service flows according to claim 1, characterized in that, The wireless access switch and the wired access switch are used for networking connections between the factory layer, the field layer, and the equipment layer, including: The wireless access switch is used to establish a connection between the industrial gateway and the factory server; The wired access switch is used to create a network between the field layer and the factory layer.

3. A network resource scheduling method for federated learning service flows, applied to the central controller in the network resource scheduling system for federated learning service flows as described in any one of claims 1-2, characterized in that, The method includes: Obtain business flow information and preset federated learning models for federated learning business processes; A resource scheduling model is established based on the business flow information and the federated learning model; The step of establishing a resource scheduling model based on the business flow information and the federated learning model includes: The federated learning model is optimized based on the preset federated learning business flow parameter constraints to obtain a multi-objective resource scheduling constraint optimization model; Extract status information from the business flow information; The remaining time slot capacity is obtained based on the service flow information and the multi-objective resource scheduling constraint optimization model. Establish a resource scheduling model based on status information and remaining time slot capacity; The step of establishing a resource scheduling model based on status information and remaining time slot capacity includes: The state information is subjected to feature extraction to obtain state features; A state space is established based on the state characteristics and the remaining time slot capacity; An action space is established based on the state information and the preset time slot allocation decision; Decision analysis is performed based on the state space and the action space, and a resource scheduling model is established based on a preset network model; The business flow information is input into the resource scheduling model to obtain a resource scheduling plan.

4. The network resource scheduling method for federated learning service flows according to claim 3, characterized in that, The preset federated learning workflow parameter constraints include: The scheduling cycle, time slot length, transmission time, transmission time slot capacity, and deterministic requirements of the federated learning service flow.

5. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the network resource scheduling method for federated learning traffic flow as described in any one of claims 3-4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the network resource scheduling method for federated learning traffic flow as described in any one of claims 3-4.

Images