CN114363923B - Industrial Internet of things resource allocation method and system based on federal edge learning - Google Patents
Industrial Internet of things resource allocation method and system based on federal edge learning Download PDFInfo
- Publication number
- CN114363923B CN114363923B CN202111444856.4A CN202111444856A CN114363923B CN 114363923 B CN114363923 B CN 114363923B CN 202111444856 A CN202111444856 A CN 202111444856A CN 114363923 B CN114363923 B CN 114363923B
- Authority
- CN
- China
- Prior art keywords
- round
- devices
- federal
- local
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 71
- 238000013468 resource allocation Methods 0.000 title claims abstract description 19
- 238000005457 optimization Methods 0.000 claims abstract description 45
- 238000005265 energy consumption Methods 0.000 claims abstract description 25
- 238000004891 communication Methods 0.000 claims abstract description 24
- 230000007774 longterm Effects 0.000 claims abstract description 19
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 14
- 238000006243 chemical reaction Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 27
- 230000005540 biological transmission Effects 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 238000003860 storage Methods 0.000 claims description 12
- 238000004220 aggregation Methods 0.000 claims description 6
- 230000002776 aggregation Effects 0.000 claims description 6
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000001174 ascending effect Effects 0.000 claims description 3
- 239000013641 positive control Substances 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims 2
- 238000006467 substitution reaction Methods 0.000 claims 2
- 238000004590 computer program Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 229920002430 Fibre-reinforced plastic Polymers 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000011151 fibre-reinforced plastic Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Landscapes
- Mobile Radio Communication Systems (AREA)
Abstract
The invention provides an industrial Internet of things resource allocation method based on federal edge learning, which comprises the following steps: acquiring industrial equipment data; obtaining an allocation result by utilizing a federal edge learning wireless communication network model under the industrial Internet of things according to the acquired industrial equipment data; wherein the wireless communication network model is globally optimized by an optimization objective that minimizes the total cost of all devices; the optimization objective of minimizing the total cost of all devices is realized by a Lyapunov optimization method and an iterative algorithm. The invention adopts Lyapunov optimization theory to realize the conversion from long-term problems to short-term problems, and introduces a virtual energy queue to avoid the problem that industrial equipment cannot work beyond the energy consumption requirement.
Description
Technical Field
The invention relates to the technical field of wireless communication, in particular to an industrial Internet of things resource allocation method and system based on federal edge learning.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Currently, with the wide deployment of industrial internet of things (Industrial Internet of Things, IIOT) devices and the rapid development of Edge artificial intelligence (Edge Artificial Intelligence, edge AI) technology, the industrial internet of things technology has been widely applied to Edge networks, and large-scale industrial data is generated at edges. However, training a Machine Learning (ML) model by a conventional centralized training method requires that raw data of all devices be summarized to a central server for calculation by wireless transmission. The method can not only greatly consume the infinite bandwidth resources, but also consume more calculation resources due to the fact that data of all devices are calculated in a central server, and a large amount of delay is generated, so that huge cost is brought. Therefore, the traditional centralized training method is not practical for the actual IIOT scene.
To solve the above problem, a distributed model training framework, federal edge learning (FederatedEdgeLearning, FEEL), is proposed. The goal of FL is to enable the device to co-learn a shared machine learning (learning) model and to co-operate with the central server while keeping all training data on the device, thereby separating the ability to execute ML from the need to upload/store data in the central server. By locally updating the model parameters, the FEEL utilizes data and computing power distributed on the device, so that the energy consumption of the central server can be reduced, the model training delay is reduced, the device data privacy is protected, and the overall cost is reduced.
Although the FEEL can reduce delay and cost in the wireless network scene of IIOT, because the wireless network bandwidth resources are limited, in each training process, user selection and bandwidth allocation need to be reasonably performed; meanwhile, in the long-term FEEL process, decisions between different rounds have dependencies, and the energy of the user is limited. Therefore, in order to better train the federal learning model at the edge IIOT, the selection of the device and the bandwidth allocation need to be reasonably performed, so that the total cost of the device for a long period is minimized and the FEEL can be better applied to the IIOT under the condition that the long-term energy consumption requirement of the device is met.
Disclosure of Invention
In order to solve the problems, the invention provides the industrial Internet of things resource allocation method and the system based on the federal edge learning, and the method and the system can improve the training accuracy performance of the federal learning, so that the FEEL can be better applied in IIOT.
According to some embodiments, the present invention employs the following technical solutions:
an industrial internet of things resource allocation method based on federal edge learning, comprising:
acquiring industrial equipment data;
obtaining an allocation result by utilizing a federal edge learning wireless communication network model under the industrial Internet of things according to the acquired industrial equipment data;
wherein the wireless communication network model is globally optimized by an optimization objective that minimizes the total cost of all devices; the optimization objective of minimizing the total cost of all devices is realized by a Lyapunov optimization method and an iterative algorithm.
Further, the optimization objective of minimizing the total cost of all devices is realized by optimizing device selection and bandwidth allocation based on time delay constraint and long-term energy consumption requirements of the devices.
Further, the lyapunov optimization method is used to achieve the conversion of long-term problems to short-term problems.
Further, the optimization objective of minimizing the total cost of all devices is obtained by utilizing the wireless communication network model of federal edge learning under the industrial Internet of things, which comprises the steps of analyzing the wireless communication network model of federal edge learning under the industrial Internet of things, and constructing a corresponding model aiming at the uplink of local calculation of the devices and transmission of model parameters to an edge base station.
Further, after the acquisition of the industrial equipment data set, the training of the data set is considered to be performed by applying federal learning for a long period of time.
Further, the federal learning includes assigning global model parameters to industrial devices participating in the federal learning, the industrial devices updating the local model based on the local data.
Further, the federal learning further includes aggregating the updated local model parameters with an edge server to obtain new global model parameters.
An industrial internet of things resource allocation system based on federal edge learning, comprising:
a data acquisition module configured to acquire industrial equipment data;
the computing module is configured to obtain an allocation result by utilizing a wireless communication network model of federal edge learning under the industrial Internet of things according to the acquired industrial equipment data;
wherein the wireless communication network model is globally optimized by an optimization objective that minimizes the total cost of all devices; the optimization objective of minimizing the total cost of all devices is realized by a Lyapunov optimization method and an iterative algorithm.
A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded and executed by a processor of a terminal device to perform the method of federal edge learning based resource allocation for industrial internet of things.
A terminal device comprising a processor and a computer readable storage medium, the processor configured to implement instructions; the computer readable storage medium is for storing a plurality of instructions adapted to be loaded by a processor and to perform the one federal edge learning-based industrial internet of things resource allocation method.
Compared with the prior art, the invention has the beneficial effects that:
the invention adopts Lyapunov optimization theory to realize the conversion from long-term problems to short-term problems, and introduces a virtual energy queue to avoid the problem that industrial equipment cannot work beyond the energy consumption requirement; by utilizing an iterative algorithm to perform reasonable equipment selection and bandwidth allocation, the improvement of accuracy performance of federal learning on the edge industrial Internet of things can be realized while the cost minimization of all industrial equipment is finally realized, so that the FEEL can train a model better on IIOT.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute an undue limitation to the application.
Fig. 1 is a model diagram of a wireless network of the present embodiment 1;
fig. 2 is a flowchart of the present embodiment 1;
FIG. 3 is a flowchart of an algorithm for optimizing industrial equipment selection and bandwidth allocation of embodiment 1;
FIG. 4 is a graph comparing the accuracy effects of federal learning under a method of jointly optimizing device selection and bandwidth allocation with a method of random device selection and bandwidth allocation;
FIG. 5 is a graph of federally learned loss functions versus a method of jointly optimizing device selection and bandwidth allocation versus a method of random device selection and bandwidth allocation.
The specific embodiment is as follows:
the invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present application. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Example 1
As shown in fig. 1, an industrial internet of things resource allocation method based on federal edge learning includes:
acquiring industrial equipment data;
obtaining an allocation result by utilizing a federal edge learning wireless communication network model under the industrial Internet of things according to the acquired industrial equipment data;
wherein the wireless communication network model is globally optimized by an optimization objective that minimizes the total cost of all devices; the optimization objective of minimizing the total cost of all devices is realized by a Lyapunov optimization method and an iterative algorithm.
The optimization objective of minimizing the total cost of all the devices is realized by optimizing the device selection and the bandwidth allocation under the conditions of time delay constraint and long-term energy consumption requirements of the devices.
The Lyapunov optimization method is used for achieving the conversion from a long-term problem to a short-term problem.
The method comprises the steps of obtaining an optimization target for minimizing total cost of all equipment by utilizing a wireless communication network model of federal edge learning under the industrial Internet of things, analyzing the wireless communication network model of federal edge learning under the industrial Internet of things, and constructing a corresponding model aiming at uplink of local calculation of equipment and transmission of model parameters to an edge base station.
After the industrial equipment data set is obtained, the data set is trained by considering that federal learning is applied for a long time.
The federal learning includes assigning global model parameters to industrial devices participating in the federal learning, the industrial devices updating the local model based on the local data.
The federal learning also includes aggregating the updated local model parameters with an edge server to obtain new global model parameters.
Specifically, as shown in fig. 2, the industrial internet of things resource allocation method based on federal edge learning includes the following steps:
step S01: constructing a system model, and constructing a base facility and equipment of a wireless network based on FEEL (field-based Internet of things) for minimizing cost;
step S02: constructing a corresponding model aiming at the uplink of local calculation of equipment and transmission of model parameters to an edge base station;
step S03: problem modeling, namely under the conditions of time delay constraint and long-term energy consumption requirement of equipment, optimizing equipment selection and bandwidth allocation to achieve an optimization target of minimizing the total cost of all the equipment;
step S04: the conversion from long-term problems to short-term problems is realized by adopting the Lyapunov optimization theory, and then an iterative algorithm is adopted to perform reasonable equipment selection and bandwidth allocation.
In the step S01, consider an industrial internet of things wireless network system (as shown in fig. 1) including a plurality of industrial devices and a base station (with a server):
in an industrial internet of things wireless network, which has a Base Station (Base Station) with an edge server and a plurality of industrial devices, in order to realize better training of an ML model in the scene, long-term use of FEEL is considered for training data.
The industrial equipment set is [ N ]]= {1,2,..Wherein x is i Representing the i-th input sample, y i Represents x i Is a mark output of (2). Such data may be generated through the use of equipment and applied to various machine learning tasks such as smart manufacturing, smart logistics, and the like.
The goal of the FL training process is to find the optimal learning model parameter ω to minimize its loss function value:wherein->f n (ω，x i ，y i ) Is a sample of device n (x i ，y i ) Error value of parameter ω.
The training iteration process of FL is typically divided into three steps:
step1: the server broadcasts the t-1 (t is larger than or equal to 1) th time global model parameters to all local devices n which select participation FL:
step2: after receiving the global model parameters broadcast by the server, the device n updates the local model according to the local data until the updated model parameters are uploaded after the local accuracy xi is reached:
step3: the server aggregates the local model parameters uploaded by all the devices, calculates and updates to obtain new global model parameters:
the above process is repeated until convergence or all iterations are completed. Each iteration described above is referred to as a round.
In said step S02, a modeling analysis is performed for each of the three steps of the FEEL iteration, focusing on the device local calculation and the uplink transmission of the model parameters to the edge base station.
Because of the heterogeneous nature of the devices, the FEEL process requires a reasonable choice of devices to achieve better performance. Within each countt, K devices need to be selected to participate in the FL process. With a n (t) ∈ {0,1} indicates whether device n was selected in round t to upload model parameters. a (t) = { a 1 (t)，..，a N (t) } represents the device selection decision within the t-th round and groups the users currently selected by round as [ G ]]＝{n∈N|a n (t)＝1}。
If device n is selected at t-th round, i.e. a n (t) =1, then the global model parameters of the last round are receivedThereafter, the local model is trained based on the local data. Let the number of CPU revolutions required to process a data sample be C n Then the number of CPU revolutions required to make one local iteration is C n |D n | a. The invention relates to a method for producing a fibre-reinforced plastic composite. Defining the computational CPU frequency of device n as f n Then the time to perform one local iteration is +.>
Thereby obtaining the time delay of local calculation asThe locally calculated energy consumption isWherein (1)>Represents the number of local iterations of a round, < >>Is based on the coefficients of the CPU frequency chip architecture.
The selected device n (a n (t) =1) after the local calculation iteration is completed, its local model parameters need to be setUploading to the edge server, and at the same time, causing energy consumption and time delay in the transmission process.
For a model upload link we consider frequency division multiple access FDMA with a total bandwidth of the channel of B. b n (t)∈[0，1]The bandwidth allocation ratio of the device n in the t-th round is shown. b (t) = { b 1 (t)，...，b N (t) } represents bandwidth allocation decisions within round t. Simultaneous presence of constraints
P n For the transmission power of device n, the transmission rate
Then the transmission delay of device n at round isThe transmission energy consumption is->Wherein L is n Uploading parameters for each model +.>Data size of the model.
Thus, the delay of device n at roundt is obtained asThe energy consumption of device n at round is +.>
After all the selected devices upload updated model parameters, the edge server performs global model aggregation.
The global model parameters areWherein dt= Σ n D n a n (t) representing the number of data samples of all devices scheduled within the tth round.
Within a round, the computation cost of device n isWherein->A unit price required for one CPU revolution; the communication cost is->Wherein->The unit price for the bandwidth connecting the edge server and the device. So the cost in each round is +.>
In said step S03, with the synchronization model aggregation in the proposed framework, all relevant devices are required to initiate model training at the beginning of each round at the same time. The device selection, bandwidth allocation are jointly optimized under a given T training round conditions while guaranteeing latency requirements and long-term device energy budget to minimize the cost of all long-term devices.
Note that: a (T) =a (0),..a (T-1), B (T) =b (0),..b (T-1), a (T) = { a 1 (t)，...a n (t)，...，a N (t)}，b(t)＝{b 1 (t)，...b n (t)，...，b N (t) }. Wherein constraint (1) is a selection decision of device n; constraint (2) (3) is a requirement for the bandwidth allocation ratio of the device; constraint (4) is a time delay requirement for device n, and devices not meeting the requirement cannot participate in the FL process; constraint (5) is a long-term energy consumption limit for device n,representing the total energy budget of device n.
In the step S04, the conversion from the long-term problem to the short-term problem is achieved by adopting the lyapunov optimization theory, and then an iterative algorithm is adopted to perform reasonable device selection and bandwidth allocation.
The existing method cannot be directly solved due to the long-term optimization objective and the limiting condition of the problem P1. The Lyapunov optimization method is considered to be adopted to convert the problem into a series of short-term certainty problems for solving; meanwhile, based on the virtual queue theory, the long-term energy consumption limiting condition (5) is converted into a stable queue limit.
Q n (t) represents a queue backlog of device n's shortage of energy in countt:
wherein Q is n (1)＝0。
The original problem P1 is rewritten into a single round problem by utilizing Lyapunov optimization theory, which is defined as
Where V is a positive control parameter to adjust the trade-off between cost and energy consumption.
Substituted son and then spread into
Note that a in the above n (t) is a binary variable, and b n (t) is in the range of [0,1 ]]The continuous variable in, and therefore the formula is a mixed integer problem, is often difficult to solve. Here we divide it into two questionsThe problem, namely the equipment selection problem and the bandwidth allocation problem, are respectively solved and then iterated.
Sub-problem 1: device selection problem
Assuming that the bandwidth ratio of all devices is
Order theThe P2 problem translates into:
s.t.Constraints(1)(4)
given a bandwidth, the system would want to choose a less costly, less energy-consuming device, and thus η, according to the equation n The smaller the device, the higher the priority, the more prone to be selected. Industrial equipment selection decisions can be made based on this priority.
First, all devices are treated as eta n The values of (2) are arranged in ascending order to form a set I; if the delay of the device n does not meet the limiting condition (4), a n (t) =0; otherwise a n (t) =1. Taking η from set i n Low top K a n (t) =1 the device gets the current device selection set G.
Sub-problem 2: bandwidth allocation problem
And selecting the set G for the device, and distributing the bandwidth. The current device-based selection problem is then:
i.e.
Here, the above formula is observed, and optimization of the above formula is equivalent to
The above equation can be determined to be a convex function based on the definition of the convex function. Thus, a CVX toolkit is introduced here for solution.
After solving the two sub-problems, iterating until all rounds are finished.
The solving algorithm comprises the following steps:
the first step: initializing energy queues
And a second step of: at t=1.2..the following steps are performed in a T inner loop:
step1: initial setupAnd calculating the priority eta, and sequencing all industrial equipment according to the eta value from small to large.
Step2: judging whether all the devices meet the time delay condition (4), if not, a is not met n (t) =0; otherwise a n (t)＝1。
Step3: take eta from small to large n The first K a of (2) n (t) =1 the device gets the current device selection set G.
Step4: for the equipment selection set G, performing bandwidth allocation by using a CVX tool to obtain optimal bandwidth allocation b * (G)。
Step5: finding the currently optimal device selection set G * Obtaining optimal equipment selection decisionAnd optimal bandwidth allocation decision b * (G*)。
Step6: the energy queue is updated according to the equation.
Step7: t→t+1, step2-Step6 are repeated until the cycle ends.
Fig. 4 is a graph showing the effect of the accuracy of federal learning under the method of joint optimization device selection and bandwidth allocation (proposed_optimal) and the method of random device selection and bandwidth allocation (original_random), and fig. 5 is a graph showing the effect of the loss function of federal learning under the method of joint optimization device selection and bandwidth allocation and the method of random device selection and bandwidth allocation. By comparing the accuracy and the loss function effect graphs obtained by the two schemes, the method for jointly optimizing the equipment selection and the bandwidth allocation can be found to be higher in federal learning accuracy, smaller in loss function and better in convergence effect than the random equipment selection and the bandwidth allocation method can obtain: for the case of the round number of 20 and the user number of 50, the accuracy of federal learning finally reaches approximately 97%, the convergence trend is realized, and the loss function is very smooth; the method of adopting random device selection and bandwidth allocation does not consider the heterogeneity among devices, and each time the random device selection may cause large device difference, and the training result is poor: for the case of the round number of 20 and the user number of 50, the accuracy of federal learning reaches approximately 85% finally, and the fluctuation in the learning process is large and the effect is poor.
Example 2
An industrial internet of things resource allocation system based on federal edge learning, comprising:
a data acquisition module configured to acquire industrial equipment data;
the computing module is configured to obtain an allocation result by utilizing a wireless communication network model of federal edge learning under the industrial Internet of things according to the acquired industrial equipment data;
wherein the wireless communication network model is globally optimized by an optimization objective that minimizes the total cost of all devices; the optimization objective of minimizing the total cost of all devices is realized by a Lyapunov optimization method and an iterative algorithm.
Example 3
A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded and executed by a processor of a terminal device to provide a federal edge learning-based industrial internet of things resource allocation method according to this embodiment 1.
Example 4
A terminal device comprising a processor and a computer readable storage medium, the processor configured to implement instructions; the computer readable storage medium is configured to store a plurality of instructions adapted to be loaded and executed by a processor to provide a federal edge learning-based industrial internet of things resource allocation method according to embodiment 1.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The foregoing description is only of the preferred embodiments of the present application and is not intended to limit the same, but rather, various modifications and variations may be made by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.
Claims (10)
1. The industrial Internet of things resource allocation method based on federal edge learning is characterized by comprising the following steps of:
acquiring industrial equipment data;
obtaining an allocation result by utilizing a federal edge learning wireless communication network model under the industrial Internet of things according to the acquired industrial equipment data;
wherein the wireless communication network model is globally optimized by an optimization objective that minimizes the total cost of all devices; the optimization objective of minimizing the total cost of all equipment is realized by using a Lyapunov optimization method and an iterative algorithm, and the optimization objective is specifically as follows:
the industrial equipment set is [ N ]]= {1,2,..Wherein x is i Representing the i-th input sample, y i Represents x i Is a mark output of (2);
the optimal learning model parameter omega is found through the FL training process to minimize the loss function value:
wherein->f n (ω，x i ，y i ) Is a sample of device n (x i ，y i ) Error value of parameter omega;
the training iterative process of the FL is divided into three steps:
step1: the server broadcasts the t-1 th global model parameters to all local devices n that choose to participate in the FL:
step2: after receiving the global model parameters broadcast by the server, the device n updates the local model according to the local data until the updated model parameters are uploaded after the local accuracy xi is reached:
step3: the server aggregates the local model parameters uploaded by all the devices, calculates and updates to obtain new global model parameters:
repeating the above process until convergence or all iteration times are completed; each iteration is referred to as a round;
modeling analysis is respectively carried out on the three steps of FEEL iteration, and the local calculation of equipment and the uplink of model parameters transmitted to an edge base station are considered in an important way;
selecting K devices to participate in the FL process in each round t; with a n (t) ∈ {0,1} represents whether device n was selected in round t to upload model parameters; a (t) = { a 1 (t)，...，a N (t) } represents the device selection decision in the t-th round, the set of users for which the current round t is selected is [ G ]]＝{n∈N|a n (t)＝1}；
If device n is selected in the t-th round, i.e. a n (t) =1, then on receipt of the global of the last roundModel parametersThen, training the local model according to the local data;
obtaining the time delay of local calculation asThe locally calculated energy consumption isWherein (1)>Represents the number of local iterations of a round, < >>Is based on the coefficient of CPU frequency chip architecture;
after the selected device n completes the local calculation iteration, the selected device n locally models parametersUploading to an edge server;
considering frequency division multiple access FDMA for a model upload link, the total bandwidth of the channel is B; b n (t)∈[0，1]Representing the bandwidth allocation proportion of the device n in the t-th round; b (t) = { b 1 (t)，...，b N (t) } represents bandwidth allocation decisions within round t; simultaneous presence of constraints
P n For the transmission power of device n, the transmission rate
The transmission delay of the device n in the round t is as followsThe transmission energy consumption isWherein L is n Uploading parameters for each model +.>Data size of the model;
the time delay of the device n in the round t is obtained as followsThe energy consumption of the device n in the round t is
After all the selected devices upload updated model parameters, the edge server carries out global model aggregation;
the global model parameters areWherein D is t ＝∑ n D n a n (t) representing the number of data samples of all devices scheduled in the t-th round;
in one round, the computational cost of device n isWherein->A unit price required for one CPU revolution; the communication cost is-> A bandwidth unit price for connecting the edge server and the device; the cost in each round is +.>
Using the synchronization model aggregation in the proposed framework, all relevant devices start model training at the beginning of each round simultaneously; under the given T training round conditions, the device selection and bandwidth allocation are jointly optimized, and the time delay requirement and the device energy budget are ensured at the same time, so that the cost of all devices is minimized;
note that: a (T) =a (0),..a (T-1), B (T) =b (0),..b (T-1), a (T) = { a 1 (t)，...an(t)，...，a N (t)}，b(t)＝{b 1 (t)，...b n (t)，...，b N (t) }; wherein constraint (1) is a selection decision of device n; constraint (2) (3) is a requirement for the bandwidth allocation ratio of the device; constraint (4) is a time delay requirement for device n, and devices not meeting the requirement do not participate in the FL process; constraint (5) is an energy consumption limit for device n,representing the total energy budget of device n;
Q n (t) represents a queue backlog of device n's shortage of energy in round t:
wherein Q is n (1)＝0；
The original problem P1 is rewritten into a single round problem by utilizing Lyapunov optimization theory, which is defined as
Where V is a positive control parameter to adjust the trade-off between cost and energy consumption;
after substitution, the expansion is as follows:
wherein a is n (t) is a binary variable, and b n (t) is in the range of [0,1 ]]The continuous variable in the device is divided into two sub problems, namely, the device selection problem and the bandwidth allocation problem are respectively solved and then iterated;
sub-problem 1: device selection problem
The bandwidth ratio of all devices is equal to
Order theConverting the P2 problem into:
s.t.Constraints(1)(4)
making an industrial equipment selection decision according to the priority;
first, all devices are treated as eta n The values of (2) are arranged in ascending order to form a set I; if the delay of the device n does not meet the limiting condition (4), a n (t) =0; otherwise a n (t) =1; taking η from set I n Low top K a n (t) =1 the device gets the current device selection set G;
sub-problem 2: bandwidth allocation problem
Selecting a set G for the equipment, and distributing bandwidth; the current problems based on device selection are:
introducing a CVX tool package to solve;
after solving the two sub-problems, iterating until all rounds are finished;
the solving algorithm comprises the following steps:
the first step: initializing energy queues
And a second step of: at t=1, 2..the following steps are performed in a T inner loop:
step1: initial setupCalculating priority eta, and sequencing all industrial equipment according to the eta value from small to large;
step2: judging whether all the devices meet the time delay condition (4), if not, a is not met n (t) =0; otherwise a n (t)＝1；
Step3: take eta from small to large n The first K a of (2) n (t) =1 the device gets the current device selection set G;
step4: for the equipment selection set G, performing bandwidth allocation by using a CVX tool to obtain optimal bandwidth allocation b * (G)；
Step5: finding the currently optimal device selection set G * Obtaining optimal equipment selection decisionAnd optimal bandwidth allocation decision b * (G*)；
Step6: updating the energy queue according to the equation;
step7: t→t+1, step2-Step6 are repeated until the cycle ends.
2. The method for allocating resources of the industrial internet of things based on federal edge learning according to claim 1, wherein the optimization objective for minimizing the total cost of all devices is realized by optimizing device selection and bandwidth allocation based on time delay constraint and long-term energy consumption requirements of the devices.
3. The industrial internet of things resource allocation method based on federal edge learning according to claim 2, wherein the lyapunov optimization method is used to implement the conversion from long-term problems to short-term problems.
4. The method for allocating resources of the industrial internet of things based on federal edge learning according to claim 1, wherein the optimization objective for minimizing the total cost of all devices is obtained by using a wireless communication network model of federal edge learning under the industrial internet of things, and the method comprises analyzing the wireless communication network model of federal edge learning under the industrial internet of things, and constructing a corresponding model for uplink of local calculation of devices and transmission of model parameters to an edge base station.
5. The method for allocating resources of the industrial internet of things based on federal edge learning according to claim 1, wherein the federal edge learning is applied for training of the data set for a long period after the industrial equipment data is acquired.
6. The method for allocating resources of an industrial internet of things based on federal edge learning of claim 5, wherein federal edge learning includes assigning global model parameters to industrial devices participating in federal edge learning, the industrial devices updating a local model based on the global model parameters.
7. The method for allocating resources of an industrial internet of things based on federal edge learning of claim 6, wherein the federal edge learning further comprises aggregating the updated local model parameters with an edge server to obtain new global model parameters.
8. Industrial Internet of things resource allocation system based on federal edge learning, which is characterized by comprising:
a data acquisition module configured to acquire industrial equipment data;
the computing module is configured to obtain an allocation result by utilizing a wireless communication network model of federal edge learning under the industrial Internet of things according to the acquired industrial equipment data;
wherein the wireless communication network model is globally optimized by an optimization objective that minimizes the total cost of all devices; the optimization objective of minimizing the total cost of all equipment is realized by using a Lyapunov optimization method and an iterative algorithm, and the optimization objective is specifically as follows:
the industrial equipment set is [ N ]]= {1,2,..Wherein x is i Representing the i-th input sample, y i Represents x i Is a mark output of (2);
the optimal learning model parameter omega is found through the FL training process to minimize the loss function value:
wherein->f n (ω，x i ，y i ) Is a sample of device n (x i ，y i ) Error value of parameter omega;
the training iterative process of the FL is divided into three steps:
step1: the server broadcasts the t-1 th global model parameters to all local devices n that choose to participate in the FL:
step2: after receiving the global model parameters broadcast by the server, the device n updates the local model according to the local data until the updated model parameters are uploaded after the local accuracy xi is reached:
step3: the server aggregates the local model parameters uploaded by all the devices, calculates and updates to obtain new global model parameters:
repeating the above process until convergence or all iteration times are completed; each iteration is referred to as a round;
modeling analysis is respectively carried out on the three steps of FEEL iteration, and the local calculation of equipment and the uplink of model parameters transmitted to an edge base station are considered in an important way;
selecting K devices to participate in the FL process in each round t; with a n (t) ∈ {0,1} represents whether device n was selected in round t to upload model parameters; a (t) = { a 1 (t)，...，a N (t) } represents the device selection decision in the t-th round, the set of users for which the current round t is selected is [ G ]]＝{n∈N|a n (t)＝1}；
If device n is selected in the t-th round, i.e. a n (t) =1, then on receipt of the global model parameters of the previous roundThen, training the local model according to the local data;
obtaining the time delay of local calculation asThe locally calculated energy consumption isWherein (1)>Represents the number of local iterations of a round, < >>Is based on the coefficient of CPU frequency chip architecture;
after the selected device n completes the local calculation iteration, the selected device n locally models parametersUploading to an edge server;
uploading chains for modelsConsidering frequency division multiple access FDMA, the total bandwidth of the channel is B; b n (t)∈[0，1]Representing the bandwidth allocation proportion of the device n in the t-th round; b (t) = { b 1 (t)，...，b N (t) } represents bandwidth allocation decisions within round t; simultaneous presence of constraints
P n For the transmission power of device n, the transmission rate
The transmission delay of the device n in the round t is as followsThe transmission energy consumption isWherein L is n Uploading parameters for each model +.>Data size of the model;
the time delay of the device n in the round t is obtained as followsThe energy consumption of the device n in the round t is
After all the selected devices upload updated model parameters, the edge server carries out global model aggregation;
the global model parameters areWherein D is t ＝∑ n D n a n (t) representing the scheduled in the t-th roundThe number of data samples for all devices;
in one round, the computational cost of device n isWherein->A unit price required for one CPU revolution; the communication cost is-> A bandwidth unit price for connecting the edge server and the device; the cost in each round is +.>
Using the synchronization model aggregation in the proposed framework, all relevant devices start model training at the beginning of each round simultaneously; under the given T training round conditions, the device selection and bandwidth allocation are jointly optimized, and the time delay requirement and the device energy budget are ensured at the same time, so that the cost of all devices is minimized;
note that: a (T) =a (0),..a (T-1), B (T) =b (0),..b (T-1), a (T) = { a 1 (t)，...a n (t)，...，a N (t)}，b(t)＝{b 1 (t)，...b n (t)，...，b N (t) }; wherein constraint (1) is a selection decision of device n; constraint (2) (3) is a requirement for the bandwidth allocation ratio of the device; constraint (4) is a time delay requirement for device n, and devices not meeting the requirement do not participate in the FL process; constraint (5) is an energy consumption limit for device n,representing the total energy budget of device n;
Q n (t) represents a queue backlog of device n's shortage of energy in round t:
wherein Q is n (1)＝0；
The original problem P1 is rewritten into a single round problem by utilizing Lyapunov optimization theory, which is defined as
Where V is a positive control parameter to adjust the trade-off between cost and energy consumption;
after substitution, the expansion is as follows:
wherein a is n (t) is a binary variable, and b n (t) is in the range of [0,1 ]]The continuous variable in the system is divided into two sub-problems, namely equipment selectionThe problem and the bandwidth allocation problem are respectively solved and then iterated;
sub-problem 1: device selection problem
The bandwidth ratio of all devices is equal to
Order theConverting the P2 problem into:
s.t.Constraints(1)(4)
making an industrial equipment selection decision according to the priority;
first, all devices are treated as eta n The values of (2) are arranged in ascending order to form a set I; if the delay of the device n does not meet the limiting condition (4), a n (t) =0; otherwise a n (t) =1; taking η from set I n Low top K a n (t) =1 the device gets the current device selection set G;
sub-problem 2: bandwidth allocation problem
Selecting a set G for the equipment, and distributing bandwidth; the current problems based on device selection are:
introducing a CVX tool package to solve;
after solving the two sub-problems, iterating until all rounds are finished;
the solving algorithm comprises the following steps:
the first step: initializing energy queues
And a second step of: at t=1, 2..the following steps are performed in a T inner loop:
step1: initial setupCalculating priority eta, and sequencing all industrial equipment according to the eta value from small to large;
step2: judging whether all the devices meet the time delay condition (4), if not, a is not met n (t) =0; otherwise a n (t)＝1；
Step3: take eta from small to large n The first K a of (2) n (t) =1 the device gets the current device selection set G;
step4: for the equipment selection set G, performing bandwidth allocation by using a CVX tool to obtain optimal bandwidth allocation b * (G)；
Step5: finding the currently optimal device selection set G * Obtaining optimal equipment selection decisionAnd optimal bandwidth allocation decision b * (G*)；
Step6: updating the energy queue according to the equation;
step7: t→t+1, step2-Step6 are repeated until the cycle ends.
9. A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded and executed by a processor of a terminal device to a federal edge learning based industrial internet of things resource allocation method according to any one of claims 1-7.
10. A terminal device comprising a processor and a computer readable storage medium, the processor configured to implement instructions; a computer readable storage medium for storing a plurality of instructions adapted to be loaded by a processor and to perform a federal edge learning-based industrial internet of things resource allocation method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111444856.4A CN114363923B (en) | 2021-11-30 | 2021-11-30 | Industrial Internet of things resource allocation method and system based on federal edge learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111444856.4A CN114363923B (en) | 2021-11-30 | 2021-11-30 | Industrial Internet of things resource allocation method and system based on federal edge learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114363923A CN114363923A (en) | 2022-04-15 |
| CN114363923B true CN114363923B (en) | 2024-03-26 |
Family
ID=81097327
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111444856.4A Active CN114363923B (en) | 2021-11-30 | 2021-11-30 | Industrial Internet of things resource allocation method and system based on federal edge learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114363923B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112804107A (en) * | 2021-01-28 | 2021-05-14 | 南京邮电大学 | Layered federal learning method for energy consumption adaptive control of equipment of Internet of things |
| CN112817653A (en) * | 2021-01-22 | 2021-05-18 | 西安交通大学 | Cloud-side-based federated learning calculation unloading computing system and method |
| CN113139663A (en) * | 2021-04-23 | 2021-07-20 | 深圳市大数据研究院 | Federal edge learning configuration information acquisition method, device, equipment and medium |
| CN113194489A (en) * | 2021-04-01 | 2021-07-30 | 西安电子科技大学 | Minimum-maximum cost optimization method for effective federal learning in wireless edge network |
| CN113406974A (en) * | 2021-08-19 | 2021-09-17 | 南京航空航天大学 | Learning and resource joint optimization method for unmanned aerial vehicle cluster federal learning |
| CN113504999A (en) * | 2021-08-05 | 2021-10-15 | 重庆大学 | Scheduling and resource allocation method for high-performance hierarchical federated edge learning |
| CN113537514A (en) * | 2021-07-27 | 2021-10-22 | 北京邮电大学 | High-energy-efficiency federal learning framework based on digital twins |
-
2021
- 2021-11-30 CN CN202111444856.4A patent/CN114363923B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112817653A (en) * | 2021-01-22 | 2021-05-18 | 西安交通大学 | Cloud-side-based federated learning calculation unloading computing system and method |
| CN112804107A (en) * | 2021-01-28 | 2021-05-14 | 南京邮电大学 | Layered federal learning method for energy consumption adaptive control of equipment of Internet of things |
| CN113194489A (en) * | 2021-04-01 | 2021-07-30 | 西安电子科技大学 | Minimum-maximum cost optimization method for effective federal learning in wireless edge network |
| CN113139663A (en) * | 2021-04-23 | 2021-07-20 | 深圳市大数据研究院 | Federal edge learning configuration information acquisition method, device, equipment and medium |
| CN113537514A (en) * | 2021-07-27 | 2021-10-22 | 北京邮电大学 | High-energy-efficiency federal learning framework based on digital twins |
| CN113504999A (en) * | 2021-08-05 | 2021-10-15 | 重庆大学 | Scheduling and resource allocation method for high-performance hierarchical federated edge learning |
| CN113406974A (en) * | 2021-08-19 | 2021-09-17 | 南京航空航天大学 | Learning and resource joint optimization method for unmanned aerial vehicle cluster federal learning |
Non-Patent Citations (3)
| Title |
|---|
| Content-based Vehicle Selection and Resource Allocation for Federated Learning in IoV;Siyu Wang等;《2021 IEEE Wireless Communications and Networking Conference Workshops (WCNCW)》;20210507;全文 * |
| 基于联邦学习的工业物联网资源分配方法研究;纪秀朝;《中国优秀硕士学位论文全文数据库(电子期刊)》;20230601;全文 * |
| 智能生态网络：知识驱动的未来价值互联网基础设施;雷凯;黄硕康;方俊杰;黄济乐;谢英英;彭波;;应用科学学报;20200130(01);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114363923A (en) | 2022-04-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110968426B (en) | Edge cloud collaborative k-means clustering model optimization method based on online learning | |
| CN110366193B (en) | Slice deployment method and device for network arrangement layer bearing of smart grid | |
| CN108924198A (en) | A kind of data dispatching method based on edge calculations, apparatus and system | |
| CN114997337B (en) | Information fusion method, data communication method, information fusion device, data communication device, electronic equipment and storage medium | |
| CN113905347B (en) | Cloud edge end cooperation method for air-ground integrated power Internet of things | |
| CN113115459B (en) | Multi-scale and multi-dimensional resource allocation method for power Internet of things mass terminal | |
| CN114626306B (en) | Method and system for guaranteeing freshness of regulation and control information of park distributed energy | |
| CN109445386A (en) | A kind of most short production time dispatching method of the cloud manufacturing operation based on ONBA | |
| CN113472597A (en) | Distributed convolutional neural network fine-grained parameter transmission scheduling method and device | |
| CN115002123B (en) | System and method for rapidly adapting task offloading based on mobile edge computation | |
| CN113406974A (en) | Learning and resource joint optimization method for unmanned aerial vehicle cluster federal learning | |
| Li et al. | Data analytics for fog computing by distributed online learning with asynchronous update | |
| Sharara et al. | A recurrent neural network based approach for coordinating radio and computing resources allocation in cloud-ran | |
| CN114339891A (en) | Edge unloading resource allocation method and system based on Q learning | |
| CN114363923B (en) | Industrial Internet of things resource allocation method and system based on federal edge learning | |
| Wang et al. | Solving system-level synthesis problem by a multi-objective estimation of distribution algorithm | |
| Ding et al. | A multiagent meta-based task offloading strategy for mobile-edge computing | |
| CN113824650B (en) | Parameter transmission scheduling algorithm and system in distributed deep learning system | |
| CN113747500B (en) | High-energy-efficiency low-delay workflow application migration method based on generation of countermeasure network in complex heterogeneous mobile edge calculation | |
| CN113010296B (en) | Formalized model based task analysis and resource allocation method and system | |
| CN114942799A (en) | Workflow scheduling method based on reinforcement learning under cloud edge environment | |
| Lyu et al. | Rethinking Resource Management in Edge Learning: A Joint Pre-training and Fine-tuning Design Paradigm | |
| CN114022731A (en) | Federal learning node selection method based on DRL | |
| CN109746918B (en) | Optimization method for delay of cloud robot system based on joint optimization | |
| CN113543271A (en) | Effective capacity-oriented resource allocation method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |