CN113805142B - Building floor indoor positioning method based on federal learning - Google Patents
Building floor indoor positioning method based on federal learning Download PDFInfo
- Publication number
- CN113805142B CN113805142B CN202111088881.3A CN202111088881A CN113805142B CN 113805142 B CN113805142 B CN 113805142B CN 202111088881 A CN202111088881 A CN 202111088881A CN 113805142 B CN113805142 B CN 113805142B
- Authority
- CN
- China
- Prior art keywords
- model
- client
- server
- local
- global
- 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 87
- 238000012549 training Methods 0.000 claims abstract description 105
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 67
- 238000013135 deep learning Methods 0.000 claims abstract description 13
- 230000008569 process Effects 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 7
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000006854 communication Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/0252—Radio frequency fingerprinting
- G01S5/02521—Radio frequency fingerprinting using a radio-map
- G01S5/02524—Creating or updating the radio-map
- G01S5/02525—Gathering the radio frequency fingerprints
- G01S5/02526—Gathering the radio frequency fingerprints using non-dedicated equipment, e.g. user equipment or crowd-sourcing
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Remote Sensing (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Radar, Positioning & Navigation (AREA)
- Collating Specific Patterns (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention provides a building floor indoor positioning method based on federal learning, which utilizes a distributed deep learning technology jointly participated by an edge server and a plurality of mobile clients to construct a radio frequency fingerprint positioning model. The server firstly initializes the model and performs centralized pre-training by using a small amount of fingerprint data, each client uses the local fingerprint data to perform further model training, then the trained local model is transmitted to the server, and the server aggregates the local models collected from each client to obtain a global model for radio frequency fingerprint positioning. The invention adopts the federal learning training method with pre-training and combines the radio frequency fingerprint positioning method based on the convolutional neural network to distribute the data acquisition and model training tasks to each client, so that the model training cost is dispersed, the data acquisition and storage cost is reduced, the position data privacy of the client is protected, and the training efficiency and the positioning effect are ensured.
Description
Technical Field
The invention relates to the technical field of thermoelectric gas turbines, in particular to a building floor indoor positioning method based on federal learning.
Background
Indoor positioning needs are increasing with the popularity of mobile and internet of things devices. The indoor positioning method can be roughly divided into: a radio frequency Fingerprint based method (finger of Arrival, AOA), a Time of Arrival (TOA), a Time difference of Arrival (Time Difference of Arrival, TDoA), a round trip Time based method (Return Time of Flight, RToF) and a Phase of Arrival based method (PoA). The wireless radio frequency signals available for implementing these methods include: wiFi signals, bluetooth signals, UWB signals, zigBee signals, RFID signals, photo-electric signals, magnetic signals, etc.
Among these several types of methods, some require additional complex hardware support, and some require strict time synchronization, thereby greatly increasing indoor positioning costs. The indoor positioning method based on the radio frequency fingerprint firstly requires collecting sufficient radio frequency fingerprint data with position labels (such as WiFi received signal strength groups) in an offline construction stage and constructing a fingerprint map database or training a deep learning positioning model according to the radio frequency fingerprint data; and in the online use stage, the user is required to submit the current radio frequency fingerprint sample for comparison of the input fingerprint database or prediction of the input positioning model, so that the position of the user is obtained. In recent years, with the advent of the artificial intelligence era, deep learning technology has been dominant in the field of indoor positioning of radio frequency fingerprints.
However, achieving accurate indoor positioning in a dynamically varying large indoor environment using conventional centralized deep learning techniques remains a challenge, mainly summarized as the following problems:
(1) Model training overhead big problem: indoor environments tend to be dynamically changeable, such as furniture movement, personnel walking, signal source increase and decrease, and the like, which can cause significant changes in path attenuation, environment reflection, multipath propagation, and the like of wireless radio frequency signals, so that original fingerprint training data and an existing positioning model can not accurately describe a relevant area. In order to cope with the indoor environment change, the deep learning model is required to be repeatedly trained frequently, so that a large model training overhead is generated.
(2) High data acquisition cost: because the indoor environment is dynamically changeable, maintaining high accuracy of indoor positioning has to require frequent updating of the radio frequency fingerprint data for model training calibration, which can result in large data acquisition costs. Especially for large three-dimensional indoor environments, such as multi-building multi-floor complexes of campuses, factories and the like, the data acquisition cost generated by frequently and largely updating fingerprint training data is often too high. In addition to the labor cost, the communication cost of continuously uploading and summarizing a large amount of raw fingerprint data updates to the training server is often high.
(3) High data storage cost and privacy cost: the mobile crowd sensing method proposed in recent years can effectively reduce the data acquisition cost in a larger indoor environment, but a client participating in sensing copies locally acquired radio frequency fingerprint data to a training server to generate certain data storage cost at the server and the client at the same time. In addition, the radio frequency fingerprint data with the position tag contains the position privacy information of the collector, so that the corresponding larger data privacy cost can reduce the willingness of the client to share the local fingerprint data with the server.
Disclosure of Invention
The embodiment of the invention provides a building floor indoor positioning method based on federal learning, which is used for solving the problems in the prior art.
In order to achieve the above purpose, the present invention adopts the following technical scheme.
A federal learning-based building floor indoor positioning method, comprising:
s1, a server establishes a server local radio frequency fingerprint data set with a self-position tag;
s2, the server establishes a deep learning network model based on the local radio frequency fingerprint data set of the server; the deep learning network model comprises a coding and decoding model and a convolutional neural network model;
s3, the server performs concentrated pre-training on the coding and decoding models based on the server local radio frequency fingerprint data set, and sends the concentrated pre-trained coding and decoding models to each client, so that each client performs federal training on the received concentrated pre-trained coding and decoding models simultaneously based on the self-built client local radio frequency fingerprint data set with the client self-position label to obtain a first local model;
s4, the server receives the first local models sent by each client, and aggregates the plurality of first local models to obtain a global model of the coding and decoding model;
s5, the server sends the global model of the coding and decoding model to each client so that each client carries out federal training on the global model of the coding and decoding model received by each client simultaneously based on a self-built client local radio frequency fingerprint data set with the self-location tag of the client to obtain an updated first local model, and sends the updated first local model to the server; the server aggregates the received updated first local model to obtain a global model of the updated coding and decoding model; repeating the process of the step until the global model of the coding and decoding model converges;
s6, the server performs concentrated pre-training on the convolutional neural network model based on the server local radio frequency fingerprint data set, and sends the concentrated pre-trained convolutional neural network model to each client, so that each client performs federal training on the concentrated pre-trained convolutional neural network model sent by the server based on the self-built client local radio frequency fingerprint data set with the client self-location tag to obtain a second local model;
s7, the server receives the second local models sent by each client, and aggregates the plurality of second local models to obtain a global model of the convolutional neural network model;
s8, the server sends the global model of the convolutional neural network model to each client so that each client carries out federal training on the global model of the convolutional neural network model received by each client based on a self-built client local radio frequency fingerprint data set with the own position label of the client to obtain an updated second local model, and sends the updated second local model to the server; the server aggregates the received updated second local model to obtain a global model of the updated convolutional neural network model; repeating the process of the step until the global model of the convolutional neural network model converges;
the global model of the converged codec model and the global model of the convolutional neural network model are used to predict the building number and floor number where the user is located.
Preferably, the server local radio frequency fingerprint data set comprises centralized pre-training data for use by the server; the client local rf fingerprint data set includes local training data that each client uses alone.
Preferably, the step S1 further includes preprocessing data in the server local rf fingerprint data set, and specifically includes:
the server obtains a minimum RSS value of the received local data of the radio frequency fingerprint;
the server obtains a global minimum RSS value based on the minimum RSS value of the received local data of the radio frequency fingerprint;
server through type
Performing power processing on the RSS value of the radio frequency fingerprint data set; wherein Min is a global minimum RSS value; alpha >1 is a constant, such as the natural logarithm e.
Preferably, the learning rate is 0.001 in the process of centralized pre-training and federal training of the codec model and the convolutional neural network model.
Preferably, the convolutional neural network model has a one-dimensional convolutional kernel architecture, the convolutional kernel architecture comprises three convolutional operations, the first convolutional operation has a convolutional kernel size of 22, 99 convolutional kernels in total, the second convolutional operation has a convolutional kernel size of 22, 66 convolutional kernels in total, and the third convolutional operation has a convolutional kernel size of 22, and 33 convolutional kernels in total; the step size of each convolution operation is 1.
Preferably, the method further comprises:
and S7, the server sends the global model of the converged codec model and the global model of the convolutional neural network model to one or more clients, so that the one or more clients can predict building numbers and floor numbers through the global model of the converged codec model and the global model of the convolutional neural network model.
In a second aspect, the present invention provides a federally learned building floor indoor positioning method, comprising:
e1, each client establishes a client local radio frequency fingerprint data set with a self position label;
each client receives the centralized pre-trained coding and decoding model sent by the server, and each client performs federal training on the centralized pre-trained coding and decoding model received by each client based on the local radio frequency fingerprint data set of each client to obtain a first local model and sends the first local model to the server, so that the server can aggregate the first local model sent by each client to obtain a global model of the coding and decoding model;
each client receives the global model of the coding and decoding model sent by the server, and each client carries out federal training on the global model of the coding and decoding model received by each client based on the local radio frequency fingerprint data set of each client to obtain an updated first local model and sends the updated first local model to the server, so that the server can aggregate the updated first local model sent by each client to obtain the global model of the updated coding and decoding model; repeating the process of the step until the global model of the coding and decoding model converges;
each client receives the concentrated pre-trained convolutional neural network model sent by the server, each client performs federal training on the concentrated pre-trained convolutional neural network model received by each client based on the local radio frequency fingerprint data set of each client to obtain a second local model, and sends the second local model to the server, so that the server can aggregate the second local model sent by each client to obtain a global model of the convolutional neural network model;
each client receives the global model of the convolutional neural network model sent by the server, and each client carries out federal training on the global model of the convolutional neural network model received by each client based on the local radio frequency fingerprint data set of each client to obtain an updated second local model and sends the updated second local model to the server, so that the server can aggregate the updated second local model sent by each client to obtain the global model of the updated convolutional neural network model; repeating the process of the step until the global model of the convolutional neural network model converges;
the global model of the converged codec model and the global model of the convolutional neural network model are used to predict the building number and floor number where the user is located.
According to the technical scheme provided by the embodiment of the invention, the building floor indoor positioning method based on federal learning provided by the invention trains the building floor positioning model through the federal learning training method with centralized pre-training, and then predicts the building number and floor number of the user through the positioning model, so that the data acquisition and storage cost can be effectively reduced, and the position privacy of the client can be ensured not to be leaked. Meanwhile, the method adopts a positioning mode of firstly extracting features through a self-encoder and then carrying out multi-label classification through a one-dimensional convolutional neural network, and effectively ensures the accuracy of building floor positioning on the basis of federal learning.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a process flow diagram of a federally learned building floor indoor positioning method provided by the invention;
FIG. 2 is a flow chart of a preferred embodiment of a federally learned building floor indoor positioning method according to the present invention;
FIG. 3 is a convolutional neural network model for building and floor classification in a federal learning-based building floor indoor positioning method provided by the invention;
FIG. 4 is a schematic diagram of a communication flow for obtaining a global minimum Received Signal Strength (RSS) value according to the federal learning-based building floor indoor positioning method provided by the present invention;
FIG. 5 is a schematic diagram of a centralized pre-training and federal training overall flow of a positioning model of a federal learning-based building floor indoor positioning method according to the present invention;
FIG. 6 is a schematic diagram of a single-round federal training process for a federal learning-based building floor indoor positioning method according to the present invention;
fig. 7 is a schematic diagram of a complete interaction process between a display server and a client of a building floor indoor positioning method based on federal learning.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only for explaining the present invention and are not to be construed as limiting the present invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or coupled. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
For the purpose of facilitating an understanding of the embodiments of the invention, reference will now be made to the drawings of several specific embodiments illustrated in the drawings and in no way should be taken to limit the embodiments of the invention.
Referring to fig. 1, the invention provides a building floor indoor positioning method based on federal learning, which displays a processing procedure at one side of a server, and comprises the following steps:
s1, a server establishes a server local radio frequency fingerprint data set with a self-position tag;
s2, the server establishes a deep learning network model based on the local radio frequency fingerprint data set of the server; the deep learning network model comprises a coding and decoding model and a convolutional neural network model;
s3, the server performs concentrated pre-training on the coding and decoding models based on the server local radio frequency fingerprint data set, and sends the concentrated pre-trained coding and decoding models to each client, so that each client performs federal training on the received concentrated pre-trained coding and decoding models simultaneously based on the self-built client local radio frequency fingerprint data set with the client self-position label to obtain a first local model;
s4, the server receives the first local models sent by each client, and aggregates the plurality of first local models to obtain a global model of the coding and decoding model;
s5, the server sends the global model of the coding and decoding model to each client so that each client carries out federal training on the global model of the coding and decoding model received by each client simultaneously based on a self-built client local radio frequency fingerprint data set with the self-location tag of the client to obtain an updated first local model, and sends the updated first local model to the server; the server aggregates the received updated first local model to obtain a global model of the updated coding and decoding model; repeating the process of the step for iteration for a plurality of times until the global model of the coding and decoding model converges;
s6, the server performs concentrated pre-training on the convolutional neural network model based on the server local radio frequency fingerprint data set, and sends the concentrated pre-trained convolutional neural network model to each client, so that each client performs federal training on the concentrated pre-trained convolutional neural network model sent by the server based on the self-built client local radio frequency fingerprint data set with the client self-location tag to obtain a second local model;
s7, the server receives the second local models sent by each client, and aggregates the plurality of second local models to obtain a global model of the convolutional neural network model;
s8, the server sends the global model of the convolutional neural network model to each client so that each client carries out federal training on the global model of the convolutional neural network model received by each client based on a self-built client local radio frequency fingerprint data set with the own position label of the client to obtain an updated second local model, and sends the updated second local model to the server; the server aggregates the received updated second local model to obtain a global model of the updated convolutional neural network model; the process of the step is repeated for a plurality of times for iteration until the global model of the convolutional neural network model converges.
The number of iterative processes described above may be appropriately set.
The global model of the converged codec model and the global model of the convolutional neural network model are used to predict the building number and floor number where the user is located.
It should be understood that the federal training process described above refers to each client training the received local model, global model, and aggregated global model on the fly relatively independently, and all clients performing the training process simultaneously.
In the preferred embodiment provided by the invention, the radio frequency fingerprint data set with the position tag built by the server and the client is divided into two major parts: a portion of the pre-training data is centralized pre-training data used by the edge server; the other part is local training data used by each client independently. The data distribution between the clients may have two cases of independent co-distribution or non-independent co-distribution: the independent same distribution reflects the condition that each client uniformly samples in the whole area; the non-independent co-distribution reflects the non-uniform sampling by each client based on the respective location preference.
Further, in step S2, a convolutional neural network model based on multi-label classification is established, and a network model diagram is shown in fig. 3. Wherein the former part is the self-encoder part and the latter part is the convolution kernel.
In some improved embodiments, the influence of the RSS value fluctuating with the environment on the weak fingerprint signal is reduced, and in step S1, there is a process of preprocessing the data of the self-built rf fingerprint data set, where the server and each client communicate with each other to find the global minimum RSS value in the rf fingerprint data, and the communication process is shown in fig. 4. The method specifically comprises the following steps:
the server and each client find out the minimum RSS value of the received local data of the radio frequency fingerprint, and the client sends the minimum RSS value to the server;
the server obtains a global minimum RSS value based on the minimum RSS value of the local data of the radio frequency fingerprint received by each client;
server through type
Performing unified power processing on RSS values of the radio frequency fingerprint data set; wherein Min is a global minimum RSS value; alpha >1 is a constant, such as the natural logarithm e.
In the preferred embodiment provided by the present invention, the model training phase of steps S3 and S6 is divided into a centralized pre-training and federal training process. The self-encoder of the FedCNN-BFC method model is first trained. For this purpose, a codec model is first constructed, and then the codec model is subjected to centralized pre-training and federal training in succession. After training is completed, the coding part of the FedCNN-BFC model is initialized by using the encoder part parameters obtained by training, and the rest part of the model is randomly initialized. And finally, carrying out centralized pre-training and federal training on the FedCNN-BFC model. The flow of this step is depicted in fig. 5.
After the steps are completed, the server distributes the aggregated global model to each client for the user to carry out building numbering and floor numbering positioning.
In a second aspect, the present invention provides a building floor indoor positioning method based on federal learning, which displays a processing procedure on one side of a client, and specifically includes the following steps:
e1, each client establishes a client local radio frequency fingerprint data set with a self position label;
each client receives the centralized pre-trained coding and decoding model sent by the server, and each client performs federal training on the centralized pre-trained coding and decoding model received by each client based on the local radio frequency fingerprint data set of each client to obtain a first local model and sends the first local model to the server, so that the server can aggregate the first local model sent by each client to obtain a global model of the coding and decoding model;
each client receives the global model of the coding and decoding model sent by the server, and each client carries out federal training on the global model of the coding and decoding model received by each client based on the local radio frequency fingerprint data set of each client to obtain an updated first local model and sends the updated first local model to the server, so that the server can aggregate the updated first local model sent by each client to obtain the global model of the updated coding and decoding model; repeating the process of the step until the global model of the coding and decoding model converges;
each client receives the concentrated pre-trained convolutional neural network model sent by the server, each client performs federal training on the concentrated pre-trained convolutional neural network model received by each client based on the local radio frequency fingerprint data set of each client to obtain a second local model, and sends the second local model to the server, so that the server can aggregate the second local model sent by each client to obtain a global model of the convolutional neural network model;
each client receives the global model of the convolutional neural network model sent by the server, and each client carries out federal training on the global model of the convolutional neural network model received by each client based on the local radio frequency fingerprint data set of each client to obtain an updated second local model and sends the updated second local model to the server, so that the server can aggregate the updated second local model sent by each client to obtain the global model of the updated convolutional neural network model; repeating the steps until the global model of the convolutional neural network model converges.
It should be understood that the federal training process described above refers to each client training the received local model, global model, and aggregated global model on the fly relatively independently, and all clients performing the training process simultaneously.
The global model of the converged codec model and the global model of the convolutional neural network model are used to predict the building number and floor number where the user is located.
Fig. 6 shows the complete interaction process of the server and the client.
The invention also provides an embodiment for displaying the effect of positioning by executing the positioning method provided by the invention.
Experiments were performed with the selection of UJIIndenorLoc datasets to simulate centralized pre-training and federal training for the positioning model. The training dataset is divided into two parts: a small part of data is used for centralized pre-training of the server; most of the other data was used for federal training of clients, which were distributed in experiments to 15 clients, 900 pieces of data each. The client data distribution simulates the independent co-distribution and non-independent co-distribution conditions respectively: the independent same-distribution mode assumes that each client performs uniform random sampling in the whole area; the non-independent co-distributed allocation approach assumes that each client performs non-uniform sampling based on its respective location preference.
In step two, the FedCNN-BFC model extracts 520 feature compression of each piece of data from the encoder section as 64, and then inputs the extracted feature compression into the convolution section. The convolution kernel of the convolution part uses one-dimensional convolution and is divided into three one-dimensional convolution processes. The kernel size of the first one-dimensional convolution is 22, and 99 convolution kernels are used; the kernel size of the second one-dimensional convolution is 22, and 66 convolution kernels are used; the third one-dimensional convolution has a kernel size of 22, and has 33 convolution kernels, and the step length of each convolution operation is 1.
The loss function used in training the codec model is MSELoss and the loss function used in training the FedCNN-BFC ensemble model is BCELoss.
In order to verify the positioning effect of the proposed method, the proposed federal learning-based building floor classification method FedCNN-BFC is compared with the existing centralized deep learning-based building floor classification method, and the method is named as CenDNN-BFC (Centralized DNN for Building and Floor Classification). Under the condition that the client holds independent and same-distribution data and dependent and same-distribution data, the positioning accuracy of the distributed federal learning method provided by the invention is very close to that of the existing centralized deep learning method through comparison. Considering that the method can reduce the data acquisition and storage cost and protect the position privacy of the client, the method has more advantages in solving the problem of building floor classification of indoor positioning of radio frequency fingerprints.
Table 1 positioning accuracy when client data are independently and simultaneously distributed
| Method | Drop0％ | Drop10％ | Drop20％ | Drop30％ |
| CenDNN-BFC | 0.993 | 0.993 | 0.993 | 0.993 |
| FedCNN-BFC | 0.992 | 0.992 | 0.990 | 0.990 |
TABLE 2 positioning accuracy when clients are non-independent and distributed simultaneously
| Method | Drop0％ | Drop10％ | Drop20％ | Drop30％ |
| CenDNN-BFC | 0.993 | 0.993 | 0.993 | 0.993 |
| FedCNN-BFC | 0.976 | 0.975 | 0.968 | 0.965 |
Table 1 shows the test effect of the positioning model after model training is completed when each client holds independent and uniformly distributed data, 15 clients participate in the whole training, 10 clients participate in each training round are selected, and the training is divided into four client training and out-of-order conditions, and the out-of-order conditions are 0% -30%. The drop-out situation is specifically described in fig. 5. Table 2 shows the test effect of the positioning model after model training is completed when each client holds the independent data with the same distribution, and the test effect is equally divided into four client training and fall-behind conditions. As can be seen from the table, the building floor classification method based on federal learning has the test accuracy very close to that of the centralized method on the premise of reducing the data acquisition and storage cost and protecting the position privacy of the client. This represents an advantage of the proposed method of the present invention.
In summary, the building storey indoor positioning method based on federal learning provided by the invention is used for constructing a radio frequency fingerprint positioning model by using a distributed deep learning technology jointly participated by one edge server and a plurality of mobile clients aiming at a three-dimensional indoor environment containing multiple buildings and multiple storeys. During the location fingerprint data acquisition process, the server side only needs to provide a very small amount of training data, while the vast majority of training data comes from the participation contributions of the client side. In the positioning model training process, a server firstly performs model initialization and performs centralized pre-training by using a small amount of fingerprint data. After the centralized pre-training is completed, the server distributes the resulting model to each client. Based on the received pre-trained model, each client uses its local fingerprint data for further model training, and then transmits the trained local model to the server. At the end of this round of training, the server aggregates the local models collected from each client to obtain a global model for radio frequency fingerprint localization. In order to ensure the training quality, the server sends the obtained model to each client again to perform the model training of the next round, and updates the local model until a certain training round number is reached. The invention adopts the federal learning training method with pre-training and combines the radio frequency fingerprint positioning method based on the convolutional neural network to distribute the data acquisition and model training tasks to each client, so that the model training cost is dispersed, the data acquisition and storage cost is reduced, the position data privacy of the client is protected, and the training efficiency and the positioning effect are ensured.
Those of ordinary skill in the art will appreciate that: the drawing is a schematic diagram of one embodiment and the modules or flows in the drawing are not necessarily required to practice the invention.
From the above description of embodiments, it will be apparent to those skilled in the art that the present invention may be implemented in software plus a necessary general hardware platform. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the embodiments or some parts of the embodiments of the present invention.
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 apparatus or system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, with reference to the description of method embodiments in part. The apparatus and system embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.
Claims (7)
1. A federal learning-based building floor indoor positioning method, comprising:
s1, a server establishes a server local radio frequency fingerprint data set with a self-position tag;
s2, the server establishes a deep learning network model based on the local radio frequency fingerprint data set of the server; the deep learning network model comprises a coding and decoding model and a convolutional neural network model;
s3, the server performs concentrated pre-training on the coding and decoding models based on the server local radio frequency fingerprint data set, and sends the concentrated pre-trained coding and decoding models to each client, so that each client performs federal training on the received concentrated pre-trained coding and decoding models simultaneously based on the self-built client local radio frequency fingerprint data set with the client self-position label to obtain a first local model;
s4, the server receives the first local models sent by each client, and aggregates the plurality of first local models to obtain a global model of the coding and decoding model;
s5, the server sends the global model of the coding and decoding model to each client so that each client carries out federal training on the global model of the coding and decoding model received by each client simultaneously based on a self-built client local radio frequency fingerprint data set with the self-location tag of the client to obtain an updated first local model, and sends the updated first local model to the server; the server aggregates the received updated first local model to obtain a global model of the updated coding and decoding model; repeating the procedure of the step until the global model of the coding and decoding model converges;
s6, the server performs concentrated pre-training on the convolutional neural network model based on the server local radio frequency fingerprint data set, and sends the concentrated pre-trained convolutional neural network model to each client, so that each client performs federal training on the concentrated pre-trained convolutional neural network model sent by the server based on the self-built client local radio frequency fingerprint data set with the client self-location tag to obtain a second local model;
s7, the server receives the second local models sent by each client, and aggregates the plurality of second local models to obtain a global model of the convolutional neural network model;
s8, the server sends the global model of the convolutional neural network model to each client so that each client carries out federal training on the global model of the convolutional neural network model received by each client based on a self-built client local radio frequency fingerprint data set with the own position label of the client to obtain an updated second local model, and sends the updated second local model to the server; the server aggregates the received updated second local model to obtain a global model of the updated convolutional neural network model; repeating the procedure of the step until the global model of the convolutional neural network model converges;
and the converged global model of the coding and decoding model and the global model of the convolutional neural network model are used for predicting the building number and the floor number of the user.
2. The method of claim 1, wherein the server local rf fingerprint data set includes server-used centralized pre-training data; the client local radio frequency fingerprint data set includes local training data used by each client alone.
3. The method according to claim 1, wherein step S1 further comprises preprocessing data in the server local rf fingerprint dataset, and specifically comprises:
the server obtains a minimum RSS value of the received local data of the radio frequency fingerprint;
the server obtains a global minimum RSS value based on the minimum RSS value of the received local data of the radio frequency fingerprint;
server through type
Performing power processing on the RSS value of the radio frequency fingerprint data set; wherein Min is a global minimum RSS value; alpha >1 is a constant, such as the natural logarithm e.
4. A method according to any one of claims 1 to 3, wherein the learning rate is 0.001 during the centralized pre-training and federal training of the codec model and convolutional neural network model.
5. A method according to any one of claims 1 to 3, wherein the convolutional neural network model has a one-dimensional convolutional kernel architecture comprising three convolutional operations, a first convolutional operation having a convolutional kernel size of 22 for a total of 99 convolutional kernels, a second convolutional operation having a convolutional kernel size of 22 for a total of 66 convolutional kernels, and a third convolutional operation having a convolutional kernel size of 22 for a total of 33 convolutional kernels; the step size of each convolution operation is 1.
6. A method according to any one of claims 1 to 3, further comprising:
and S7, the server sends the converged global model of the coding and decoding model and the global model of the convolutional neural network model to one or more clients, so that the one or more clients can predict building numbers and floor numbers through the converged global model of the coding and decoding model and the global model of the convolutional neural network model.
7. A federal learning-based building floor indoor positioning method, comprising:
e1, each client establishes a client local radio frequency fingerprint data set with a self position label;
each client receives the centralized pre-trained coding and decoding model sent by the server, and each client performs federal training on the centralized pre-trained coding and decoding model received by each client based on the local radio frequency fingerprint data set of each client to obtain a first local model and sends the first local model to the server, so that the server can aggregate the first local model sent by each client to obtain a global model of the coding and decoding model;
each client receives the global model of the coding and decoding model sent by the server, and each client carries out federal training on the global model of the coding and decoding model received by each client based on the local radio frequency fingerprint data set of each client to obtain an updated first local model and sends the updated first local model to the server, so that the server can aggregate the updated first local model sent by each client to obtain the global model of the updated coding and decoding model; repeating the process of the step until the global model of the coding and decoding model converges;
each client receives the concentrated pre-trained convolutional neural network model sent by the server, each client performs federal training on the concentrated pre-trained convolutional neural network model received by each client based on the local radio frequency fingerprint data set of each client to obtain a second local model, and sends the second local model to the server, so that the server can aggregate the second local model sent by each client to obtain a global model of the convolutional neural network model;
each client receives the global model of the convolutional neural network model sent by the server, and each client carries out federal training on the global model of the convolutional neural network model received by each client based on the local radio frequency fingerprint data set of each client to obtain an updated second local model and sends the updated second local model to the server, so that the server can aggregate the updated second local model sent by each client to obtain the global model of the updated convolutional neural network model; repeating the process of the step until the global model of the convolutional neural network model converges;
and the converged global model of the coding and decoding model and the global model of the convolutional neural network model are used for predicting the building number and the floor number of the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111088881.3A CN113805142B (en) | 2021-09-16 | 2021-09-16 | Building floor indoor positioning method based on federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111088881.3A CN113805142B (en) | 2021-09-16 | 2021-09-16 | Building floor indoor positioning method based on federal learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113805142A CN113805142A (en) | 2021-12-17 |
| CN113805142B true CN113805142B (en) | 2023-11-07 |
Family
ID=78941388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111088881.3A Active CN113805142B (en) | 2021-09-16 | 2021-09-16 | Building floor indoor positioning method based on federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113805142B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114666734B (en) * | 2022-03-22 | 2022-11-22 | 广州市梦享网络技术有限公司 | Positioning method and device for deep learning radio frequency signal fingerprint identification algorithm |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111866869A (en) * | 2020-07-07 | 2020-10-30 | 兰州交通大学 | Federal learning indoor positioning privacy protection method facing edge calculation |
| CN111931949A (en) * | 2019-05-13 | 2020-11-13 | 国际商业机器公司 | Communication in a federated learning environment |
| CN112464290A (en) * | 2020-12-17 | 2021-03-09 | 浙江工业大学 | Vertical federal learning defense method based on self-encoder |
| CN112862011A (en) * | 2021-03-31 | 2021-05-28 | 中国工商银行股份有限公司 | Model training method and device based on federal learning and federal learning system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190012592A1 (en) * | 2017-07-07 | 2019-01-10 | Pointr Data Inc. | Secure federated neural networks |
-
2021
- 2021-09-16 CN CN202111088881.3A patent/CN113805142B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111931949A (en) * | 2019-05-13 | 2020-11-13 | 国际商业机器公司 | Communication in a federated learning environment |
| CN111866869A (en) * | 2020-07-07 | 2020-10-30 | 兰州交通大学 | Federal learning indoor positioning privacy protection method facing edge calculation |
| CN112464290A (en) * | 2020-12-17 | 2021-03-09 | 浙江工业大学 | Vertical federal learning defense method based on self-encoder |
| CN112862011A (en) * | 2021-03-31 | 2021-05-28 | 中国工商银行股份有限公司 | Model training method and device based on federal learning and federal learning system |
Non-Patent Citations (2)
| Title |
|---|
| FedLoc: Federated learning framework for data-driven cooperative localization and location data processing;Yin F等;《IEEE Open Journal of Signal Processing》(第1期);第187-215页 * |
| 基于迁移学习深度卷积神经网络的 配电网故障区域定位;孟子超;《南方电网技术》;第13卷(第7期);第25-33页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113805142A (en) | 2021-12-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ibrahim et al. | CNN based indoor localization using RSS time-series | |
| JP7154295B2 (en) | Motion detection based on machine learning of radio signal characteristics | |
| Dorfan et al. | Tree-based recursive expectation-maximization algorithm for localization of acoustic sources | |
| CN108628904A (en) | A kind of path code, Similar Track search method and device and electronic equipment | |
| CN110531313A (en) | A kind of near-field signals source localization method based on deep neural network regression model | |
| Zhang et al. | Multilayer probability extreme learning machine for device-free localization | |
| CN109918684A (en) | Model training method, interpretation method, relevant apparatus, equipment and storage medium | |
| Bhattarai et al. | Geomagnetic field based indoor landmark classification using deep learning | |
| CN112532746B (en) | Cloud edge cooperative sensing method and system | |
| Khatab et al. | A fingerprint technique for indoor localization using autoencoder based semi-supervised deep extreme learning machine | |
| Cui et al. | A robust mobile robot indoor positioning system based on Wi-Fi | |
| CN113805142B (en) | Building floor indoor positioning method based on federal learning | |
| Cha et al. | A hierarchical auxiliary deep neural network architecture for large-scale indoor localization based on Wi-Fi fingerprinting | |
| Mohapatra et al. | Precision local positioning mechanism in underground mining using IoT-enabled WiFi platform | |
| Yang et al. | EKF–GPR-based fingerprint renovation for subset-based indoor localization with adjusted cosine similarity | |
| CN114334036A (en) | Model training method, related device, equipment and storage medium | |
| US20220386264A1 (en) | Indoor target positioning method based on improved convolutional neural network model | |
| CN107124761B (en) | Cellular network wireless positioning method fusing PSO and SS-ELM | |
| CN111461306B (en) | Feature evaluation method and device | |
| CN111276138B (en) | Method and device for processing voice signal in voice wake-up system | |
| CN110320493B (en) | Indoor positioning method, device, electronic equipment and computer storage medium | |
| Chen et al. | Harris hawks optimization algorithm and BP neural network for ultra-wideband indoor positioning | |
| Bai et al. | RFedRNN: An end-to-end recurrent neural network for radio frequency path fingerprinting | |
| CN113344212B (en) | Model training method, device, computer equipment and readable storage medium | |
| US11888544B2 (en) | Selection of physics-specific model for determination of characteristics of radio frequency signal propagation |
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 |