Smart bracelet dial layout control method, device, equipment and storage medium

Through multi-sensor data fusion and transfer learning technology, the smart bracelet dial design builds dynamic user portraits and adjusts dial elements, solving the problem that dial design cannot be personalized in the existing technology and improving the user experience.

CN118734718BActive Publication Date: 2024-11-12SHENZHEN SMART CARE TECH LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411203443.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-12
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

The existing smart bracelet dial design is difficult to flexibly adjust according to user individual differences and dynamic needs, cannot meet user personalized needs, and lacks an effective intelligent recommendation mechanism.

Method used

Process user data through multi-sensor fusion, perform clustering analysis and time series analysis, build dynamic user portraits, and use generative adversarial networks and variational autoencoder to select and adjust dial elements from the preset dial element library, combined with transfer learning methods Form a personalized migration model and control the dial layout.

Benefits of technology

It realizes automatic selection and adjustment of dial elements based on user data, and generates a dial design that meets user needs, which significantly improves the user experience and meets users' dynamic needs and preferences.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118734718B_ABST
    Figure CN118734718B_ABST
Patent Text Reader

Abstract

The present invention provides a dial layout control method, device, equipment and storage medium for a smart bracelet, wherein the method comprises: processing user data through multi-sensor fusion to obtain user feature data. Using cluster analysis and time series analysis, a dynamic user portrait is constructed. According to the dynamic user portrait, elements are selected from a dial element library, and combined and adjusted using a generative adversarial network and a variational autoencoder to obtain a candidate dial design. The source domain knowledge is transferred to the target domain through transfer learning, and fine-tuned to form a personalized transfer model. According to the personalized transfer model, the candidate dial designs are evaluated and sorted to obtain a target dial design, and the dial layout is controlled. The method automatically selects and adjusts the dial elements according to the collected data, generates a dial design that meets the user's needs, and continuously optimizes according to user feedback, which can significantly improve the user experience of the smart bracelet and better meet the user's dynamic needs and preferences.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a dial layout control method, device, equipment and storage medium for a smart bracelet. Background Art

[0002] As a portable smart device, smart bracelets are widely used in the fields of health monitoring, exercise recording and smart reminders. Existing smart bracelets usually collect users' physiological data, behavioral data and environmental data through sensors and display the corresponding information on the dial. However, the existing technology has great deficiencies in the personalization and dynamic adjustment of the layout of the smart bracelet dial.

[0003] The existing smart bracelet dial layout is usually based on a preset template, which is difficult to flexibly adjust according to the individual differences and dynamic needs of users. This fixed dial layout method cannot fully consider the user's life pattern, behavioral periodicity and long-term trend, resulting in poor user experience and failure to meet the personalized needs of users. At the same time, the existing technology also lacks an effective mechanism for intelligent recommendation of dial design. It usually recommends through simple rules or static models, which is difficult to adapt to the personalized requirements and changes of users. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problem that the above-mentioned existing smart bracelet dial design lacks an effective mechanism for intelligent recommendation and is difficult to adapt to the personalized requirements and changes of users. The first aspect of the present invention provides a dial layout control method for a smart bracelet, and the dial layout control method for the smart bracelet includes:

[0005] Perform multi-sensor fusion processing on the user physiological data, behavior data and environmental data collected by the smart bracelet to obtain fused data, and perform preprocessing and feature extraction on the fused data to obtain user feature data;

[0006] Performing cluster analysis on the target users according to the user characteristic data to obtain the user activity type and life pattern, and performing time series analysis on the user characteristic data to obtain the periodic pattern and long-term trend of the user behavior;

[0007] Constructing an initial user profile based on the user's activity type, life pattern, periodic pattern, and long-term trend, and updating the initial user profile based on explicit feedback from the target user to obtain a dynamic user profile;

[0008] Selecting a dial element from a preset dial element library according to the dynamic user portrait, and combining and adjusting the selected dial elements using a generative adversarial network and a variational autoencoder to obtain a candidate dial design;

[0009] Using a transfer learning method, the source domain knowledge of the preset source domain dial recommendation model is transferred to the target domain of the target user to obtain a transfer learning model, and the transfer learning model is fine-tuned according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model;

[0010] The candidate dial designs are evaluated and ranked according to the personalized migration model to obtain a target dial design, and the dial layout of the smart bracelet is controlled according to the target dial design.

[0011] Optionally, in a first implementation of the first aspect of the present invention, the smart bracelet collects the user physiological data, behavior data and environmental data of the target user through multi-sensor fusion processing to obtain fused data, and pre-processes and extracts features of the fused data to obtain user feature data, including:

[0012] Applying data alignment and time synchronization algorithms to the user's physiological data, behavioral data, and environmental data collected by the smart bracelet to obtain multimodal data with consistent time series, and using an adaptive weighted fusion algorithm on the multimodal data with consistent time series to obtain preliminary fused data;

[0013] Applying a hybrid filtering algorithm combining Kalman filtering and particle filtering to the preliminary fused data to obtain fused data after noise suppression and outlier removal, and performing segmented statistical feature extraction and frequency domain transformation on the fused data to obtain time-frequency domain features;

[0014] Applying nonlinear principal component analysis and independent component analysis to the time-frequency domain features to obtain a feature representation after dimensionality reduction, and performing time series modeling on the feature representation using a recursive neural network to obtain a dynamic feature sequence;

[0015] Physiological state indicators, behavioral pattern indicators and environmental adaptability indicators are calculated according to the dynamic feature sequence to obtain user feature data.

[0016] Optionally, in a second implementation of the first aspect of the present invention, performing cluster analysis on users according to the user feature data to obtain user activity types and life patterns, and performing time series analysis on the user feature data to obtain periodic patterns and long-term trends of user behaviors include:

[0017] Calculate the distance between the user feature data and the preset user group cluster centers to obtain the similarity between the target user and each cluster center, and determine the activity type and life pattern of the target user based on the cluster characteristics of the cluster center with the highest similarity;

[0018] Applying sliding window technology and adaptive segmentation algorithm to the user feature data to obtain time series segments, and performing fast Fourier transform and continuous wavelet transform on the time series segments to obtain multi-scale spectrum features, and using peak detection and harmonic analysis based on the multi-scale spectrum features to obtain a periodic pattern of user behavior;

[0019] The exponentially weighted moving average method is applied to the user feature data to obtain a smoothed time series, and the smoothed time series is subjected to seasonal decomposition and trend extraction to obtain trend components. The Prophet time series prediction model is used based on the trend components to perform long-term trend prediction and change point detection to obtain the long-term trend of user behavior.

[0020] Optionally, in a third implementation of the first aspect of the present invention, selecting a dial element from a preset dial element library according to the dynamic user portrait, and combining and adjusting the selected dial elements using a generative adversarial network and a variational autoencoder to obtain a candidate dial design includes:

[0021] Performing feature vectorization processing on the dynamic user portrait to obtain a user preference vector, and performing similarity calculation on a preset dial element library according to the user preference vector to obtain a preliminary matching dial element set;

[0022] Applying a multi-objective optimization algorithm to the initially matched dial element set to obtain a dial element combination, and applying a conditional generative adversarial network to the dial element combination to obtain an initial candidate dial design;

[0023] Performing latent space encoding on the initial candidate dial design using a variational autoencoder to obtain a latent representation of the dial design, and performing interpolation and perturbation operations on the latent representation to obtain a diversified set of latent representations;

[0024] The diverse set of latent representations is reconstructed using a decoder of a variational autoencoder to obtain multiple candidate dial designs.

[0025] Optionally, in a fourth implementation manner of the first aspect of the present invention, applying a multi-objective optimization algorithm to the preliminarily matched dial element set to obtain a dial element combination, and applying a conditional generative adversarial network to the dial element combination to obtain an initial candidate dial design includes:

[0026] Extracting features from the preliminarily matched dial element set to obtain element feature vectors, and constructing a dial element complementarity matrix based on the element feature vectors to obtain a complementary relationship between the elements;

[0027] Applying a genetic algorithm to the complementary relationship to obtain an initial dial element combination population, and performing Pareto sorting on the initial dial element combination population to obtain a non-dominated solution set;

[0028] Applying a simulated annealing algorithm to the non-dominated solution set to obtain a locally optimal combination of dial elements, and performing a crossover mutation operation on the locally optimal combination of dial elements to obtain a diversified combination of dial elements;

[0029] Constructing a condition vector for the diverse combination of dial elements to obtain design constraints, inputting the design constraints into a generator of a conditional generative adversarial network to obtain an initial candidate dial design sample, and applying a discriminator of the conditional generative adversarial network to the initial candidate dial design sample to obtain a authenticity score;

[0030] The generator and the discriminator are alternately trained according to the authenticity score to obtain an optimized conditional generative adversarial network, and the optimized conditional generative adversarial network is used to generate multiple initial candidate dial designs.

[0031] Optionally, in a fifth implementation of the first aspect of the present invention, the use of a transfer learning method to migrate the source domain knowledge of a preset source domain dial recommendation model to the target domain of the target user to obtain a transfer learning model, and fine-tuning the transfer learning model according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model includes:

[0032] Performing structural analysis on a preset source domain watch face recommendation model to obtain source domain knowledge representation, constructing a knowledge graph based on the source domain knowledge representation to obtain a source domain knowledge graph, and applying a graph neural network to the source domain knowledge graph to obtain a graph embedding representation;

[0033] Extracting features from the dynamic user portrait and historical interaction data of the target user to obtain a target domain feature vector, and aligning the target domain feature vector with the graph embedding representation to obtain a cross-domain mapping matrix;

[0034] Migrating the parameters of the source domain dial recommendation model according to the cross-domain mapping matrix to obtain an initial transfer learning model, and applying domain adversarial training to the initial transfer learning model to obtain a domain invariant feature representation;

[0035] A multi-task learning framework is constructed based on the domain-invariant feature representation, and the multi-task learning framework is fine-tuned according to the dynamic user profile and historical interaction data of the target user to obtain a personalized transfer model.

[0036] Optionally, in a sixth implementation of the first aspect of the present invention, evaluating and sorting the candidate dial designs according to the personalized migration model to obtain a target dial design, and controlling the dial layout of the smart bracelet according to the target dial design includes:

[0037] Evaluate and sort the candidate watch face designs according to the personalized migration model to obtain a target watch face design;

[0038] Decomposing the target dial design into elements to obtain a dial layout element set, and constructing a layout constraint graph according to the dial layout element set to obtain a spatial relationship between the elements;

[0039] Positioning and calculating the dial layout elements according to the spatial relationship to obtain element coordinate information, and generating layout control instructions according to the element coordinate information to obtain a dial layout control solution;

[0040] The dial layout control scheme is converted into a device-specific control signal to obtain a display control signal of the smart bracelet, and the display control signal is sent to the display module of the smart bracelet to realize the dial layout control.

[0041] A second aspect of the present invention provides a dial layout control device for a smart bracelet, the dial layout control device for the smart bracelet comprising:

[0042] The data processing module is used to perform multi-sensor fusion processing on the user physiological data, behavior data and environmental data collected by the smart bracelet to obtain fused data, and to perform preprocessing and feature extraction on the fused data to obtain user feature data;

[0043] A user analysis module, used to perform cluster analysis on target users based on the user feature data to obtain user activity types and life patterns, and to perform time series analysis on the user feature data to obtain periodic patterns and long-term trends of user behavior;

[0044] A portrait generation module is used to construct an initial user portrait according to the user's activity type, life pattern, periodic pattern and long-term trend, and update the initial user portrait according to the explicit feedback information of the target user to obtain a dynamic user portrait;

[0045] A candidate dial determination module, configured to select dial elements from a preset dial element library according to the dynamic user portrait, and combine and adjust the selected dial elements using a generative adversarial network and a variational autoencoder to obtain a candidate dial design;

[0046] A transfer learning module, used to transfer the source domain knowledge of the preset source domain dial recommendation model to the target domain of the target user by using a transfer learning method to obtain a transfer learning model, and fine-tune the transfer learning model according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model;

[0047] The layout control module is used to evaluate and sort the candidate dial designs according to the personalized migration model to obtain a target dial design, and control the dial layout of the smart bracelet according to the target dial design.

[0048] A third aspect of the present invention provides a dial layout control device for a smart bracelet, comprising: a memory and at least one processor, the memory storing instructions, the memory and the at least one processor being interconnected through a line; the at least one processor calls the instructions in the memory so that the dial layout control device of the smart bracelet executes the steps of the above-mentioned dial layout control method of the smart bracelet.

[0049] A fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, the computer executes the steps of the above-mentioned smart bracelet dial layout control method.

[0050] The dial layout control method, device, equipment and storage medium of the above-mentioned smart bracelet obtain user feature data by processing user data through multi-sensor fusion. Use cluster analysis and time series analysis to build a dynamic user portrait. According to the dynamic user portrait, select elements from the dial element library, combine and adjust them using generative adversarial networks and variational autoencoders to obtain candidate dial designs. Transfer the source domain knowledge to the target domain through transfer learning, and fine-tune to form a personalized transfer model. Evaluate and sort the candidate dial designs according to the personalized transfer model to obtain the target dial design, and perform dial layout control. This method automatically selects and adjusts the dial elements based on the collected data, generates a dial design that meets user needs, and continuously optimizes according to user feedback, which can significantly improve the user experience of the smart bracelet and better meet the dynamic needs and preferences of users.

[0051] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a schematic diagram of a first embodiment of a method for controlling a dial layout of a smart bracelet in an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of an embodiment of a dial layout control device for a smart bracelet in an embodiment of the present invention;

[0055] Figure 3 Schematic diagram of an embodiment of a dial layout control device of a smart bracelet in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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.

[0057] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0058] To facilitate understanding of this embodiment, a method for controlling the dial layout of a smart bracelet disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps:

[0059] 101. Perform multi-sensor fusion processing on the user physiological data, behavior data and environmental data collected by the smart bracelet to obtain fused data, and perform preprocessing and feature extraction on the fused data to obtain user feature data;

[0060] In one embodiment of the present invention, the multi-sensor fusion processing is performed on the user physiological data, behavior data and environmental data of the target user collected by the smart bracelet to obtain fused data, and the fused data is preprocessed and feature extracted to obtain user feature data, including: applying data alignment and time synchronization algorithms to the user physiological data, behavior data and environmental data collected by the smart bracelet to obtain multimodal data with consistent time series, and using an adaptive weighted fusion algorithm on the multimodal data with consistent time series to obtain preliminary fused data; applying a hybrid filtering algorithm combining Kalman filtering and particle filtering to the preliminary fused data to obtain fused data after noise suppression and outlier removal, and performing segmented statistical feature extraction and frequency domain transformation on the fused data to obtain time-frequency domain features; applying nonlinear principal component analysis and independent component analysis to the time-frequency domain features to obtain feature representation after dimensionality reduction, and using a recursive neural network to perform time series modeling on the feature representation to obtain a dynamic feature sequence; calculating physiological state indicators, behavior pattern indicators and environmental adaptability indicators according to the dynamic feature sequence to obtain user feature data.

[0061] Specifically, the physiological data, behavioral data and environmental data of the target user are collected through multiple sensors of the smart bracelet. These data include heart rate, blood oxygen, acceleration, gyroscope, temperature, humidity and other information. Due to the differences in sampling frequency and data format of different sensors, the system applies data alignment and time synchronization algorithms to process these raw data. Specifically, interpolation technology is used to upsample low-frequency data or downsample high-frequency data, and timestamp information is used for alignment to obtain multimodal data with consistent time series. Next, an adaptive weighted fusion algorithm is used for these multimodal data with consistent time series. This algorithm takes into account the reliability and importance of different sensor data and dynamically adjusts the weights of each data source. For example, the weight is determined based on the signal-to-noise ratio of the data or the consistency with historical data. Through this step, preliminary fused data is obtained. In order to further improve the data quality, a hybrid filtering algorithm combining Kalman filtering and particle filtering is applied to the preliminary fused data. Kalman filtering is suitable for processing linear Gaussian systems, while particle filtering can process nonlinear non-Gaussian systems. By combining these two methods, noise is effectively suppressed and outliers are eliminated, thereby obtaining more reliable fused data. Subsequently, these filtered fusion data are subjected to segmented statistical feature extraction and frequency domain transformation. The segmented statistical features include statistics such as mean, variance, kurtosis, and skewness, while the frequency domain transformation uses fast Fourier transform (FFT) or wavelet transform to obtain the frequency characteristics of the data. The purpose of this step is to extract the time-frequency domain features of the data and provide richer information for subsequent analysis. After obtaining the time-frequency domain features, nonlinear principal component analysis (NLPCA) and independent component analysis (ICA) are applied to perform feature dimensionality reduction and extraction. NLPCA can capture the nonlinear structure in the data, while ICA can separate independent signal sources. Through the combination of these two methods, a reduced-dimensional feature representation is obtained, which retains the main information of the original data while removing redundancy and noise. In order to capture the temporal dependency of the features, a recurrent neural network (RNN) is used to perform temporal modeling on the reduced-dimensional feature representation. The characteristics of RNN are that it can process sequence data and remember previous information. Through this step, a dynamic feature sequence is obtained, which not only contains the information of the current moment, but also the influence of historical information. Finally, based on this dynamic feature sequence, the physiological state index, behavioral pattern index and environmental adaptability index are calculated. Physiological state indicators include heart rate variability, stress level, etc.; behavioral pattern indicators include activity intensity, gait characteristics, etc.; environmental adaptability indicators include body temperature regulation ability, environmental change response, etc. The calculation of these indicators involves multiple methods such as threshold judgment, pattern recognition, and trend analysis. By combining these indicators, comprehensive user feature data is finally obtained, which reflects the user's physiological state, behavioral pattern and environmental adaptability.

[0062] 102. Perform cluster analysis on target users based on user feature data to obtain user activity types and life patterns, and perform time series analysis on user feature data to obtain periodic patterns and long-term trends of user behavior;

[0063] In one embodiment of the present invention, clustering analysis is performed on users according to the user characteristic data to obtain user activity types and life patterns, and time series analysis is performed on the user characteristic data to obtain periodic patterns and long-term trends of user behavior, including: distance calculation is performed between the user characteristic data and the preset user group cluster centers to obtain the similarity between the target user and each cluster center, and based on the cluster cluster characteristics of the cluster center with the highest similarity, the activity type and life pattern of the target user are determined; sliding window technology and adaptive segmentation algorithm are applied to the user characteristic data to obtain time series fragments, and fast Fourier transform and continuous wavelet transform are performed on the time series fragments to obtain multi-scale spectrum characteristics, and peak detection and harmonic analysis are used according to the multi-scale spectrum characteristics to obtain periodic patterns of user behavior; exponentially weighted moving average method is applied to the user characteristic data to obtain smoothed time series, and seasonal decomposition and trend extraction are performed on the smoothed time series to obtain trend components, and long-term trend prediction and change point detection are performed according to the trend components using the Prophet time series prediction model to obtain long-term trend prediction and change point detection to obtain long-term trends of user behavior.

[0064] Specifically, the distance between the feature vector of the target user and each preset cluster center is calculated using methods such as Euclidean distance or Mahalanobis distance, so as to obtain the similarity between the target user and each cluster center. Then, based on the cluster characteristics of the cluster center with the highest similarity, the activity type and life pattern of the target user are determined. This method avoids the computational overhead of a complete cluster analysis of a single user, and at the same time utilizes the predefined user group characteristics to quickly classify the target user. Subsequently, the sliding window technology and the adaptive segmentation algorithm are applied to the user feature data to obtain time series fragments. The sliding window technology divides the data by sliding a fixed-size window on the time series, while the adaptive segmentation algorithm dynamically adjusts the length of the segment according to the characteristics of the data. The combination of these two methods can effectively process time patterns of different scales. Fast Fourier transform (FFT) and continuous wavelet transform (CWT) are performed on the obtained time series fragments to obtain multi-scale spectral features. FFT is suitable for analyzing signals with obvious periodicity, while CWT can capture frequency changes at different time scales. Based on the obtained multi-scale spectral features, the peak detection algorithm is used to identify significant peaks in the spectrum, which correspond to the main periods in the time series. At the same time, harmonic analysis is used to identify the multiple relationships between these peaks, thereby obtaining the periodic patterns of user behavior. These periodic patterns can reflect the regular behaviors of users in their daily lives, such as daily, weekly or monthly activity patterns. In order to analyze the long-term trend of user behavior, the exponentially weighted moving average method is first applied to the user feature data to obtain a smoothed time series. The exponentially weighted moving average method gives higher weights to recent data, which can effectively smooth short-term fluctuations while retaining long-term trend information. The smoothed time series is subjected to seasonal decomposition and trend extraction to obtain trend components. Seasonal decomposition decomposes the time series into three parts: trend, seasonality and residual, while trend extraction further separates long-term change trends. Finally, the Prophet time series prediction model is used to analyze the extracted trend components. The Prophet model is an additive model that can handle time series data with multiple periodicities and can automatically handle missing values ​​and outliers. The Prophet model is used to predict long-term trends, and its built-in change point detection function is used to identify significant change points in the time series. These change points may represent major changes in user behavior or lifestyle. Through this series of analyses, the long-term trend of user behavior is finally obtained.

[0065] 103. Construct an initial user profile based on user activity types, life patterns, periodic patterns, and long-term trends, and update the initial user profile based on explicit feedback from target users to obtain a dynamic user profile;

[0066] In one embodiment of the present invention, an initial user profile is first constructed based on user activity type, life pattern, periodic pattern and long-term trend. This process uses a multi-dimensional feature vector to represent user features, where each dimension corresponds to a specific attribute or behavior pattern. Specifically, the user activity type is encoded as a discrete category feature, the life pattern is converted into a time distribution vector, the periodic pattern is represented as a combination of frequency and intensity, and the long-term trend is encoded as trend slope and change point information. These features are standardized and normalized by feature engineering technology to ensure that features of different dimensions are comparable. Next, a multi-layer perceptron (MLP) neural network model is used to fuse these processed features into a unified user representation. The input layer of the MLP model receives all features, and through nonlinear transformations of multiple hidden layers, a compact user vector representation is finally generated at the output layer. This user vector is the initial user profile, which integrates various user behaviors and pattern information. In order to enable the user profile to be dynamically updated, the system designs a feedback mechanism to collect explicit feedback information from the target user. This feedback information includes explicit and implicit feedback such as the user's rating of the recommended content, click behavior, and usage time. The system uses online learning algorithms, such as online gradient descent, to update user profiles in real time. Specifically, whenever new user feedback is received, the system calculates the error between the current user profile prediction result and the actual feedback, and then uses this error to adjust the weight parameters of the MLP model. In order to balance the importance of new and old information, a forgetting mechanism is introduced. This mechanism uses an exponential decay function to reduce the weight of old data, ensuring that the user profile can timely reflect the user's latest preferences and behavior changes. At the same time, in order to prevent overfitting of a single feedback, regularization techniques are used to constrain the amplitude of model updates. In addition, the system also implements an anomaly detection module to identify sudden changes in user behavior. When an anomaly is detected, the system temporarily increases the weight of the new feedback so that the user profile can quickly adapt to the user's new behavior pattern. This module uses moving average and standard deviation to define the range of normal behavior, and any behavior outside this range is considered a potential anomaly.

[0067] 104. Selecting dial elements from a preset dial element library according to the dynamic user portrait, and combining and adjusting the selected dial elements using a generative adversarial network and a variational autoencoder to obtain a candidate dial design;

[0068] In one embodiment of the present invention, the step of selecting dial elements from a preset dial element library according to the dynamic user portrait, and combining and adjusting the selected dial elements using a generative adversarial network and a variational autoencoder to obtain a candidate dial design includes: performing feature vectorization processing on the dynamic user portrait to obtain a user preference vector, and performing similarity calculation on the preset dial element library according to the user preference vector to obtain a preliminary matching set of dial elements; applying a multi-objective optimization algorithm to the preliminary matching set of dial elements to obtain a dial element combination, and applying a conditional generative adversarial network to the dial element combination to obtain an initial candidate dial design; performing latent space encoding on the initial candidate dial design using a variational autoencoder to obtain a latent representation of the dial design, and performing interpolation and perturbation operations on the latent representation to obtain a diverse set of potential representations; and reconstructing the diverse set of potential representations using a decoder of a variational autoencoder to obtain multiple candidate dial designs.

[0069] Specifically, the dynamic user portrait is first feature vectorized to obtain the user preference vector. This step uses dimensionality reduction techniques, such as principal component analysis (PCA) or t-SNE, to compress the high-dimensional user portrait into a low-dimensional preference vector. Next, the preset dial element library is similarity calculated based on this user preference vector. Here, a measurement method such as cosine similarity or Euclidean distance is used to calculate the similarity between the user preference vector and the feature vector of each dial element, so as to obtain a preliminary matching dial element set. Subsequently, a multi-objective optimization algorithm, such as NSGA-II (non-dominated sorting genetic algorithm II), is applied to the preliminary matching dial element set to obtain the optimal combination of dial elements. The algorithm considers multiple objectives such as aesthetics, functionality, and user preferences at the same time, and finds the Pareto optimal solution set through an iterative optimization process. In this process, each candidate solution represents a combination of dial elements, and new candidate solutions are generated through crossover and mutation operations, and non-dominated sorting and crowding distance calculations are used to select the next generation population. After obtaining the optimized dial element combination, a conditional generative adversarial network (cGAN) is used to generate the initial candidate dial design. The generator of cGAN receives a combination of dial elements as conditional input, and inputs random noise to generate a dial design that meets the conditions. The discriminator is responsible for distinguishing whether the generated dial design is real and whether it meets the given conditions. Through adversarial training of the generator and the discriminator, the system is able to generate initial candidate dial designs that are both in line with user preferences and visually attractive. Next, the initial candidate dial design is encoded in the latent space using a variational autoencoder (VAE). The encoder of the VAE maps the dial design to a low-dimensional latent space, which is usually assumed to follow a Gaussian distribution. During the encoding process, the VAE learns not only the mean of the latent vector, but also its variance, thereby capturing the uncertainty and diversity of the dial design. After obtaining the latent representation of the dial design, it is interpolated and perturbed to obtain a diverse set of latent representations. The interpolation operation is to select two points in the latent space and then perform linear interpolation between the two points to explore continuous changes in the latent space. The perturbation operation is to add small random noise to the latent representation to produce small changes. The combination of these two operations can generate a series of similar but distinctive latent representations. Finally, the decoder of VAE is used to reconstruct the diverse set of latent representations to obtain multiple candidate dial designs. The decoder maps the points in the latent space back to the original dial design space, thereby generating a series of candidate dial designs that are both in line with user preferences and diverse.

[0070] Furthermore, the multi-objective optimization algorithm is applied to the preliminary matched dial element set to obtain a dial element combination, and a conditional generative adversarial network is applied to the dial element combination to obtain an initial candidate dial design, including: extracting features from the preliminary matched dial element set to obtain element feature vectors, and constructing a dial element complementarity matrix based on the element feature vectors to obtain a complementary relationship between elements; applying a genetic algorithm to the complementary relationship to obtain an initial dial element combination population, and performing Pareto sorting on the initial dial element combination population to obtain a non-dominated solution set; applying a simulated annealing algorithm to the non-dominated solution set to obtain a local minimum. The optimal dial element combination is obtained, and a cross-mutation operation is performed on the locally optimal dial element combination to obtain a diversified dial element combination; a conditional vector is constructed for the diversified dial element combination to obtain a design constraint condition, and the design constraint condition is input into the generator of the conditional generative adversarial network to obtain an initial candidate dial design sample, and the discriminator of the conditional generative adversarial network is applied to the initial candidate dial design sample to obtain a authenticity score; the generator and the discriminator are alternately trained according to the authenticity score to obtain an optimized conditional generative adversarial network, and the optimized conditional generative adversarial network is used to generate multiple initial candidate dial designs.

[0071] Specifically, firstly, the feature extraction of the preliminary matched dial element set is performed to obtain the element feature vector. This step uses a pre-trained convolutional neural network (such as ResNet or VGG) to extract the visual features of the dial elements, and combines the functional attributes and style information of the elements to form a comprehensive feature representation. Then, the dial element complementarity matrix is ​​constructed based on these element feature vectors to calculate the complementary relationship between the elements. The complementarity is calculated based on the similarity and difference of the feature vectors, using metrics such as cosine similarity, while considering the functional complementarity of the elements. Then, a genetic algorithm is applied to the obtained complementary relationship to generate an initial dial element combination population. The chromosome encoding of the genetic algorithm represents the selection and arrangement of the dial elements, and the initial population randomly generates multiple dial element combinations. The evaluation function considers factors such as complementarity between elements, user preference matching, and design aesthetics. Next, a multi-objective evaluation is performed on the initial population, and a non-dominated sorting method is used to perform Pareto sorting to obtain a non-dominated solution set. This step ensures a balance between multiple objectives (such as aesthetics, functionality, and user preference matching). The simulated annealing algorithm is applied to the non-dominated solution set to obtain the locally optimal combination of dial elements. The simulated annealing algorithm selects the initial solution from the non-dominated solution set, generates new solutions through random perturbations, and decides whether to accept the new solution based on the energy function and temperature parameters. This process allows the algorithm to conduct a wider exploration in the search space and avoid being trapped in the local optimum. After obtaining the local optimal solution, a crossover mutation operation is performed on it to generate a diverse combination of dial elements. The crossover operation combines the features of different solutions, while the mutation operation introduces random changes to increase the diversity of the population. Subsequently, a condition vector is constructed for the diverse combination of dial elements to obtain the design constraints. The condition vector contains information such as the type, position, and size of the dial elements, as well as user preferences and design style requirements. These condition vectors are fed as input to the generator of the pre-trained conditional generative adversarial network (cGAN). The generator generates initial candidate dial design samples based on these conditions and random noise. The generated samples are then fed into the discriminator of the pre-trained cGAN for evaluation. The discriminator evaluates the authenticity of the samples and the degree of condition matching, and outputs a authenticity score. This score reflects the quality of the generated samples and whether they meet the given conditional constraints. Finally, according to the score of the discriminator, high-quality candidate dial designs are screened out. A score threshold can be set to retain only designs that exceed the threshold, or the top N designs with the highest scores can be selected. In order to increase diversity, algorithms such as maximizing marginal relevance (MMR) can also be used to select designs that are very different from each other while ensuring quality. By running this generation and evaluation process multiple times, using pre-trained cGAN, inputting different conditional vectors and random noise, multiple high-quality initial candidate dial designs are finally generated. These designs meet user preferences and functional requirements, and are innovative and diverse.During the generation process, the specific properties of the generated results, such as the overall style, color scheme, or functional layout, can be controlled by adjusting the parameters in the conditional vector.

[0072] 105. Use the transfer learning method to transfer the source domain knowledge of the preset source domain dial recommendation model to the target domain of the target user to obtain a transfer learning model, and fine-tune the transfer learning model according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model;

[0073] In one embodiment of the present invention, the source domain knowledge of the preset source domain dial recommendation model is transferred to the target domain of the target user by using a transfer learning method to obtain a transfer learning model, and the transfer learning model is fine-tuned according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model, including: performing a structural analysis on the preset source domain dial recommendation model to obtain a source domain knowledge representation, and constructing a knowledge graph according to the source domain knowledge representation to obtain a source domain knowledge graph, applying a graph neural network to the source domain knowledge graph to obtain a graph embedding representation; performing feature extraction on the dynamic user portrait and historical interaction data of the target user to obtain a target domain feature vector, and aligning the target domain feature vector with the graph embedding representation to obtain a cross-domain mapping matrix; migrating the parameters of the source domain dial recommendation model according to the cross-domain mapping matrix to obtain an initial transfer learning model, and applying domain adversarial training to the initial transfer learning model to obtain a domain-invariant feature representation; constructing a multi-task learning framework based on the domain-invariant feature representation, and fine-tuning the multi-task learning framework according to the dynamic user portrait and historical interaction data of the target user to obtain a personalized transfer model.

[0074] Specifically, the preset source domain dial recommendation model is first structurally analyzed to obtain the source domain knowledge representation. This process includes analyzing the model's hierarchical structure, weight distribution, and activation pattern, thereby extracting the key features and decision rules captured by the model. Based on these analysis results, a knowledge graph is constructed to represent the concepts, relationships, and attributes of the source domain in the form of a graph. As a structured representation of source domain knowledge, this knowledge graph contains information on dial design elements, user preferences, design rules, and other aspects. Next, the constructed source domain knowledge graph is processed using a graph neural network (GNN). GNN can effectively learn node representations and edge relationships in graph structured data, and fuse local and global information in the graph through message passing and aggregation operations. Specifically, models such as graph convolutional networks (GCNs) or graph attention networks (GATs) can be used to perform multi-level feature extraction and aggregation on the knowledge graph, and finally obtain a low-dimensional graph embedding representation. This graph embedding representation condenses the essence of source domain knowledge and lays the foundation for subsequent knowledge transfer. At the same time, feature extraction is performed on the dynamic user portraits and historical interaction data of the target users to obtain the target domain feature vector. This step can use deep neural networks or traditional feature engineering methods to extract key features that can represent the personalized needs and behavior patterns of target users. The extracted features may include multiple dimensions such as user preferences, usage habits, and interaction patterns. In order to associate the source domain knowledge with the target domain data, an alignment operation needs to be performed. This process involves mapping and aligning the target domain feature vector with the previously obtained graph embedding representation. A cross-domain mapping matrix can be learned using regularized linear mapping or nonlinear mapping methods such as orthogonal process regression (OPR) or deep alignment network (DAN). This matrix can transform the feature space of the target domain into a representation space compatible with the source domain knowledge. The parameters of the source domain dial recommendation model are migrated using the learned cross-domain mapping matrix. This step usually involves initializing or fine-tuning the model parameters so that the source domain model can have a good starting point on the target domain. Specifically, the mapping matrix can be used to adjust the weights of the source domain model to adapt it to the feature distribution of the target domain, thereby obtaining an initial transfer learning model. In order to further improve the adaptability of the model in the target domain, domain adversarial training is applied to the initial transfer learning model. This process introduces a domain discriminator, which attempts to distinguish whether the feature is from the source domain or the target domain. The model is trained to deceive this discriminator and learn domain-invariant feature representations. This adversarial training mechanism helps to eliminate the distribution differences between the source and target domains, allowing the model to better generalize to the target domain. Finally, based on the obtained domain-invariant feature representations, a multi-task learning framework is constructed. This framework considers multiple related tasks at the same time, such as dial style prediction, functional element recommendation, user satisfaction prediction, etc. Multi-task learning can make full use of the correlation between tasks and improve the generalization ability and performance of the model.In this framework, some low-level feature extraction layers can be shared between different tasks, while high-level layers are customized for each specific task. This multi-task learning framework is fine-tuned using the dynamic user profiles and historical interaction data of the target users. The fine-tuning process uses optimization algorithms such as gradient descent to adjust the model parameters for the specific data of the target users. This step can use mini-batch or online learning methods to adapt to the dynamic changes in user preferences. At the same time, regularization techniques such as L1 / L2 regularization or early stopping can be introduced to prevent overfitting and maintain the generalization ability of the model.

[0075] 106. Evaluate and sort the candidate dial designs according to the personalized migration model to obtain the target dial design, and control the dial layout of the smart bracelet according to the target dial design.

[0076] In one embodiment of the present invention, the evaluating and sorting the candidate dial designs according to the personalized migration model to obtain a target dial design, and performing dial layout control on the smart bracelet according to the target dial design includes: evaluating and sorting the candidate dial designs according to the personalized migration model to obtain a target dial design; performing element decomposition on the target dial design to obtain a set of dial layout elements, and constructing a layout constraint diagram based on the set of dial layout elements to obtain the spatial relationship between the elements; performing positioning calculations on the dial layout elements according to the spatial relationship to obtain element coordinate information, and generating layout control instructions based on the element coordinate information to obtain a dial layout control scheme; converting the dial layout control scheme into a device-specific control signal to obtain a display control signal of the smart bracelet, and sending the display control signal to the display module of the smart bracelet to implement dial layout control.

[0077] Specifically, the personalized transfer model is first used to evaluate and sort the candidate dial designs. In this process, each candidate dial design is converted into a feature vector and input into the personalized transfer model. The model calculates the fitness score of each design based on the user's preferences, historical behavior, and current context. These scores reflect the degree of match between the design and the user's needs, including multiple dimensions such as aesthetics, functionality, and personalization. Subsequently, the candidate designs are sorted in descending order according to these scores, and the design with the highest score is selected as the target dial design. Next, the selected target dial design is element-decomposed to obtain a set of dial layout elements. This step involves image segmentation and feature recognition technology to decompose the overall design into independent visual and functional elements, such as time display, date, step counter, heart rate monitoring, etc. Each element is labeled with its attribute information such as type, size, style, etc. Based on the decomposed set of dial layout elements, a layout constraint graph is constructed. This graph structure represents the spatial relationship and layout constraints between elements. The nodes in the graph represent individual layout elements, and the edges represent the relative position relationship, alignment requirements, or spacing constraints between elements. For example, the time display may need to be located in the center, while other functional modules are distributed around it. When constructing this constraint graph, factors such as the aesthetic principles of design, functional importance, and user usage habits are considered. Then, according to the spatial relationships defined in the layout constraint graph, the dial layout elements are accurately positioned and calculated. This process can use a force-directed graph drawing algorithm or a constraint satisfaction problem solver. The force-directed algorithm converts the layout problem into a physical simulation, and determines the final position by simulating the repulsion and attraction between elements. The constraint satisfaction problem solver expresses the layout requirements as a series of mathematical constraints and finds a solution that satisfies all constraints through an optimization algorithm. The output of this step is the precise coordinate information of each element on the dial. Based on the calculated element coordinate information, specific layout control instructions are generated. These instructions describe in detail how to place and render each dial element on the display interface of the smart bracelet. The instruction set includes parameters such as the position, size, rotation angle, stacking order, and transparency of the element. At the same time, the animation effects of dynamic elements, such as the movement of the second hand or the transition effect of data update, also need to be considered. These instructions constitute a complete dial layout control scheme. The last step is to convert the dial layout control scheme into a display control signal specific to the smart bracelet. This conversion process needs to take into account the hardware specifications of the bracelet, such as screen resolution, color depth, refresh rate, etc. The control signal includes low-level graphics drawing commands, bitmap data, or vector graphics instructions, depending on the display system architecture of the bracelet. The instructions also need to be optimized during conversion to ensure smooth display effects under limited processing power and memory conditions. The generated display control signal is then sent to the display module of the smart bracelet through the device's communication interface. This involves Bluetooth Low Energy (BLE) or other wireless communication protocols, and the efficiency and reliability of data transmission need to be considered.After receiving these signals, the display module parses and performs corresponding rendering operations to present the designed dial layout on the screen.

[0078] In this embodiment, the user's data is processed by multi-sensor fusion to obtain user feature data. Cluster analysis and time series analysis are used to construct a dynamic user portrait. According to the dynamic user portrait, elements are selected from the dial element library, and the generative adversarial network and variational autoencoder are used to combine and adjust to obtain candidate dial designs. The source domain knowledge is transferred to the target domain through transfer learning, and fine-tuned to form a personalized migration model. According to the personalized migration model, the candidate dial designs are evaluated and sorted to obtain the target dial design, and the dial layout is controlled. This method automatically selects and adjusts the dial elements according to the collected data, generates a dial design that meets user needs, and continuously optimizes according to user feedback, which can significantly improve the user experience of the smart bracelet and better meet the dynamic needs and preferences of users.

[0079] The above describes the dial layout control method of the smart bracelet in the embodiment of the present invention. The following describes the dial layout control device of the smart bracelet in the embodiment of the present invention. Figure 2 , an embodiment of the dial layout control device of the smart bracelet in the embodiment of the present invention includes:

[0080] The data processing module 201 is used to perform multi-sensor fusion processing on the user physiological data, behavior data and environmental data collected by the smart bracelet to obtain fused data, and to perform preprocessing and feature extraction on the fused data to obtain user feature data;

[0081] A user analysis module 202 is used to perform cluster analysis on target users based on the user feature data to obtain user activity types and life patterns, and to perform time series analysis on the user feature data to obtain periodic patterns and long-term trends of user behavior;

[0082] A portrait generation module 203 is used to construct an initial user portrait according to the user's activity type, life pattern, periodic pattern and long-term trend, and to update the initial user portrait according to explicit feedback information from the target user to obtain a dynamic user portrait;

[0083] A candidate watch face determination module 204 is used to select watch face elements from a preset watch face element library according to the dynamic user portrait, and combine and adjust the selected watch face elements using a generative adversarial network and a variational autoencoder to obtain a candidate watch face design;

[0084] The transfer learning module 205 is used to transfer the source domain knowledge of the preset source domain dial recommendation model to the target domain of the target user by using the transfer learning method to obtain a transfer learning model, and fine-tune the transfer learning model according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model;

[0085] The layout control module 206 is used to evaluate and sort the candidate dial designs according to the personalized migration model to obtain a target dial design, and control the dial layout of the smart bracelet according to the target dial design.

[0086] In an embodiment of the present invention, the dial layout control device of the smart bracelet runs the dial layout control method of the smart bracelet, and the dial layout control device of the smart bracelet processes the user's data through multi-sensor fusion to obtain user feature data. A dynamic user portrait is constructed using cluster analysis and time series analysis. According to the dynamic user portrait, elements are selected from the dial element library, and combined and adjusted using a generative adversarial network and a variational autoencoder to obtain a candidate dial design. The source domain knowledge is transferred to the target domain through transfer learning, and a personalized transfer model is formed by fine-tuning. According to the personalized transfer model, the candidate dial design is evaluated and sorted to obtain the target dial design, and the dial layout control is performed. This method automatically selects and adjusts the dial elements according to the collected data, generates a dial design that meets the user's needs, and continuously optimizes according to user feedback, which can significantly improve the user experience of the smart bracelet and better meet the user's dynamic needs and preferences.

[0087] above Figure 2 From the perspective of modular functional entities, the dial layout control device of the smart bracelet in the embodiment of the present invention is described in detail. The dial layout control device of the smart bracelet in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0088] Figure 3It is a structural schematic diagram of a dial layout control device of a smart bracelet provided by an embodiment of the present invention. The dial layout control device 300 of the smart bracelet may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the dial layout control device 300 of the smart bracelet. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the dial layout control device 300 of the smart bracelet to implement the steps of the dial layout control method of the above-mentioned smart bracelet.

[0089] The smart bracelet dial layout control device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 3 The structure of the dial layout control device of the smart bracelet shown does not constitute a limitation on the dial layout control device of the smart bracelet provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0090] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the dial layout control method of the smart bracelet.

[0091] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0092] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0093] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for controlling the dial layout of a smart bracelet, characterized in that: The dial layout control method of the smart bracelet includes: Perform multi-sensor fusion processing on the user physiological data, behavior data and environmental data collected by the smart bracelet to obtain fused data, and perform preprocessing and feature extraction on the fused data to obtain user feature data; Performing cluster analysis on the target users according to the user characteristic data to obtain the user activity type and life pattern, and performing time series analysis on the user characteristic data to obtain the periodic pattern and long-term trend of the user behavior; Constructing an initial user profile based on the user's activity type, life pattern, periodic pattern, and long-term trend, and updating the initial user profile based on explicit feedback from the target user to obtain a dynamic user profile; Selecting dial elements from a preset dial element library according to the dynamic user portrait, and combining and adjusting the selected dial elements using a generative adversarial network and a variational autoencoder to obtain a candidate dial design; selecting dial elements from a preset dial element library according to the dynamic user portrait, and combining and adjusting the selected dial elements using a generative adversarial network and a variational autoencoder to obtain a candidate dial design includes: performing feature vectorization processing on the dynamic user portrait to obtain a user preference vector, and performing similarity calculation on the preset dial element library according to the user preference vector to obtain a preliminary matching dial element set; applying a multi-objective optimization algorithm to the preliminary matching dial element set to obtain a dial element combination, and applying a conditional generative adversarial network to the dial element combination to obtain an initial candidate dial design; performing latent space encoding on the initial candidate dial design using a variational autoencoder to obtain a potential representation of the dial design, and performing interpolation and perturbation operations on the potential representation to obtain a diverse set of potential representations; reconstructing the diverse set of potential representations using a decoder of a variational autoencoder to obtain multiple candidate dial designs; The method of applying a multi-objective optimization algorithm to the preliminary matched dial element set to obtain a dial element combination, and applying a conditional generative adversarial network to the dial element combination to obtain an initial candidate dial design includes: extracting features from the preliminary matched dial element set to obtain element feature vectors, and constructing a dial element complementarity matrix based on the element feature vectors to obtain a complementary relationship between elements; applying a genetic algorithm to the complementary relationship to obtain an initial dial element combination population, and performing Pareto sorting on the initial dial element combination population to obtain a non-dominated solution set; applying a simulated annealing algorithm to the non-dominated solution set to obtain a locally optimal dial element combination population. A dial element combination is constructed, and a cross-mutation operation is performed on the locally optimal dial element combination to obtain a diversified dial element combination; a condition vector is constructed for the diversified dial element combination to obtain a design constraint condition, the design constraint condition is input into a generator of a conditional generative adversarial network to obtain an initial candidate dial design sample, and a discriminator of a conditional generative adversarial network is applied to the initial candidate dial design sample to obtain a authenticity score; the generator and the discriminator are alternately trained according to the authenticity score to obtain an optimized conditional generative adversarial network, and a plurality of initial candidate dial designs are generated using the optimized conditional generative adversarial network; The source domain knowledge of the preset source domain dial recommendation model is transferred to the target domain of the target user by using a transfer learning method to obtain a transfer learning model, and the transfer learning model is fine-tuned according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model; the source domain knowledge of the preset source domain dial recommendation model is transferred to the target domain of the target user by using a transfer learning method to obtain a transfer learning model, and the transfer learning model is fine-tuned according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model, including: performing structural analysis on the preset source domain dial recommendation model to obtain a source domain knowledge representation, and constructing a knowledge graph based on the source domain knowledge representation, Obtain a source domain knowledge graph, apply a graph neural network to the source domain knowledge graph, and obtain a graph embedding representation; perform feature extraction on the dynamic user profile and historical interaction data of the target user to obtain a target domain feature vector, and align the target domain feature vector with the graph embedding representation to obtain a cross-domain mapping matrix; migrate the parameters of the source domain dial recommendation model according to the cross-domain mapping matrix to obtain an initial transfer learning model, and apply domain adversarial training to the initial transfer learning model to obtain a domain-invariant feature representation; construct a multi-task learning framework based on the domain-invariant feature representation, and fine-tune the multi-task learning framework according to the dynamic user profile and historical interaction data of the target user to obtain a personalized migration model; The candidate dial designs are evaluated and sorted according to the personalized migration model to obtain a target dial design, and the dial layout of the smart bracelet is controlled according to the target dial design. The evaluating and sorting of the candidate dial designs according to the personalized migration model to obtain a target dial design, and the dial layout of the smart bracelet is controlled according to the target dial design, including: evaluating and sorting the candidate dial designs according to the personalized migration model to obtain a target dial design; performing element decomposition on the target dial design to obtain a dial layout element set, and constructing a layout constraint diagram according to the dial layout element set to obtain the spatial relationship between the elements; positioning and calculating the dial layout elements according to the spatial relationship to obtain element coordinate information, and generating layout control instructions according to the element coordinate information to obtain a dial layout control scheme; converting the dial layout control scheme into a device-specific control signal to obtain a display control signal of the smart bracelet, and sending the display control signal to the display module of the smart bracelet to realize dial layout control.

2. The dial layout control method of the smart bracelet according to claim 1, characterized in that: The multi-sensor fusion processing is performed on the physiological data, behavior data and environmental data of the target user collected by the smart bracelet to obtain fused data, and the fused data is preprocessed and feature extracted to obtain user feature data, including: Applying data alignment and time synchronization algorithms to the user's physiological data, behavioral data, and environmental data collected by the smart bracelet to obtain multimodal data with consistent time series, and using an adaptive weighted fusion algorithm on the multimodal data with consistent time series to obtain preliminary fused data; Applying a hybrid filtering algorithm combining Kalman filtering and particle filtering to the preliminary fused data to obtain fused data after noise suppression and outlier removal, and performing segmented statistical feature extraction and frequency domain transformation on the fused data to obtain time-frequency domain features; Applying nonlinear principal component analysis and independent component analysis to the time-frequency domain features to obtain a feature representation after dimensionality reduction, and performing time series modeling on the feature representation using a recursive neural network to obtain a dynamic feature sequence; Physiological state indicators, behavioral pattern indicators and environmental adaptability indicators are calculated according to the dynamic feature sequence to obtain user feature data.

3. The dial layout control method of the smart bracelet according to claim 2, characterized in that: The cluster analysis of the target users according to the user characteristic data to obtain the user activity type and life pattern, and the time series analysis of the user characteristic data to obtain the periodic pattern and long-term trend of the user behavior include: Calculate the distance between the user feature data and the preset user group cluster centers to obtain the similarity between the target user and each cluster center, and determine the activity type and life pattern of the target user based on the cluster characteristics of the cluster center with the highest similarity; Applying sliding window technology and adaptive segmentation algorithm to the user feature data to obtain time series segments, and performing fast Fourier transform and continuous wavelet transform on the time series segments to obtain multi-scale spectrum features, and using peak detection and harmonic analysis based on the multi-scale spectrum features to obtain a periodic pattern of user behavior; The exponentially weighted moving average method is applied to the user feature data to obtain a smoothed time series, and the smoothed time series is subjected to seasonal decomposition and trend extraction to obtain trend components. The Prophet time series prediction model is used based on the trend components to perform long-term trend prediction and change point detection to obtain the long-term trend of user behavior.

4. A dial layout control device for a smart bracelet, characterized in that: The dial layout control device of the smart bracelet includes: The data processing module is used to perform multi-sensor fusion processing on the user physiological data, behavior data and environmental data collected by the smart bracelet to obtain fused data, and to perform preprocessing and feature extraction on the fused data to obtain user feature data; A user analysis module, used to perform cluster analysis on target users based on the user feature data to obtain user activity types and life patterns, and to perform time series analysis on the user feature data to obtain periodic patterns and long-term trends of user behavior; A portrait generation module is used to construct an initial user portrait according to the user's activity type, life pattern, periodic pattern and long-term trend, and update the initial user portrait according to the explicit feedback information of the target user to obtain a dynamic user portrait; A candidate dial determination module is used to select dial elements from a preset dial element library according to the dynamic user portrait, and use a generative adversarial network and a variational autoencoder to combine and adjust the selected dial elements to obtain a candidate dial design; the selecting dial elements from a preset dial element library according to the dynamic user portrait, and using a generative adversarial network and a variational autoencoder to combine and adjust the selected dial elements to obtain a candidate dial design includes: performing feature vectorization processing on the dynamic user portrait to obtain a user preference vector, and combining and adjusting the preset dial elements according to the user preference vector The method further comprises: performing similarity calculation on the dial element library to obtain a preliminary matching set of dial elements; applying a multi-objective optimization algorithm to the preliminary matching set of dial elements to obtain a dial element combination, and applying a conditional generative adversarial network to the dial element combination to obtain an initial candidate dial design; performing latent space encoding on the initial candidate dial design using a variational autoencoder to obtain a latent representation of the dial design, and performing interpolation and perturbation operations on the latent representation to obtain a diversified set of latent representations; reconstructing the diversified set of latent representations using a decoder of a variational autoencoder to obtain multiple candidate dial designs; The method of applying a multi-objective optimization algorithm to the preliminary matched dial element set to obtain a dial element combination, and applying a conditional generative adversarial network to the dial element combination to obtain an initial candidate dial design includes: extracting features from the preliminary matched dial element set to obtain element feature vectors, and constructing a dial element complementarity matrix based on the element feature vectors to obtain a complementary relationship between elements; applying a genetic algorithm to the complementary relationship to obtain an initial dial element combination population, and performing Pareto sorting on the initial dial element combination population to obtain a non-dominated solution set; applying a simulated annealing algorithm to the non-dominated solution set to obtain a locally optimal dial element combination population. A dial element combination is constructed, and a cross-mutation operation is performed on the locally optimal dial element combination to obtain a diversified dial element combination; a condition vector is constructed for the diversified dial element combination to obtain a design constraint condition, the design constraint condition is input into a generator of a conditional generative adversarial network to obtain an initial candidate dial design sample, and a discriminator of a conditional generative adversarial network is applied to the initial candidate dial design sample to obtain a authenticity score; the generator and the discriminator are alternately trained according to the authenticity score to obtain an optimized conditional generative adversarial network, and a plurality of initial candidate dial designs are generated using the optimized conditional generative adversarial network; A transfer learning module is used to transfer the source domain knowledge of a preset source domain dial recommendation model to the target domain of the target user by using a transfer learning method to obtain a transfer learning model, and fine-tune the transfer learning model according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model; the transfer learning method is used to transfer the source domain knowledge of a preset source domain dial recommendation model to the target domain of the target user to obtain a transfer learning model, and fine-tune the transfer learning model according to the dynamic user portrait and the historical interaction data of the target user to obtain a personalized transfer model, including: performing structural analysis on the preset source domain dial recommendation model to obtain a source domain knowledge representation, and constructing a knowledge representation based on the source domain knowledge representation. Recognize the graph to obtain the source domain knowledge graph, apply the graph neural network to the source domain knowledge graph to obtain the graph embedding representation; extract features from the dynamic user portrait and historical interaction data of the target user to obtain the target domain feature vector, and align the target domain feature vector with the graph embedding representation to obtain a cross-domain mapping matrix; migrate the parameters of the source domain dial recommendation model according to the cross-domain mapping matrix to obtain an initial transfer learning model, and apply domain adversarial training to the initial transfer learning model to obtain a domain-invariant feature representation; construct a multi-task learning framework based on the domain-invariant feature representation, and fine-tune the multi-task learning framework according to the dynamic user portrait and historical interaction data of the target user to obtain a personalized migration model; A layout control module is used to evaluate and sort the candidate dial designs according to the personalized migration model to obtain a target dial design, and to control the dial layout of the smart bracelet according to the target dial design; the evaluating and sorting the candidate dial designs according to the personalized migration model to obtain a target dial design, and controlling the dial layout of the smart bracelet according to the target dial design includes: evaluating and sorting the candidate dial designs according to the personalized migration model to obtain a target dial design; performing element decomposition on the target dial design to obtain a dial layout element set, and constructing a layout constraint diagram based on the dial layout element set to obtain the spatial relationship between the elements; performing positioning calculation on the dial layout elements according to the spatial relationship to obtain element coordinate information, and generating layout control instructions based on the element coordinate information to obtain a dial layout control scheme; converting the dial layout control scheme into a device-specific control signal to obtain a display control signal of the smart bracelet, and sending the display control signal to the display module of the smart bracelet to realize dial layout control.

5. A dial layout control device for a smart bracelet, characterized in that: The dial layout control device of the smart bracelet includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instruction in the memory so that the dial layout control device of the smart bracelet executes the steps of the dial layout control method of the smart bracelet as described in any one of claims 1-3.

6. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by the processor, the steps of the dial layout control method of the smart bracelet as described in any one of claims 1-3 are implemented.