A user state evaluation method and system based on heart rate characteristics

Abstract

The application discloses a user state evaluation method and system based on heart rate characteristics, relates to the technical field of data analysis, and comprises the following steps: collecting heart rate time series data; reconstructing the phase space of the heart rate time series data, and extracting chaotic attractor features representing heart rate complex dynamic characteristics; constructing a heart rate dynamics graph according to the chaotic attractor features, and performing feature aggregation on the dynamics graph by using a graph convolution network to obtain global topological dynamic features; inputting the global topological dynamic features, an attention mechanism and user individual baseline features into a pre-trained neural network to output an intermediate state value reflecting the current state of the user; and adaptively adjusting the intermediate state value to generate a final state evaluation result. The application not only improves the accuracy and robustness of state evaluation, but also provides scientific and reliable technical support for intelligent health management and risk early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a user state evaluation method and system based on heart rate characteristics. BACKGROUND

[0002] With the popularity of wearable devices and physiological signal monitoring technology, heart rate signals, as an important reflection of human health and psychological state, have attracted high attention from the fields of medicine, health management and artificial intelligence. Existing health evaluation methods mainly rely on heart rate, heart rate variability and other indicators, combined with statistical characteristics or machine learning models, to preliminarily judge the user's physical state, emotional changes and health risks. However, human heart rate signals are a complex nonlinear time series signal, whose change law is comprehensively regulated by multiple factors such as the nervous system, endocrine, environmental stress, emotional fluctuations, etc., showing high dynamicity and individual difference. Single indicators or static statistical models are difficult to accurately reveal the deep dynamic mechanism.

[0003] In recent years, dynamic system theory, chaos theory, time series modeling, deep learning and other technologies have been gradually applied to physiological signal analysis, aiming to explore the complex rules and personalized characteristics in heart rate data. However, in practical applications, how to simultaneously consider the multi-scale characteristics of data, individual physiological differences, context relevance and the adaptive ability of the model remains a difficulty faced by the academic and industrial communities. Current systems mainly rely on linear analysis, single-scale modeling or end-to-end deep learning, which are difficult to achieve high precision, strong interpretability and continuous self-optimization, limiting the further promotion and landing of heart rate signals in personalized health monitoring and intelligent intervention scenarios. Therefore, it is urgent to develop new heart rate signal modeling and evaluation methods to meet the efficient, intelligent and dynamic evaluation needs of user state. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a user state evaluation method based on heart rate characteristics to solve the problems of dynamic and accurate evaluation and individualized adaptation in the prior art.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a user state evaluation method based on heart rate characteristics, which comprises collecting heart rate data of a user and performing time series synchronization on the heart rate data to obtain heart rate time series data;

[0008] Performing phase space reconstruction on the heart rate time series data to extract chaotic attractor features representing the complex dynamic characteristics of heart rate;

[0009] A heart rate dynamics graph is constructed based on the chaotic attractor features, and a graph convolutional network is used to aggregate the features of the dynamics graph to obtain global topological dynamic features.

[0010] By acquiring contextual labels related to the user's state and embedding these labels into an attention mechanism, the global topological dynamic features, the attention mechanism, and the user's individual baseline features are input into a pre-trained neural network, which outputs an intermediate state value that reflects the user's current state.

[0011] The intermediate state values ​​are adaptively adjusted to generate the final state evaluation result.

[0012] As a preferred embodiment of the user status assessment method based on heart rate characteristics described in this invention, the heart rate time-series data is represented as follows: ;in, Indicates the first Heart rate values ​​at each time point This represents the total number of data points;

[0013] The phase space reconstruction includes performing feature analysis on the heart rate time-series data and generating J sets of data pairs with delay time and embedding dimension based on the feature analysis results; wherein, the j-th set of data pairs with delay time and embedding dimension is represented as follows: ; This represents the delay time of the j-th data pair; This represents the number of embedding dimensions for the j-th data pair;

[0014] For each set of data Using the delayed embedding method, heart rate data is embedded into... In the phase space, we obtain the reconstructed vector:

[0015]

[0016] in, This indicates that for the j-th data pair, the first... One reconstructed phase space vector; This represents all valid indexes that can be constructed for the j-th data pair; This indicates that under the j-th data set, in the j-th... The data in the first dimension of a reconstructed phase space; This indicates that under the j-th data set, in the j-th... The data in the Nth dimension of a reconstructed phase space;

[0017] After integrating the j-th data pair, the reconstructed trajectory point set is as follows: ;

[0018] in, represents the total number of effective trajectory points; the trajectory point sets under all data pairs are integrated to obtain a set .

[0019] As a preferred scheme of the user state evaluation method based on heart rate features, the feature analysis comprises: calculating a mutual information function based on the heart rate time series data, and selecting a first minimum value of a mutual information curve as a delay time; after selecting one delay time data, a minimum embedding dimension when a false nearest neighbor ratio falls to a preset threshold is selected as an embedding dimension based on a false nearest neighbor method.

[0020] According to all completed data pairs, the heart rate time series data corresponding to the reconstruction vector are removed from the heart rate time series data, and the mutual information function is recalculated to select the delay time and the embedding dimension is selected by using the false nearest neighbor method in the removed heart rate time series data.

[0021] In the selection, the data pairs satisfy the constraint that the number is not less than a preset value.

[0022] After the constraint is satisfied, in the generation, the change of the minimum value of the mutual information is analyzed, if the first minimum value of the mutual information calculated next time is less than a preset value, the next group of data pairs is continuously generated until the first minimum value of the mutual information calculated next time is not less than the preset value or a preset maximum generation number is reached.

[0023] As a preferred scheme of the user state evaluation method based on heart rate features, the chaotic attractor feature comprises: in the trajectory point set, a dynamic feature reflecting heart rate system complexity, chaos and nonlinear regulation law; in each trajectory point, a chaotic feature vector is reflected.

[0024] Chaotic attractor features are extracted for each group of data pairs, and the chaotic attractor feature of the jth trajectory point set after reconstruction is denoted as , including the chaotic feature vector of each trajectory point; the chaotic feature sets under all parameter groups are integrated to form a global multi-dimensional chaotic feature matrix:

[0025]

[0026] Each column is a group of features, and each row corresponds to a chaotic description dimension.

[0027] As a preferred scheme of the user state evaluation method based on heart rate features, the heart rate dynamics diagram comprises: for each dimension of the trajectory point set , respectively, construct a graph structure; each reconstructed phase space trajectory point in the trajectory point set is a node; the node feature is the corresponding chaotic feature vector; the weight between two nodes is defined as the Euclidean distance between the two chaotic feature vectors;

[0028] By introducing dynamic timing constraints, only connecting adjacent or time-adjacent nodes, J time heart rate dynamics graphs are formed;

[0029] The feature aggregation includes, using principal component analysis, identifying key nodes in the nodes of the J time heart rate dynamics graphs, respectively, to obtain J sets of key nodes;

[0030] All key nodes are connected in a new graph structure according to the time sequence, and the chaotic feature vectors are recalculated using the key nodes arranged in the time sequence, and the Euclidean distance between the chaotic feature vectors is used to represent the weight between two nodes.

[0031] As a preferred scheme of the user state evaluation method based on heart rate characteristics, wherein: the attention mechanism includes inputting the identified user state information and real-time heart rate variability data into the pre-trained Bayesian network; output the attention weight distribution result of the current state of the user;

[0032] The intermediate state value is a vector, the direction represents the emotional direction, and the numerical value represents the emotional score.

[0033] As a preferred scheme of the user state evaluation method based on heart rate characteristics, wherein: the adaptive adjustment includes setting the numerical value range of the intermediate state value in the 0-100 as the predicted interval; if the numerical value of the intermediate state value exceeds the predicted interval, the intervention is automatically triggered and the heart rate data after the intervention is collected, the feedback data is used to optimize the neural network and the individual baseline feature, and the closed-loop adaptive optimization of the evaluation model is realized; if the numerical value of the intermediate state value exceeds the predicted interval, the state evaluation result under each emotion is obtained according to the intermediate state value through the preset mapping function of each state type, as the current user state;

[0034] Wherein, the initial value of the individual baseline feature is a preset standard value; the intervention process includes adjusting the baseline feature of the individual and the weight of the neural network respectively according to the minimum step, optimizing the parameters by using the annealing algorithm until the numerical value of the intermediate state value exceeds the predicted interval, and completing the optimization.

[0035] Secondly, the application provides a user state evaluation system based on heart rate characteristics, comprising a collection unit that collects heart rate data of a user and synchronizes the heart rate data in time sequence to obtain heart rate time sequence data.

[0036] The feature extraction unit reconstructs the phase space of the heart rate time series data and extracts chaotic attractor features that characterize the complex dynamics of heart rate.

[0037] The feature aggregation unit constructs a heart rate dynamics graph based on the chaotic attractor features, and uses a graph convolutional network to aggregate features of the dynamics graph to obtain global topological dynamic features.

[0038] The analysis unit acquires contextual labels related to the user's state and embeds these labels into an attention mechanism. It then inputs the global topological dynamic features, the attention mechanism, and the user's individual baseline features into a pre-trained neural network and outputs intermediate state values ​​that reflect the user's current state.

[0039] The output unit performs adaptive adjustments to the intermediate state values ​​to generate the final state evaluation result.

[0040] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the user state assessment method based on heart rate characteristics as described in the first aspect of the present invention.

[0041] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the user state assessment method based on heart rate characteristics as described in the first aspect of the present invention.

[0042] The beneficial effects of this invention are as follows: By modeling the multi-scale dynamic characteristics and reconstructing the phase space of heart rate data, the complex nonlinear regulatory laws inherent in heart rate signals can be fully explored, improving the ability to identify the user's true state. The introduction of key node identification and topology analysis effectively enhances the model's sensitivity to abnormal changes and key physiological events. Combining attention mechanisms with individual baseline characteristics achieves adaptive fusion of contextual information and user differences, making the assessment results more personalized and interpretable. This invention can automatically trigger intervention when the state is abnormal and utilize intervention feedback for closed-loop optimization of model and baseline parameters, significantly improving the system's self-learning and continuous evolution capabilities. The overall method not only improves the accuracy and robustness of state assessment but also provides scientific and reliable technical support for intelligent health management and risk warning. Attached Figure Description

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort based on these drawings.

[0044] Figure 1 Flow chart of the user state evaluation method based on heart rate features. DETAILED DESCRIPTION

[0045] In order to make the above objectives, features and advantages of the present application more apparent, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0046] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the concept of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0047] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is separate or alternative to other embodiments.

[0048] Reference Signs List Figure 1 For one embodiment of the present application, the embodiment provides a user state evaluation method based on heart rate features, comprising the following steps:

[0049] S1: collecting heart rate data of a user, and performing time sequence synchronization on the heart rate data to obtain heart rate time sequence data.

[0050] Further, the heart rate time sequence data is expressed as: ; wherein, represents the heart rate value at the th time point, represents the total number of data points.

[0051] By collecting and synchronizing the user's heart rate data, complete, continuous and standardized heart rate time series data can be obtained, providing basic data support for subsequent dynamic feature extraction, time series modeling and state evaluation. This step helps to eliminate time offset, missing or redundancy in the data collection process, ensuring that the obtained heart rate time series data is consistent and accurate, thereby laying a solid foundation for systematically analyzing heart rate variation rules and accurately reflecting the user's real physiological state. This design provides a high-quality and reliable data source for subsequent complex feature analysis, modeling and evaluation, improving the scientificity and effectiveness of the entire evaluation process.

[0052] S2: Phase space reconstruction is performed on the heart rate time series data to extract chaotic attractor features representing the complex dynamics of heart rate.

[0053] The phase space reconstruction includes performing feature analysis on the heart rate time series data, and generating J sets of delay time and embedding dimension data pairs based on the feature analysis results; wherein the jth set of delay time and embedding dimension data pair is represented as ; represents the delay time of the jth set of data pair; represents the embedding dimension number of the jth set of data pair.

[0054] For each set of data pair , the delay embedding method is used to embed the heart rate data into dimensional phase space to obtain the reconstructed vector:

[0055]

[0056] wherein, represents the jth set of data pair under the th reconstructed phase space vector; represents the jth set of data pair under all valid indices that can be constructed. represents the jth set of data under the th reconstructed phase space, the first dimension data; represents the jth set of data under the th reconstructed phase space, the Nth dimension data. The integrated trajectory point set under the jth set of data pair is obtained as: . Wherein, represents the total number of valid trajectory points; the trajectory point sets under all data pairs are integrated to obtain the set .

[0057] Further, based on the heart rate time series data, a mutual information function is calculated, and a first minimum value of a mutual information curve is selected as a delay time; after selecting one delay time data, based on the nearest neighbor method, the minimum embedding dimension when the nearest neighbor ratio falls below a preset threshold is selected as the embedding dimension.

[0058] According to all completed data pairs, the heart rate time series data corresponding to the reconstruction vector is removed from the heart rate time series data, and the mutual information function is recalculated to select the delay time and the embedding dimension using the nearest neighbor method.

[0059] In the selection process, the data pairs meet the constraint that the number is not less than a preset value. After meeting the constraint, the data pairs are generated, the change of the minimum value of the mutual information is analyzed, if the first minimum value of the mutual information calculated next time is less than a preset value, the next group of data pairs is generated, until the first minimum value of the mutual information calculated next time is not less than a preset value or the maximum generation number is reached.

[0060] It is to be noted that by reconstructing the heart rate time series data in multiple groups of parameters, the complex dynamic behavior of the heart rate signal at different scales can be fully captured. The mutual information function is used to select the optimal delay time, and the nearest neighbor method is used to determine the embedding dimension, which helps to ensure the accuracy and scientificity of the reconstructed phase space, and avoids feature omission or redundancy. The progressive generation of multiple groups of parameter data pairs not only enhances the adaptability to the multi-scale and nonlinear structure of the heart rate signal, but also improves the richness and discriminability of the dynamic characteristics. By removing the used data after each reconstruction, the information overlap can be reduced, and the independence and representativeness of the features can be improved. Dynamic adjustment of the data pair generation process can prevent invalid expansion of parameters or redundancy of feature dimensions, and ensure that the final feature set not only meets the comprehensiveness, but also has high efficiency.

[0061] The chaotic attractor feature includes a trajectory point set, reflecting the complexity, chaos and nonlinear regulation law of the heart rate system; in each trajectory point, a chaotic feature vector is reflected. In this embodiment, the chaotic attractor feature refers to a set of dynamic features capable of reflecting the complexity, chaos and nonlinear regulation law of the heart rate system, which are extracted from the chaotic attractor trajectory formed in the high-dimensional phase space after the phase space reconstruction of the heart rate time series data, including but not limited to: correlation dimension, maximum Lyapunov exponent, approximate entropy, sample entropy, average radius of attractor trajectory, traversal length, spatial density distribution, and topological structure parameters of attractor network, etc. This set of features, as the key input for user heart rate state evaluation and identification, can comprehensively and quantitatively reveal the nonlinear dynamics mechanism behind the heart rate changes. Among them, the correlation dimension is used to quantify the fractal complexity of the attractor trajectory, reflecting the dynamic change degree of freedom of the heart rate system. The maximum Lyapunov exponent measures the sensitivity of the attractor to initial perturbation, revealing the chaotic characteristics of heart rate regulation. If it is positive, it indicates that the system has chaotic behavior. Approximate entropy and sample entropy are used to characterize the complexity and unpredictability of the attractor, and the larger the value represents the more disordered and unpredictable the heart rate sequence. Attractor trajectory statistics include the average radius, traversal length, density distribution, etc. of the attractor in the phase space, which are used to reflect the spatial distribution structure and activity level of the trajectory.

[0062] The attractor network topological feature includes the node distribution, clustering coefficient, average path length, etc. of the network constructed based on the attractor trajectory, which is used to reveal the multiscale correlation and regulation mechanism of the heart rate dynamics system.

[0063] For each group of data pairs, the chaotic attractor feature is extracted, and the chaotic attractor feature of the jth reconstructed trajectory point set is denoted as , including the chaotic feature vector of each trajectory point. The chaotic features of all parameter groups are integrated to form a global multi-dimensional chaotic feature matrix:

[0064]

[0065] Wherein, each column is a group of features, and each row corresponds to a chaotic description dimension. By extracting the chaotic attractor feature of the reconstructed trajectory point set under each group of parameters and integrating it into a global multi-dimensional chaotic feature matrix, comprehensive and rich data support can be provided for subsequent key node screening and time series dynamics graph structure reconstruction. As a comprehensive expression carrier of multiple parameters and multiple scale dynamics features, this matrix enables the subsequent principal component analysis, key node identification, etc. to systematically find the core area and abnormal change points at different dynamics scales based on multi-dimensional information, improving the scientificity and effectiveness of feature screening and structure optimization. This design lays a solid foundation for more accurate state evaluation and model adaptive optimization.

[0066] S3: constructing a heart rate dynamics graph based on the chaotic attractor features, and performing feature aggregation on the dynamics graph by using a graph convolution network to obtain global topological dynamic features.

[0067] The heart rate dynamics graph includes, for each dimension of the trajectory point set , a graph structure is constructed respectively; each reconstructed phase space trajectory point in the trajectory point set is a node; the node feature is the corresponding chaotic feature vector; the weight between two nodes is defined as the Euclidean distance between the chaotic feature vectors of the two nodes.

[0068] By introducing dynamic temporal constraints, only connecting adjacent or temporally close nodes, J temporal heart rate dynamics graphs are formed. "Only connecting adjacent or temporally close nodes" means that when constructing the heart rate dynamics graph, only connections between reconstructed trajectory points that are continuous in time series or have a time interval less than a preset threshold are established. Specifically, assuming that the time series indexes of the trajectory points are and , if (where is a set time proximity threshold, usually 1 or less than a certain small constant), an edge is established between the two nodes in the dynamics graph.

[0069] The feature aggregation includes, using principal component analysis, identifying key nodes for the nodes in the J temporal heart rate dynamics graphs respectively to obtain J groups of key nodes. In this embodiment, each key node is associated with a non-key node that is directly connected in the respective temporal heart rate dynamics graph and has a weight greater than a preset value. When constructing a new graph structure, the relationship between each two connected nodes is analyzed, if after adding "a node associated with any of the two nodes" both nodes can establish a high-weight relationship with the newly added node, the newly added node is retained. Thus, the supplement of the nodes is realized.

[0070] All key nodes are connected in the new graph structure according to the time sequence, and the chaotic feature vectors are recalculated using the key nodes arranged in the time sequence. The weight between two nodes is represented by the Euclidean distance between the chaotic feature vectors.

[0071] It is to be noted that by constructing a heart rate dynamics graph based on chaotic attractor features and introducing dynamic temporal constraints, only connecting temporally close nodes, the continuous evolution law of the heart rate signal under multi-scale dynamics can be truly and finely reflected, and the redundant or meaningless association between irrelevant nodes is avoided. Each node takes the chaotic feature vector as an attribute, and the Euclidean distance as an edge weight, which ensures the expressive power of the graph structure for complex dynamic patterns.

[0072] The key nodes are screened by principal component analysis, which can automatically identify the most representative areas for dynamic evolution in multidimensional information, effectively highlighting abnormal points or regulatory cores. At the same time, by supplementing non-key nodes that are strongly associated with key nodes, the integrity of the local topology and global structure is corrected, avoiding information omission caused by simple screening.

[0073] Finally, the key nodes are reconnected in time sequence and the features are calculated, which not only improves the description ability of the new graph structure to the main line of dynamics, but also provides a more targeted and discriminative high-dimensional feature basis for subsequent feature aggregation and global state evaluation. This design ensures that the evaluation model can not only grasp the global evolution trend, but also sensitively perceive local anomalies, achieving the goal of high-precision, strong-explanation personalized dynamic evaluation.

[0074] S4: Obtain the previous text label related to the user state and embed the previous text label into the attention mechanism; input the global topology dynamic feature, the attention mechanism and the user individual baseline feature into the pre-trained neural network to output an intermediate state value reflecting the current state of the user.

[0075] The attention mechanism includes using a pre-trained Bayesian network to input the identified user state information (the identification result of the user state before the current time) and real-time heart rate variability data (here, a data sequence of a relatively short time interval, representing data sequences in the current and previous time period); output the attention weight distribution result of the current state of the user.

[0076] By introducing the previous text label related to the user state and embedding it into the attention mechanism, the model can effectively capture the dynamic relationship between historical state changes and current physiological signals, thereby enhancing the continuous perception of user behavior and psychological trends. By using a pre-trained Bayesian network to model historical state information and real-time heart rate variability data together, the attention weight of different features to the current state is output, realizing automatic allocation and causal inference of feature contribution.

[0077] Inputting the global topology dynamic feature, the attention weight and the user individual baseline feature into the neural network together can realize deep fusion of multi-modal and multi-scale information, improve the accurate discrimination and individualized adaptation ability of the current state of the user. The intermediate state value is output in vector form, so that the model can not only give a directional judgment of the emotion type, but also quantify the emotion or health risk degree, which is convenient for subsequent decision-making and dynamic intervention. This design ensures that the model's evaluation of the user's state is comprehensive and detailed, with continuity, individualization and interpretability, laying a solid foundation for efficient and intelligent health management.

[0078] The intermediate state value is a vector, the direction represents the emotion direction, and the numerical value represents the emotion score.

[0079] S5: adaptively adjusting the state of the intermediate state value to generate a final state evaluation result.

[0080] Further, the adaptive adjustment includes that if the numerical value of the intermediate state value exceeds the interval conforming to the prediction, an intervention is automatically triggered and heart rate data after the intervention is collected, the feedback data is used to optimize the neural network and the individual baseline feature, and closed-loop adaptive optimization of the evaluation model is realized; if the numerical value of the intermediate state value exceeds the interval conforming to the prediction, the state evaluation result under each emotion is obtained according to the intermediate state value through a preset mapping function (a preset conversion function that converts a vector into an evaluation result of each emotion and reflects a "component" of each emotion) of each state type, as the current user state.

[0081] The initial value of the individual baseline feature is a preset standard value; the intervention process includes that the baseline feature of the individual and the weight of the neural network are respectively adjusted by a minimum step, and a parameter optimization is performed by using an annealing algorithm until the numerical value of the intermediate state value exceeds the interval conforming to the prediction, and the optimization is completed.

[0082] Combining the annealing algorithm for parameter fine-tuning can balance between convergence efficiency and global optimization, so that the model can quickly adapt to the latest changes of the user state and realize closed-loop adaptive optimization. The preset mapping function converts the intermediate state vector into a multi-component and fine-grained quantitative result of various emotions or health risks, which provides a more operable basis for clinical intervention, health reminders and the like. Not only the sensitivity and adaptive ability of the system to abnormal states are improved, but also the individualized evolution and continuous optimization of the evaluation result in long-term operation are ensured, and finally the efficient, intelligent and reliable user state management goal is realized.

[0083] The embodiment also provides a user state evaluation system based on heart rate features, which comprises:

[0084] A collection unit collects heart rate data of a user and performs time sequence synchronization on the heart rate data to obtain heart rate time sequence data.

[0085] A feature extraction unit performs phase space reconstruction on the heart rate time sequence data to extract a chaotic attractor feature representing heart rate complex dynamic characteristics.

[0086] A feature aggregation unit constructs a heart rate dynamics graph according to the chaotic attractor feature and aggregates features of the dynamics graph by using a graph convolution network to obtain a global topological dynamic feature.

[0087] The analysis unit obtains a preceding context label related to the user state and embeds the preceding context label into an attention mechanism; inputs the global topology dynamic feature, the attention mechanism and the user individual baseline feature into a pre-trained neural network, and outputs an intermediate state value reflecting the current state of the user.

[0088] The output unit adaptively adjusts the state of the intermediate state value to generate a final state evaluation result.

[0089] The embodiment also provides a computer device suitable for the user state evaluation method based on heart rate features, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the user state evaluation method based on heart rate features proposed in the above embodiment.

[0090] The computer device can be a terminal, and the computer device comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, a trackball or a touchpad arranged on the shell of the computer device. In addition, the input device can also be an external keyboard, a touchpad or a mouse, etc.

[0091] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement a user state evaluation method based on heart rate characteristics as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk, or an optical disk.

[0092] To sum up, the present application is fused with complex dynamic modeling, graph neural network, attention mechanism and adaptive optimization, etc. The method first fully excavates the chaotic attractor features of heart rate time series data at different scales through multi-parameter phase space reconstruction, and constructs a multi-dimensional heart rate dynamics graph. The principal component analysis is used to automatically identify the key nodes, and the continuity and local anomalies of heart rate changes are accurately described by combining with dynamic time series constraints. The pre-trained Bayesian network is used to fuse the historical state and real-time heart rate variability data, output the attention weight for different features, realize the causal inference and information distribution.

[0093] The model inputs the global topological dynamic features, attention mechanism and individual baseline features into the neural network, and outputs the intermediate state value reflecting the emotional direction and score. For state abnormalities, the system automatically triggers intelligent intervention, and through incremental feedback and annealing algorithm, the neural network weight and individual baseline parameter are realized closed-loop adaptive optimization.

[0094] The accuracy, individualization and dynamic response capability of user state evaluation are significantly improved, which can not only reflect the complex physiological and psychological changes in detail, but also continuously optimize, and has good application prospect and popularization value.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than limiting the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A user state evaluation method based on heart rate features, characterized in that: The method comprises the following steps: Collecting heart rate data of a user and performing time synchronization on the heart rate data to obtain heart rate time series data; Performing phase space reconstruction on the heart rate time series data to extract chaotic attractor features representing the complex dynamic characteristics of the heart rate; Constructing a heart rate dynamics graph according to the chaotic attractor features and performing feature aggregation on the dynamics graph by using a graph convolution network to obtain global topological dynamic features; Obtaining a pre-context label related to the state of the user and embedding the pre-context label in an attention mechanism; inputting the global topological dynamic features, the attention mechanism and individual baseline features of the user into a pre-trained neural network to output an intermediate state value reflecting the current state of the user; Performing adaptive adjustment on the intermediate state value to generate a final state evaluation result; The heart rate timing data is represented as: ; wherein, represents a heart rate value at the th time point, represents the total number of data points; The phase space reconstruction comprises: performing feature analysis on the heart rate time series data, and generating J sets of data pairs of delay time and embedding dimension number according to a result of the feature analysis; wherein a jth set of data pairs of delay time and embedding dimension number is represented as ; denotes the delay time of the jth set of data pairs; denotes the embedding dimension number of the jth set of data pairs. For each set of data pairs , using the delay embedding method, the heart rate data is embedded into dimensional phase space to obtain the reconstructed vector: in, This indicates that for the j-th data pair, the first... One reconstructed phase space vector; This represents all valid indexes that can be constructed for the j-th data pair; This indicates that under the j-th data set, in the j-th... The data in the first dimension of a reconstructed phase space; This indicates that under the j-th data set, in the j-th... The data in the Nth dimension of a reconstructed phase space; The integrated jth group of data pairs, the reconstructed trajectory point set is obtained: ; wherein, denotes the total number of valid trajectory points; integrating the trajectory point sets under all data pairs to obtain the set ; The chaotic attractor features include dynamic features representing the complexity, chaos and nonlinear regulation law of the heart rate system in a set of trajectory points, and a chaotic feature vector is reflected in each trajectory point; The chaotic attractor feature is extracted for each group of data pairs, and the chaotic attractor feature of the jth group of trajectory point sets after reconstruction is denoted as , including the chaotic feature vector of each trajectory point; all chaotic feature sets under the parameters are integrated to form a global multi-dimensional chaotic feature matrix: where each column is a set of features and each row corresponds to a chaos description dimension; The heart rate dynamics graph comprises, for each dimension of the trajectory point set , respectively, construct a graph structure; each reconstructed phase space trajectory point in the trajectory point set is a node; the node feature is the corresponding chaotic feature vector; the weight between two nodes is defined as the Euclidean distance between the two chaotic feature vectors; By introducing dynamic time sequence constraints, only adjacent or time sequence similar nodes are connected to form J time sequence heart rate dynamics graphs; The feature aggregation includes using principal component analysis to identify key nodes in the nodes of the J time sequence heart rate dynamics graphs to obtain J sets of key nodes; All key nodes are connected in a new graph structure according to the time sequence, and the chaotic feature vectors are recalculated by using the key nodes arranged according to the time sequence, and the weight between two nodes is represented by the Euclidean distance between the chaotic feature vectors. 2.The user state assessment method based on heart rate features according to claim 1, wherein: The feature analysis includes calculating a mutual information function based on the heart rate time series data, and selecting the first minimum value of the mutual information curve as the delay time; After selecting one delay time data, the minimum embedding dimension when the false nearest neighbor ratio falls below a preset threshold is selected as the embedding dimension based on the false nearest neighbor method; According to all completed data pairs, the heart rate time series data corresponding to the reconstruction vectors are removed from the heart rate time series data, and the delay time and embedding dimension are recalculated by using the false nearest neighbor method on the removed heart rate time series data; When selecting, the constraint that the number of data pairs is not less than a preset value is met; After meeting the constraint, when generating the data pairs, the change of the minimum value of the mutual information is analyzed, if the first minimum value of the mutual information calculated next time is less than a preset value, the next group of data pairs is generated, and the process is repeated until the first minimum value of the mutual information calculated next time is not less than a preset value or the maximum generation number is reached. 3.The user state assessment method based on heart rate features according to claim 2, wherein: The attention mechanism includes inputting the identified user state information and real-time heart rate variability data into a pre-trained Bayesian network; and outputting an attention weight distribution result of the current state of the user; The intermediate state value is a vector, the direction of which represents the emotional direction, and the numerical value represents the emotional score. 4.The user state assessment method based on heart rate features according to claim 3, wherein: The adaptive adjustment includes setting the numerical value range of the intermediate state value in the interval of 0-100 as the interval conforming to the prediction. If the numerical value of the intermediate state value exceeds the interval consistent with the prediction, an intervention is triggered automatically and heart rate data after the intervention is collected, the feedback data is used to optimize the neural network and individual baseline characteristics, and closed-loop adaptive optimization of the evaluation model is realized; If the numerical value of the intermediate state value exceeds the interval consistent with the prediction, the state evaluation result under each emotion is obtained according to the intermediate state value through the mapping function preset for each state type, and is used as the current user state. The initial value of the individual baseline characteristics is a preset standard value. The intervention process includes adjusting the baseline characteristics of the individual and the weights of the neural network by the minimum step length respectively, optimizing the parameters by using the annealing algorithm until the numerical value of the intermediate state value exceeds the interval consistent with the prediction, and completing the optimization.

5. A user state assessment system based on heart rate characteristics, based on the user state assessment method based on heart rate characteristics according to any one of claims 1 to 4, characterized in that: The method comprises the following steps: The acquisition unit acquires heart rate data of the user and performs time sequence synchronization on the heart rate data to obtain heart rate time sequence data. The feature extraction unit performs phase space reconstruction on the heart rate time sequence data to extract chaotic attractor features representing heart rate complex dynamic characteristics. The feature aggregation unit constructs a heart rate dynamics graph according to the chaotic attractor features and uses a graph convolution network to aggregate features of the dynamics graph to obtain global topological dynamic features. The analysis unit obtains a previous text label related to the user state and embeds the previous text label into an attention mechanism, inputs the global topological dynamic features, the attention mechanism and individual baseline characteristics of the user into a pre-trained neural network, and outputs an intermediate state value reflecting the current state of the user. The output unit performs adaptive adjustment on the intermediate state value to generate a final state evaluation result. 6.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the user state evaluation method based on heart rate characteristics according to any one of claims 1-4.

7. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the user state evaluation method based on heart rate characteristics according to any one of claims 1-4.

Images