Model training method and device based on wearable system features, equipment and medium

Through digital twin technology and correlation screening and cluster evaluation methods, the problems of massive data selection and quantitative analysis are solved, the user's portrait is improved, the system burden is reduced and the model accuracy is improved.

CN118885809APending Publication Date: 2024-11-01CHINA MOBILE COMM LTD RES INST +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410918902.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-10
Publication Date
2024-11-01

AI Technical Summary

Technical Problem

In the process of fusion of static data and dynamic data, we face the problem of massive data being unable to adaptively select and quantitative analysis, which leads to increased system burden and affects the refined optimization of user portraits.

Method used

Using a method based on digital twin technology, the user group portrait model is constructed and initialized, and massive data is adaptively selected and quantitatively analyzed through correlation screening and cluster evaluation, and weight coefficient assignment and cluster evaluation are combined with static and dynamic features. and optimize group portrait models.

Benefits of technology

It effectively avoids excessive consumption of computing power resources and loss of small sample data, improves model accuracy, makes the group portrait model more suitable for testing users, and reduces the system burden.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118885809A_ABST
    Figure CN118885809A_ABST
Patent Text Reader

Abstract

The invention provides a model training method and device based on wearable system features, equipment and a medium, and the method applied to a cloud server comprises the steps: constructing and initializing a group portrait model of a user according to a digital twin technology; a training data set corresponding to the group portrait model is obtained, the training data set comprises static information and motion states of multiple users and first feature data of the wearable system in each motion state, and the first feature data comprises multiple first user features and first feature parameters corresponding to each first user feature; performing correlation screening and clustering evaluation on the first feature data according to the static information and the motion state, and training a group portrait model; and sending the trained group portrait model to a terminal of a test user. According to the embodiment of the invention, self-adaptive selection and quantitative analysis are carried out on the mass data at the cloud server, so that excessive consumption of computing power resources can be avoided, and loss of small sample data can also be avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data service technology, and in particular to a model training method, device, equipment and medium based on wearable system characteristics. Background Art

[0002] In the Internet era, user portraits are usually generated based on Internet user behavior habits and demographic informatics data, and play an important role in Internet business recommendations. Behavioral habits include shopping, browsing, searching, etc., and demographic informatics data includes age, gender, etc.

[0003] However, the user data that Internet user portraits are based on, whether it is demographic data such as age and gender, or behavioral habit data, are all obtained based on a period of time, and the ability to understand the real-time status of users is insufficient. That is, in different real-time states such as running, working, and leisure, the user status and needs may be different, and there is the possibility of further refined optimization.

[0004] With the development of the Internet of Things, especially the body area network represented by wearable systems that monitor human vital signs in real time, it provides an important means to make up for the real-time nature of Internet user portraits. Therefore, dynamic portraits are proposed on the basis of static portraits. User portraits not only include static user characteristics based on relatively long-term data in the Internet era (static user portraits: including age, gender, etc.), but also include massive dynamic characteristics (dynamic user portraits) generated in real time by user vital signs and behaviors in the body area network era.

[0005] However, before the introduction of dynamic features, the number of static features was limited, and they were often described by exhausting all parameters. After the introduction of dynamic features, the combination of parameters with static features produced a large number of feature parameters, which could not be selected adaptively, bringing burden to subsequent analysis and increasing the burden on the system. Summary of the invention

[0006] The purpose of the present invention is to provide a model training method, device, equipment and medium based on wearable system characteristics, which is used to solve the problem of inability to adaptively select and quantitatively analyze massive data in the process of fusion of static data and dynamic data.

[0007] In order to solve the above technical problems, an embodiment of the present invention provides a model training method based on wearable system features, which is applied to a cloud server. The method includes:

[0008] Based on digital twin technology, build and initialize the user group portrait model;

[0009] Acquire a training data set corresponding to the group portrait model, the training data set including static information, motion states, and first feature data of a wearable system in each of the motion states of a plurality of users, the first feature data including a plurality of first user features and a first feature parameter corresponding to each of the first user features;

[0010] Performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model;

[0011] The trained group portrait model is sent to the test user's terminal.

[0012] Optionally, performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, and training the group portrait model includes:

[0013] According to the static information and the motion state, the first feature data is respectively subjected to correlation screening to obtain second feature data, where the second feature data includes a plurality of second user features and a second feature parameter corresponding to each of the second user features;

[0014] Assigning a value to the second feature data to obtain a primary weight coefficient of each of the second user features;

[0015] According to a preset arrangement order, clustering and cluster evaluation are performed on the second feature parameters corresponding to each of the second user features in turn to obtain a secondary weight coefficient of each of the second user features;

[0016] The second user features in the second feature data are sorted according to the first-level weight coefficient and the second-level weight coefficient to train the group portrait model.

[0017] Optionally, according to the static information and the motion state, respectively performing correlation screening on the first characteristic data to obtain second characteristic data includes:

[0018] Performing correlation calculation based on the static information and the first feature data to obtain a first correlation coefficient corresponding to each of the first user features;

[0019] Filter the first feature data according to the first correlation coefficient to obtain third feature data, where the third feature data includes a plurality of third user features and a third feature parameter corresponding to each of the third user features;

[0020] Performing correlation calculation based on the motion state and the third feature data to obtain a second correlation coefficient corresponding to each of the third user features;

[0021] The third characteristic data is screened according to the second correlation coefficient to obtain second characteristic data.

[0022] Optionally, assigning a value to the second feature data to obtain a primary weight coefficient of each second user feature includes:

[0023] generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0024] Calculate according to the first correlation coefficient and the second correlation coefficient to obtain a comprehensive correlation coefficient of each of the second user features;

[0025] According to the descending order of the comprehensive correlation coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a primary weight coefficient of each second user feature.

[0026] Optionally, clustering and clustering evaluation are performed on the second feature parameters corresponding to each of the second user features in sequence according to a preset arrangement order to obtain a secondary weight coefficient of each of the second user features, including:

[0027] According to a preset arrangement order, sequentially performing adaptive density clustering on the second feature parameters of the plurality of users corresponding to the second user feature, to obtain clustering data of each second user feature;

[0028] Calculating a silhouette coefficient according to the clustering data to obtain a silhouette coefficient of each of the second user features;

[0029] The second user features are assigned values ​​according to the silhouette coefficients to obtain a secondary weight coefficient of each of the second user features.

[0030] Optionally, according to a preset arrangement order, adaptive density clustering is performed on the second feature parameters of the plurality of users corresponding to the second user feature in sequence to obtain clustering data of each second user feature, including:

[0031] Normalizing the second characteristic parameter to obtain normalized data;

[0032] According to a preset arrangement order and a density clustering algorithm, second feature parameters of the plurality of users corresponding to each of the second user features are calculated in turn to determine a type of the second feature parameter, the type including a core point and a non-core point;

[0033] The second feature parameters are clustered according to the core points to obtain cluster data of each of the second user features.

[0034] Optionally, the clustering data includes a plurality of clustering clusters;

[0035] The step of calculating the silhouette coefficient according to the clustering data to obtain the silhouette coefficient of each of the second user features includes:

[0036] Performing similarity calculation between the plurality of clusters in the clustering data of each second user feature to obtain a similarity coefficient of each second user feature;

[0037] Calculating dissimilarity between the plurality of clusters in the clustering data of each second user feature to obtain a dissimilarity coefficient of each second user feature;

[0038] A silhouette coefficient is calculated according to the similarity coefficient and the dissimilarity coefficient to obtain a silhouette coefficient of each of the second user features.

[0039] Optionally, assigning a value to the second user feature according to the silhouette coefficient to obtain a secondary weight coefficient of each second user feature includes:

[0040] generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0041] According to the descending order of the silhouette coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a secondary weight coefficient of each second user feature.

[0042] Optionally, the method further comprises:

[0043] Receiving sample data sent by the terminal;

[0044] The sample data is added to the training data set to update the training data set.

[0045] The embodiment of the present invention further provides a model training method based on wearable system features, which is applied to a terminal, and the method includes:

[0046] Get the trained group portrait model sent by the cloud server;

[0047] Acquire a test data set of a test user, wherein the test data set includes static information, motion state, and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features;

[0048] The group portrait model is corrected according to the test data set and the error correction algorithm, and the group portrait model is optimized.

[0049] Optionally, the group portrait model is corrected according to the test data set and the error correction algorithm, including:

[0050] According to the static information in the test data set and the fourth characteristic data, the group portrait model is corrected once according to an error correction algorithm;

[0051] According to the static information, the motion state and the fourth characteristic data in the test data set, the group portrait model is corrected secondary according to an error correction algorithm.

[0052] Optionally, the group portrait model is corrected once according to the static information and the fourth feature data in the test data set according to an error correction algorithm, including:

[0053] Inputting the static information of the test user into the group portrait model to obtain fifth feature data of multiple users corresponding to the static information, wherein the fifth feature data includes multiple test user features and a fifth feature parameter corresponding to each of the test user features;

[0054] Respectively calculating the mean of the fifth feature parameters of multiple users corresponding to each of the test user features to obtain the first prediction feature parameter corresponding to each of the test user features;

[0055] According to the first prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected once according to an error correction algorithm.

[0056] Optionally, according to the static information, the motion state and the fourth feature data in the test data set, the group portrait model is corrected twice according to an error correction algorithm, including:

[0057] Inputting the static information and the motion state of the test user into the modified group portrait model, clustering the test user, and obtaining the user category of the test user;

[0058] Acquire sixth feature data of a plurality of users belonging to the user category, the sixth feature data comprising a plurality of test user features and a sixth feature parameter corresponding to each of the test user features;

[0059] Respectively calculating the mean of the sixth feature parameters of multiple users corresponding to each of the test user features to obtain a second prediction feature parameter corresponding to each of the test user features;

[0060] According to the second prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected secondary according to an error correction algorithm.

[0061] Optionally, the method further comprises:

[0062] Comparing the data in the test data set with the data in the training data set to determine whether there is data in the test data set that is different from the training data set, wherein the training data set is a data set for training the group portrait model;

[0063] When there is data in the test data set that is different from the training data set, determining the data as sample data, and counting the number of the sample data;

[0064] When the amount of the sample data is greater than a first threshold, the sample data is sent to a cloud server.

[0065] An embodiment of the present invention further provides a network device, including a transceiver and a processor, wherein the processor is configured to:

[0066] Based on digital twin technology, build and initialize the user group portrait model;

[0067] Acquire a training data set corresponding to the group portrait model, the training data set including static information, motion states, and first feature data of a wearable system in each of the motion states of a plurality of users, the first feature data including a plurality of first user features and a first feature parameter corresponding to each of the first user features;

[0068] Performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model;

[0069] The transceiver is used to send the trained group portrait model to the terminal of the test user.

[0070] An embodiment of the present invention further provides a network device, including a transceiver and a processor, wherein the transceiver is used to obtain a trained group portrait model sent by a cloud server;

[0071] The processor is configured to:

[0072] Acquire a test data set of a test user, wherein the test data set includes static information, motion state, and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features;

[0073] The group portrait model is corrected according to the test data set and the error correction algorithm, and the group portrait model is optimized.

[0074] The embodiment of the present invention further provides a model training device based on wearable system features, which is applied to a cloud server and includes:

[0075] The first building module is used to build and initialize a user group portrait model based on the digital twin technology;

[0076] A first acquisition module is used to acquire a training data set corresponding to the group portrait model, wherein the training data set includes static information and motion states of multiple users and first feature data of the wearable system in each of the motion states, wherein the first feature data includes multiple first user features and a first feature parameter corresponding to each of the first user features;

[0077] A first training module, configured to perform correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model;

[0078] The first sending module is used to send the trained group portrait model to the terminal of the test user.

[0079] The embodiment of the present invention further provides a model training device based on wearable system features, which is applied to a terminal and includes:

[0080] The second acquisition module is used to obtain the trained group portrait model sent by the cloud server;

[0081] A third acquisition module is used to acquire a test data set of a test user, wherein the test data set includes static information, motion state and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features;

[0082] The first correction module is used to correct the group portrait model according to the test data set and the error correction algorithm, and optimize the group portrait model.

[0083] An embodiment of the present invention also provides a network device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the program implements the model training method based on wearable system features as described in any one of the above items.

[0084] An embodiment of the present invention also provides a readable storage medium, comprising: a program is stored on the readable storage medium, and when the program is executed by a processor, the steps of the model training method based on wearable system features as described in any one of the above items are implemented.

[0085] An embodiment of the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the model training method based on wearable system features as described in any of the above items.

[0086] At least one of the above technical solutions of the present invention has the following beneficial effects:

[0087] In the above scheme, when training the group portrait model, for the massive data in the training data set, the cloud server is first used to perform correlation screening and clustering evaluation on the training data set, that is, adaptive selection and quantitative analysis of the massive data can avoid excessive consumption of computing resources and loss of small sample data. The trained group portrait model is then sent to the test user's terminal, and the group portrait model is optimized on the terminal side. First, the group portrait model is corrected once according to the test user's static information, and then, the group portrait model is corrected twice according to the test user's dynamic information, so that the optimized group portrait model is more suitable for test user applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 A schematic diagram of a flow chart of a model training method based on wearable system features according to one embodiment of the present invention;

[0089] Figure 2 This is a flow chart of a model training method based on wearable system features according to an embodiment of the present invention;

[0090] Figure 3 The second flowchart of the model training method based on the wearable system characteristics according to the embodiment of the present invention is shown;

[0091] Figure 4 The third flowchart of the model training method based on the wearable system characteristics according to the embodiment of the present invention is shown;

[0092] Figure 5 A schematic flow chart of a model training method based on wearable system features according to another embodiment of the present invention;

[0093] Figure 6 The fourth flowchart of the model training method based on the wearable system characteristics according to the embodiment of the present invention is shown;

[0094] Figure 7 The fifth flowchart of the model training method based on the wearable system characteristics according to the embodiment of the present invention is shown;

[0095] Figure 8 The sixth flowchart of the model training method based on the wearable system characteristics according to the embodiment of the present invention is shown;

[0096] Fig. 9 A schematic diagram of the structure of a network device according to an embodiment of the present invention;

[0097] Fig.10 It is a schematic diagram of the structure of a terminal according to an embodiment of the present invention;

[0098] Fig.11 It is a structural schematic diagram of a model training device based on wearable system characteristics according to one embodiment of the present invention;

[0099] Fig.12 This is a schematic structural diagram of a model training device based on wearable system features according to another embodiment of the present invention. DETAILED DESCRIPTION

[0100] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0101] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0102] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0103] It is worth noting that the technology described in the embodiments of the present application is not limited to the Long Term Evolution (LTE) / LTE-Advanced (LTE-A) system, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency Division Multiple Access (SC-FDMA) and other systems. The terms "system" and "network" in the embodiments of the present application are often used interchangeably, and the described technology can be used for the above-mentioned systems and radio technologies as well as for other systems and radio technologies. The following description describes a new radio (NR) system for example purposes, and NR terms are used in most of the following descriptions, but these technologies can also be applied to applications other than NR system applications, such as the 6th Generation (6G) communication system.

[0104] The following description provides examples and does not limit the scope, applicability, or configuration set forth in the claims. Changes may be made to the functions and arrangements of the elements discussed without departing from the spirit and scope of the present disclosure. Various examples may appropriately omit, replace, or add various procedures or components. For example, the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. In addition, features described with reference to certain examples may be combined in other examples.

[0105] like Figure 1As shown, an embodiment of the present invention provides a model training method based on wearable system features, which is applied to a cloud server, and the method includes:

[0106] Step S101, constructing and initializing a user group portrait model based on digital twin technology;

[0107] In step S101, with the development of the Internet, the application of digital twin technology is becoming more and more extensive, and the application of user portraits based on digital twin technology is also becoming more and more extensive. The Internet can accurately push users based on user data and behavioral habits in user portraits. In the embodiment of the present invention, based on digital twin technology, a group portrait model of multiple users is constructed and initialized.

[0108] Step S102, obtaining a training data set corresponding to the group portrait model, wherein the training data set includes static information, motion states, and first feature data of the wearable system in each of the motion states of multiple users, wherein the first feature data includes multiple first user features and a first feature parameter corresponding to each of the first user features.

[0109] In step S102, the static information includes but is not limited to one or more of information such as age, gender, height and weight, and the motion state includes but is not limited to walking, running, going up and down stairs, etc. The wearable system is a system that can be worn by the user. When the user wears it, it can monitor the user in real time, obtain multiple first user features of the user and the first feature parameters corresponding to each of the first user features. For example, in the medical field, the wearable system can be a wearable gait monitoring system, which is convenient for obtaining the real-time gait number of the patient. At this time, the first user feature includes but is not limited to one or more of time domain features, space domain features, energy domain features and frequency domain features, wherein the time domain features include but are not limited to the average gait cycle, cycle duration, etc., the space domain features include but are not limited to step length, displacement, displacement speed, etc., the energy domain features include but are not limited to driving ability, braking ability, etc., and the frequency domain features include but are not limited to ground count, etc.

[0110] It should be noted that the motion state may be pre-set or determined based on the first characteristic parameter collected by the wearable system.

[0111] Step S103, performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model;

[0112] In step S103, by training the group portrait model with massive static and dynamic data in the training data set, a more accurate portrait model can be obtained with higher accuracy. However, massive data also brings the problem of information overload, which brings a burden to the system. Therefore, in the embodiment of the present invention, the massive data in the training data set is firstly subjected to correlation screening and clustering evaluation, that is, adaptive selection and quantitative analysis of the massive data is performed, which can avoid excessive consumption of computing resources and the loss of small sample data.

[0113] Step S104: sending the trained group portrait model to the test user's terminal.

[0114] In step S104, the trained group portrait model is sent to the terminal of the test user, and the group portrait model is personalized and optimized on the terminal side, so that the optimized group portrait model is more suitable for the test user application.

[0115] In an embodiment of the present invention, during the model training process, correlation screening and clustering evaluation are first performed on massive data, that is, adaptive selection and quantitative analysis are performed on massive data, which can avoid excessive consumption of computing resources and reduce system burden, as well as avoid the loss of small sample data and improve model accuracy. The trained group portrait model is then sent to the test user's terminal, and the group portrait model is personalized optimized for the test user on the terminal side, so that the optimized group portrait model is more suitable for test user applications.

[0116] Optionally, performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, and training the group portrait model includes:

[0117] According to the static information and the motion state, the first feature data is respectively subjected to correlation screening to obtain second feature data, where the second feature data includes a plurality of second user features and a second feature parameter corresponding to each of the second user features;

[0118] Assigning a value to the second feature data to obtain a primary weight coefficient of each of the second user features;

[0119] According to a preset arrangement order, clustering and cluster evaluation are performed on the second feature parameters corresponding to each of the second user features in turn to obtain a secondary weight coefficient of each of the second user features;

[0120] The second user features in the second feature data are sorted according to the first-level weight coefficient and the second-level weight coefficient to train the group portrait model.

[0121] In an embodiment of the present invention, a method for processing massive data in a training data set is described. First, correlation screening is performed on the training data set. The first feature data is screened according to the correlation between static information and the first feature data and the correlation between different motion states and the first feature data, and only data with higher correlation is retained, taking into account both static and dynamic dual correlations, greatly reducing the amount of data, and avoiding the loss of important small sample data. Then, the screened second feature data is assigned a value to obtain a first-level weight coefficient corresponding to the second user feature. Then, according to a preset arrangement order, the second feature data is clustered, clustered, evaluated, and assigned a second time to obtain a second-level weight coefficient of the second user feature, wherein, in general, the preset arrangement order is in descending order according to the first-level weight coefficient; secondly, the first-level weight coefficient and the second-level weight coefficient are calculated to obtain a comprehensive weight coefficient of the second user feature, wherein the comprehensive weight coefficient = the first-level weight coefficient x the second-level weight coefficient. Finally, the second user feature and its corresponding second feature parameter are sorted in descending order according to the comprehensive weight coefficient, and the training data set is updated. The group portrait model is trained based on the updated training data set, so that the trained group portrait model is more accurate.

[0122] Optionally, according to the static information and the motion state, respectively performing correlation screening on the first characteristic data to obtain second characteristic data includes:

[0123] Performing correlation calculation based on the static information and the first feature data to obtain a first correlation coefficient corresponding to each of the first user features;

[0124] Filter the first feature data according to the first correlation coefficient to obtain third feature data, where the third feature data includes a plurality of third user features and a third feature parameter corresponding to each of the third user features;

[0125] Performing correlation calculation based on the motion state and the third feature data to obtain a second correlation coefficient corresponding to each of the third user features;

[0126] The third characteristic data is screened according to the second correlation coefficient to obtain second characteristic data.

[0127] In the embodiment of the present invention, Figure 2 As shown, when the training data set is screened for relevance, the static information and the first feature data of each user are first traversed to calculate the relevance therebetween. The calculation formula is: Among them, X i,k represents the i-th item of data in the static information of the k-th user, Y j,k represents the first feature parameter corresponding to the j-th first user feature of the k-th user, Represents the average value of the ith item of data in the static information (for example, the ith item of data is age, represents the average age of all users in the training data set), represents the average value of the first feature parameters of all users in the training data set corresponding to the jth first user feature in the first feature data, N is the number of users in the training data set, r i,j represents the first correlation coefficient between the i-th item of data in the static information and the j-th first user feature;

[0128] Then, for each data in the static information, calculate the mean of its correlation coefficient, and the calculation formula is: L is the number of first user features, is the mean correlation coefficient of the i-th data in the static information;

[0129] Secondly, select the data with the largest mean correlation coefficient as the most relevant data in the static information. Finally, under the most relevant data, select the correlation coefficient r i,j The first feature data that is greater than the second threshold is used as the third feature data.

[0130] The third characteristic data is screened again. First, each motion state and the third characteristic data of each user are traversed to calculate the correlation between them. The calculation formula is: Among them, X m,k represents the motion data of the kth user in the mth motion state (for example, the mth motion state is running, and the corresponding motion data is the running speed), Y j,k represents the third feature parameter corresponding to the j-th third user feature of the k-th user, represents the average value of the motion data of all users in the training data set in the mth motion state, represents the average value of the third feature parameters of all users in the training data set corresponding to the jth third user feature in the third feature data, N is the number of users in the training data set, and r m,j represents the second correlation coefficient between the m-th motion state and the j-th third user feature;

[0131] Then, the third user feature and the corresponding third feature parameter whose correlation coefficient in each motion state is greater than the third threshold are selected and determined as the second feature data.

[0132] The embodiment of the present invention takes into account both static and dynamic dual correlations to screen the massive data in the training data set, retaining only data with higher correlation, thereby greatly reducing the amount of data.

[0133] Optionally, assigning a value to the second feature data to obtain a primary weight coefficient of each second user feature includes:

[0134] generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0135] Calculate according to the first correlation coefficient and the second correlation coefficient to obtain a comprehensive correlation coefficient of each of the second user features;

[0136] According to the descending order of the comprehensive correlation coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a primary weight coefficient of each second user feature.

[0137] In the embodiment of the present invention, if the number of second user features is M, then a sequence S1 of equally spaced values ​​of M is generated in the range (0,1] and arranged in descending order. Then, the second user features and their corresponding second feature parameters are sorted in descending order according to the comprehensive correlation coefficient, where the comprehensive correlation coefficient = the first correlation coefficient x the second correlation coefficient. Finally, each value in S1 corresponding to the order of the second user feature is used as the first-level weight coefficient of the second user feature.

[0138] Optionally, clustering and clustering evaluation are performed on the second feature parameters corresponding to each of the second user features in sequence according to a preset arrangement order to obtain a secondary weight coefficient of each of the second user features, including:

[0139] According to a preset arrangement order, sequentially performing adaptive density clustering on the second feature parameters of the plurality of users corresponding to the second user feature, to obtain clustering data of each second user feature;

[0140] Calculating a silhouette coefficient according to the clustering data to obtain a silhouette coefficient of each of the second user features;

[0141] The second user features are assigned values ​​according to the silhouette coefficients to obtain a secondary weight coefficient of each of the second user features.

[0142] In the embodiment of the present invention, adaptive density clustering is performed on the data to obtain cluster data, and data clustering of user groups is achieved based on the distribution density of the data. The method does not need to pre-specify the number of clusters, and is automatically generated completely according to the distribution law of the data itself, which can better achieve the division of gait portraits of the crowd, and the elastic adaptive selection of clustering granularity can not only avoid the occupation of resources by parameters that have no clustering separability, but also avoid a one-size-fits-all approach and avoid filtering of small samples but important information;

[0143] Then, cluster evaluation is performed on the cluster data to obtain the silhouette coefficient of each second user feature, and then, a value is assigned according to the silhouette coefficient to obtain the secondary weight coefficient of each second user feature.

[0144] Optionally, according to a preset arrangement order, adaptive density clustering is performed on the second feature parameters of the plurality of users corresponding to the second user feature in sequence to obtain clustering data of each second user feature, including:

[0145] Normalizing the second characteristic parameter to obtain normalized data;

[0146] According to a preset arrangement order and a density clustering algorithm, second feature parameters of the plurality of users corresponding to each of the second user features are calculated in turn to determine a type of the second feature parameter, the type including a core point and a non-core point;

[0147] The second feature parameters are clustered according to the core points to obtain cluster data of each of the second user features.

[0148] In the embodiment of the present invention, Figure 3 As shown, in order to facilitate subsequent calculations, firstly select a second user feature in turn, normalize the second feature parameters of all users corresponding to the second user feature, and normalize the data range to the range of 0-1. The normalization formula is: Among them, Y m is the second feature parameter of the mth user corresponding to the selected second user feature, Y min is the minimum second feature parameter corresponding to the selected second user feature, Y max is the maximum second feature parameter corresponding to the selected second user feature, Normalized data of the second feature parameter of the mth user corresponding to the selected second user feature;

[0149] Then, according to the preset arrangement order, a second user feature is selected in turn, and clustering calculation is performed on the second feature parameters of all users corresponding to the second user feature. The specific clustering calculation process is:

[0150] The first step is to set the basic parameter values ​​of density clustering: neighborhood radius ε, used to determine the search range; the minimum number of points MinPts, used to determine whether the search point is a core point;

[0151] The second step is to traverse the second feature parameters of all users corresponding to the currently selected second user feature, calculate the distance between other second feature parameters and the currently selected second feature parameter, and retain the second feature parameters whose distance is less than ε;

[0152] Step 3: If the number of retained parameters corresponding to the currently selected second feature parameter is greater than or equal to MinPts, the currently selected second feature parameter is determined to be a core point; if the number of retained parameters corresponding to the currently selected second feature parameter is less than MinPts, the currently selected second feature parameter is determined to be a non-core point;

[0153] Step 4: If the second characteristic parameter is a non-core point, further determine whether the point is located in the ε region of other core points. If yes, determine it as a boundary point; if not, determine it as a noise point.

[0154] The fifth step is to select a core point in turn after determining the types of all the second characteristic parameters, recursively visit all its density-reachable points (i.e., the second characteristic parameters in the ε region), and divide them into a cluster category to determine the cluster cluster until all core points and boundary points are divided into corresponding cluster clusters, and only noise points are not classified.

[0155] In addition, after the clustering calculation of the second feature parameters corresponding to all the second user features is completed, it is also necessary to calculate the boundary mean of the most relevant indicators between different clusters for subsequent application. The calculation formula is: Among them, X r,e,m represents the most relevant index boundary value between the rth cluster and the eth cluster corresponding to the mth selected second user feature, is the obtained boundary mean.

[0156] Optionally, the clustering data includes a plurality of clustering clusters;

[0157] The step of calculating the silhouette coefficient according to the clustering data to obtain the silhouette coefficient of each of the second user features includes:

[0158] Performing similarity calculation between the plurality of clusters in the clustering data of each second user feature to obtain a similarity coefficient of each second user feature;

[0159] Calculating dissimilarity between the plurality of clusters in the clustering data of each second user feature to obtain a dissimilarity coefficient of each second user feature;

[0160] A silhouette coefficient is calculated according to the similarity coefficient and the dissimilarity coefficient to obtain a silhouette coefficient of each of the second user features.

[0161] In the embodiment of the present invention, Figure 4 As shown, intra-cluster similarity calculation and intra-cluster step similarity calculation are performed for multiple clusters of each second user feature to obtain the silhouette coefficient corresponding to each second user feature. The specific calculation process is:

[0162] The first step is to select a second user feature in turn, and obtain the cluster and the data within the cluster corresponding to the second user feature;

[0163] In the second step, for the selected second user feature, select a cluster in turn and calculate its intra-cluster similarity coefficient. The calculation formula is: Among them, C i is the number of data in the i-th cluster in the m-th selected second user feature, d(i,j) is the j-th data in the i-th cluster, a m (i) is the calculated intra-cluster similarity coefficient of the mth selected second user feature;

[0164] The third step is to calculate the inter-cluster dissimilarity coefficient for the selected second user feature. The calculation formula is: Among them, b m (i) represents the minimum average distance between a certain data in the m-th selected second user feature and the data in other clusters;

[0165] The fourth step is to calculate the clustering silhouette coefficient for the selected second user feature. The calculation formula is: if|C i |>1, where s m (i) represents the silhouette coefficient of the mth selected second user feature.

[0166] Optionally, assigning a value to the second user feature according to the silhouette coefficient to obtain a secondary weight coefficient of each second user feature includes:

[0167] generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0168] According to the descending order of the silhouette coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a secondary weight coefficient of each second user feature.

[0169] In the embodiment of the present invention, if the number of second user features is M, then a sequence S2 of equally spaced values ​​of M is generated in the range (0, 1] and arranged in descending order. Then, the second user features and their corresponding second feature parameters are sorted in descending order according to the silhouette coefficients. Finally, each value in S2 corresponding to the order of the second user feature is used as the secondary weight coefficient of the second user feature.

[0170] Optionally, the method further comprises:

[0171] Receiving sample data sent by the terminal;

[0172] The sample data is added to the training data set to update the training data set.

[0173] In an embodiment of the present invention, when sample data sent by the terminal of the test user is received, it means that the terminal side has collected new sample data about the test user, adds the sample data to the training data set, updates the training data set, and retrains the group portrait model. The iterative update makes the group portrait model more refined.

[0174] like Figure 5 As shown, an embodiment of the present invention further provides a model training method based on wearable system features, which is applied to a terminal, and the method includes:

[0175] Step S501, obtaining the trained group portrait model sent by the cloud server;

[0176] In step S501, the terminal periodically downloads the group portrait model from the cloud server;

[0177] Step S502, obtaining a test data set of a test user, wherein the test data set includes static information, motion state, and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features;

[0178] In step S502, one terminal corresponds to one test user, and a test data set of the test user is obtained, wherein the static information includes but is not limited to one or more of information such as age, gender, height and weight, and the motion state includes but is not limited to walking, running, going up and down stairs, etc. The wearable system is a system wearable by the test user. When the test user wears it, it can monitor the test user in real time, obtain multiple fourth user features of the test user and the fourth feature parameters corresponding to each fourth user feature. For example, in the medical field, the wearable system can be a wearable gait monitoring system, which is convenient for obtaining the real-time gait number of the patient. At this time, the fourth user feature includes but is not limited to one or more of time domain features, space domain features, energy domain features and frequency domain features, wherein the time domain features include but are not limited to average gait cycle, cycle duration, etc., the space domain features include but are not limited to step length, displacement, displacement speed, etc., the energy domain features include but are not limited to driving ability, braking ability, etc., and the frequency domain features include but are not limited to landing counts, etc.

[0179] It should be noted that the motion state can be pre-set or determined according to the fourth characteristic parameter collected by the wearable system.

[0180] Step S503, correcting the group portrait model according to the test data set and the error correction algorithm, and optimizing the group portrait model.

[0181] In step S503, correction, i.e., personalized optimization, is performed based on the test data set of the test user and the group portrait model of the error correction algorithm.

[0182] In the embodiment of the present invention, the terminal and the cloud server cooperate with each other, the terminal regularly obtains the group portrait model and optimizes the model through personalization through correction, so that the optimized group portrait model is more suitable for test users and the result is more accurate.

[0183] Optionally, the group portrait model is corrected according to the test data set and the error correction algorithm, including:

[0184] According to the static information in the test data set and the fourth characteristic data, the group portrait model is corrected once according to an error correction algorithm;

[0185] According to the static information, the motion state and the fourth characteristic data in the test data set, the group portrait model is corrected secondary according to an error correction algorithm.

[0186] In an embodiment of the present invention, firstly, a static correction is made to the group portrait model according to the static information and the fourth characteristic data in the test data set and an error correction algorithm, so that the group portrait model is more suitable for the test user. Then, a dynamic secondary correction is made to the group portrait model according to the error correction algorithm in combination with the motion state in the test data set. Through the static and dynamic two-level correction, the feature selection of the limited resources of the terminal is taken into account, so that the model is more accurate and more personalized.

[0187] Optionally, the group portrait model is corrected once according to the static information and the fourth feature data in the test data set according to an error correction algorithm, including:

[0188] Inputting the static information of the test user into the group portrait model to obtain fifth feature data of multiple users corresponding to the static information, wherein the fifth feature data includes multiple test user features and a fifth feature parameter corresponding to each of the test user features;

[0189] Respectively calculating the mean of the fifth feature parameters of multiple users corresponding to each of the test user features to obtain the first prediction feature parameter corresponding to each of the test user features;

[0190] According to the first prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected once according to an error correction algorithm.

[0191] In the embodiment of the present invention, Figure 6 As shown in the figure, a static correction is made to the group portrait model. The correction process is as follows:

[0192] The first step is to input the static information of the test user into the group portrait model to obtain the fifth feature data corresponding to the group portrait model and the static information of the test user;

[0193] In the second step, the fifth characteristic parameters corresponding to each test user characteristic in the fifth characteristic data are averaged, and the calculation formula is: Among them, c i is the fifth feature parameter of the i-th test user feature corresponding to the j-th test user feature, N is the number of test user features, is the first predicted feature parameter of the j-th test user feature;

[0194] The third step is to perform error correction based on the first prediction feature parameter and the test feature parameter in the actual fourth feature data of the test user to obtain a prediction error value. c is the test feature parameter in the fourth feature data, Δ is the prediction error;

[0195] The fourth step is to input the prediction error into the deep learning error correction algorithm to obtain the corrected error prediction value. Where x is the static information in the model, F(·) is the deep learning error correction algorithm, Δ pred To correct the error prediction value;

[0196] Step 5: According to the correction error prediction value Δ pred Calculate with the actual error value to obtain the loss value. The calculation formula is: in, represents the parameter corresponding to the kth test user feature in the corrected error prediction value of the test user, Δ k Representation and The corresponding true error value, K is the number of clusters, N is the number of test user features, and L is the final loss value;

[0197] Step 6: During the correction process, determine whether the currently calculated loss value is lower than the average loss value of the previous fourth threshold (e.g., 10) times:

[0198] If the loss value decreases, it means that the model is still converging. The parameters of the deep learning error correction model are updated through the gradient descent method. In order to avoid the model training time being too long, the maximum training round is set, that is, the training is stopped after the maximum training round is reached;

[0199] If the loss value no longer decreases during the calculation of the fourth threshold, it indicates that the model can no longer converge, and the model update (correction) is stopped. Taking the fourth threshold equal to 10 as an example, the specific calculation formula is: Among them, L t The loss value calculated for the current training, is the average of the previous 10 loss values, and count is the number of consecutive counts of loss values ​​that no longer decrease during training, or a stop condition is triggered after reaching 10.

[0200] Optionally, according to the static information, the motion state and the fourth feature data in the test data set, the group portrait model is corrected twice according to an error correction algorithm, including:

[0201] Inputting the static information and the motion state of the test user into the modified group portrait model, clustering the test user, and obtaining the user category of the test user;

[0202] Acquire sixth feature data of a plurality of users belonging to the user category, the sixth feature data comprising a plurality of test user features and a sixth feature parameter corresponding to each of the test user features;

[0203] Respectively calculating the mean of the sixth feature parameters of multiple users corresponding to each of the test user features to obtain a second prediction feature parameter corresponding to each of the test user features;

[0204] According to the second prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected secondary according to an error correction algorithm.

[0205] In the embodiment of the present invention, Figure 7 As shown in the figure, the group portrait model is dynamically corrected twice, and the correction process is:

[0206] The first step is to input the static information and the motion state of the test user into the modified group portrait model, and cluster the test user according to the cluster boundary calculated above to obtain the user category of the test user;

[0207] The second step is to obtain the sixth characteristic data of multiple users belonging to the above user category in the group portrait model;

[0208] The third step is to calculate the mean of the sixth feature parameter within the 95% confidence interval corresponding to each test user feature in the sixth feature data, obtain the second predicted feature parameter corresponding to each test user feature, and then generate an individual twin static portrait;

[0209] The fourth step is to obtain the corrected error prediction value Δ based on the error correction algorithm according to the second prediction feature parameter in the individual twin static portrait and the test feature parameter in the fourth feature data of the test user. pred , and then obtain the corrected characteristic parameters based on the second predicted characteristic parameters to achieve secondary correction of the group portrait model. The calculation formula is: c corrected =c initial +Δ pred , where c initial represents the second prediction feature parameter, c corrected Represents the corrected characteristic parameters.

[0210] Optionally, the method further comprises:

[0211] Using the second corrected group portrait model, static information of the test user and feature data of the wearable system are obtained, wherein the feature data includes a plurality of user features and feature parameters corresponding to each of the user features;

[0212] Performing correlation calculation based on the static information and the characteristic data of the wearable system to obtain a third correlation coefficient;

[0213] The characteristic data of the wearable system are sorted according to the third correlation coefficient.

[0214] In an embodiment of the present invention, when applying the group portrait model after secondary correction, the data of the test user is obtained, and the correlation between each user feature and the static information is calculated respectively, and the user features and corresponding parameters are sorted according to the correlation, which is beneficial to provide important indicators for the test user and is more convenient.

[0215] Optionally, the method further comprises:

[0216] Comparing the data in the test data set with the data in the training data set to determine whether there is data in the test data set that is different from the training data set, wherein the training data set is a data set for training the group portrait model;

[0217] When there is data in the test data set that is different from the training data set, determining the data as sample data, and counting the number of the sample data;

[0218] When the amount of the sample data is greater than a first threshold, the sample data is sent to a cloud server.

[0219] In an embodiment of the present invention, the data of the test user is acquired in real time through the wearable system, and the test data set is continuously updated. At the same time, the latest data in the test data set is compared with the data in the training data set. If they are different, the latest data in the test data set is recorded as newly added sample data, and the number of newly added sample data is counted. When the number of sample data is greater than a first threshold, the model update mechanism is started, and the sample data is sent to the cloud server, so that the cloud server retrains the group portrait model according to the sample data, and the group portrait model is made more refined through iterative updates.

[0220] like Figure 8 As shown, the method provided by the embodiment of the present invention can continuously iterate and update the group portrait model through the cooperation between the cloud and the terminal, so that the group portrait model is more refined. The iterative update process is described below through an embodiment:

[0221] Step S801, the cloud server trains a group portrait model according to the training data set;

[0222] Step S802, the terminal of the test user periodically obtains a group portrait model from the cloud server;

[0223] Step S803, the terminal obtains the test data set of the test user in real time, and modifies the group portrait model according to the test data set;

[0224] Step S804, determine whether there is data in the test data set obtained by the terminal that is different from the training data set? If yes, go to step S805; if not, go to step S803;

[0225] Step S805, determining the data in the test data set that is different from the training data set as new sample data;

[0226] Step S806, determine whether the number of sample data is greater than the first threshold. If yes, go to step S807; if no, go to step S804;

[0227] Step S807, the terminal sends sample data to the cloud server;

[0228] Step S808: The cloud server updates the training data set according to the sample data.

[0229] like Fig. 9 As shown, the embodiment of the present invention further provides a network device 900, including a transceiver 901 and a processor 902, wherein the processor 902 is configured to:

[0230] Based on digital twin technology, build and initialize the user group portrait model;

[0231] Acquire a training data set corresponding to the group portrait model, the training data set including static information, motion states, and first feature data of a wearable system in each of the motion states of a plurality of users, the first feature data including a plurality of first user features and a first feature parameter corresponding to each of the first user features;

[0232] Performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model;

[0233] The transceiver 901 is used to send the trained group portrait model to the terminal of the test user.

[0234] Optionally, the processor 902 is further configured to:

[0235] According to the static information and the motion state, the first feature data is respectively subjected to correlation screening to obtain second feature data, where the second feature data includes a plurality of second user features and a second feature parameter corresponding to each of the second user features;

[0236] Assigning a value to the second feature data to obtain a primary weight coefficient of each of the second user features;

[0237] According to a preset arrangement order, clustering and cluster evaluation are performed on the second feature parameters corresponding to each of the second user features in turn to obtain a secondary weight coefficient of each of the second user features;

[0238] The second user features in the second feature data are sorted according to the first-level weight coefficient and the second-level weight coefficient to train the group portrait model.

[0239] Optionally, the processor 902 is further configured to:

[0240] Performing correlation calculation based on the static information and the first feature data to obtain a first correlation coefficient corresponding to each of the first user features;

[0241] Filter the first feature data according to the first correlation coefficient to obtain third feature data, where the third feature data includes a plurality of third user features and a third feature parameter corresponding to each of the third user features;

[0242] Performing correlation calculation based on the motion state and the third feature data to obtain a second correlation coefficient corresponding to each of the third user features;

[0243] The third characteristic data is screened according to the second correlation coefficient to obtain second characteristic data.

[0244] Optionally, the processor 902 is further configured to:

[0245] generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0246] Calculate according to the first correlation coefficient and the second correlation coefficient to obtain a comprehensive correlation coefficient of each of the second user features;

[0247] According to the descending order of the comprehensive correlation coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a primary weight coefficient of each second user feature.

[0248] Optionally, the processor 902 is further configured to:

[0249] According to a preset arrangement order, sequentially performing adaptive density clustering on the second feature parameters of the plurality of users corresponding to the second user feature, to obtain clustering data of each second user feature;

[0250] Calculating a silhouette coefficient according to the clustering data to obtain a silhouette coefficient of each of the second user features;

[0251] The second user features are assigned values ​​according to the silhouette coefficients to obtain a secondary weight coefficient of each of the second user features.

[0252] Optionally, the processor 902 is further configured to:

[0253] Normalizing the second characteristic parameter to obtain normalized data;

[0254] According to a preset arrangement order and using a density clustering algorithm, the second feature parameters of the plurality of users corresponding to each of the second user features are calculated in turn to determine the type of the second feature parameters;

[0255] The second feature parameters are clustered according to the core points to obtain cluster data of each of the second user features.

[0256] Optionally, the processor 902 is further configured to:

[0257] Performing similarity calculation between the plurality of clusters in the clustering data of each second user feature to obtain a similarity coefficient of each second user feature;

[0258] Calculating dissimilarity between the plurality of clusters in the clustering data of each second user feature to obtain a dissimilarity coefficient of each second user feature;

[0259] A silhouette coefficient is calculated according to the similarity coefficient and the dissimilarity coefficient to obtain a silhouette coefficient of each of the second user features.

[0260] Optionally, the processor 902 is further configured to:

[0261] generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0262] According to the descending order of the silhouette coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a secondary weight coefficient of each second user feature.

[0263] Optionally, the transceiver 901 is further configured to receive sample data sent by the terminal;

[0264] The processor 902 is further configured to add the sample data to the training data set to update the training data set.

[0265] like Fig.10 As shown, an embodiment of the present invention further provides a terminal 1000, including a transceiver 1001 and a processor 1002, wherein the transceiver 1001 is used to obtain a trained group portrait model sent by a cloud server;

[0266] The processor 1002 is configured to:

[0267] Acquire a test data set of a test user, wherein the test data set includes static information, motion state, and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features;

[0268] The group portrait model is corrected according to the test data set and the error correction algorithm, and the group portrait model is optimized.

[0269] Optionally, the processor 1002 is further configured to:

[0270] According to the static information in the test data set and the fourth characteristic data, the group portrait model is corrected once according to an error correction algorithm;

[0271] According to the static information, the motion state and the fourth characteristic data in the test data set, the group portrait model is corrected secondary according to an error correction algorithm.

[0272] Optionally, the processor 1002 is further configured to:

[0273] Inputting the static information of the test user into the group portrait model to obtain fifth feature data of multiple users corresponding to the static information, wherein the fifth feature data includes multiple test user features and a fifth feature parameter corresponding to each of the test user features;

[0274] Respectively calculating the mean of the fifth feature parameters of multiple users corresponding to each of the test user features to obtain the first prediction feature parameter corresponding to each of the test user features;

[0275] According to the first prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected once according to an error correction algorithm.

[0276] Optionally, the processor 1002 is further configured to:

[0277] Inputting the static information and the motion state of the test user into the modified group portrait model, clustering the test user, and obtaining the user category of the test user;

[0278] Acquire sixth feature data of a plurality of users belonging to the user category, the sixth feature data comprising a plurality of test user features and a sixth feature parameter corresponding to each of the test user features;

[0279] Respectively calculating the mean of the sixth feature parameters of multiple users corresponding to each of the test user features to obtain a second prediction feature parameter corresponding to each of the test user features;

[0280] According to the second prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected secondary according to an error correction algorithm.

[0281] Optionally, the processor 1002 is further configured to:

[0282] Comparing the data in the test data set with the data in the training data set to determine whether there is data in the test data set that is different from the training data set, wherein the training data set is a data set for training the group portrait model;

[0283] When there is data in the test data set that is different from the training data set, determining the data as sample data, and counting the number of the sample data;

[0284] The transceiver 1001 is further configured to send the sample data to a cloud server when the amount of the sample data is greater than a first threshold.

[0285] like Fig.11 As shown, an embodiment of the present invention further provides a model training device based on wearable system features, which is applied to a cloud server and includes:

[0286] The first construction module 1101 is used to construct and initialize a group portrait model of users based on the digital twin technology;

[0287] A first acquisition module 1102 is used to acquire a training data set corresponding to the group portrait model, wherein the training data set includes static information and motion states of multiple users and first feature data of the wearable system in each of the motion states, wherein the first feature data includes multiple first user features and a first feature parameter corresponding to each of the first user features;

[0288] A first training module 1103 is used to perform correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, and train the group portrait model;

[0289] The first sending module 1104 is used to send the trained group portrait model to the terminal of the test user.

[0290] Optionally, the first training module 1103 includes:

[0291] A first screening unit is used to perform correlation screening on the first feature data according to the static information and the motion state to obtain second feature data, where the second feature data includes a plurality of second user features and a second feature parameter corresponding to each of the second user features;

[0292] A first assignment unit, used for assigning a value to the second feature data to obtain a primary weight coefficient of each second user feature;

[0293] A first clustering unit, configured to cluster and cluster evaluate the second feature parameters corresponding to each of the second user features in sequence according to a preset arrangement order, to obtain a secondary weight coefficient of each of the second user features;

[0294] The first sorting unit is used to sort the second user features in the second feature data according to the first-level weight coefficient and the second-level weight coefficient to train the group portrait model.

[0295] Optionally, the first screening unit comprises:

[0296] A first calculation unit, configured to perform correlation calculation based on the static information and the first feature data to obtain a first correlation coefficient corresponding to each of the first user features;

[0297] a second screening unit, configured to screen the first feature data according to the first correlation coefficient to obtain third feature data, wherein the third feature data includes a plurality of third user features and a third feature parameter corresponding to each of the third user features;

[0298] A second calculation unit, configured to perform correlation calculation based on the motion state and the third feature data to obtain a second correlation coefficient corresponding to each of the third user features;

[0299] The third screening unit is used to screen the third characteristic data according to the second correlation coefficient to obtain second characteristic data.

[0300] Optionally, the first assignment unit includes:

[0301] A first generating unit is used to generate an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0302] a third calculation unit, configured to perform calculation according to the first correlation coefficient and the second correlation coefficient to obtain a comprehensive correlation coefficient of each of the second user features;

[0303] The second assignment unit is used to assign values ​​to the second user features in sequence according to the order of the comprehensive correlation coefficients from large to small through the equally spaced numerical sequence to obtain a primary weight coefficient of each of the second user features.

[0304] Optionally, the first clustering unit includes:

[0305] A second clustering unit, configured to perform adaptive density clustering on the second feature parameters of the plurality of users corresponding to the second user feature in sequence according to a preset arrangement order, to obtain clustering data of each of the second user features;

[0306] a fourth calculation unit, configured to calculate a silhouette coefficient according to the clustering data to obtain a silhouette coefficient of each of the second user features;

[0307] The third assignment unit is used to assign values ​​to the second user features according to the silhouette coefficient to obtain a secondary weight coefficient of each second user feature.

[0308] Optionally, the second clustering unit includes:

[0309] A first processing unit, used for performing normalization processing on the second characteristic parameter to obtain normalized data;

[0310] A third clustering unit is used to calculate the second feature parameters of the plurality of users corresponding to each of the second user features in turn according to a preset arrangement order and a density clustering algorithm, and determine a type of the second feature parameters, wherein the type includes a core point and a non-core point;

[0311] The first division unit is used to perform cluster division on the second feature parameter according to the core point to obtain cluster data of each second user feature.

[0312] Optionally, the fourth computing unit includes:

[0313] a fifth calculation unit, configured to perform similarity calculation between the plurality of clusters in the clustering data of each second user feature, to obtain a similarity coefficient of each second user feature;

[0314] a sixth calculation unit, configured to calculate dissimilarity between the plurality of clusters in the clustering data of each second user feature, to obtain a dissimilarity coefficient of each second user feature;

[0315] A seventh calculation unit is used to calculate a silhouette coefficient according to the similarity coefficient and the dissimilarity coefficient to obtain a silhouette coefficient of each of the second user features.

[0316] Optionally, the third assignment unit includes:

[0317] A second generating unit is used to generate an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small;

[0318] The fourth assignment unit is used to assign values ​​to the second user features in sequence according to the order of the silhouette coefficients from large to small through the equally spaced numerical sequence to obtain a secondary weight coefficient of each second user feature.

[0319] Optionally, the device further comprises:

[0320] A first receiving module, used for receiving sample data sent by the terminal;

[0321] The first updating module is used to add the sample data to the training data set to update the training data set.

[0322] It should be noted that the embodiment of the device is a device corresponding to the embodiment of the above-mentioned method. All implementation methods in the embodiment of the above-mentioned method are applicable to the embodiment of the device and can achieve the same technical effect. To avoid repetition, they will not be described here.

[0323] like Fig.12 As shown, an embodiment of the present invention further provides a model training device based on wearable system features, which is applied to a terminal and includes:

[0324] The second acquisition module 1201 is used to acquire the trained group portrait model sent by the cloud server;

[0325] A third acquisition module 1202 is used to acquire a test data set of a test user, wherein the test data set includes static information, motion state, and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features;

[0326] The first correction module 1203 is used to correct the group portrait model according to the test data set and the error correction algorithm, and optimize the group portrait model.

[0327] Optionally, the first correction module 1203 includes:

[0328] A first correction unit, configured to correct the group portrait model once according to the static information in the test data set and the fourth feature data according to an error correction algorithm;

[0329] The second correction unit is used to perform secondary correction on the group portrait model according to the static information, the motion state and the fourth characteristic data in the test data set according to an error correction algorithm.

[0330] Optionally, the first correction unit includes:

[0331] A first acquisition unit is used to input the static information of the test user into the group portrait model to obtain fifth feature data of multiple users corresponding to the static information, wherein the fifth feature data includes multiple test user features and a fifth feature parameter corresponding to each of the test user features;

[0332] An eighth calculation unit, configured to respectively perform mean calculation on the fifth feature parameters of the plurality of users corresponding to each of the test user features, to obtain a first prediction feature parameter corresponding to each of the test user features;

[0333] The third correction unit is used to correct the group portrait model according to the first prediction feature parameter and the test feature parameter in the fourth feature data according to an error correction algorithm.

[0334] Optionally, the second correction unit includes:

[0335] A second division unit is used to input the static information and the motion state of the test user into the modified group portrait model, perform cluster division on the test user, and obtain the user category of the test user;

[0336] A second acquisition unit, configured to acquire sixth feature data of a plurality of users belonging to the user category, the sixth feature data comprising a plurality of test user features and a sixth feature parameter corresponding to each of the test user features;

[0337] A ninth calculation unit, configured to respectively perform mean calculation on the sixth feature parameters of the plurality of users corresponding to each of the test user features, to obtain a second prediction feature parameter corresponding to each of the test user features;

[0338] The fourth correction unit is used to perform secondary correction on the group portrait model according to the second prediction feature parameter and the test feature parameter in the fourth feature data according to an error correction algorithm.

[0339] Optionally, the device further comprises:

[0340] A first judgment module is used to compare the data in the test data set with the data in the training data set to determine whether there is data in the test data set that is different from the training data set, wherein the training data set is a data set for training the group portrait model;

[0341] A first determination module is used to determine the data as sample data and count the number of the sample data when there is data different from the training data set in the test data set;

[0342] The second sending module is used to send the sample data to the cloud server when the amount of the sample data is greater than the first threshold.

[0343] It should be noted that the embodiment of the device is a device corresponding to the embodiment of the above-mentioned method. All implementation methods in the embodiment of the above-mentioned method are applicable to the embodiment of the device and can achieve the same technical effect. To avoid repetition, they will not be described here.

[0344] An embodiment of the present invention also provides a network device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, the program implements the model training method based on wearable system features as described in any one of the above items, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0345] The embodiment of the present invention also provides a readable storage medium, including: a program is stored on the readable storage medium, and when the program is executed by the processor, the steps of the model training method based on wearable system features as described in any of the above items are implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the computer-readable storage medium is such as a read-only memory (Read-Only Memory, referred to as ROM), a random access memory (Random Access Memory, referred to as RAM), a disk or an optical disk, etc.

[0346] An embodiment of the present invention also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the steps of the model training method based on wearable system characteristics as described in any of the above items are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be repeated here.

[0347] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0348] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A model training method based on wearable system features, characterized in that: Applied to a cloud server, the method includes: Based on digital twin technology, build and initialize the user group portrait model; Acquire a training data set corresponding to the group portrait model, the training data set including static information, motion states, and first feature data of a wearable system in each of the motion states of a plurality of users, the first feature data including a plurality of first user features and a first feature parameter corresponding to each of the first user features; Performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model; The trained group portrait model is sent to the test user's terminal.

2. The model training method based on wearable system features according to claim 1, characterized in that: The first feature data is respectively subjected to correlation screening and cluster evaluation according to the static information and the motion state, and the group portrait model is trained, including: According to the static information and the motion state, the first feature data is respectively subjected to correlation screening to obtain second feature data, where the second feature data includes a plurality of second user features and a second feature parameter corresponding to each of the second user features; Assigning a value to the second feature data to obtain a primary weight coefficient of each of the second user features; According to a preset arrangement order, clustering and cluster evaluation are performed on the second feature parameters corresponding to each of the second user features in turn to obtain a secondary weight coefficient of each of the second user features; The second user features in the second feature data are sorted according to the first-level weight coefficient and the second-level weight coefficient to train the group portrait model.

3. The model training method based on wearable system features according to claim 2 is characterized in that: According to the static information and the motion state, the first characteristic data is respectively subjected to correlation screening to obtain second characteristic data, including: Performing correlation calculation based on the static information and the first feature data to obtain a first correlation coefficient corresponding to each of the first user features; Filter the first feature data according to the first correlation coefficient to obtain third feature data, where the third feature data includes a plurality of third user features and a third feature parameter corresponding to each of the third user features; Performing correlation calculation based on the motion state and the third feature data to obtain a second correlation coefficient corresponding to each of the third user features; The third characteristic data is screened according to the second correlation coefficient to obtain second characteristic data.

4. The model training method based on wearable system features according to claim 3 is characterized in that: Assigning a value to the second feature data to obtain a primary weight coefficient of each second user feature includes: generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small; Calculate according to the first correlation coefficient and the second correlation coefficient to obtain a comprehensive correlation coefficient of each of the second user features; According to the descending order of the comprehensive correlation coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a primary weight coefficient of each second user feature.

5. The model training method based on wearable system features according to claim 2, characterized in that: According to a preset arrangement order, clustering and clustering evaluation are sequentially performed on the second feature parameters corresponding to each of the second user features to obtain a secondary weight coefficient of each of the second user features, including: According to a preset arrangement order, sequentially performing adaptive density clustering on the second feature parameters of the plurality of users corresponding to the second user feature, to obtain clustering data of each second user feature; Calculating a silhouette coefficient according to the clustering data to obtain a silhouette coefficient of each of the second user features; The second user features are assigned values ​​according to the silhouette coefficients to obtain a secondary weight coefficient of each of the second user features.

6. The model training method based on wearable system features according to claim 5 is characterized in that: According to a preset arrangement order, adaptive density clustering is performed on the second feature parameters of the plurality of users corresponding to the second user feature in sequence to obtain clustering data of each second user feature, including: Normalizing the second characteristic parameter to obtain normalized data; According to a preset arrangement order and a density clustering algorithm, second feature parameters of the plurality of users corresponding to each of the second user features are calculated in turn to determine a type of the second feature parameter, the type including a core point and a non-core point; The second feature parameters are clustered according to the core points to obtain cluster data of each of the second user features.

7. The model training method based on wearable system features according to claim 5, characterized in that: The clustering data includes a plurality of clustering clusters; The step of calculating the silhouette coefficient according to the clustering data to obtain the silhouette coefficient of each of the second user features includes: Performing similarity calculation between the plurality of clusters in the clustering data of each second user feature to obtain a similarity coefficient of each second user feature; Calculating dissimilarity between the plurality of clusters in the clustering data of each second user feature to obtain a dissimilarity coefficient of each second user feature; A silhouette coefficient is calculated according to the similarity coefficient and the dissimilarity coefficient to obtain a silhouette coefficient of each of the second user features.

8. The model training method based on wearable system features according to claim 5, characterized in that: Assigning a value to the second user feature according to the silhouette coefficient to obtain a secondary weight coefficient of each second user feature includes: generating an equally spaced numerical sequence according to the number of the second user features, wherein the number of numerical values ​​in the equally spaced numerical sequence is the same as the number of the second user features, and the numerical values ​​in the equally spaced numerical sequence are sorted from large to small; According to the descending order of the silhouette coefficients, the second user features are assigned values ​​in sequence through the equally spaced numerical sequence to obtain a secondary weight coefficient of each second user feature.

9. The model training method based on wearable system features according to claim 1, characterized in that: The method further comprises: Receiving sample data sent by the terminal; The sample data is added to the training data set to update the training data set.

10. A model training method based on wearable system features, characterized in that: Applied to a terminal, the method comprises: Get the trained group portrait model sent by the cloud server; Acquire a test data set of a test user, wherein the test data set includes static information, motion state, and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features; The group portrait model is corrected according to the test data set and the error correction algorithm, and the group portrait model is optimized.

11. The model training method based on wearable system features according to claim 10, characterized in that: Correcting the group portrait model according to the test data set and the error correction algorithm includes: According to the static information in the test data set and the fourth characteristic data, the group portrait model is corrected once according to an error correction algorithm; According to the static information, the motion state and the fourth characteristic data in the test data set, the group portrait model is corrected secondary according to an error correction algorithm.

12. The model training method based on wearable system features according to claim 11, characterized in that: According to the static information in the test data set and the fourth characteristic data, the group portrait model is corrected once according to an error correction algorithm, including: Inputting the static information of the test user into the group portrait model to obtain fifth feature data of multiple users corresponding to the static information, wherein the fifth feature data includes multiple test user features and a fifth feature parameter corresponding to each of the test user features; Respectively calculating the mean of the fifth feature parameters of the plurality of users corresponding to each of the test user features to obtain the first prediction feature parameter corresponding to each of the test user features; According to the first prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected once according to an error correction algorithm.

13. The model training method based on wearable system features according to claim 11, characterized in that: According to the static information, the motion state and the fourth feature data in the test data set, the group portrait model is corrected twice according to an error correction algorithm, including: Inputting the static information and the motion state of the test user into the modified group portrait model, clustering the test user, and obtaining the user category of the test user; Acquire sixth feature data of a plurality of users belonging to the user category, the sixth feature data comprising a plurality of test user features and a sixth feature parameter corresponding to each of the test user features; Respectively calculating the mean of the sixth feature parameters of multiple users corresponding to each of the test user features to obtain a second prediction feature parameter corresponding to each of the test user features; According to the second prediction feature parameter and the test feature parameter in the fourth feature data, the group portrait model is corrected secondary according to an error correction algorithm.

14. The model training method based on wearable system features according to claim 10, characterized in that: The method further comprises: Comparing the data in the test data set with the data in the training data set to determine whether there is data in the test data set that is different from the training data set, wherein the training data set is a data set for training the group portrait model; When there is data in the test data set that is different from the training data set, determining the data as sample data, and counting the number of the sample data; When the amount of the sample data is greater than a first threshold, the sample data is sent to a cloud server.

15. A network device, characterized in that: The invention comprises a transceiver and a processor, wherein the processor is configured to: Based on digital twin technology, build and initialize the user group portrait model; Acquire a training data set corresponding to the group portrait model, the training data set including static information, motion states, and first feature data of a wearable system in each of the motion states of a plurality of users, the first feature data including a plurality of first user features and a first feature parameter corresponding to each of the first user features; Performing correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model; The transceiver is used to send the trained group portrait model to the terminal of the test user.

16. A terminal, characterized in that: It includes a transceiver and a processor, wherein the transceiver is used to obtain the trained group portrait model sent by the cloud server; The processor is configured to: Acquire a test data set of a test user, wherein the test data set includes static information, motion state, and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features; The group portrait model is corrected according to the test data set and the error correction algorithm, and the group portrait model is optimized.

17. A model training device based on wearable system features, characterized in that: Applied to cloud servers, including: The first building module is used to build and initialize a user group portrait model based on the digital twin technology; A first acquisition module is used to acquire a training data set corresponding to the group portrait model, wherein the training data set includes static information and motion states of multiple users and first feature data of the wearable system in each of the motion states, wherein the first feature data includes multiple first user features and a first feature parameter corresponding to each of the first user features; A first training module, configured to perform correlation screening and cluster evaluation on the first feature data according to the static information and the motion state, respectively, to train the group portrait model; The first sending module is used to send the trained group portrait model to the terminal of the test user.

18. A model training device based on wearable system features, characterized in that: Applied to terminals, including: The second acquisition module is used to obtain the trained group portrait model sent by the cloud server; A third acquisition module is used to acquire a test data set of a test user, wherein the test data set includes static information, motion state and fourth feature data of the wearable system in each motion state of the test user, wherein the fourth feature data includes multiple test user features and a test feature parameter corresponding to each of the test user features; The first correction module is used to correct the group portrait model according to the test data set and the error correction algorithm, and optimize the group portrait model.

19. A network device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the model training method based on wearable system features as described in any one of claims 1 to 9 or implements the model training method based on wearable system features as described in any one of claims 10 to 14.

20. A readable storage medium, characterized in that: include: The readable storage medium stores a program, and when the program is executed by the processor, it implements the steps of the model training method based on wearable system characteristics as described in any one of claims 1 to 9 or implements the steps of the model training method based on wearable system characteristics as described in any one of claims 10 to 14.

21. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of the model training method based on wearable system characteristics as described in any one of claims 1 to 9 or the steps of the model training method based on wearable system characteristics as described in any one of claims 10 to 14.

Citation Information

Cited By

  • Control method and system based on wearable intelligent equipment

    CN119960603A