Smart watch blood pressure value calibration method and system based on particle swarm support vector
The smartwatch blood pressure calibration method based on particle swarm support vectors addresses the shortcomings of smartwatch blood pressure monitoring devices in terms of accuracy and energy consumption. By employing kernel principal component analysis and adaptive energy management, the method improves detection accuracy and adapts to inputs from different devices, achieving efficient blood pressure calibration.
Patent Information
- Application Number
- CN202511518682.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-23
- Publication Date
- 2026-01-20
AI Technical Summary
Existing smartwatch blood pressure monitoring devices have shortcomings in detection accuracy and energy consumption management. Traditional methods cannot adapt to individual differences and complex physiological changes. Deep learning models are limited in terms of computing resources and hardware configuration. Existing calibration methods are energy-intensive and difficult to improve.
A smartwatch blood pressure calibration method based on particle swarm optimization and support vectors is adopted. Through data cleaning, feature extraction, parameter optimization, and fitting function selection, multiple kernel functions are optimized using the particle swarm optimization algorithm to design a kernel function suitable for blood pressure calibration. Combined with kernel principal component analysis and adaptive energy management, the adaptability and accuracy of the model are improved.
It achieves efficient and adaptable blood pressure calibration, improves detection accuracy, meets the power consumption requirements of smartwatches for long-term use, adapts to input from different devices, and balances prediction accuracy and power consumption.
Smart Images
Figure CN121359896A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of blood pressure detection devices, and particularly relates to a smart watch blood pressure value calibration method and system based on a particle swarm support vector. BACKGROUND
[0002] With the popularization of medical knowledge, people gradually realize the importance of blood pressure as a health indicator. High or low blood pressure not only causes a variety of diseases, but also is an early signal of many diseases. Traditional blood pressure measurement methods mainly include mercury and electronic blood pressure meters. These devices can provide relatively accurate measurement results, but they are usually large in size and complex to operate, which makes them difficult to be a convenient and portable tool for daily life. Therefore, in order to meet the needs of people to implement blood pressure detection, smart watches begin to integrate blood pressure monitoring functions, aiming to provide a way to quickly obtain blood pressure values anytime and anywhere.
[0003] Traditional smart watches mainly rely on linear regression, empirical models and other basic algorithms when implementing blood pressure monitoring. However, this method has obvious limitations: it can only provide rough estimates, and its adaptability to individual differences and complex physiological changes is limited; at the same time, due to the lack of understanding of the deep features of PPG signals, its accuracy often cannot meet the clinical standards.
[0004] In recent years, with the development of deep learning technology, advanced algorithms such as neural networks have also been introduced into the field of blood pressure detection. This kind of method can automatically learn complex nonlinear mapping relationships from a large number of PPG signals by constructing a multi-layer artificial neural network, so as to more accurately predict blood pressure values. For example, convolutional neural networks (CNN) can capture local patterns in time series data, while long short-term memory networks (LSTM) are good at processing sequence information with long-term dependencies. The combination of these two networks provides a more powerful tool for blood pressure monitoring. However, blood pressure detection based on deep learning also faces some challenges: training and running complex neural network models requires a large amount of computing resources, which is a considerable burden for smart watches that are battery-powered and have limited computing power. At the same time, different models of watches have different hardware configurations, and the signals collected by PPG sensors are also different. This means that even the most advanced deep learning models need to be adjusted and optimized for specific devices to ensure their performance and accuracy, which greatly limits the generality and ease of use of the model.
[0005] In the document "A new approach for daily life blood-pressure estimation using smart watch. [C]. Biomedical Signal Processing and Control, 2022: volume 75, 103616.", a scheme for calibrating the results of smart watch operation is proposed to adapt to individuals. The scheme first calculates the PAT feature by synchronously collecting electrocardiogram (ECG) and photoplethysmogram (PPG) signals, and then extracts blood pressure related information by combining PPG time-frequency domain features and pulse interrelation matrix. Finally, two blood pressure estimation modes are proposed: a two-step method without calibration and a personal calibration method based on transfer learning. Although this method realizes blood pressure detection based on PPG and ECG, it cannot improve the existing mature scheme of smart wearable devices. At the same time, calibration according to personal data requires a long usage time, and the complex algorithm used consumes too much energy for smart wearable devices.
[0006] The present application proposes a particle swarm support vector-based smart watch blood pressure numerical calibration method and system, which performs secondary calibration based on the results of smart watch operation, uses particle swarm algorithm to optimize and enhance a variety of kernel functions, and designs a new kernel function suitable for blood pressure calibration to fit the blood pressure calibration, so that it can efficiently use information of different scales to guide blood pressure calibration. SUMMARY
[0007] The present application aims to provide a particle swarm support vector-based smart watch blood pressure numerical calibration method and system to solve the problems of existing blood pressure detection devices in detection accuracy and energy consumption management in the background art.
[0008] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: The present application proposes a particle swarm support vector-based smart watch blood pressure numerical calibration method, including the following steps: Data cleaning; distinguish data types of human index data in the smart watch, delete incomplete data and remove abnormal data by using standard score method; Feature extraction; mapping transformation is performed on the cleaned data, the data is converted to the value range available for kernel principal component analysis, and the features of different data are extracted by using kernel principal component analysis. According to the extraction result, the human index features with analysis value are selected to enter the next stage, and the redundant data is removed; Parameter optimization; the selected human index features are used to establish an initial SVR model as a calibration model, the calibration model is used to input the selected features and output the calibrated features, and the particle swarm optimization algorithm is used to find the optimal hyperparameters for the initial SVR model; the initial SVR model uses four different kernel functions, each kernel function represents a different fitting method, and four kernel functions are set for each particle to have specific hyperparameters, and the four kernel functions are linear kernel function, polynomial kernel function, RBF kernel function and BPC kernel function; the SVR model with optimal hyperparameters is obtained in the training fitting process corresponding to the fitting function of the four kernel functions; Fitting function selection; design an electric power adaptive mechanism to select a suitable fitting function in the fitting method of the four different kernel functions, and balance the prediction accuracy and energy consumption of the calibration model.
[0009] Preferably, the standard score method removes abnormal data, specifically as follows:
[0010] Where X is the observed value, μ is the mean of the data set, and σ is the standard deviation of the data set. Calculate the standard score of each data point, and set a threshold value. If the Z score of a certain data point exceeds this range, it is considered to be an outlier and is removed from the data set.
[0011] Preferably, the feature extraction is specifically as follows: Map the processed data to a high-dimensional space, map through a kernel function, and the kernel function includes a linear kernel function, a polynomial kernel function and an RBF kernel function; use the kernel function Calculate the similarity between all sample pairs x i , x j Use the similarity to construct a kernel matrix K , ; Calculate the centralized version of the kernel matrix K centered , specifically calculate the average value of each row and each column, get the average value of the entire matrix, and assign it to each element to form a matrix with all elements equal;
[0012] Where, is the average value of each row, is the average value of each column; Decompose K centered to get the eigenvalues and eigenvectors, sort them according to the size of the eigenvalues, and select the first kThe eigenvector corresponding to the largest eigenvalue is taken as the final feature representation.
[0013] Preferably, the particle swarm optimization algorithm is specified as follows: Let the size of the particle swarm be N, and each particle in the particle swarm represents a set of SVR model hyperparameters, and the position of each particle is represented as a vector , where d is the dimension number; For each different particle, run the SVR algorithm and take the mean square error as the fitness value of the particle; if the fitness value of the current particle is better than its previous fitness value, update the individual optimal position of the particle p best,i Find the particle with the highest fitness value among all particles and update its position as the global optimal position g best ; Update the speed and position of each particle, specifically:
[0014]
[0015] , where w is the inertia weight, controlling the influence of the speed, c 1 and c 2 are learning factors, respectively adjusting the influence of individual experience and group experience on the new speed, r 1 and r 2 are random numbers uniformly distributed in the range of [0, 1].
[0016] Preferably, The linear kernel function is specified as:
[0017] The polynomial kernel function is specified as:
[0018] The RBF kernel function is specified as:
[0019] The BPC kernel function is specified as:
[0020] , where is the Euclidean distance between numerical features, is the Hamming distance between classification features, and Respectively represent the sample x And y The numerical part in the sample.
[0021] Further, the four kernel functions set specific hyperparameters for each particle, as follows: The hyperparameters of the linear kernel are C and epsilon; wherein C controls the penalty degree of the error term, and epsilon defines the insensitive region of SVR; The hyperparameters of the polynomial kernel are d, gamma and r; wherein d is the order of the polynomial, gamma controls the influence degree of the inner product, and r is a constant term; The hyperparameters of the RBF kernel are C, gamma and epsilon; wherein C controls the penalty degree of the error term, gamma defines the width of the radial basis function kernel, and epsilon defines the insensitive region of SVR; The hyperparameters of the BPC kernel are C, gamma1, gamma2 and d; wherein C controls the penalty degree of the error term, gamma1 controls the influence degree of the numerical feature distance, gamma2 controls the influence degree of the categorical feature distance, and d controls the complexity of the polynomial part.
[0022] Further, for the SVR model with the optimal hyperparameters, a design function Is fitted to the given data, and the loss is calculated based on the epsilon-insensitive loss function during the fitting process; the final objective function is to maximize:
[0023] Wherein, And Are Lagrange multipliers; Based on the optimal Lagrange multipliers, a decision function Is constructed:
[0024] Wherein, b Is calculated by the data points satisfying the KKT condition.
[0025] Preferably, the fitting function is selected, as follows: The fitting function of the four different kernel functions is obtained during the fitting process, wherein the fitting mode power consumption of the linear kernel function, the polynomial kernel function, the RBF kernel function and the BPC kernel function increases in turn; when the device power decreases, gradually switch from the fitting mode with high power consumption to the fitting mode with low power consumption.
[0026] The present application proposes an intelligent watch blood pressure value calibration system based on particle swarm support vector in the second aspect, comprising: The data cleaning module is used for distinguishing data types of human body index data in the smart watch, deleting incomplete data and removing abnormal data by using a standard score method. The feature extraction module is used for performing mapping transformation on the cleaned data, extracting features of different data by using kernel principal component analysis, and selecting human body index features with analysis value according to the extraction result. The calibration model generation module is used for establishing an initial SVR model as a calibration model by using the selected human body index features, and the initial SVR model uses four different kernel functions, and each kernel function represents a different fitting mode. The parameter optimization module is used for setting specific hyperparameters for each particle by using different kernel functions, and finding optimal hyperparameters for the initial SVR model by using a particle swarm optimization algorithm. The fitting function selection module is used for selecting a fitting function in the fitting mode of different kernel functions, and balancing the prediction accuracy and energy consumption of the SVR model.
[0027] In a third aspect, the present application provides a smart watch, wherein the smart watch is installed with the smart watch blood pressure value calibration system based on a particle swarm support vector.
[0028] In a fourth aspect, the present application provides a blood pressure measuring device, wherein the blood pressure measuring device comprises at least one processor, and the at least one processor executes the smart watch blood pressure value calibration system based on a particle swarm support vector, so as to realize the smart watch blood pressure value calibration method based on a particle swarm support vector.
[0029] Compared with the prior art, the present application has the following beneficial effects: (1) The present application provides a smart watch blood pressure value calibration method with high efficiency and strong adaptability, and a method based on multi-kernel support vector regression (SVR) optimization and adaptive energy consumption management is adopted to overcome the deficiencies of existing devices in precision and energy consumption management. In order to improve data quality, the Z-score method is used to filter abnormal values and perform mapping transformation on data in the data cleaning stage. In the feature extraction link, KPCA is used to identify the core features of different data groups; linear kernel, polynomial kernel and radial basis function are used as three kernel functions to capture the complex nonlinear relationship in the data, thereby improving the generalization ability of the model. For the parameter optimization part, the particle swarm optimization algorithm (PSO) is introduced to find the optimal hyperparameters for the SVR model. Through the above method, a practical blood pressure calibration model is successfully established, and the related parameters of the model are determined. The model can adjust the model to adapt to different application scenarios by using kernel principal component analysis according to different inputs of the watch. The calibration process of the present application not only improves the accuracy of blood pressure monitoring, but also meets the practical needs of long-term use of the smart watch.
[0030] (2) The method of the present application takes the SVR model as the core as the calibration model, the smart watch takes the original error blood pressure as the necessary main input, takes other human indicators such as heart rate, age, weight and the like as optional auxiliary inputs, inputs the indicators into the SVR model, the model outputs the calibrated blood pressure value after operation, the model selects the core indicators based on KPCA, removes the redundant indicators, determines the indicators participating in operation, simultaneously adapts to various inputs of different devices, and determines the specific hyperparameter value of the model based on the particle swarm algorithm.
[0031] (3) The method of the present application uses four different kernel functions for the calibration model, each kernel function represents a different fitting mode, and adapts to different application scenarios. Among the four kernel functions, a kernel function BPC based on the blood pressure calibration scene is innovatively proposed, so as to better adapt to the demand of blood pressure monitoring. In the fitting mode selection aspect, a power self-adaptive mechanism is designed, and the most suitable fitting function is dynamically selected according to the power state. BRIEF DESCRIPTION OF DRAWINGS
[0032] Figure 1 It is a flowchart of the smart watch blood pressure value calibration method based on particle swarm support vector in the present application. Figure 2 It is a flowchart of the particle swarm optimization SVR algorithm in the present application. DETAILED DESCRIPTION
[0033] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.
[0034] Embodiment 1: The overall process of the arbitrary smart watch blood pressure detection calibration method based on particle swarm support vector regression designed in the present application is as shown in Figure 1 There are four stages: data cleaning, feature extraction, parameter optimization and fitting function selection.
[0035] Stage 1: Data cleaning.
[0036] The present application considers adapting the blood pressure detection calibration method to any smart watch in the prior art. The smart watch equipped with the system can certainly obtain the original blood pressure value, and may also obtain various other human body data. The model of the present application accepts various human body indicators that can be quantified as numerical values, including blood pressure, weight, gender, height, heart rate, blood lipids, heart rate variability, etc., among which the blood pressure value data is a necessary item, and the other data is an optional item. The method is divided into two parts: model establishment and model use. The model establishment requires multiple sets of smart watch blood pressure data and corresponding real blood pressure data. Therefore, in the data cleaning part, first, remove the incomplete data and blank data mixed in the data collection, then use the Z-score method to remove abnormal data points, ensure the quality of the input data, and finally map the data to the value range available for kernel principal component analysis.
[0037] In this stage, the data categories and numerical types that the source device can provide are first determined, and then the incomplete data in the data table is determined. The incomplete data is deleted, and then the problem data is removed according to the standard score method. The standard score is calculated as follows:
[0038] Where X is the observed value, μ is the mean of the data set (mean), and σ is the standard deviation of the data set. The standard score of each data point is calculated, and a threshold value (the present application defaults to 3) is set. If the Z-score of a certain data point exceeds this range, it is considered to be an outlier and is removed from the data set. Finally, the text type in the source device is mapped to the numerical type, and all the numerical values are mapped to 0-1.
[0039] Stage 2: Feature extraction.
[0040] In the feature extraction part, kernel principal component analysis (KPCA) is used to select the features of different data. The most relevant blood pressure and other indicators are selected from the data indicators, and redundant indicators are removed. Only the selected indicators will be used in the subsequent stage. In kernel principal component analysis, linear kernel, polynomial kernel and radial basis function (RBF) three kernel functions are applied to capture the non-linear relationship in the data and improve the generalization ability of the model.
[0041] This stage maps the features converted from stage 1 to a high-dimensional space, and the mapping method is determined by the kernel function. The kernel function includes the following: Linear kernel:
[0042] Polynomial kernel:
[0043] RBF kernel:
[0044] Compute the kernel matrix between all sample pairs according to the selected kernel function K , now assume there is a nonlinear mapping : X→F. For any two points x i , x j in the input space, map them to high dimensional space , compute the similarity between two points using kernel function, and use the similarity to construct the kernel matrix K .
[0045] Specifically, use the kernel function to compute the similarity between all sample pairs, thus constructing the kernel matrix K, where .
[0046] Next, compute the centered version of the kernel matrix K centered , in the formula is the average of each row, is the average of each column, and finally get the average of the whole matrix and assign it to each element, forming a matrix with all elements equal. Eliminate the influence of the non-zero mean of the data on the subsequent PCA analysis.
[0047]
[0048] Perform eigen decomposition to get eigenvalues and eigenvectors, sort them according to the size of eigenvalues, select the eigenvectors corresponding to the top k largest eigenvalues as the final feature representation.
[0049] Stage 3: Parameter optimization.
[0050] In the parameter optimization part, the particle swarm optimization algorithm (PSO) is used to find the optimal hyperparameters for SVR. According to different kernel functions, set specific hyperparameters for each particle, hyperparameter range and upper and lower bounds, to ensure efficient and accurate search process. At the same time, a new kernel function BPC based on blood pressure calibration scene is proposed, which better fits the blood pressure calibration scene.
[0051] As shown in Figure 2 , this stage first defines the particle swarm, each particle in the particle swarm represents a set of possible solutions (i.e. the hyperparameters of SVR). Set the size of the particle swarm as N, the position of each particle is represented as a vector , where d is the dimension number (i.e. the number of hyperparameters to be optimized). For SVR, different kernel functions have different hyperparameters: The hyperparameters of linear kernel are C and epsilon. The C parameter controls the penalty strength of the error term, which is the trade-off between the model fitting degree of training data and the generalization ability. The epsilon parameter defines the insensitive region of SVR, that is, the error between the predicted value and the true value within the epsilon range will not be punished.
[0052] The hyperparameters of polynomial kernel are d, gamma and r, where d is the order of the polynomial. The gamma controls the influence degree of inner product. The r is the constant term.
[0053] The hyperparameters of RBF kernel are C, gamma and epsilon, where the C parameter controls the penalty strength of the error term, the gamma parameter defines the width of the radial basis function (RBF) kernel, which determines the influence range of a single sample. The epsilon parameter defines the insensitive region of SVR.
[0054] The hyperparameters of BPC kernel are C, gamma1, gamma2 and d, where the C parameter controls the penalty strength of the error term, the gamma1 parameter controls the influence degree of numerical feature distance. The gamma2 parameter controls the influence degree of categorical feature distance, and the d controls the complexity of the polynomial part.
[0055] The general form of BPC kernel is:
[0056] Where, is the Euclidean distance between numerical features (such as high and low blood pressure, heart rate, time), is the Hamming distance between categorical features, X num and Y num represent the numerical part in samples x and y respectively.
[0057] The kernel function itself is a mapping relationship, the present application maps the data to a hyperplane, and performs kernel principal component analysis and particle swarm optimization algorithm on the hyperplane. Specifically, the kernel function used in stages 2 and 3 is the same, and the kernel function equation is also the same. Different kernels have different hyperparameters, which affect the particle swarm optimization algorithm. Most of the proposed hyperparameters can be found in the specific method part of the particle swarm optimization algorithm. Since there are three kernel functions for principal component analysis and four particle swarm optimization algorithms, after kernel principal component analysis using RBF kernel, the selected indicators will be applied to the particle swarm optimization algorithm of RBF kernel and BPC kernel at the same time, and the other two are one-to-one.
[0058] For each different particle, run the SVR algorithm and use the mean squared error as the fitness value of the particle.
[0059] If the fitness value of the current particle is better than its previous best, update the individual best position of the particle p best,i .
[0060] Find the particle with the highest fitness value among all particles and update its position as the global best position g best .
[0061] Update the velocity and position of each particle according to the formula:
[0062]
[0063] where, w is the inertia weight, controlling the influence of the velocity, c 1 and c 2 are learning factors, adjusting the influence of individual experience and group experience on new velocity, respectively, r 1 and r 2 are random numbers uniformly distributed in the range [0,1].
[0064] After determining the hyperparameters and the dataset, the SVR algorithm can calculate a fixed function that fits the dataset, which is a mapping of input data and true values, but it is not necessarily optimal, and the hyperparameters need to be adjusted continuously. The process of adjusting the hyperparameters is the work of the particle swarm algorithm.
[0065] The purpose of the SVR algorithm in this phase is to find a function that fits the given data, the training data is in the form of During the fitting process, the loss needs to be calculated, for this purpose, an ε-insensitive loss function is defined, if the difference between the predicted value and the actual value is less than ε, the loss is not calculated.
[0066] The final objective function is to maximize:
[0067] where, and are Lagrange multipliers, after obtaining the optimal Lagrange multiplier, the decision function is constructed , where b is calculated by the data points that satisfy the KKT (Karush-Kuhn-Tucher) condition:
[0068] Stage 4: fitting function selection.
[0069] Four different kernel functions are obtained in the fitting process, wherein the linear kernel has the lowest calculation difficulty and consumes the least power, the polynomial kernel is slightly higher, the RBF term kernel and the BPC kernel are the highest, and the best kernel function is selected according to the actual fitting situation. The power consumption is determined by the complexity of the model itself, and the more complex the kernel function, the more difficult it is to calculate and the more power it consumes. When the power of the smart watch is reduced, gradually switch from the fitting mode with high power consumption to the fitting mode with low power consumption. The effect of calibrating the blood pressure value of the smart watch is achieved.
[0070] The above is only used to help understand the method of the present application and its core essence, but the protection scope of the present application is not limited thereto. For those skilled in the art, within the technical scope disclosed by the present application, equivalent replacement or change according to the technical scheme and inventive concept of the present application should be covered within the protection scope of the present application. In summary, the content of the specification should not be understood as a limitation of the present application.
Claims
1. A particle swarm support vector-based smart watch blood pressure value calibration method, characterized in that, It comprises the following steps: Data cleaning; the human body index data in the smart watch is distinguished by data type, incomplete data is deleted, and standard score method is used to remove abnormal data; Feature extraction; the mapped and transformed data after cleaning is converted to the value domain available for kernel principal component analysis, the features of different data are extracted by kernel principal component analysis, and the human body index features with analysis value are selected according to the extraction result; Parameter optimization; an initial SVR model is established by using the selected human body index features by kernel principal component analysis as a calibration model, the calibration model is used to input the selected features and output the features after calibration, and a particle swarm optimization algorithm is used to find the optimal hyperparameters for the initial SVR model; the initial SVR model uses four different kernel functions, each kernel function represents a different fitting mode, four kernel functions set specific hyperparameters for each particle, and the four kernel functions are linear kernel function, polynomial kernel function, RBF kernel function and BPC kernel function; the SVR model with optimal hyperparameters obtains the fitting functions corresponding to the four kernel functions in the training fitting process; Fitting function selection; an electric power adaptive mechanism is designed to select a suitable fitting function in the fitting modes of the four different kernel functions, and the prediction accuracy and energy consumption of the calibration model are balanced. 2.The particle swarm support vector-based smart watch blood pressure value calibration method according to claim 1, characterized in that, The feature extraction is specifically as follows: The processed data is mapped to a high-dimensional space, and the mapping is performed by using a kernel function, which includes a linear kernel function, a polynomial kernel function and an RBF kernel function The similarity between all sample pairs is calculated x i , x j A kernel matrix is constructed using the similarity K , ; Computing a centered version of the kernel matrix K centered The average value of each row and each column is specifically computed, the average value of the entire matrix is obtained and assigned to each element, forming a matrix with all elements equal; wherein is the average value of each row, is the average value of each column; right K centered Eigenvalue decomposition yields eigenvalues and eigenvectors. These eigenvalues are then sorted by size, and the top eigenvalues are selected. k The eigenvector corresponding to the largest eigenvalue is used as the final feature representation. 3.The particle swarm support vector-based smart watch blood pressure value calibration method according to claim 1, characterized in that, The particle swarm optimization algorithm is specifically as follows: Let the size of the particle swarm be N, each particle in the particle swarm represents a set of SVR model hyperparameters, and the position of each particle is represented as a vector wherein d is the dimension number; For each different particle, run the SVR algorithm and take the mean square error as the fitness value of the particle; if the fitness value of the current particle is better than its previous fitness value, update the individual optimal position of the particle p best,i ; find the particle with the highest fitness value among all particles and update its position as the global optimal position g best ; updating the velocity of each particle and position in particular: where, w is the inertial weight, controlling the influence of the velocity, c 1 and c 2 are learning factors, respectively adjusting the influence of individual experience and swarm experience on the new velocity, r 1 and r 2 are random numbers uniformly distributed in the range [0, 1]. 4.The particle swarm support vector-based smart watch blood pressure value calibration method according to claim 3, characterized in that, The BPC kernel function is specifically as follows: wherein, is the Euclidean distance between numerical features, is the Hamming distance between categorical features, and represent the numerical part in samples x and y respectively. 5.The particle swarm support vector-based smart watch blood pressure value calibration method according to claim 4, characterized in that, The four kernel functions set specific hyperparameters for each particle, which are specifically as follows: The hyperparameters of the linear kernel are C and epsilon; wherein C controls the punishment degree of the error term, and epsilon defines the insensitive region of SVR; The hyperparameters of the polynomial kernel are d, gamma and r; wherein d is the order of the polynomial, gamma controls the influence degree of the inner product, and r is a constant term; The hyperparameters of the RBF kernel are C, gamma and epsilon; Wherein C controls the punishment degree of the error term, gamma defines the width of the radial basis function kernel, and epsilon defines the insensitive region of SVR; The hyperparameters of the BPC kernel are C, gamma1, gamma2 and d; wherein C controls the punishment degree of the error term, gamma1 controls the influence degree of the numerical feature distance, gamma2 parameter controls the influence degree of the distance of the categorical feature, and d controls the complexity of the polynomial part. 6.The particle swarm support vector-based smart watch blood pressure value calibration method according to claim 3, characterized in that, For the SVR model with the optimal hyperparameters determined, a design function is fitted to the given data, with the loss computed based on the ε-insensitive loss function during the fitting process; the final objective function is to maximize: wherein and is a Lagrange multiplier; constructing a decision function based on an optimal lagrangian multiplier : wherein, b Computed by the data points satisfying the KKT conditions. 7.The particle swarm support vector-based smart watch blood pressure value calibration method according to claim 6, characterized in that, The fitting function selection is specifically as follows: The fitting functions of the four different kernel functions are obtained in the fitting process, wherein the power consumption of the fitting modes of the linear kernel function, the polynomial kernel function, the RBF kernel function and the BPC kernel function increases in turn; when the device power decreases, the fitting mode gradually shifts from the fitting mode with high power consumption to the fitting mode with low power consumption.
8. The particle swarm support vector-based smartwatch blood pressure value calibration system applied to the method of any one of claims 1-7, characterized in that, It comprises: A data cleaning module for distinguishing the human body index data in the smart watch by data type, deleting incomplete data, and removing abnormal data by using standard score method; A feature extraction module for mapping and transforming the cleaned data, extracting the features of different data by kernel principal component analysis, and selecting human body index features with analysis value according to the extraction result; The calibration model generation module is configured to establish an initial SVR model as a calibration model by using the selected human index features, and the initial SVR model uses four different kernel functions, each of which represents a different fitting mode; The parameter optimization module is configured to set specific hyperparameters for each particle by using different kernel functions, and to find optimal hyperparameters for the initial SVR model by using a particle swarm optimization algorithm; the SVR model with the optimal hyperparameters is used to obtain fitting functions corresponding to the four kernel functions in the training fitting process; The fitting function selection module is configured to select a fitting function from the fitting modes of different kernel functions, and to balance the prediction accuracy and energy consumption of the SVR model.
9. A smart watch characterized by The smart watch is provided with the smart watch blood pressure value calibration system based on the particle swarm support vector as claimed in claim 8.
10. Blood pressure measuring device, characterized in that The blood pressure measuring device comprises at least one processor; and the at least one processor executes the smart watch blood pressure value calibration method based on the particle swarm support vector as claimed in any one of claims 1-7.