WO2020259718A1

WO2020259718A1 - Federated learning-based indoor positioning method, apparatus, and terminal device, and storage medium

Info

Publication number: WO2020259718A1
Application number: PCT/CN2020/107331
Authority: WO
Inventors: 程勇; 刘洋; 陈天健
Original assignee: 深圳前海微众银行股份有限公司
Priority date: 2019-09-20
Filing date: 2020-08-06
Publication date: 2020-12-30
Also published as: CN110632554A

Abstract

A federated learning-based indoor positioning method, apparatus, and terminal device, and a computer-readable storage medium. Establishing an indoor positioning training data set for terminal devices (S100); the terminal devices perform model training on the basis of the indoor positioning training data set so as to obtain a model parameter update (S200); transforming the model parameter update to an overall model parameter for terminal devices to undergo indoor positioning (S300).

Description

Indoor positioning method, device, terminal equipment and medium based on federated learning

This application claims the priority of the Chinese patent application filed on September 20, 2019, with application number 201910898051.3, and titled "Indoor positioning method, device, terminal equipment and medium based on federal learning", the entire content of which is incorporated by reference In this application.

Technical field

This application relates to the field of Fintech (financial technology) technology, and in particular to an indoor positioning method, device, terminal device, and computer-readable storage medium based on federated learning.

Background technique

Wi-Fi signals exist in most indoor environments. For example, offices, teaching buildings, restaurants, cafes, shopping malls, supermarkets, airports, train stations and subway cars already have Wi-Fi signals. Coverage, so the technology of using Wi-Fi signals for indoor positioning has ushered in a wide range of applications.

However, in the existing method of indoor positioning based on Wi-Fi signals, it is necessary to construct a Wi-Fi location fingerprint library intensively, that is, to obtain Wi-Fi location fingerprint information by measuring at a large number of target locations to build Wi-Fi. Location fingerprint database, in this way, not only need to spend a lot of time and energy to build the location fingerprint database, resulting in the overall measurement of the indoor location is very limited, in addition, in the use of crowdsourcing (through a large number of users' mobile terminals to collect Wi- When the location fingerprint database is constructed using Fi location fingerprint information), the user’s private information, such as the place and location information the user has been to, will also be leaked.

Technical solutions

The main purpose of this application is to provide an indoor positioning method, device, terminal equipment, and computer readable storage medium based on federated learning, which aims to solve the existing indoor positioning method, which can measure the indoor position as a whole is limited and easy to leak Technical issues of user privacy.

In order to achieve the above objective, this application provides an indoor positioning method based on federated learning. The indoor positioning method based on federated learning includes:

Construct indoor positioning training data sets for each terminal device;

Each of the terminal devices performs model training based on the indoor positioning training data set to obtain model parameter updates;

The update of the model parameters is converted into global model parameters for each terminal device to perform indoor positioning.

Further, before the step of constructing an indoor positioning training data set of each terminal device, the method further includes:

Obtain the location information of the location of each terminal device and the location fingerprint information collected by each terminal device.

Further, the location information includes: wide area location information and indoor location information,

The step of constructing the indoor positioning training data set owned by each terminal device includes:

Sequentially extracting target indoor position information corresponding to each piece of position fingerprint information collected by the terminal device from the indoor position information;

Using the location fingerprint information, the target indoor location information, and the wide area location information to which the target indoor location information belongs respectively as a piece of indoor positioning training data;

All the indoor positioning training data are collected to construct an indoor positioning training data set of each terminal device.

Further, the wide area location information is physical location information, and the indoor location information is coordinate information using the wide area location information as a reference point.

Further, the step of combining each of the terminal devices to perform model training based on the indoor positioning training data set to obtain model parameter updates includes:

Detecting a model update request for each terminal device to perform model training;

According to the detected model update request, each terminal device performs model training locally on each terminal device to obtain model parameter updates.

Further, the step of performing model training on each of the terminal devices locally according to the detected model update request includes:

Detecting whether the global model parameter is included in the model update request;

If so, each of the terminal devices uses the global model parameters to perform model training locally;

If not, each terminal device uses the indoor positioning training data set to perform model training locally.

Further, the step of converting the model parameter update into a global model parameter for each of the terminal devices to perform indoor positioning includes:

Performing preset fusion processing on the model parameter update, and convert the model parameter update into a global model parameter;

The global model parameters are distributed to each of the terminal devices, so that each of the terminal devices can perform indoor positioning based on model training locally.

In addition, in order to achieve the above objective, the present application also provides an indoor positioning device based on federated learning, and the indoor positioning device based on federated learning includes:

The building module is used to build the indoor positioning training data set owned by each terminal device;

The training module is used to perform model training based on the indoor positioning training data set to obtain model parameter updates;

The positioning module is used for converting the model parameter update into a global model parameter for each of the terminal devices to perform indoor positioning.

The present application also provides a terminal device. The terminal device includes a memory, a processor, and an indoor positioning program based on federated learning that is stored in the memory and can be run on the processor. When the indoor positioning program is executed by the processor, the steps of the indoor positioning method based on federated learning as described above are realized.

The present application also provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the indoor positioning method based on federated learning as described above are realized.

The indoor positioning method, device, terminal device, and computer-readable storage medium based on federated learning proposed in this application construct an indoor positioning training data set owned by each terminal device; each terminal device is based on the indoor positioning training data set Model training is performed to obtain model parameter updates; the model parameter updates are converted into global model parameters for indoor positioning by each terminal device. Build its own indoor positioning training data set based on the user’s own terminal equipment, and then combine each user’s terminal equipment to perform machine learning model training based on their own indoor positioning data and obtain the model parameters of their respective models Update, and finally perform a centralized conversion process through the model parameter update obtained from the model training of the terminal equipment owned by each user to obtain the global model parameters required by all user terminal equipment for indoor positioning, so that each user terminal equipment is locally The global model parameters obtained by the conversion can be used for model training to obtain the indoor location information of the user terminal device itself when the model training is completed.

This application realizes that there is no need to concentrate on building a location fingerprint database, but to combine the data owned by each terminal device itself for model training, avoiding the need to spend a lot of time and energy to build a location fingerprint database, and to centralize each terminal device The model parameters obtained by model training are updated to obtain global model parameters, which greatly expands the indoor position that can be measured by the terminal device. Moreover, since each terminal device only trains the machine learning model locally, they have not disclosed their respective locations. Owning indoor positioning training data, thus, will not cause the leakage of user privacy information.

Description of the drawings

FIG. 1 is a schematic diagram of the hardware operation structure involved in the solution of the embodiment of the present application;

2 is a schematic flowchart of a first embodiment of an indoor positioning method based on federated learning according to the present application;

3 is a detailed flowchart of step S200 in an embodiment of an indoor positioning method based on federated learning according to the present application;

4 is a schematic diagram of an application scenario of an embodiment of an indoor positioning method based on federated learning according to the present application;

Fig. 5 is a schematic structural diagram of an indoor positioning device based on federated learning in this application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Embodiments of the invention

It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application.

As shown in Fig. 1, Fig. 1 is a schematic structural diagram of a hardware operating environment involved in a solution of an embodiment of the present application.

It should be noted that Fig. 1 can be a structural diagram of the hardware operating environment of the terminal device. The terminal device in the embodiment of the present application may be a terminal device such as a PC and a portable computer.

As shown in FIG. 1, the terminal device may include a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to implement connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a magnetic disk memory. Optionally, the memory 1005 may also be a storage device independent of the foregoing processor 1001.

Those skilled in the art can understand that the structure of the terminal device shown in FIG. 1 does not constitute a limitation on the terminal device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange different components.

As shown in FIG. 1, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a processing program for distributed tasks. Among them, the operating system is a program that manages and controls the hardware and software resources of the sample terminal equipment, supports the processing of distributed tasks, and the operation of other software or programs.

In the terminal device shown in FIG. 1, the user interface 1003 is mainly used for data communication with various terminals; the network interface 1004 is mainly used for connecting to a background server and performing data communication with the background server; and the processor 1001 can be used for calling the memory 1005 The indoor positioning program based on federated learning stored in and perform the following operations:

Construct indoor positioning training data sets for each terminal device;

Further, the processor 1001 may call an indoor positioning program based on federated learning stored in the memory 1005, and before executing the step of constructing an indoor positioning training data set of each terminal device, it may also perform the following operations:

Further, the processor 1001 may call an indoor positioning program based on federated learning stored in the memory 1005, and also perform the following operations:

Detecting and controlling the model update request for each of the terminal devices to perform model training;

According to the detected model update request, each of the terminal devices performs model training locally to obtain model parameter updates.

Performing preset fusion processing on the model parameter update into a global model parameter;

Based on the above structure, various embodiments of the indoor positioning method based on federated learning in this application are proposed.

Please refer to FIG. 2, which is a schematic flowchart of a first embodiment of an indoor positioning method based on federated learning in this application.

The embodiment of the application provides an embodiment of an indoor positioning method based on federated learning. It should be noted that although the logical sequence is shown in the flowchart, in some cases, the sequence may be executed in a different order than here. Steps shown or described.

The indoor positioning method based on federated learning in the embodiment of the present application is applied to the above-mentioned terminal device. The terminal device in the embodiment of the present application may be a terminal device such as a PC, a portable computer, etc., which is not specifically limited here.

The indoor positioning method based on federated learning in this embodiment includes:

Step S100: Construct an indoor positioning training data set of each terminal device.

The indoor positioning training data set owned by each terminal device is built locally in each terminal device.

Further, in another embodiment, in the foregoing step S100, before constructing the indoor positioning training data set owned by each terminal device, the indoor positioning method based on federated learning in this application further includes:

Step A: Obtain the location information of the location of each terminal device and the location fingerprint information collected by each terminal device.

Each terminal device independently collects location information and location fingerprint information of its location. The location information includes wide-area location information and indoor location information. Location fingerprint information can be collected based on Wi-Fi signals, or location fingerprint information is also based on Bluetooth Wait for other wireless signals to collect.

It should be noted that, in this embodiment, the indoor location information is: the current user terminal equipment (may be the user’s mobile terminal equipment, such as a mobile phone) has visited indoor location information (such as office spaces, teaching buildings, restaurants, cafes). , Shopping malls, supermarkets, airports, train stations, subway cars and other indoor location information); wide-area location information is: outdoor large-scale location information, for example, the location information of an office building, or the location information of a shopping mall , Wide-area location information is physical location information that can be obtained through GPS satellites, or Beidou navigation satellite systems, or mobile communication base station positioning systems, or through map information. The manifestation of wide-area location information can be in the navigation system. Latitude and longitude coordinates, or “XX city-XX road-XX number” in the map information; location fingerprint information is: Wi-Fi location fingerprint information or Bluetooth location fingerprint information measured by the current user’s mobile terminal device, Wi-Fi -Fi location fingerprint information or Bluetooth location fingerprint information has a one-to-one correspondence with indoor location information. Wi-Fi location fingerprint information or Bluetooth location fingerprint information can include RSS (Really Simple Syndication), a description and synchronization of website content One or more of the listed information such as the MAC address of the Wi-Fi AP (where AP is Access Point: wireless access node, session point or access bridge), multi-path structure), multi-path structure, and AP’s IP address .

It should be noted that in this embodiment, each terminal device may not upload the location information and fingerprint information of the location where it is collected independently to any other server, but store it locally in each terminal device, thereby avoiding user privacy. Information disclosure, such as not revealing information such as places and locations that users have been to.

It should be noted that, in this embodiment, the indoor position information of the location of the terminal device may be coordinate information with the wide area position information as the coordinate center, or the indoor position information may also use the wide area position information as the reference point Therefore, the wide-area location information can help the user’s mobile terminal to calculate the wide-area coordinate position of the indoor location information that recognizes the current mobile terminal’s location (for example, the global coordinate position in the form of longitude and latitude coordinates). That is to say, the indoor location information of the user's mobile terminal does not only contain the local coordinate location in the room (that is, the coordinate information with the wide area location information as the coordinate center, or the wide area location information as the reference point). Coordinate information).

Further, step S100, constructing an indoor positioning training data set of each terminal device, includes:

Step S101, sequentially extracting target indoor position information corresponding to each piece of position fingerprint information collected by the terminal device from the indoor position information.

From the multiple pieces of indoor location information recorded by the terminal device, the target indoor location information uniquely corresponding to each piece of location fingerprint information collected by the current terminal device is sequentially extracted.

For example, from the multiple pieces of indoor location information expressed in spatial coordinate format recorded by the current user’s mobile terminal, one piece of Wi-Fi in the form of "RSS+multipath structure" recorded by the current user’s mobile terminal is extracted. Fi location fingerprint information, which uniquely corresponds to a piece of target indoor location information, and sequentially extracts each piece of target indoor location information uniquely corresponding to each piece of Wi-Fi location fingerprint information among multiple pieces of current indoor location information.

In step S102, the location fingerprint information, the target indoor location information, and the wide area location information to which the target indoor location information belongs are respectively used as a piece of indoor positioning training data.

A piece of location fingerprint information, the target indoor location information corresponding to the current location fingerprint information, and the wide area location information to which the current target indoor location information belongs are marked as a piece of indoor positioning training data of the current terminal device.

For example, take the Wi-Fi location fingerprint information with the current form of "RSS+multipath structure", and the target indoor location information in the spatial coordinate format that uniquely corresponds to the current Wi-Fi location fingerprint information, and the current target indoor The wide-area location information identified by the coordinate center of the location information is marked as a piece of indoor positioning training data uniquely owned by the mobile terminal of the current user.

Step S103: Count all the indoor positioning training data to construct an indoor positioning training data set of each terminal device.

Count each piece of indoor positioning training data that contains location fingerprint information, indoor location information, and wide-area location information owned by the terminal device and save it in the local database of the current terminal device, thereby constructing an indoor positioning unique to the current terminal device Training data set.

In this embodiment, by constructing a training set of indoor positioning data unique to each terminal device on different terminal devices, it is possible to implement machine learning model training for each terminal device based on its own indoor positioning data training set. A terminal device can learn the indoor location information of a place that the terminal device has not been to (the indoor location recorded by other terminal devices), so that when a terminal device goes to a place that has not been before, As long as any other terminal device associated with the current terminal device has reached and recorded the indoor location information of the location where the current terminal device is located, the current terminal device can obtain accurate indoor positioning information.

Further, in this embodiment, by adding wide-area location information to the indoor positioning data of the terminal device, the effect of joint learning for multiple terminal devices is equivalent to the use of "crowdsourcing" to collect training data and then perform machine To learn the effect of model training, compared to the "crowdsourcing" approach, this embodiment does not disclose the user's private information, for example, does not disclose information such as places and locations the user has been to.

In step S200, each terminal device performs model training based on the indoor positioning training data set to obtain model parameter updates.

After each terminal device completes its own local indoor positioning training data set, each terminal device is combined to perform machine learning model training based on the indoor positioning training data set it owns, so that the model parameters after model training are obtained locally on each terminal device Update.

It should be noted that the model parameter update is a term that refers to the update of the parameters. The model parameter update can be the sent model parameter, the sent gradient value, the loss value, or the weight of the neural network.

Further, please refer to FIG. 3, which is a schematic diagram of the detailed process of step S200. In step S200, model training is performed in conjunction with each of the terminal devices based on the indoor positioning training data set to obtain model parameter updates, including:

Step S201: Detect a model update request of each terminal device for model training.

Each terminal device detects and obtains a model update request that controls its machine learning model training, where the model training request can be a start instruction that does not carry any data, or the model training request can also be carried by each terminal device independently The model parameters obtained by the model training update the global model parameters generated by the transformation.

For example, in the application scenario shown in Figure 4, after the terminal device 1, the terminal device 2 to the terminal device n have all constructed their own indoor positioning training data set, the server sends the terminal device 1 to the terminal device 1 through a point-to-point communication. The terminal device 2 to the terminal device n send out a model update request for controlling the machine learning model training, or the server can also use multicast, or multicast, or broadcast, and send it to the terminal device 1, the terminal device 2 to one or more of the terminal devices n sends the model update request, and each terminal device 1, the terminal device 2 to the terminal device n detects and obtains the model update request in real time.

Step S202: According to the detected model update request, each terminal device performs model training locally to obtain model parameter updates.

After each terminal device obtains the model update request that controls it to perform machine learning model training, it further detects whether the current model update request includes global model parameters generated by the model parameter update obtained by each terminal device independently performing model training. Thus, according to the detection result, the machine learning model is trained locally in each terminal device.

It should be noted that in this embodiment, each terminal device can also independently use its own indoor positioning training data for model training, without the need to cooperate with other terminal devices. Further, each terminal device can also update the request according to the detection model. , The training of the machine learning model is performed independently or in conjunction with other terminal devices on the confidential cloud server associated with the current terminal device. In this embodiment, the indoor positioning training data collected by each terminal device can be stored in each terminal device Locally, either stored in the confidential network storage space associated with the terminal device, or stored in the confidential cloud storage space associated with the terminal device, thereby ensuring that the privacy information of the user of the terminal device will not be leaked.

Step S300: Convert the model parameter update into a global model parameter for each terminal device to perform indoor positioning.

After each terminal device performs machine learning model training locally to obtain model parameter updates, the model parameter updates obtained by each terminal device's independent model training are subjected to fusion processing to transform and generate global model parameters, and the transformed global model parameters are passed through The model update request that controls each terminal device to perform machine learning model training is distributed to each terminal device, so that each terminal device can continue to perform model training to achieve indoor positioning.

For example, in the application scenario shown in Figure 4, the server uses multicast, or multicast, or broadcast, and sends to the terminal device 1, terminal device 2 to terminal device n at the same time to control each terminal device to start model training The model update request sent by the server to each terminal device 1, terminal device 2 to terminal device n carries the model update request of each terminal device 1, terminal device 2 to terminal device n based on its own The model parameter update obtained by model training is performed on the indoor positioning training data set, and the global model parameters generated by the fusion process are transformed, so as to detect and obtain the model update request in real time after each terminal device 1, terminal device 2 to terminal device n , By using the carried global model parameters to continue the local model training of machine learning models (such as LSTM models (long-short term memory)) to achieve indoor positioning, that is, the server and each terminal device 1, The terminal device 2 to the terminal device n repeat the model training based on the global model parameters to obtain the model parameter update, and the model parameter update fusion process is transformed into the new global model parameter step, until each terminal device 1 terminal device 2 to terminal device n performs the machine After the training of the learning model is completed, each terminal device 1 terminal device 2 to terminal device n can obtain accurate indoor position information (that is, indoor positioning).

In this embodiment, the indoor positioning training data set owned by each terminal device is constructed; each terminal device is combined to perform model training based on the indoor positioning training data set to obtain model parameter updates; and the model parameters are updated Converted into global model parameters for each terminal device to perform indoor positioning. Build its own indoor positioning training data set based on the user’s own terminal equipment, and then combine each user’s terminal equipment to perform machine learning model training based on their own indoor positioning data and obtain the model parameters of their respective models Update, and finally perform a centralized conversion process through the model parameter update obtained from the model training of the terminal equipment owned by each user to obtain the global model parameters required by all user terminal equipment for indoor positioning, so that each user terminal equipment is locally The global model parameters obtained by the conversion can be used for model training to obtain the indoor location information of the user terminal device itself when the model training is completed.

It is realized that there is no need to centrally build a location fingerprint database, but to combine the data owned by each terminal device itself for model training, which avoids spending a lot of time and energy to build a location fingerprint database, and conducts centralized model training for each terminal device The obtained model parameters are updated and transformed to obtain global model parameters, which greatly expands the indoor location that can be measured by the terminal device, and because each terminal device is only machined locally or on a confidential cloud server associated with the terminal device The training of the learning model does not disclose the indoor positioning training data owned by each, so that it will not cause the leakage of user privacy information.

Further, a second embodiment of the indoor positioning method based on federated learning in this application is proposed.

Based on the first embodiment of the indoor positioning method based on federated learning, in the second embodiment of the indoor positioning method based on federated learning of the present application, in the foregoing step S202, the terminals are combined according to the detected model update request The equipment conducts model training locally, including:

Step S2021: Detect whether the global model parameter is included in the model update request.

After each terminal device detects and obtains the model update request that controls its machine learning model training in real time, it further detects whether the current model update request includes the model parameter update obtained by each terminal device's independent model training for transformation and generation Model parameters.

Step S2022, each of the terminal devices uses the global model parameters to perform model training locally.

If the terminal device detects that the current model update request carries the global model parameters generated by transforming the model parameters obtained by each terminal device's independent model training, the global model parameters generated by the transformation are used locally in each terminal device to perform the machine Model training for learning models.

For example, in the application scenario shown in Figure 4, the terminal device 1, the terminal device 2 to the terminal device n jointly train the machine learning model for indoor positioning-the LSTM model, each terminal device 1, the terminal device 2 to The terminal device n uses its own indoor positioning training data set to train locally on the terminal device, and sends the model parameter update obtained after training to the server, and the server sends the received terminal device 1, terminal device 2 to terminal device n The incoming model parameter updates are fused (for example, to obtain a weighted average), and the global model parameters obtained after the fusion processing are sent to each terminal device through a model training request, so that each terminal device continues to operate according to the global model parameters. Local training is used for indoor positioning LSTM model.

It should be noted that the terminal device and the server can use encryption (for example, homomorphic encryption, or password sharing) to send model parameter updates and global model parameters.

Step S2023, each of the terminal devices uses the indoor positioning training data set to perform model training locally.

If the terminal device detects that the current model update request does not carry the global model parameters generated by the model parameter updates obtained by each terminal device's independent model training, it will use the indoor positioning owned by each terminal device locally. The training data set is used for model training of machine learning models.

For example, in the application scenario shown in Fig. 4, when the model update request sent by the server to each terminal device 1, terminal device 2 to terminal device n does not carry the global model parameter obtained by fusing the model parameter update ( At this time, when the model update request is equivalent to the start instruction for controlling the terminal device to perform machine learning model training), each terminal device 1, terminal device 2 to terminal device n uses the indoor positioning training data set constructed by each of them to start local Train the LSTM model for indoor positioning.

Further, step S300 in the foregoing first embodiment includes:

Step S301: Perform preset fusion processing on the model parameter update, and convert the model parameter update into a global model parameter.

Step S302: Distribute the global model parameters to each of the terminal devices, so that each of the terminal devices can perform indoor positioning locally based on model training.

For example, in the application scenario shown in Figure 4, the terminal device 1, the terminal device 2 to the terminal device n jointly train the machine learning model for indoor positioning-the LSTM model, each terminal device 1, the terminal device 2 to The terminal device n uses its own indoor positioning training data set to train locally on the terminal device, and sends the model parameter update obtained after training to the server, and the server sends the received terminal device 1, terminal device 2 to terminal device n The incoming model parameter updates are fused (for example, to obtain a weighted average), and the global model parameters obtained after the fusion processing are sent to each terminal device through a model training request, so that each terminal device 1, terminal device 2 and terminal device Device n continues to train locally the LSTM model for indoor positioning according to the global model parameters, until the machine learning model training of each terminal device 1 terminal device 2 to terminal device n is completed, and each terminal device 1 terminal device 2 to terminal device n Then you can get accurate indoor location information.

In this embodiment, the machine learning model used for indoor positioning is trained locally by each terminal device in conjunction with each terminal device, and after each terminal device performs the machine learning model training locally to obtain the model parameter update, the terminal device The model parameter updates obtained by independent model training are fused to transform and generate global model parameters, and the transformed global model parameters are distributed to each terminal device through a model update request that controls each terminal device to perform machine learning model training. Each terminal device continues to perform model training to achieve indoor positioning.

Realized, based on the combination of various terminal devices through horizontal federation learning to train the indoor positioning machine learning model, and repeated model training for indoor positioning, the indoor position that can be measured by the terminal device is not expanded, and the indoor positioning is increased. Orientation, and also improve the positioning accuracy and efficiency of indoor positioning.

In addition, please refer to Fig. 5, an embodiment of this application also proposes an indoor positioning device based on federated learning. The indoor positioning device based on federated learning in this application includes:

The building module is used to build the indoor positioning training data set of each terminal device;

A training module for each of the terminal devices to perform model training based on the indoor positioning training data set to obtain model parameter updates;

In an embodiment, the indoor positioning device based on federated learning in this application further includes:

The acquiring module is used to acquire the location information of the location of each terminal device and the location fingerprint information collected by each terminal device.

In an embodiment, the building module includes:

An extraction unit, configured to sequentially extract target indoor position information corresponding to each piece of position fingerprint information collected by the terminal device from the indoor position information;

A marking unit, configured to separately use the position fingerprint information, the target indoor position information, and the wide area position information to which the target indoor position information belongs as a piece of indoor positioning training data;

The construction unit is used to count all the indoor positioning training data to construct an indoor positioning training data set of each terminal device.

In an embodiment, the training module includes:

The detection unit is configured to detect a model update request for model training of each terminal device;

The training unit is configured to perform model training on each of the terminal devices locally according to the detected model update request to obtain model parameter updates.

In an embodiment, the joint training unit includes:

The detection subunit is used to detect whether the global model parameter is included in the model update request;

The first joint training subunit is used for each of the terminal devices to perform model training locally by using the global model parameters;

The second joint training subunit is used for each of the terminal devices to use the indoor positioning training data set to perform model training locally.

In an embodiment, the positioning module includes:

A conversion unit, configured to perform preset fusion processing on the model parameter update, and convert the model parameter update into a global model parameter;

The distributing positioning unit is configured to distribute the global model parameters to each of the terminal devices, so that each of the terminal devices can perform indoor positioning based on model training locally.

The steps of implementing the above-mentioned indoor positioning method based on federated learning during operation of each functional module of the indoor positioning device based on federated learning proposed in this embodiment will not be repeated here.

In addition, the embodiment of the present application also proposes a computer-readable storage medium, which is applied to a computer. The computer-readable storage medium may be a non-volatile computer-readable storage medium. The indoor positioning program based on federated learning is executed by the processor to implement the steps of the indoor positioning method based on federated learning.

Wherein, the steps implemented when the indoor positioning program based on federated learning running on the processor is executed can refer to each embodiment of the indoor positioning method based on federated learning of this application, which will not be repeated here.

It should be noted that in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, It also includes other elements that are not explicitly listed, or elements inherent to the process, method, article, or system. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, article or system that includes the element.

The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to enable a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the method described in each embodiment of the present application.

The above are only preferred embodiments of this application, and do not limit the scope of this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly used in other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims

An indoor positioning method based on federated learning, wherein the indoor positioning method based on federated learning includes:

Construct indoor positioning training data sets for each terminal device;

Each of the terminal devices performs model training based on the indoor positioning training data set to obtain model parameter updates;

The update of the model parameters is converted into global model parameters for each terminal device to perform indoor positioning.
The indoor positioning method based on federated learning according to claim 1, wherein before the step of constructing an indoor positioning training data set of each terminal device, the method further comprises:

Obtain the location information of the location of each terminal device and the location fingerprint information collected by each terminal device.
The indoor positioning method based on federated learning according to claim 2, wherein the position information includes at least: wide area position information and indoor position information,

The step of constructing the indoor positioning training data set owned by each terminal device includes:

Sequentially extracting target indoor position information corresponding to each piece of position fingerprint information collected by the terminal device from the indoor position information;

Using the location fingerprint information, the target indoor location information, and the wide area location information to which the target indoor location information belongs respectively as a piece of indoor positioning training data;

All the indoor positioning training data are collected to construct an indoor positioning training data set of each terminal device.
The indoor positioning method based on federated learning according to claim 3, wherein the wide area location information is physical location information, and the indoor location information is coordinate information using the wide area location information as a reference point.
The indoor positioning method based on federated learning according to claim 1, wherein the step of performing model training for each of the terminal devices based on the indoor positioning training data set to obtain model parameter updates includes:

Detecting a model update request for each terminal device to perform model training;

According to the detected model update request, each of the terminal devices performs model training locally to obtain model parameter updates.
The indoor positioning method based on federated learning according to claim 5, wherein the step of performing model training on each of the terminal devices locally according to the detected model update request includes:

Detecting whether the global model parameter is included in the model update request;

If so, each of the terminal devices uses the global model parameters to perform model training locally;

If not, each terminal device uses the indoor positioning training data set to perform model training locally.
The indoor positioning method based on federated learning according to claim 1, wherein the step of converting the model parameter update into a global model parameter for each of the terminal devices to perform indoor positioning comprises:

Performing preset fusion processing on the model parameter update, and convert the model parameter update into a global model parameter;

The global model parameters are distributed to each of the terminal devices, so that each of the terminal devices can perform indoor positioning based on model training locally.
An indoor positioning device based on federated learning, wherein the indoor positioning device based on federated learning comprises:

The building module is used to build the indoor positioning training data set of each terminal device;

A training module for each of the terminal devices to perform model training based on the indoor positioning training data set to obtain model parameter updates;

The positioning module is used for converting the model parameter update into a global model parameter for each of the terminal devices to perform indoor positioning.
A terminal device, wherein the terminal device includes a memory, a processor, and an indoor positioning program based on federated learning that is stored in the memory and can be run on the processor, and the indoor positioning based on federated learning The following steps are implemented when the program is executed by the processor:

Construct indoor positioning training data sets for each terminal device;

Each of the terminal devices performs model training based on the indoor positioning training data set to obtain model parameter updates;

The update of the model parameters is converted into global model parameters for each terminal device to perform indoor positioning.
The terminal device according to claim 9, wherein before the step of constructing an indoor positioning training data set of each terminal device, the method further comprises:

Obtain the location information of the location of each terminal device and the location fingerprint information collected by each terminal device.
The terminal device according to claim 10, wherein the location information includes at least: wide area location information and indoor location information,

The step of constructing the indoor positioning training data set owned by each terminal device includes:

Sequentially extracting target indoor position information corresponding to each piece of position fingerprint information collected by the terminal device from the indoor position information;

Using the location fingerprint information, the target indoor location information, and the wide area location information to which the target indoor location information belongs respectively as a piece of indoor positioning training data;

All the indoor positioning training data are collected to construct an indoor positioning training data set of each terminal device.
The terminal device according to claim 11, wherein the wide area location information is physical location information, and the indoor location information is coordinate information using the wide area location information as a reference point.
The terminal device according to claim 9, wherein the step of performing model training for each of the terminal devices based on the indoor positioning training data set to obtain model parameter updates includes:

Detecting a model update request for each terminal device to perform model training;

According to the detected model update request, each of the terminal devices performs model training locally to obtain model parameter updates.
The terminal device according to claim 13, wherein the step of performing model training on each of the terminal devices locally according to the detected model update request comprises:

Detecting whether the global model parameter is included in the model update request;

If so, each of the terminal devices uses the global model parameters to perform model training locally;

If not, each terminal device uses the indoor positioning training data set to perform model training locally.
9. The terminal device of claim 9, wherein the step of converting the model parameter update into a global model parameter for each of the terminal devices to perform indoor positioning comprises:

Performing preset fusion processing on the model parameter update, and convert the model parameter update into a global model parameter;

The global model parameters are distributed to each of the terminal devices, so that each of the terminal devices can perform indoor positioning based on model training locally.
A storage medium, wherein a computer program is stored on the storage medium, and when the computer program is executed by a processor, the following steps are implemented:

Construct indoor positioning training data sets for each terminal device;

Each of the terminal devices performs model training based on the indoor positioning training data set to obtain model parameter updates;

The update of the model parameters is converted into global model parameters for each terminal device to perform indoor positioning.
The storage medium of claim 16, wherein before the step of constructing an indoor positioning training data set of each terminal device, the method further comprises:

Obtain the location information of the location of each terminal device and the location fingerprint information collected by each terminal device.
17. The storage medium of claim 17, wherein the location information includes at least: wide area location information and indoor location information,

The step of constructing the indoor positioning training data set owned by each terminal device includes:

Sequentially extracting target indoor position information corresponding to each piece of position fingerprint information collected by the terminal device from the indoor position information;

Using the location fingerprint information, the target indoor location information, and the wide area location information to which the target indoor location information belongs respectively as a piece of indoor positioning training data;

All the indoor positioning training data are collected to construct an indoor positioning training data set of each terminal device.
18. The storage medium of claim 18, wherein the wide area location information is physical location information, and the indoor location information is coordinate information using the wide area location information as a reference point.
The storage medium according to claim 16, wherein the step of performing model training for each of the terminal devices based on the indoor positioning training data set to obtain model parameter updates includes:

Detecting a model update request for each terminal device to perform model training;

According to the detected model update request, each of the terminal devices performs model training locally to obtain model parameter updates.