CN112637883B - Federal learning method with robustness to wireless environment change in electric power Internet of things - Google Patents
Federal learning method with robustness to wireless environment change in electric power Internet of things Download PDFInfo
- Publication number
- CN112637883B CN112637883B CN202011432782.8A CN202011432782A CN112637883B CN 112637883 B CN112637883 B CN 112637883B CN 202011432782 A CN202011432782 A CN 202011432782A CN 112637883 B CN112637883 B CN 112637883B
- Authority
- CN
- China
- Prior art keywords
- model
- local
- neural network
- global
- wireless environment
- 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
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/06—Testing, supervising or monitoring using simulated traffic
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention provides a federal learning method with robustness to wireless environment change in the Internet of things of electric power, which comprises the following steps: building a neural network model, initializing weight parameters of the neural network model, and distributing a data set to UEs at a multi-user equipment end; training a neural network model; uploading the parameters of the trained neural network model to a wireless Access Point (AP), and simulating wireless environment factors to analyze convergence performance of a global FL model; analyzing convergence performance of the global FL model by simulating wireless environment factors; and selecting a client for executing federal learning by adopting a checking mechanism and a scheduling strategy at the access point AP to weaken the influence of corresponding wireless environment factors on the convergence of the global model, and having a robust federal learning method for wireless environment change. The invention uses a checking mechanism to detect the local FL model received by the access point and adopts a dispatching strategy to select the client for executing the federal learning to weaken the influence of infinite factors, so that the federal learning framework has robustness to the wireless environment change.
Description
Technical Field
The disclosure relates to the technical field of wireless communication networks, in particular to a federal learning method with robustness to wireless environment changes in the electric power internet of things.
Background
The electric power Internet of things is an application of the Internet of things in the smart grid, is a result of the development of information communication technology to a certain stage, effectively integrates communication infrastructure resources and electric power system infrastructure resources, improves the informatization level of the electric power system, improves the utilization efficiency of the existing infrastructure of the electric power system, and provides important technical support for links such as power grid generation, power transmission, power transformation, power distribution and power consumption.
Standard Machine learning (Machine learning) methods require training data to be concentrated on a Machine or data center. Due to latency, bandwidth, and privacy constraints, it is often not feasible to transfer a user-collected data set to a data center or cloud to implement a powerful machine learning solution. While there are hard-to-break barriers between data sources, the data required for artificial intelligence typically involves multiple fields. In most industries, data exists in island form, and due to the problems of industry competition, privacy safety, complex administrative procedures and the like, even if data integration is realized among different departments of the same company, important resistance is faced, and in reality, it is almost impossible or the required cost is huge to integrate data scattered in various places and various institutions.
Federal Learning (FL), which allows model training to be decoupled from the need to directly access the raw training data. Unlike centralized learning at a data center, FL typically operates in a wireless edge network where the communication medium is resource constrained and unreliable. The iterative algorithm running on FL requires a very low latency and high throughput connection between the computing units, but the AP typically needs to link a large number of UEs over a resource-constrained spectrum, thus allowing only a limited number of multi-user device-side UEs to send their trained weight parameters over unreliable channels in each round of global aggregation. Due to bandwidth limitations, only a portion of UEs can be scheduled for updating in each iteration. Due to the shared nature of the wireless medium, transmissions are subject to interference and cannot be guaranteed.
Disclosure of Invention
The present disclosure addresses the above-described problems by providing a federal learning method in the power internet of things that is robust to wireless environment changes.
In order to solve at least one of the above technical problems, the present disclosure proposes the following technical solutions:
the invention provides a federal learning method with robustness to wireless environment change in the electric power Internet of things, which comprises the following steps:
step one: building a neural network model for learning, initializing weight parameters of the neural network model, and distributing a data set to UEs at a multi-user equipment end;
step two: federal learning is carried out through the built neural network model so as to obtain a trained neural network model;
step three: uploading the parameters of the trained neural network model to a wireless Access Point (AP), and simulating wireless environment factors to analyze the convergence performance of the global FL model at the moment;
step four: analyzing the convergence performance of the global FL model at the moment by simulating the wireless environment factors;
step five: and selecting a client for executing federal learning by adopting a checking mechanism and a dispatching strategy at the wireless access point AP according to the acquired convergence performance condition to weaken the influence of corresponding wireless environment factors on the convergence of the global model, and finally obtaining the robust federal learning method for the wireless environment change in the electric power Internet of things.
In some embodiments, building the neural network model for learning in step one includes:
determining the type of neural network used;
determining the number of layers of the neural network used and the number of ganglion points of each layer;
and initializing the weight and bias parameters of all the neural networks.
In some embodiments, step two comprises:
determining parameters of a neural network model, training by using a data set based on the model, wherein the parameters of the neural network model comprise communication round times, the number of clients K, the client score C, the local training generation E, the local small lot B, the test data small lot, the learning rate lr and an optimization method;
distributing the original data set to K clients in a uniformly and uniformly distributed or non-uniformly and uniformly distributed mode;
the K clients learn by using the local training data sets until the total loss function reaches the minimum value, and upload the local parameter model of the neural network to the wireless access point AP through a wireless link;
after receiving the local model parameters from all the UEs, the wireless access point AP starts to aggregate K local parameter models to obtain a global average model, and transmits the model to all the UEs, and the UEs start to learn in a new round according to the updated local FL model, and learn in a plurality of communication rounds to obtain a trained model.
In some embodiments, in the step of distributing the original data set to the K clients in a uniformly and homogeneously distributed or non-uniformly and homogeneously distributed manner,
the uniform and uniform distribution is to divide the original data set into K equal parts directly, but the non-uniform and uniform distribution needs to sort the original data set according to the size of the tag data and then distribute the sorted original data set to K clients, i.e. the data set of each client is non-uniform distribution.
In some embodiments, uploading the parameters of the trained neural network model to the wireless access point AP in step three includes:
in the training process of the communication round, each user equipment side first uses its training data to train the local FL model w k ；
Each user equipment will w through the wireless cellular link k Transmitting to the AP;
the AP receives the local FL models of all the participating user equipment, updates the global FL model, and updates the global FL model w t+1 Transmitting the local FL model to all user equipment ends to optimize the local FL model; the global FL model is obtained according to all local FL models of all user equipment participating in learning, and can be expressed by the following formula:
wherein w is t+1 Is a global FL model, n, generated by a wireless access point AP k Representing the size of a K local data set of the user device, K representing a kth client, t representing a t communication round, w k Representing the local FL model.
In some embodiments, the convergence performance of the global FL model at this time is analyzed by simulating wireless environmental factors in step four, wherein the wireless environmental factors include bandwidth limitations, inter-zone interference and path loss,
bandwidth limitations in wireless environment factors are simulated by limiting the number of clients participating in federal learning,
the inter-zone interference and path loss in the wireless environment factors are simulated by adding errors to the local FL model parameters.
In some embodiments, the step of simulating the inter-zone interference and path loss in the wireless environment factors by adding an error to the local FL model parameters includes:
in each communication round, K (C E (0, 1)) clients are selected to participate in training;
in the K C clients, the influence of model parameter errors of the a K C clients, the b K C clients and the C K C clients on the recognition precision of the final aggregation model is tested and selected respectively; wherein a, b, c each are each independently of the other E (0, 1);
the error mode of client model parameter error comprises:
setting the model neural network layer number to be L layers, selecting all parameters of which M layers are wrong, traversing all parameters of the M layers to enable the parameters to be wrong with the probability of Q, wherein M is less than L, and setting the Q epsilon (0, 1) error as follows:
W＝W+randomuniform(-1,1)
wherein, random (-1, 1) is a random generation of a floating point number from-1 to 1, w represents the local FL model.
In some embodiments, the scheduling policy in step five comprises:
the method comprises the steps that a wireless Access Point (AP) uniformly and randomly selects a plurality of relevant multi-user equipment terminals UEs to update parameters in each communication round, and a special sub-channel is allocated to each selected UE to send training parameters.
In some embodiments, the verification mechanism is used to check for data errors in a local FL model received at the wireless access point AP;
the wireless access point AP does not update the global FL model by using the received data error local FL model;
the wireless access point AP will directly update the global FL model with the remaining correct local FL model.
The beneficial effects of the present disclosure are: the invention uses a checking mechanism to detect the local FL model received by the access point and adopts a dispatching strategy to select the client for executing the federal learning to weaken the influence of infinite factors, so that the federal learning framework has robustness to the wireless environment change.
In addition, in the technical solutions of the present disclosure, the technical solutions may be implemented by adopting conventional means in the art, which are not specifically described.
Drawings
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings that are required in the description of the embodiments or the prior art will be briefly described, and it is apparent that the drawings in the following description are some embodiments of the present disclosure, and other drawings may be obtained according to the drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic structural diagram of a federal learning method with robustness to wireless environmental changes in an electric power internet of things according to an embodiment of the present disclosure;
fig. 2 is a flowchart of a federal learning method in the disclosed power internet of things that is robust to wireless environmental changes.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more apparent, the present disclosure will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are illustrative of some, but not all embodiments of the disclosure and are not intended to limit the disclosure. All other embodiments, which can be made by one of ordinary skill in the art without inventive effort, based on the embodiments in this disclosure are intended to be within the scope of this disclosure.
It should be noted that the terms "comprises" and "comprising," along with any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or server that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
UEs in the present application are user facilities, i.e. multi-user equipment terminals, access points: the access point, FL, federated learning, refers to federal learning and the UE is herein a client, i.e., user Equipment.
Example 1:
1-2 of the accompanying drawings, an embodiment of the present application provides a federal learning method with robustness to wireless environment changes in the electric power internet of things, including the following steps:
step one: s1, building a neural network model for learning, initializing weight parameters of the neural network model, and distributing a data set to UEs at a multi-user equipment end;
step two: s2, performing federal learning through the built neural network model to obtain a trained neural network model;
step three: s3, uploading parameters of the trained neural network model to a wireless Access Point (AP), and simulating wireless environment factors to analyze convergence performance of the global FL model at the moment;
step four: s4, analyzing the convergence performance of the global FL model at the moment by simulating the wireless environment factors;
step five: s5, selecting a client for executing federal learning at the wireless access point AP by adopting a checking mechanism and a scheduling strategy according to the acquired convergence performance condition to weaken the influence of corresponding wireless environment factors on the global model convergence, and finally obtaining the federal learning method with robustness to wireless environment change in the electric power Internet of things.
As shown in fig. 1, (a) in fig. 1 shows that each UE calculates a single update based on its locally stored data, and (B) in fig. 1 the access point AP aggregates the updates received from UEs to build a new global model, and (C) in fig. 1 sends the new model back to UEs, and the process is repeated.
The building of the neural network model for learning in the first step comprises the following steps:
determining the type of neural network used; wherein, the type of the neural network can be MLP or CNN;
determining the number of layers of the neural network used and the number of ganglion points of each layer;
and initializing the weight and bias parameters of all the neural networks.
Specifically, the second step includes:
determining parameters of a neural network model, namely selecting a general model, training by using a data set based on the model, and specifically determining the number of communication rounds, the number K of clients, the number C of clients, the local training generation E, the local small lot B, the small lot of test data, the learning rate lr and the optimization method; the number of communication rounds is also called a communication round, which refers to one training and updating exchange between an AP and related UEs, the number K of clients refers to the number of UEs associated with the AP, and the fraction C of clients refers to the number of UEs which are involved in federal learning to the number of clients;
distributing the original data set to K clients in a uniformly and uniformly distributed or non-uniformly and uniformly distributed mode;
the K clients learn by using the local training data sets until the total loss function reaches the minimum value, and upload the local parameter model of the neural network to the wireless access point AP through a wireless link;
after receiving the local model parameters from all the UEs, the wireless access point AP starts to aggregate K local parameter models to obtain a global average model, and transmits the model to all the UEs, and the UEs start to learn in a new round according to the updated local FL model, and learn in a plurality of communication rounds to obtain a trained model.
Illustrating: a cellular network comprising an AP and a set of associated K multi-user equipment-side UEs cooperatively performing FL algorithm for data analysis and inference, e.g., wireless access point assuming the AP has within its Thiessen polygonK associated multi-user UEs are evenly distributed, and in this network a fixed number of frequency spectrum is divided equally into N radio access channels, where N < K. For a generic user equipment k, it is considered to be equipped with a base station having n K A local data set of sample points. For the Mnist data set, the data set is disordered, 60000/K label data and 10000/K test data are sequentially distributed to a client, the client score is C, namely, the number of user equipment participating in federal learning is m=KxC, the local training generation is local_ep, the local small lot is local_bs, the learning rate is lr, and the communication round is epochs. The task is to classify a handwritten word using the MNIST dataset or to classify a color image using the CIFAR-10 dataset.
The neural network model comprises two types of MLP and CNN, wherein the MLP model comprises an input layer, a hidden layer and an output layer, the number of the hidden layers and the number of nodes are selected to be proper according to the requirement, the number of the nodes of the input layer and the number of the nodes of the output layer are determined according to a specific data set and classification conditions, but for the two classification problems, the number of the nodes of the output layer can be 10. The CNN model contains a build-up of volumes activated using the Relu function, fully connected layers activated using the Relu function, and one softmax output layer. The number of convolution layers and the number of channels, the number of pooling layers, the number of channels and the pooling method used select proper values according to the requirements.
Specifically, the step of minimizing the overall loss function includes: at each AP, the goal is to learn a statistical model from the data present on the K relevant multi-user device-side UEs, i.e., the AP needs to fit a vector wεR d In order to minimize a specific loss function by using the entire data set of all multi-user equipment-side UEs under its service. Formally, such tasks can be expressed as:
the formula is a multi-client minimum loss function, is a loss function of an AP (access point) end, aims to minimize the sum of loss functions of all user equipment ends participating in learning, and can be used as a termination condition of local model training.
Wherein n is k ＝|D K I indicates the size of the user device k local data set,
represents the size, w, of the entire dataset t Is composed of A P The global FL model generated, f (w i ) Is a loss function, different loss functions can be defined for different FL algorithms;
k represents the number of UEs, i.e. the number of UEs associated with the AP;
in particular, the constraint is to ensure that once the FL algorithm converges, all user devices and APs will share the same FL model for their learning task, i.e., the downloading of the global FL model to all clients;
the purpose of this formulation is to minimize the sum of the loss functions of all clients participating in the learning, which can be used as termination conditions for the local model training.
In the FL algorithm, a local FL model w for each user k k Is dependent on the global model w t While the global model w t Is dependent on the local FL models of all users. Local FL model w k Is dependent on the learning algorithm. Global model w t The update of (2) is given by:
in the training process, each user first trains the local FL model w using its training data k W is then transmitted over the wireless cellular link k To the AP. Once the AP receives the local FL models for all participating users, it will update the global FL model based on the above equation and will update the global FL model w t+1 To all users to optimize the local FL model. Over time, APs and multi-user equipment end UEsTheir optimal FL models can be found and used to minimize the loss function.
Preferably, in the step of distributing the original data set to the K clients in a uniformly and uniformly distributed or non-uniformly and uniformly distributed manner,
the uniform and uniform distribution is to divide the original data set into K equal parts directly, but the non-uniform and uniform distribution needs to sort the original data set according to the size of the tag data and then distribute the sorted original data set to K clients, i.e. the data set of each client is non-uniform distribution.
Preferably, uploading the parameters of the trained neural network model to the wireless access point AP in step three includes:
in the training process of the communication round, each user equipment side first uses its training data to train the local FL model w k ；
Each user equipment will w through the wireless cellular link k Transmitting to the AP;
the AP receives the local FL models of all the participating user equipment, updates the global FL model, and updates the global FL model w t+1 Transmitting the local FL model to all user equipment ends to optimize the local FL model; the global FL model is obtained according to all local FL models of all user equipment participating in learning, and can be expressed by the following formula:
wherein w is t+1 Is a global FL model, n, generated by a wireless access point AP k Representing the size of a K local data set of the user device, K representing a kth client, t representing a t communication round, w k Representing the local FL model.
Preferably, the convergence performance of the global FL model at this time is analyzed by simulating wireless environment factors in step four, wherein the wireless environment factors include bandwidth limitation, section interference and path loss,
simulating a bandwidth limitation in a wireless environment factor by limiting the number of clients participating in federal learning;
the inter-zone interference and path loss in the wireless environment factors are simulated by adding errors to the local FL model parameters.
In particular, for bandwidth limitations, the wireless access point AP should only select a subset of the multi-user equipment UEs and update its parameters at the same time to keep the communication time within an acceptable range. For this reason, the scheduling policy plays a key role in allocating the resource-limited radio channels to the appropriate multi-user equipments UEs. The invention adopts random scheduling strategy, and allocates wireless channels with limited resources to proper multi-user equipment terminals UEs, including in each communication round, the wireless access point AP uniformly and randomly selects a plurality of related multi-user equipment terminals UEs to update parameters, and allocates a dedicated sub-channel to each selected UE to transmit training parameters.
Wherein the step of simulating the inter-zone interference and the path loss in the wireless environment factor by adding an error to the local FL model parameter comprises:
in each communication round, K (C E (0, 1)) clients are selected to participate in training;
in the K C clients, the influence of model parameter errors of the a K C clients, the b K C clients and the C K C clients on the recognition precision of the final aggregation model is tested and selected respectively; wherein a, b, c each are each independently of the other E (0, 1); specifically, the ratio is user_error_rate_list= [ a, b, c ]; a, b, c E (0, 1) and are different from each other
The error mode of client model parameter error comprises:
setting the model neural network layer number to be L layers, selecting all parameters in which M layers are wrong, traversing all the M layers, and enabling each parameter to have Q probability error, wherein M < L, Q E (0, 1), specifically setting as layer_error=M/L, M < L, value_error_prob=Q, and setting as Q E (0, 1) error:
W＝W+randomuniform(-1,1)
wherein, random (-1, 1) is a random generation of a floating point number from-1 to 1, w represents the local FL model.
Specifically, for interval interference, path loss, and instability, a verification mechanism is used to check for data errors in the local FL model received at the wireless access point AP; assume each local FL model w k Will be sent as a single packet in the uplink, where C (w k ) =0 indicates that the local FL model received by the AP contains data errors; otherwise, we have C (w k )＝1。
In the system, as long as the received local FL model contains errors, the wireless access point AP will not update the global FL model with the received data erroneous local FL model;
the wireless access point AP will directly update the global FL model with the remaining correct local FL model.
When C (w) k ) When =0, i.e. the local FL model parameters contain errors, the AP will discard the local FL model containing errors, update the global model with the remaining correct local FL model, otherwise C (w k ) The ap updates the global FL model with the integrated local FL model=1, and performs training for the next communication round. Thus, the global model update formula can be given by:
this formula simulates the process of the verification mechanism at the AP point discarding the erroneous packet, i.e., the global average of the local FL model with the correct parameters. K represents K clients, i.e. also user equipment in the present application.
Note that there is a constant alternation between communication and computation during the training phase. In this regard, retransmission of the corrupted packet may be of no benefit because each uplink transmission of the local update will be followed by a globally averaged downlink transmission, and after receipt of this transmission the multi-user device end UEs will refresh its reference parameters and begin to use the local data to solve the new sub-problem. Therefore, if retransmission is requested when there is an error, random queuing delay and retransmission delay may be increased, so that learning delay is increased, and convergence performance of FL is further affected.
The principle of the invention is as follows:
the FL system in the present invention optimizes the global model by repeating the following process: (1) the multi-user equipment-side UEs perform local calculations using their local data to minimize a predefined empirical risk function, i.e., a loss function, and update the weights of the training to the AP, (2) the AP gathers updates from the multi-user equipment-side UEs and consults the FL unit to generate an improved global model, (3) re-distributes the output of the FL model to the multi-user equipment-side UEs, which use the global model as a reference for further local training.
In the local FL model uploading stage, the bandwidth, interval interference, path loss and instability of wireless environment factors are simulated, and finally the global aggregation model is influenced, so that the global model is distributed to the UEs of the multi-user equipment end, and the local learning of the UEs of the multi-user equipment end is influenced. By adopting a scheduling strategy and performing a checking mechanism at the AP, the constructed federal learning framework has robustness to wireless environment changes.
As shown in the figure, according to the simulation result, the loss function value in the training process is continuously reduced along with the increase of the communication rounds, the recognition accuracy obtained by testing the overall aggregated model parameters is continuously increased, and the result can be seen to have certain fluctuation. This is a signal that federal learning is normally ongoing, indicating that the global FL model derived from the local FL model is gradually fitting data, i.e., that the federal learning framework is learning, through uninterrupted parameter updates between the AP and the multi-user equipment UEs.
As shown in fig. 2, after the influence of the wireless factor is simulated, in the communication round which is just started, the difference between the recognition accuracy obtained by testing the globally aggregated model by the loss function value in the training process under the condition of different packet error rates is very small, and even better than the general federal learning algorithm, because the tiny change is equivalent to playing a role of adjusting the parameters of the weight matrix, so that the model has better performance. With the increase of communication rounds, the performance of federal learning with larger error rate gradually worsens or even fails, and the more the error parameters are, the larger the curve fluctuation amplitude is, and the larger the influence on the aggregation model performance is.
Specifically, a verification mechanism is adopted to discard parameter packets with errors and random scheduling strategies in the model, the performance of the overall aggregated model is greatly improved, and the loss function value in the training process and the recognition precision obtained by testing the overall aggregated model are converged according to a certain trend.
The greater the error rate, the greater the number of clients containing errors, so the more local model parameter updates are discarded by the checking mechanism, i.e. the fewer the local models participating in the global aggregation, at this time, no retransmission delay or queuing delay is increased. Therefore, the larger the error rate, the greater the impact on the model convergence performance, i.e. the more communication rounds are required to achieve the same recognition accuracy than the general federal learning algorithm, and secondly, it can be observed that the larger the error rate, the larger the fluctuation of the three curves thereof is, which is caused by the instability of the global FL model due to the fewer number of local models participating in the global aggregation. The more these results and participating users, the better the learning effect tends to be.
Thus, the present invention works by first constructing a federal learning method by K clients; on the basis, the limitation and sharing property of the wireless medium are simulated by limiting the number of clients participating in federal learning and adding a packet error in the update and uploading stage of a local federal learning model; and then, selecting a client for learning by adopting a dispatching strategy, discarding the parameter model with errors at an access point through a checking mechanism, and the like to obtain a federal learning framework with robustness to wireless environment changes in the electric power Internet of things. The invention solves the problem that in the wireless edge network, the Federal learning performance is rapidly deteriorated due to the resource limitation of the communication medium and the sharing property of the wireless medium, including the parameter update loss from the user equipment due to the resource limitation of the communication medium; the sharing property of the wireless medium causes that communication receives interval interference, average path loss of a communication link, large-scale fading and the like, and the parameter updating encounters unavoidable faults and the like.
It should be noted that: the embodiment sequence of the present disclosure is only for description, and does not represent the advantages and disadvantages of the embodiments. And the foregoing description has described specific embodiments of this specification. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for embodiments of the apparatus, device and storage medium, the description is relatively simple as it is substantially similar to the method embodiments, with reference to the description of portions of the method embodiments being relevant.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (5)
1. The federal learning method with robustness to wireless environment change in the electric power Internet of things is characterized by comprising the following steps of:
step one: building a neural network model for learning, initializing weight parameters of the neural network model, and distributing a data set to UEs at a multi-user equipment end;
step two: federal learning is carried out through the built neural network model so as to obtain a trained neural network model;
step three: uploading the parameters of the trained neural network model to a wireless Access Point (AP);
step four: analyzing the convergence performance of the global FL model at the moment by simulating the wireless environment factors;
step five: selecting a client for executing federal learning by adopting a checking mechanism and a scheduling strategy at a wireless Access Point (AP) according to the acquired convergence performance condition to weaken the influence of corresponding wireless environment factors on the convergence of a global model, and finally obtaining a robust federal learning method for wireless environment change in the electric power Internet of things; the scheduling policy refers to allocating a wireless channel with limited resources to a proper multi-user equipment UEs, including that in each communication round, the wireless access point AP uniformly and randomly selects a plurality of related multi-user equipment UEs to update parameters, and allocates a dedicated sub-channel to each selected UEs to transmit training parameters; the checking mechanism is used to check data errors in the local FL model received at the wireless access point AP; the wireless access point AP does not update the global FL model by using the received local FL model with data errors; the wireless access point AP will directly update the global FL model with the remaining correct local FL model;
step three, uploading the parameters of the trained neural network model to the wireless access point AP includes:
in the training process of the communication round, each user equipment side first uses its training data to train the local FL model w k ；
Each user equipment end transmits w through a wireless cellular link k Transmitting to the AP;
the AP receives the local FL models of all the participating user equipment, updates the global FL model, and updates the global FL model w t+1 Transmitting the local FL model to all user equipment ends to optimize the local FL model; the global FL model is obtained according to all local FL models of all user equipment terminals participating in learning, and can be represented by the following formula:
wherein the method comprises the steps of，w t+1 Is a global FL model, n, generated by a wireless access point AP k Representing the size of a K local data set of the user equipment, K representing a K-th client, t representing a t-th communication round, w k Representing a local FL model;
analyzing convergence performance of the global FL model at the time by simulating wireless environment factors including bandwidth limitation, inter-zone interference and path loss,
bandwidth limitations in wireless environment factors are simulated by limiting the number of clients participating in federal learning, and inter-zone interference and path loss in wireless environment factors are simulated by adding errors to the local FL model parameters.
2. The federal learning method with robustness to wireless environment variation in the power internet of things according to claim 1, wherein building the neural network model for learning in step one comprises:
determining the type of neural network used;
determining the number of layers of the neural network used and the number of ganglion points of each layer;
and initializing the weight and bias parameters of all the neural networks.
3. The method for federal learning in power internet of things having robustness to wireless environmental changes of claim 1,
the second step comprises:
determining parameters of a neural network model, training by using a data set based on the model, wherein the parameters of the neural network model comprise the number of communication rounds, the number of clients K, the client score C, the local training generation E, the local small lot B, the test data small lot, the learning rate lr and an optimization method;
distributing the original data set to K clients in a uniformly and uniformly distributed or non-uniformly and uniformly distributed mode;
the K clients learn by using the local training data sets until the overall loss function reaches the minimum value, and upload the local parameter model of the neural network to the wireless access point AP through a wireless link;
after receiving the local model parameters from all the UEs, the wireless access point AP starts to aggregate K local parameter models to obtain a global average model, and transmits the model to all the UEs, and the UEs start to learn in a new round according to the updated local FL model, and learn in a plurality of communication rounds to obtain a trained model.
4. The method for federal learning in power internet of things having robustness to wireless environment changes according to claim 3, wherein in the step of distributing the original data set to K clients in a uniformly co-distributed or non-uniformly co-distributed manner,
the uniform and uniform distribution is to divide the original data set into K equal parts directly, but the non-uniform and uniform distribution needs to sort the original data set according to the size of the tag data and then distribute the sorted original data set to K clients, i.e. the data set of each client is non-uniform distribution.
5. The federal learning method for robustness to wireless environment variation in power internet of things according to claim 1, wherein in the step of simulating inter-zone interference and path loss in a wireless environment factor by adding an error to a local FL model parameter, comprising:
in each communication round, K clients are selected from K clients to participate in training;
in the K C clients, the influence of model parameter errors of the a K C clients, the b K C clients and the C K C clients on the recognition precision of the final aggregation model is tested and selected respectively; wherein a, b, c each are each independently of the other E (0, 1);
the error mode of client model parameter error comprises:
setting the model neural network layer number to be L layers, selecting M layers to be wrong, traversing all parameters of the M layers to be wrong, and enabling each parameter to be wrong with the probability of Q, wherein M is smaller than L, and setting the Q epsilon (0, 1) to be wrong is as follows:
W＝W+randomuniform(-1,1)
wherein, random form (-1, 1) is a random generation of a floating point number of-1 to 1, and w represents a local FL model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011432782.8A CN112637883B (en) | 2020-12-09 | 2020-12-09 | Federal learning method with robustness to wireless environment change in electric power Internet of things |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011432782.8A CN112637883B (en) | 2020-12-09 | 2020-12-09 | Federal learning method with robustness to wireless environment change in electric power Internet of things |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112637883A CN112637883A (en) | 2021-04-09 |
| CN112637883B true CN112637883B (en) | 2023-04-28 |
Family
ID=75309394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011432782.8A Active CN112637883B (en) | 2020-12-09 | 2020-12-09 | Federal learning method with robustness to wireless environment change in electric power Internet of things |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112637883B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111654027B (en) * | 2020-06-15 | 2022-04-29 | 江苏方天电力技术有限公司 | Power distribution Internet of things intelligent decision method based on reinforcement learning |
| CN113762533B (en) * | 2021-09-07 | 2023-06-06 | 北京航空航天大学 | Federal learning method with robustness to data pollution in wireless edge network |
| WO2023096214A1 (en) * | 2021-11-25 | 2023-06-01 | 엘지전자 주식회사 | Method for performing federated learning in wireless communication system, and apparatus therefor |
| CN114554459B (en) * | 2022-01-19 | 2023-01-06 | 苏州大学 | Internet of vehicles federal learning client selection method assisted by near-end strategy optimization |
| CN117135693B (en) * | 2023-10-27 | 2024-01-23 | 四川长虹新网科技有限责任公司 | Real-time service distribution method based on federal learning under multi-AP environment |
| CN117692940B (en) * | 2024-02-04 | 2024-04-26 | 北京大业桥科技股份有限公司 | Microwave system performance detection method based on microwave link |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110493826A (en) * | 2019-08-28 | 2019-11-22 | 重庆邮电大学 | A kind of isomery cloud radio access network resources distribution method based on deeply study |
| CN111190487A (en) * | 2019-12-30 | 2020-05-22 | 中国科学院计算技术研究所 | Method for establishing data analysis model |
| CN111754000A (en) * | 2020-06-24 | 2020-10-09 | 清华大学 | Quality-aware edge intelligent federal learning method and system |
| CN111860581A (en) * | 2020-06-11 | 2020-10-30 | 南京邮电大学 | Federal learning training method based on model dispersion |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11481617B2 (en) * | 2019-01-22 | 2022-10-25 | Adobe Inc. | Generating trained neural networks with increased robustness against adversarial attacks |
-
2020
- 2020-12-09 CN CN202011432782.8A patent/CN112637883B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110493826A (en) * | 2019-08-28 | 2019-11-22 | 重庆邮电大学 | A kind of isomery cloud radio access network resources distribution method based on deeply study |
| CN111190487A (en) * | 2019-12-30 | 2020-05-22 | 中国科学院计算技术研究所 | Method for establishing data analysis model |
| CN111860581A (en) * | 2020-06-11 | 2020-10-30 | 南京邮电大学 | Federal learning training method based on model dispersion |
| CN111754000A (en) * | 2020-06-24 | 2020-10-09 | 清华大学 | Quality-aware edge intelligent federal learning method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112637883A (en) | 2021-04-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112637883B (en) | Federal learning method with robustness to wireless environment change in electric power Internet of things | |
| Qin et al. | Federated learning and wireless communications | |
| Drainakis et al. | Federated vs. centralized machine learning under privacy-elastic users: A comparative analysis | |
| Zhang et al. | Double coded caching in ultra dense networks: Caching and multicast scheduling via deep reinforcement learning | |
| CN111629380B (en) | Dynamic resource allocation method for high concurrency multi-service industrial 5G network | |
| Yoshida et al. | MAB-based client selection for federated learning with uncertain resources in mobile networks | |
| Xie et al. | Adaptive online decision method for initial congestion window in 5G mobile edge computing using deep reinforcement learning | |
| CN113141317B (en) | Streaming media server load balancing method, system, computer equipment and terminal | |
| Cui et al. | Optimal rate adaption in federated learning with compressed communications | |
| Yang et al. | Learning the interference graph of a wireless network | |
| Taïk et al. | Data-quality based scheduling for federated edge learning | |
| Balakrishnan et al. | Deep reinforcement learning based traffic-and channel-aware OFDMA resource allocation | |
| Liu et al. | Time minimization in hierarchical federated learning | |
| Pase et al. | On the convergence time of federated learning over wireless networks under imperfect CSI | |
| CN116389270A (en) | DRL (dynamic random link) joint optimization client selection and bandwidth allocation based method in federal learning | |
| Imani et al. | iSample: Intelligent client sampling in federated learning | |
| Chen et al. | Optimizing communication and computational resource allocations in network slicing using twin-GAN-Based DRL for 5G hybrid C-RAN | |
| Liu et al. | Federated and meta learning over non-wireless and wireless networks: A tutorial | |
| Zheng et al. | Federated learning in heterogeneous networks with unreliable communication | |
| Ma et al. | Paving the way for distributed artificial intelligence over the air | |
| Peng et al. | Data-driven spectrum partition for multiplexing URLLC and eMBB | |
| CN116542319A (en) | Self-adaptive federation learning method and system based on digital twin in edge computing environment | |
| CN115695429A (en) | Non-IID scene-oriented federal learning client selection method | |
| Srinivasan et al. | Maximizing throughput in MIMO networks with variable rate streams | |
| Chen et al. | Enhanced hybrid hierarchical federated edge learning over heterogeneous networks |
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 |