Intelligent ambulatory blood pressure monitoring and early warning system

Abstract

The invention relates to the field of blood pressure monitoring, in particular to an intelligent ambulatory blood pressure monitoring and early warning system which comprises a data acquisition module, a data transmission module, a data processing and analysis module, a user interface module, an alarm notification module, a cloud storage and service module and a feedback and support module. According to the scheme, an end-to-end frequency specificity double-attention adversarial network is adopted to comprehensively analyze the dynamic change of blood pressure, and a variational mode decomposition technology is adopted to adaptively extract components of different frequency bands from a time sequence of ambulatory blood pressure data; a double-attention mechanism is used for capturing the relation between areas in spatial distribution in a blood circulation system, an adversarial network is used for learning an overall distribution mode of the blood circulation system, extraction of global features is improved, an improved mixed loss function is used for training a network model, and the overall distribution mode of the blood circulation system is optimized. And the prediction accuracy of the end-to-end frequency-specific double-attention adversarial network is improved.

Description

Technical Field

[0001] The present invention relates to the field of blood pressure monitoring, and in particular to an intelligent dynamic blood pressure monitoring and early warning system. Background Art

[0002] Blood pressure is closely related to human health. It directly affects the function of the blood circulation system. Changes in blood pressure are closely related to the heart, blood vessels, kidneys, brain and eyes. Therefore, dynamic monitoring and prediction of blood pressure are very important. However, existing blood pressure monitoring equipment has limitations in blood pressure monitoring and prediction, resulting in insufficient accuracy in auxiliary diagnosis. In addition, the current monitoring model fails to effectively capture the spatiotemporal dynamic characteristics of blood pressure, and thus cannot provide accurate diagnostic support. Traditional loss functions are also insufficient when training specific blood pressure monitoring models, and cannot improve the prediction accuracy of the model. Summary of the invention

[0003] In view of the above situation, the present invention provides an intelligent dynamic blood pressure monitoring and early warning system. In view of the problem that the existing blood pressure monitoring equipment cannot effectively analyze and process the changes in blood pressure and lacks accuracy in helping diagnosis, an end-to-end frequency-specific dual-attention adversarial network is used to comprehensively analyze the dynamic changes of blood pressure. The network uses variational mode decomposition technology to adaptively extract components of different frequency bands from the time series of dynamic blood pressure data, which helps to capture the frequency-dependent characteristics of various regions in the blood circulation system, thereby providing deeper biological insights into blood pressure changes. In view of the fact that the existing monitoring model fails to effectively capture the spatiotemporal dynamic characteristics of blood pressure, a dual-attention mechanism is used in the end-to-end frequency-specific dual-attention adversarial network, which can capture the blood The relationship between various spatially distributed areas in the circulatory system can provide a more comprehensive analysis in space. The adversarial network is used to learn the overall distribution pattern of the blood circulation system to improve the extraction of global features. In addition, in order to improve the ability to extract global features, the adversarial network is used to learn the overall distribution pattern of the blood circulation system to improve the model's prediction accuracy for blood pressure changes; in view of the fact that traditional loss functions are also insufficient in training specific blood pressure monitoring models and cannot improve the model's prediction accuracy, the end-to-end frequency-specific dual-attention adversarial network uses an improved hybrid loss function to train the network model to improve the prediction accuracy of the end-to-end frequency-specific dual-attention adversarial network, helping doctors to promptly detect and diagnose potential health problems.

[0004] The intelligent dynamic blood pressure monitoring and early warning system provided by the present invention includes a data acquisition module, a data transmission module, a data processing and analysis module, a user interface module, an alarm notification module, a cloud storage and service module, and a feedback and support module;

[0005] The data acquisition module uses an ambulatory blood pressure monitor to perform 24-hour ambulatory blood pressure monitoring to obtain the user's blood pressure data;

[0006] The data transmission module uses Bluetooth, Wi-Fi and other wireless communication technologies to transmit the collected blood pressure data to the mobile application and cloud storage and service module;

[0007] The data processing and analysis module performs preprocessing operations on the collected blood pressure data to obtain initial data, and uses an end-to-end frequency-specific dual-attention adversarial network to perform in-depth analysis on the initial data to predict future blood pressure trends;

[0008] The user interface module provides an intuitive and easy-to-understand interface to display blood pressure readings, information from the alarm notification module and the feedback and support module, allowing users to view their history and set reminders and goals;

[0009] When the alarm notification module detects that the blood pressure is outside the normal range, it sends a warning notification to the user and his / her designated contact via SMS, email, or APP push, and provides suggestions and measures based on the severity;

[0010] The cloud storage and service module stores the user's blood pressure data in a secure cloud environment for a long time, supports cross-platform access, and provides an API interface for integration with other medical information systems;

[0011] The feedback and support module allows users to provide feedback on their usage experience, provide technical support and consulting services, and answer user questions.

[0012] Furthermore, the data processing and analysis module uses an end-to-end frequency-specific dual attention adversarial network to perform in-depth analysis on the initial data. The end-to-end frequency-specific dual attention adversarial network specifically includes the following steps:

[0013] Step S1: Variational mode decomposition, the non-stationary time series signal in the initial data is processed using variational mode decomposition to obtain initial time series data;

[0014] Step S2: Generator construction, using the dual attention module and the encoder-decoder network to process and analyze the initial time series data to obtain the predicted time series data;

[0015] Step S3: Discriminator construction, building an N-layer GRU architecture to distinguish between real time series data and predicted time series data;

[0016] Step S4: loss function, using a hybrid loss function for training, the hybrid loss function includes adversarial loss and supervised loss;

[0017] Step S5: Model training, the generator and the discriminator are trained in an alternating manner to extract global features from the initial time series data and capture important distribution and shape information;

[0018] Furthermore, in step S1, variational mode decomposition specifically includes the following steps:

[0019] Step S11: Signal decomposition: decompose the non-stationary time series signal in the initial data into K intrinsic mode functions, denoted as , each intrinsic mode function corresponds to a center frequency ;

[0020] Step S12: Variational constraints, minimizing an objective function to determine and their center frequencies , find a set of intrinsic mode functions, so that the sum of this set of intrinsic mode functions is equal to the non-stationary time series signal before processing;

[0021] Step S13: Lagrange multiplier method, using Lagrange multipliers The Lagrangian function is constructed by combining the parameters of the intrinsic mode function. The formula used is as follows: ;

[0022] In the formula, is the quadratic penalty parameter, Indicates time, It's about time The derivative of Dirac function, is an imaginary unit, is a complex exponential function, is the convolution operator, represents the intrinsic mode function, represents the center frequency, is the number of intrinsic mode functions, represents the original non-stationary time series signal, represents the Euclidean norm squared, represents the Lagrange multiplier, represents the inner product operation, represents the Lagrangian function;

[0023] Step S14: Iterative solution, using the alternating direction multiplier method to iteratively update the intrinsic mode function and center frequency , until convergence, the decomposed time series data is obtained, the formula used is as follows: ; ;

[0024] In the formula, represents the number of iterations, Indicates that except The sum of all other intrinsic mode functions except the intrinsic mode functions, Indicates The intrinsic mode functions, Represents the original time series signal The Fourier transform of represents the Lagrange multiplier The Fourier transform of represents a frequency variable, Indicates The updated value of the intrinsic mode function at the n+1th iteration in the frequency domain, represents the integral function, For the The center frequency corresponding to the intrinsic mode function is updated at the n+1th iteration;

[0025] Furthermore, in step S4, the loss function specifically includes the following steps:

[0026] Step S41: Generator loss. The adversarial loss includes the generator loss and the discriminator loss. The formula used for the generator loss function is as follows: ; ;

[0027] In the formula, represents the prediction score of the discriminator for the input real time series data, Represents real data, represents the Sigmoid function, is the index, represents the predicted time series data, represents the Sigmoid cross entropy loss function, Represents the time series data predicted by the discriminator The prediction score of is the generator loss function;

[0028] Step S42: Discriminator loss. The discriminator is used to distinguish between real time series data and predicted time series data. The discriminator loss function is used to minimize the error of the discriminator in distinguishing between true and false time series data. The formula used is as follows: ;

[0029] In the formula, represents the discriminator loss function, Represents the discriminator for real time series data The prediction score of

[0030] Step S43: supervision loss, which consists of absolute error loss, trend loss and distribution loss;

[0031] Step S44: Total generator loss. The total loss function of the generator combines the adversarial loss and the supervised loss, and each loss term is equipped with a weight parameter. The formula used is as follows: ;

[0032] In the formula, is the weight of the absolute error loss, is the weight of trend loss, is the weight of the distribution loss, is the weight of the generator adversarial loss, is the absolute error loss function, represents the trend loss function, represents the distribution loss function, is the total generator loss function.

[0033] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0034] (1) In view of the fact that existing blood pressure monitoring equipment cannot effectively analyze and process blood pressure changes and lacks accuracy in helping diagnosis, an end-to-end frequency-specific dual-attention adversarial network is used to comprehensively analyze the dynamic changes of blood pressure. The network uses variational mode decomposition technology to adaptively extract components of different frequency bands from the time series of dynamic blood pressure data, which helps to capture the frequency-dependent characteristics of various regions in the blood circulation system, thereby providing deeper biological insights into blood pressure changes;

[0035] (2) As the existing monitoring models fail to effectively capture the spatiotemporal dynamic characteristics of blood pressure, a dual attention mechanism is used in an end-to-end frequency-specific dual attention adversarial network. This mechanism can capture the relationship between the spatially distributed regions in the blood circulatory system, thereby providing a more comprehensive analysis in space. The adversarial network is used to learn the overall distribution pattern of the blood circulatory system to improve the extraction of global features. In addition, in order to improve the ability to extract global features, the adversarial network is used to learn the overall distribution pattern of the blood circulatory system. The adversarial training strategy enables the network to generate more realistic blood pressure time series data, thereby improving the model's prediction accuracy of blood pressure change trends.

[0036] (3) In view of the shortcomings of traditional loss functions in training specific blood pressure monitoring models and the inability to improve the prediction accuracy of the model, the end-to-end frequency-specific dual-attention adversarial network uses an improved hybrid loss function to train the network model. This loss function combines adversarial loss and supervised loss, which can more effectively train the network model and improve the accuracy of prediction. Through this hybrid loss function, the network can simultaneously optimize the consistency of generated data and real data, as well as the difference between predicted values ​​and actual values ​​during the training process, so that the system can provide timely warning notifications and help doctors promptly discover and diagnose potential health problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of the intelligent dynamic blood pressure monitoring and early warning system provided by the present invention;

[0038] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0040] Example 1, see Figure 1 , the intelligent dynamic blood pressure monitoring and early warning system provided by the present invention includes a data acquisition module, a data transmission module, a data processing and analysis module, a user interface module, an alarm notification module, a cloud storage and service module and a feedback and support module;

[0041] The data acquisition module uses an ambulatory blood pressure monitor to perform 24-hour ambulatory blood pressure monitoring to obtain the user's blood pressure data;

[0042] The data transmission module uses Bluetooth, Wi-Fi and other wireless communication technologies to transmit the collected blood pressure data to the mobile application and cloud storage and service module;

[0043] The data processing and analysis module performs preprocessing operations on the collected blood pressure data to obtain initial data, and uses an end-to-end frequency-specific dual-attention adversarial network to perform in-depth analysis on the initial data to predict future blood pressure trends;

[0044] The user interface module provides an intuitive and easy-to-understand interface to display blood pressure readings, information from the alarm notification module and the feedback and support module, allowing users to view their history and set reminders and goals;

[0045] When the alarm notification module detects that the blood pressure is outside the normal range, it sends a warning notification to the user and his / her designated contact via SMS, email, or APP push, and provides suggestions and measures based on the severity;

[0046] The cloud storage and service module stores the user's blood pressure data in a secure cloud environment for a long time, supports cross-platform access, and provides an API interface for integration with other medical information systems;

[0047] The feedback and support module allows users to provide feedback on their usage experience, provide technical support and consulting services, and answer user questions.

[0048] Embodiment 2. This embodiment is based on the above embodiment. In the specific implementation, when the data acquisition module takes dynamic blood pressure data, hypertensive patients from Xinzhuang Community Health Service Center in Minhang District, Shanghai from October 2023 to September 2024 are selected for research, and a basic information survey is conducted on the hypertensive patients. The survey content includes the general condition, health characteristics and health behaviors of the patients. A TM-2430 dynamic blood pressure monitor is used for the research subjects to conduct 24-hour dynamic blood pressure monitoring. Manual measurement is performed once to confirm that the sphygmomanometer is working normally. Blood pressure is measured once every 20 minutes during the day (6:00-22:00) and once every 30 minutes at night (22:00-6:00). The valid reading is set to more than 70% of the reading to be obtained. At least 20 daytime blood pressure readings and at least 7 nighttime blood pressure readings are calculated as effective monitoring. By investigating information content, dynamic blood pressure data, and performing data analysis and processing, the medication situation and blood pressure variability of the study population are obtained.

[0049] Embodiment 3, based on the above embodiment, the data processing and analysis module uses an end-to-end frequency-specific dual attention adversarial network to perform in-depth analysis on the initial data. The end-to-end frequency-specific dual attention adversarial network specifically includes the following steps:

[0050] Step S1: Variational mode decomposition, the non-stationary time series signal in the initial data is processed using variational mode decomposition to obtain initial time series data;

[0051] Step S2: Generator construction, using the dual attention module and the encoder-decoder network to process and analyze the initial time series data to obtain the predicted time series data;

[0052] Step S3: Discriminator construction, building an N-layer GRU architecture to distinguish the real time series from the predicted time series;

[0053] Step S4: loss function, using a hybrid loss function for training, the hybrid loss function includes adversarial loss and supervised loss;

[0054] Step S5: Model training, the generator and the discriminator are trained in an alternating manner to extract global features from the initial time series data and capture important distribution and shape information.

[0055] By performing the above operations, in order to address the problem that existing blood pressure monitoring equipment cannot effectively analyze and process changes in blood pressure and lacks accuracy in assisting diagnosis, an end-to-end frequency-specific dual-attention adversarial network is used to comprehensively analyze the dynamic changes of blood pressure. The network uses variational mode decomposition technology to adaptively extract components of different frequency bands from the time series of dynamic blood pressure data, which helps to capture the frequency-dependent characteristics of various regions in the circulatory system, thereby providing deeper biological insights into blood pressure changes.

[0056] Embodiment 4: This embodiment is based on the above embodiment. In step S1, variational mode decomposition specifically includes the following steps:

[0057] Step S11: Signal decomposition: decompose the non-stationary time series signal in the initial data into K intrinsic mode functions, denoted as , each intrinsic mode function corresponds to a center frequency ;

[0058] Step S12: Variational constraints, minimizing an objective function to determine and their center frequencies , find a set of intrinsic mode functions, so that the sum of this set of intrinsic mode functions is equal to the non-stationary time series signal before processing;

[0059] Step S13: Lagrange multiplier method, using Lagrange multipliers The Lagrangian function is constructed by combining the parameters of the intrinsic mode function. The formula used is as follows: ;

[0060] In the formula, is the quadratic penalty parameter, Indicates time, It's about time The derivative of Dirac function, is an imaginary unit, is a complex exponential function, is the convolution operator, represents the intrinsic mode function, represents the center frequency, is the number of intrinsic mode functions, represents the original non-stationary time series signal, represents the Euclidean norm squared, represents the Lagrange multiplier, represents the inner product operation, represents the Lagrangian function;

[0061] Step S14: Iterative solution, using the alternating direction multiplier method to iteratively update the intrinsic mode function and center frequency , until convergence, the decomposed time series data is obtained, the formula used is as follows: ; ;

[0062] In the formula, represents the number of iterations, Indicates that except The sum of all other intrinsic mode functions except the intrinsic mode functions, Indicates The intrinsic mode functions, Represents the original time series signal The Fourier transform of represents the Lagrange multiplier The Fourier transform of represents a frequency variable, Indicates The updated value of the intrinsic mode function at the n+1th iteration in the frequency domain, represents the integral function, For the The center frequency corresponding to the intrinsic mode function is updated at the n+1th iteration.

[0063] By performing the above operations, in view of the fact that the existing monitoring model fails to effectively capture the spatiotemporal dynamic characteristics of blood pressure, a dual attention mechanism is used in an end-to-end frequency-specific dual attention adversarial network, which can capture the relationship between various spatially distributed regions in the blood circulatory system, thereby providing a more comprehensive analysis in space. The adversarial network is used to learn the overall distribution pattern of the blood circulatory system to improve the extraction of global features. In addition, in order to improve the ability to extract global features, the adversarial network is used to learn the overall distribution pattern of the blood circulatory system. The adversarial training strategy enables the network to generate more realistic blood pressure time series data, thereby improving the model's prediction accuracy of blood pressure change trends.

[0064] Embodiment 5, based on the above embodiment, in step S2, the generator is constructed, specifically comprising the following steps:

[0065] Step S21: Temporal attention layer, in the dual attention module of the generator, a one-dimensional convolutional network is used to process the decomposed time series data to obtain the initial features, and the channel attention module in CBAM is used to perform average pooling and maximum pooling operations to reduce the dimension of the initial features, and a multi-layer perceptron is used to generate channel attention and output the intermediate map ;

[0066] Step S22: Spatial attention layer, using the spatial attention module in CBAM to map the intermediate Perform maximum pooling and average pooling operations, and connect the processing results to the convolution layer to obtain spatial attention features;

[0067] Step S23: Encoding layer, inputting the spatial attention features into the encoding layer of the encoding-decoding network, mapping them into the high-dimensional latent space, and obtaining high-dimensional features;

[0068] Step S24: Decoding layer, using gated recurrent units to construct a decoding layer in the encoding-decoding network, decoding the high-dimensional features to obtain predicted time series data.

[0069] Embodiment 6, based on the above embodiment, in step S4, the loss function specifically includes the following steps:

[0070] Step S41: Generator loss. The adversarial loss includes the generator loss and the discriminator loss. The formula used for the generator loss function is as follows: ; ;

[0071] In the formula, represents the prediction score of the discriminator for the input real time series data, Represents real data, the real time series data label is 1, and the predicted time series data label is 0. represents the Sigmoid function, is the index, represents the predicted time series data, represents the Sigmoid cross entropy loss function, Represents the time series data predicted by the discriminator The prediction score of is the generator loss function;

[0072] Step S42: Discriminator loss. The discriminator is used to distinguish between real time series data and predicted time series data. The discriminator loss function is used to minimize the error of the discriminator in distinguishing between true and false time series data. The formula used is as follows: ;

[0073] In the formula, represents the discriminator loss function, Represents the discriminator for real time series data The prediction score of

[0074] Step S43: supervision loss, which consists of absolute error loss, trend loss and distribution loss;

[0075] Step S44: Total generator loss. The total loss function of the generator combines the adversarial loss and the supervised loss, and each loss term is equipped with a weight parameter. The formula used is as follows: ;

[0076] In the formula, is the weight of the absolute error loss, is the weight of trend loss, is the weight of the distribution loss, is the weight of the generator adversarial loss, is the absolute error loss function, represents the trend loss function, represents the distribution loss function, is the total generator loss function.

[0077] Embodiment 7, based on the above embodiment, in step S43, the trend loss is used to indicate the fluctuation direction of the time series signal, and the formula used is as follows: ; ;

[0078] In the formula, is the time step index, and Indicated in and The time series data to be predicted, and Indicated in and Real time series data with time steps, is the direction change function, and are two variables, which represent the real time series data or predicted time series data of different time steps;

[0079] By performing the above operations, the end-to-end frequency-specific dual-attention adversarial network uses an improved hybrid loss function to train the network model, which combines adversarial loss and supervised loss to improve the prediction accuracy. This hybrid loss function can optimize the consistency of generated data and real data, as well as the difference between predicted values ​​and actual values ​​during the training process, so that the system can provide timely warning notifications and help doctors discover and diagnose potential health problems in a timely manner.

[0080] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0081] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0082] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. Intelligent dynamic blood pressure monitoring and early warning system, characterized by: It includes data acquisition module, data transmission module, data processing and analysis module, user interface module, alarm notification module, cloud storage and service module and feedback and support module; The data acquisition module uses an ambulatory blood pressure monitor to perform 24-hour ambulatory blood pressure monitoring to obtain the user's blood pressure data; The data transmission module uses Bluetooth, Wi-Fi and other wireless communication technologies to transmit the collected blood pressure data to the mobile application and cloud storage and service module; The data processing and analysis module performs preprocessing operations on the collected blood pressure data to obtain initial data, and uses an end-to-end frequency-specific dual-attention adversarial network to perform in-depth analysis on the initial data to predict future blood pressure trends; The user interface module provides an intuitive and easy-to-understand interface to display blood pressure readings, information from the alarm notification module and the feedback and support module, allowing users to view their history and set reminders and goals; When the alarm notification module detects that the blood pressure is outside the normal range, it sends a warning to the user and his / her designated contact via SMS, email, APP push, etc., and provides suggestions and measures based on the severity; The cloud storage and service module stores the user's blood pressure data in a secure cloud environment for a long time, supports cross-platform access, and provides an API interface for integration with other medical information systems; The feedback and support module allows users to provide feedback on their usage experience, provide technical support and consulting services, and answer user questions.

2. The intelligent dynamic blood pressure monitoring and early warning system according to claim 1, characterized in that: The data processing and analysis module uses an end-to-end frequency-specific dual attention adversarial network to perform in-depth analysis on the initial data. The end-to-end frequency-specific dual attention adversarial network specifically includes the following steps: Step S1: Variational mode decomposition, the non-stationary time series signal in the initial data is processed using variational mode decomposition to obtain initial time series data; Step S2: Generator construction, using the dual attention module and the encoding-decoding network to process and analyze the initial time series data to obtain the generated time series data; Step S3: Discriminator construction, building an N-layer GRU architecture to distinguish between real and generated time series; Step S4: loss function, using a hybrid loss function for training, the hybrid loss function includes adversarial loss and supervised loss; Step S5: Model training, the generator and the discriminator are trained in an alternating manner to extract global features from the initial time series data and capture important distribution and shape information.

3. The intelligent dynamic blood pressure monitoring and early warning system according to claim 2 is characterized in that: In step S1, variational mode decomposition specifically includes the following steps: Step S11: Signal decomposition: decompose the non-stationary time series signal in the initial data into K intrinsic mode functions, denoted as , each intrinsic mode function corresponds to a center frequency ; Step S12: Variational constraints, minimizing an objective function to determine and their center frequencies , find a set of intrinsic mode functions, so that the sum of this set of intrinsic mode functions is equal to the non-stationary time series signal before processing; Step S13: Lagrange multiplier method, using Lagrange multipliers The Lagrangian function is constructed by combining the parameters of the intrinsic mode function. The formula used is as follows: ； In the formula, is the quadratic penalty parameter, Indicates time, It's about time The derivative of Dirac function, is an imaginary unit, is a complex exponential function, is the convolution operator, represents the intrinsic mode function, represents the center frequency, is the number of intrinsic mode functions, represents the original non-stationary time series signal, represents the Euclidean norm squared, represents the Lagrange multiplier, represents the inner product operation, represents the Lagrangian function; Step S14: Iterative solution, using the alternating direction multiplier method to iteratively update the intrinsic mode function and center frequency , until convergence, and obtain the decomposed time series data.

4. The intelligent dynamic blood pressure monitoring and early warning system according to claim 2, characterized in that: In step S4, the loss function specifically includes the following steps: In step S4, the loss function specifically includes the following steps: Step S41: Generator loss. The adversarial loss includes the generator loss and the discriminator loss. The formula used for the generator loss function is as follows: ； ； In the formula, represents the prediction score of the discriminator for the input real time series data, Represents real data, represents the Sigmoid function, is the index, represents the predicted time series data, represents the Sigmoid cross entropy loss function, Represents the time series data predicted by the discriminator The prediction score of is the generator loss function; Step S42: discriminator loss, the discriminator is used to distinguish between real time series data and predicted time series data, and the discriminator loss function is used to minimize the error of the discriminator in distinguishing between true and false time series data; Step S43: supervision loss, which consists of absolute error loss, trend loss and distribution loss; Step S44: Total generator loss. The total loss function of the generator combines the adversarial loss and the supervised loss, and each loss term is equipped with a weight parameter.

Images