A data analysis method for intelligent blood pressure monitor based on adaptive algorithm

Abstract

The present application relates to the technical field of medical data analysis, in particular to a kind of intelligent sphygmomanometer data analysis method based on adaptive algorithm.The present application infers user getting-up time by receiving continuous blood pressure measurement data containing time stamp and synchronously collected behavior data, and presets and dynamically adjusts morning blood pressure analysis window accordingly, extracts corresponding blood pressure segment;Abnormality is excluded and interpolation is completed to the extracted blood pressure data, form complete continuous time series, and calculate change rate sequence and recursive quantitative feature;Combine short-term and long-term historical data, respectively generate statistical features, and dynamically update baseline reference value through individual adaptive model;Further, by comparing the current blood pressure peak value with the rising rate with the reference baseline, morning blood pressure abnormal event is identified.The method can improve the individual adaptability of the analysis window and the accuracy of blood pressure abnormality identification, and is suitable for fine analysis and risk warning of continuous blood pressure data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data analysis, in particular to an intelligent sphygmomanometer data analysis method based on an adaptive algorithm. BACKGROUND

[0002] With the continuous development of medical health monitoring devices and wearable technology, individualized health analysis based on continuous physiological data has gradually become an important direction of smart medical care. Blood pressure, as one of the key vital signs, its fluctuation can reflect the physiological regulation process of the human body at different time periods and in different states. Through dynamic analysis of blood pressure data, it not only helps early identification and risk assessment of diseases, but also provides support for long-term management of chronic diseases.

[0003] In the prior art, most blood pressure monitoring methods rely on user manual measurement or use fixed time periods for data collection and analysis, lacking the ability to perceive and adapt to individual behavior characteristics, resulting in inaccurate analysis window setting, insufficient timeliness and representativeness of the results. At the same time, user measurement behavior has certain randomness, and there may be problems such as abnormal values, uneven measurement intervals or data missing in blood pressure data, which further affect the analysis effect. Traditional statistical methods lack flexibility in dealing with such problems, making it difficult to adapt to individual differences and dynamic changes in physiological state.

[0004] Therefore, there is an urgent need for a method that combines behavior data and physiological data, accurately extracts key period blood pressure information, dynamically analyzes and identifies abnormalities through adaptive algorithms, to improve the analysis accuracy of blood pressure data and the intelligent level of health monitoring systems. SUMMARY

[0005] The present application provides an intelligent sphygmomanometer data analysis method based on an adaptive algorithm, which infers user wake-up time through behavior data, dynamically adjusts the morning blood pressure analysis window, cleanses and extracts blood pressure features, and updates the adaptive baseline combining individual short-term and long-term statistical characteristics, to realize accurate identification of morning blood pressure abnormalities and improve the intelligence and accuracy of individualized blood pressure data analysis.

[0006] To achieve the above purpose, the present application provides the following technical solution:

[0007] An intelligent sphygmomanometer data analysis method based on an adaptive algorithm, comprising:

[0008] Receiving continuous blood pressure measurement data containing timestamps and synchronously collected behavior data, the blood pressure measurement data covering the night sleep period to the post-wake-up period;

[0009] Infer the user wake-up time based on the behavior data, preset and adjust the morning blood pressure analysis window, and extract the blood pressure measurement data within the analysis window to form the morning blood pressure segment;

[0010] Data cleaning is performed on the morning blood pressure segment to obtain a continuous blood pressure time series, and based on the blood pressure time series, a blood pressure change rate sequence is calculated, and morning blood pressure dynamic features including maximum blood pressure rise rate, mean, standard deviation, and recursive quantitative analysis features are extracted;

[0011] Based on the morning blood pressure segment data in the first preset period, short-term statistical features are generated, and based on the morning blood pressure segment data in the second preset period, long-term statistical features are generated, and the reference baseline of the short-term and long-term statistical features is dynamically adjusted through an individual adaptive learning model;

[0012] The current morning blood pressure peak value is compared with the long-term reference baseline, and the blood pressure rise rate is dynamically compared with the short-term reference baseline. If both exceed the preset threshold, it is determined that the morning blood pressure is abnormal.

[0013] Further, the process of inferring the user's wake-up time includes:

[0014] An SVM classification model is trained based on historical behavior data to determine wake-up behavior pattern features;

[0015] Real-time user behavior data is received and behavior feature vectors are extracted, and the behavior feature vectors are input into the SVM classification model to determine the wake-up behavior and output the user's wake-up time.

[0016] Further, the process of presetting and adjusting the morning blood pressure analysis window includes:

[0017] The inferred user wake-up time is used as a reference base point, and a first preset duration is extended forward to determine a pre-wake-up measurement interval, and a second preset duration is extended backward to determine a post-wake-up measurement interval;

[0018] The pre-wake-up and post-wake-up measurement intervals are combined to form a morning blood pressure analysis window, and the window boundaries are dynamically fine-tuned based on blood pressure measurement data, behavior data, and blood pressure measurement density.

[0019] Further, the process of dynamically fine-tuning the window boundaries includes:

[0020] Based on blood pressure measurement data, behavior data, and blood pressure measurement density, an input feature vector for predicting the optimal analysis window boundaries is constructed;

[0021] The input feature vector is input into a pre-trained LightGBM regression model to output the optimal adjustment amount of the window start and end points;

[0022] The preliminary analysis window boundaries are fine-tuned according to the adjustment amount to form the final morning blood pressure analysis window.

[0023] Further, the process of data cleaning on the morning blood pressure segment includes:

[0024] The first sliding window is used to traverse the blood pressure time sequence, the blood pressure change rate in the first sliding window is calculated, and the measurement point whose change rate exceeds the preset threshold is identified as an abnormal value and is removed;

[0025] For the sequence after removing the abnormal value, whether there is missing measurement data is detected, and based on the time stamp and blood pressure value of the front and rear effective measurement points, a linear interpolation method is used to estimate the missing value and fill in, forming a complete and continuous blood pressure time sequence.

[0026] Further, the process of extracting recursive quantitative analysis features includes:

[0027] Based on the continuous blood pressure time sequence of the morning blood pressure segment, phase space reconstruction is performed to generate a trajectory matrix;

[0028] The recurrence plot is calculated based on the trajectory matrix to describe the similarity between blood pressure states;

[0029] Recursive quantitative analysis features are extracted from the recurrence plot as supplementary indicators of the morning blood pressure dynamic features.

[0030] Further, the process of dynamic adjustment includes:

[0031] In the second sliding time window, the statistical features of the user's daily morning blood pressure segment are collected;

[0032] The feature distribution of the current window and the historical window is compared,

[0033] When the distribution difference is greater than the difference threshold, it is determined that a concept drift event occurs, and the short-term and long-term reference baseline parameters are updated, so that the abnormality determination rule adapts to the long-term change trend of the individual physiological state.

[0034] Further, the determination condition of the morning blood pressure abnormal event is that the blood pressure peak value deviation exceeds 10% of the long-term reference baseline, and the blood pressure rising rate exceeds 2 times the standard deviation of the short-term reference baseline.

[0035] The beneficial effects of the present application are:

[0036] 1. By fusing behavior data and continuous blood pressure data, the user's getting-up time is intelligently identified, and the analysis window is adaptively adjusted to accurately extract the morning blood pressure features. The system further combines individual short-term and long-term blood pressure feature baselines to dynamically identify abnormal changes in blood pressure peak value and rising rate, effectively improving the accuracy and timeliness of abnormal event identification, and providing a scientific basis for individualized intervention.

[0037] 2. The method employs multiple techniques such as data cleaning, interpolation, recursive quantitative analysis, sliding window adaptive learning, and machine learning models to enhance fault tolerance for non-ideal measurement data. Combining LightGBM and SVM models improves the accuracy of window boundary recognition and wake-up event determination, enabling the system to adapt to different users' habits and blood pressure fluctuation patterns. BRIEF DESCRIPTION OF DRAWINGS

[0038] The accompanying drawings are included to provide a further understanding of the present application, and are incorporated in and constitute a part of this specification, illustrate embodiments of the present application and serve to explain the present application, and do not limit the present application. In the drawings:

[0039] Figure 1 is a flow chart of an intelligent sphygmomanometer data analysis method based on an adaptive algorithm provided by the present application. DETAILED DESCRIPTION

[0040] The preferred embodiments of the present application will be described below with reference to the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and do not limit the present application.

[0041] Embodiment One

[0042] An intelligent sphygmomanometer data analysis method based on an adaptive algorithm, as shown in Figure 1 , comprises:

[0043] S100: receiving continuous blood pressure measurement data containing time stamps and synchronously collected behavior data, the blood pressure measurement data covering the night sleep period to the post-wake-up period;

[0044] Specifically, the blood pressure measurement data of the user is continuously collected by the intelligent sphygmomanometer device, and the time stamp of each measurement is automatically recorded during the measurement process, forming a continuous blood pressure data sequence with time markers. This sequence covers the user's night sleep period until the morning wake-up period for several hours, ensuring that the dynamic change process of blood pressure before and after wake-up is obtained. At the same time, the device also integrates an acceleration sensor and a light sensor for synchronously collecting the behavior data of the user. The acceleration sensor is used to capture the change of the user's physical activity state, and the light sensor is used to perceive the ambient light intensity, which, in combination with the two, can infer the transition process of the user from a resting state to an active state. In addition, to improve the accuracy of subsequent analysis, the system also records the occurrence time of the blood pressure measurement event and counts the measurement frequency and interval in each time period as basic data for blood pressure measurement density in subsequent analysis.

[0045] S200: inferring the wake-up time of the user based on the behavior data, presetting and adjusting the wake-up blood pressure analysis window, and extracting the blood pressure measurement data within the analysis window to form a wake-up blood pressure segment;

[0046] Further, the process of inferring the user's getting-up time comprises:

[0047] training an SVM classification model based on historical behavior data to determine getting-up behavior pattern features;

[0048] receiving user behavior data in real time and extracting a behavior feature vector, and inputting the behavior feature vector into the SVM classification model to determine the getting-up behavior and output the user's getting-up time.

[0049] Specifically, in the initialization phase, the behavior data of the user for several days is collected as training samples. The behavior data includes acceleration information and light intensity data. By analyzing the activity intensity change, posture switching frequency and environmental light change of the user at different time periods, typical behavior features reflecting the transition of the user from a sleep state to a wake state are extracted. Subsequently, the feature vector set is constructed using the above labeled historical behavior data, and the SVM classification algorithm is used for offline training to obtain a classification model capable of identifying the "getting-up behavior" state. During the training process, multiple dimensions such as the activity intensity mean, the change rate, the light sudden increase feature, and the shortening of the static duration are selected as feature inputs, and the actual getting-up time of the user is used as a label for supervised learning. In actual operation, the system receives the current behavior data of the user in real time, and extracts the behavior feature vector of the latest period of time in a time sliding window manner. The feature vector is input into the pre-trained SVM model for classification and judgment, and when multiple consecutive time windows are identified as "getting-up behavior" state, it is determined that the user has gotten up, and the time point is output as the user's getting-up time.

[0050] By introducing the SVM model to intelligently infer the user's getting-up time, the individualized getting-up behavior pattern can be fully utilized based on historical and real-time behavior data, and compared with the traditional method based on fixed time threshold or simple rule judgment, the method has higher timing accuracy and individual adaptability.

[0051] Further, the process of presetting and adjusting the morning blood pressure analysis window comprises:

[0052] Taking the inferred user's getting-up time as a reference base point, extending forward by a first preset time length to determine a pre-getting-up measurement interval, and extending backward by a second preset time length to determine a post-getting-up measurement interval;

[0053] Combining the pre-getting-up and post-getting-up measurement intervals to form a morning blood pressure analysis window, and dynamically fine-tuning the window boundary based on blood pressure measurement data, behavior data, and blood pressure measurement density.

[0054] Specifically, the system obtains the user's getting-up time point inferred by the aforementioned method as the time anchor point for the morning blood pressure analysis. With this time point as the reference, the system extends forward by a first preset time length (e.g., 30 minutes) to determine a "pre-getting-up measurement interval", and extends backward by a second preset time length (e.g., 60 minutes) to determine a "post-getting-up measurement interval". The two measurement intervals are combined to form a preliminary "morning blood pressure analysis window". This window covers a key physiological change period from before the user wakes up to after the user gets up, which is conducive to capturing the dynamic process of the rapid change of the morning blood pressure. On this basis, to further improve the individual adaptability and data quality of the analysis window, the system introduces the distribution characteristics of the blood pressure measurement data, the activity intensity information (such as the physical activity level calculated from the acceleration change) in the behavior data, and the blood pressure measurement density (the number of measurements per unit time), and constructs a dynamic adjustment model for fine-tuning the window boundaries.

[0055] By taking the individual getting-up time as the anchor point, setting the preliminary analysis window in combination with the fixed time length before and after getting up, and further integrating the blood pressure measurement density, the activity intensity information in the behavior data, and other information to dynamically fine-tune the window boundaries, the analysis period range can be adaptively adjusted, so as to more accurately cover the key change stage of the blood pressure from the resting state to the active state.

[0056] Further, the process of dynamically fine-tuning the window boundaries includes:

[0057] Based on the blood pressure measurement data, the behavior data, and the blood pressure measurement density, an input feature vector for predicting the optimal analysis window boundaries is constructed;

[0058] The input feature vector is input into a pre-trained LightGBM regression model to output the optimal adjustment amount of the window start point and the end point;

[0059] According to the adjustment amount, the preliminary analysis window boundaries are finely corrected to form the final morning blood pressure analysis window.

[0060] Specifically, based on the preliminary morning blood pressure analysis window, the system collects multi-source data in this period, including: blood pressure measurement data (measurement value, timestamp), behavior data (acceleration data from wearable devices, light intensity) and blood pressure measurement density (measurement frequency distribution characteristics per unit time). The system extracts features from the above data to construct an input feature vector for boundary prediction. The extracted features can include but are not limited to: average acceleration amplitude every 5 minutes before and after getting up, mean and variance of measurement point time interval, acceleration change rate, skewness coefficient of measurement distribution, number of light intensity abrupt points, etc. Then, the input feature vector is input into the pre-off-line trained LightGBM regression model. The model is trained with a large number of labeled samples, and its output is two continuous values, which correspond to the optimal adjustment amount (unit: minute) of the start and end points of the morning blood pressure analysis window, respectively, i.e. the predicted time length that should be extended forward or contracted backward based on the current preliminary boundary. According to the boundary adjustment amount output by the model, the system modifies the start and end points of the preliminary analysis window respectively. For example, if the start point needs to be extended by 3 minutes and the end point needs to be extended by 5 minutes, the final analysis window boundary will cover the additional 8-minute interval outside the preliminary window.

[0061] By introducing the LightGBM regression model to make data-driven fine adjustment of the analysis window boundary, using blood pressure measurement data, behavior data and measurement density to construct a feature vector, the physiological and behavioral change patterns of individuals during the getting-up process can be accurately captured, the start and end points of the window can be intelligently optimized, the matching degree of the morning blood pressure analysis window and the actual blood pressure change process can be significantly improved, and the accuracy of subsequent blood pressure dynamic feature extraction and abnormality identification can be improved.

[0062] S300: Data cleaning is performed on the morning blood pressure segment to obtain a continuous blood pressure time series, and based on the blood pressure time series, a blood pressure change rate sequence is calculated, and morning blood pressure dynamic features including maximum blood pressure rise rate, mean, standard deviation and recursive quantitative analysis features are extracted;

[0063] Further, the process of data cleaning on the morning blood pressure segment includes:

[0064] A first sliding window is used to traverse the blood pressure time series, calculate the blood pressure change rate in the first sliding window, and identify and remove measurement points with a change rate exceeding a preset threshold as outliers;

[0065] For the sequence after removing outliers, it is detected whether there is missing measurement data, and based on the timestamps and blood pressure values of the preceding and following valid measurement points, a linear interpolation method is used to estimate and fill in the missing values, forming a complete and continuous blood pressure time series.

[0066] Specifically, the system receives blood pressure time series data screened through the morning blood pressure analysis window, each data containing a blood pressure value and a corresponding timestamp. To identify possible measurement outliers, the system traverses the sequence using a first sliding window of fixed length (e.g., 5 minutes or containing 3-5 measurement points). Within each sliding window, the system calculates the blood pressure rate of change based on adjacent measurement points, i.e., the blood pressure change value per unit time, and compares the rate with a pre-set threshold. If the rate before and after a measurement point changes abruptly and its absolute value exceeds the threshold (e.g., 20 mmHg / min), the measurement point is determined to be an outlier and is recorded in a rejection list. After traversal is completed, the system removes all data points marked as outliers from the original blood pressure sequence to obtain a preliminary cleaned blood pressure time series. Then, the system performs continuity detection on the sequence to identify whether there are time periods with missing data, such as parts without records for more than a pre-set maximum measurement interval (e.g., 10 minutes). For each missing section, the system finds the nearest valid measurement points before and after the section, respectively denoted as the front point and the rear point, and calculates the estimated blood pressure values in the time period based on the linear interpolation method. The system fills the interpolated blood pressure values into the missing interval to reconstruct a time-continuous and complete blood pressure time series.

[0067] By detecting and removing measurement points with abnormal rate of change through the sliding window, non-physiological fluctuations caused by measurement errors or sudden interference can be effectively excluded, improving the authenticity and stability of the blood pressure data; at the same time, missing data is filled by the linear interpolation method to ensure the continuity and integrity of the time series, laying a solid data foundation for subsequent blood pressure rate of change calculation and dynamic feature extraction.

[0068] Further, the process of extracting recursive quantitative analysis features includes:

[0069] Based on the continuous blood pressure time series of the morning blood pressure segment, phase space reconstruction is performed to generate a trajectory matrix;

[0070] Based on the trajectory matrix, a recurrence plot is calculated to describe the similarity between blood pressure states;

[0071] Recursive quantitative analysis features are extracted from the recurrence plot as supplementary indicators of the morning blood pressure dynamic features.

[0072] Specifically, the system performs phase space reconstruction based on the continuous blood pressure time series of the morning blood pressure segment after data cleaning . Phase space reconstruction is a method of mapping one-dimensional time series to multi-dimensional space to reveal its dynamic characteristics. By setting the embedding dimension and the time delay , a trajectory matrix is constructed:

[0073] ;

[0074] Then, the system calculates the Euclidean distance between each pair of vectors in the trajectory matrix and constructs a recurrence plot , which is defined as follows:

[0075] ;

[0076] wherein, denotes the distance threshold, denotes the Heaviside step function, which is used to determine whether two state points are "similar" (i.e., whether the distance is within the threshold). The resulting recurrence plot is a binary matrix reflecting the similarity between the system states at different time points. Next, the system extracts various recurrence quantitative analysis features from the recurrence plot, including but not limited to: recurrence rate, determinism, average diagonal length, entropy, longest diagonal length, and white noise ratio. The above recurrence features are aggregated into a vector, which is used as a supplementary indicator of the morning blood pressure dynamic features, enhancing the ability to describe the nonlinear or chaotic characteristics hidden in the blood pressure trend.

[0077] By phase space reconstruction of the morning blood pressure time series and generation of the recurrence plot, the potential dynamic structure and similarity between states in the blood pressure change process can be revealed. Further extraction of recurrence quantitative analysis features helps to capture the nonlinear characteristics and hidden patterns in the blood pressure regulation process, serving as a powerful supplement to traditional statistical indicators, enhancing the expression ability of blood pressure dynamic features and the sensitivity of individual state changes, thereby improving the accuracy of anomaly detection and trend assessment.

[0078] S400: Based on the morning blood pressure segment data in the first preset period, generate short-term statistical features, and based on the morning blood pressure segment data in the second preset period, generate long-term statistical features, and dynamically adjust the reference baseline of the short-term and long-term statistical features through an individual adaptive learning model;

[0079] Further, the process of dynamic adjustment includes:

[0080] In the second sliding time window, collect the statistical features of the user's daily morning blood pressure segment;

[0081] Compare the feature distribution of the current window and the historical window,

[0082] When detecting that the distribution difference is greater than the difference threshold, it is determined that a concept drift event has occurred, and the reference baseline parameters of the short-term and long-term are updated, so that the abnormality determination rule adapts to the long-term change trend of the individual physiological state.

[0083] Specifically, the system sets a second sliding time window, e.g. the last 30 days, to dynamically assess the trend of the user's morning blood pressure characteristics. Within this time window, the system extracts pre-set statistical features from the morning blood pressure segment every day, including but not limited to blood pressure peak, mean, standard deviation, maximum rising rate, and recursive quantitative analysis features, etc., to form a continuous daily feature sequence. Then, the system compares the statistical feature distribution in the current time window with the feature distribution in the earlier historical time window. The comparison method can be based on distribution similarity measurement indicators such as KL divergence or Wasserstein distance, etc. When the system detects that the feature distribution difference between the current window and the historical window exceeds the pre-set difference threshold (e.g. KL divergence is greater than 0.5), it is considered that the user's blood pressure characteristics have concept drift, i.e. the user's physiological state or life behavior has changed significantly. At this time, the system triggers the update mechanism to adaptively correct the previously generated short-term reference baseline (such as the average of the last 7 days) and the long-term reference baseline (such as the mean, variance, etc. statistics of the last 30 days). The update can use exponential moving average, weighted update or recursive least squares, etc. method, so that the baseline parameters are closer to the true state of the current user.

[0084] By dynamically monitoring the statistical feature distribution of the user's morning blood pressure segment within the sliding time window, the concept drift event caused by the change of individual blood pressure characteristics over time can be detected in time, and the short-term and long-term reference baseline parameters are dynamically adjusted accordingly, so that the blood pressure abnormality judgment model has the ability of continuous self-adaptation, adapts to the long-term evolution trend of the user's physiological state, improves the sensitivity and accuracy of individualized monitoring, and avoids false positives and false negatives caused by static threshold setting.

[0085] S500: Deviation comparison of the current morning blood pressure peak with the long-term reference baseline, and dynamic comparison of the blood pressure rising rate with the short-term reference baseline, if both exceed the pre-set threshold, it is determined as a morning blood pressure abnormality event.

[0086] Further, the judgment condition of the morning blood pressure abnormality event is that the blood pressure peak deviation exceeds 10% of the long-term reference baseline, and the blood pressure rising rate exceeds 2 times the standard deviation of the short-term reference baseline.

[0087] Specifically, the system dynamically maintains two individualized reference baselines based on the foregoing process:

[0088] Based on the peak value of morning blood pressure in the second preset period (such as the past 30 days), the mean value thereof is calculated as a long-term reference baseline, and based on the blood pressure rising rate sequence in the first preset period (such as the past 7 days), the mean value and standard deviation thereof are calculated as a short-term reference baseline. Subsequently, the system obtains two key indicators in the current morning blood pressure segment: the current blood pressure peak value: all valid blood pressure values in the current analysis window are analyzed, and the maximum value is extracted; and the current blood pressure rising rate: the steepest part of the blood pressure rising in the current morning blood pressure segment is calculated, and the maximum blood pressure change rate is obtained.

[0089] Then, abnormality determination is performed. First, the current blood pressure peak value is compared with the mean value of the long-term reference baseline, and if the deviation (current value-baseline value) exceeds 10% of the baseline value, it is considered that the index is abnormal; then, the current maximum blood pressure rising rate is compared with the mean value and standard deviation of the short-term reference baseline, and if the value exceeds the mean value + 2 times the standard deviation of the short-term reference baseline, it is also determined to be abnormal. When the above two conditions are met at the same time, the system determines that there is an abnormal event of morning blood pressure at present, and can further trigger subsequent actions such as alarm, data marking or sending health prompt suggestions to the user.

[0090] Through the comparison and determination mechanism, the system can effectively identify abnormal blood pressure rising that may be related to high blood pressure risk, abnormal sympathetic nerve activity or cardiovascular events, and realize dynamic blood pressure monitoring with high sensitivity and high individual adaptability.

[0091] Embodiment Two

[0092] The intelligent blood pressure meter data analysis method based on an adaptive algorithm described in the application is applied to the daily health management of a user who is receiving hypertension control treatment. The user uses a matching wearable device (including a blood pressure meter and a behavior sensing module) to continuously record blood pressure data and behavior data during the night to the morning.

[0093] The specific application process is as follows:

[0094] Data acquisition stage: the user wears the device at night, and the device automatically acquires blood pressure data at a set time interval (such as every 15 minutes), while recording behavior data, including three-axis acceleration and ambient light intensity, and attaching accurate time stamps.

[0095] Wake-up time recognition: the system calls the historically trained SVM model to recognize the wake-up event in real time based on the behavior data. For example, the user continuously appears acceleration fluctuation and ambient light intensity mutation at about 06:40 in the morning, and the system accurately recognizes that the time is the wake-up time.

[0096] Analysis window construction and fine-tuning: with 06:40 as the center, extending 30 minutes forward and 60 minutes backward, an initial analysis window is formed. Subsequently, the system constructs a feature vector based on blood pressure measurement density and acceleration changes, calls a pre-trained LightGBM model to fine-tune the analysis window boundaries, and finally locks 06:10-07:50 as the morning blood pressure analysis window.

[0097] Blood pressure sequence cleaning and dynamic feature extraction: abnormal values in the analysis window are removed, missing values are completed by linear interpolation to form a continuous blood pressure sequence; the maximum rising rate, mean, and standard deviation are calculated, and recursive quantitative analysis features are extracted to fully describe the blood pressure change trend.

[0098] Baseline update and anomaly detection: the system automatically maintains a 7-day short-term and a 30-day long-term reference baseline, and detects whether the distribution changes have concept drift through a sliding window mechanism. If so, the baseline parameters are adjusted adaptively.

[0099] Abnormal event determination and feedback: if the daily blood pressure peak is 151 mmHg, which is 10% higher than the long-term reference baseline (135 mmHg), and the blood pressure rising rate is 6.2 mmHg / min, which is much higher than the short-term baseline (3.0±1.2 mmHg / min), the system determines that the morning blood pressure is abnormal. This abnormal event can be automatically pushed to the user's mobile phone App for warning and reminder, and the user is advised to record abnormal factors such as diet, mood, or whether to miss taking antihypertensive drugs.

[0100] Finally, it should be noted that the above only describes the preferred embodiments of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for analyzing data of an intelligent blood pressure monitor based on an adaptive algorithm, characterized in that: include: receiving continuous blood pressure measurement data including time stamps and synchronously collected behavioral data, wherein the blood pressure measurement data covers a period from a nighttime sleep period to a period after waking up; Infer the user's wake-up time based on behavioral data, preset and adjust the morning blood pressure analysis window, and extract blood pressure measurement data within the analysis window to form a morning blood pressure segment; The morning blood pressure segment data was cleaned to obtain a continuous blood pressure time series. Based on the blood pressure time series, the blood pressure change rate series was calculated, and the dynamic characteristics of morning blood pressure, including the maximum blood pressure rise rate, mean, standard deviation, and recursive quantitative analysis characteristics, were extracted. Generate short-term statistical features based on morning blood pressure data within a first preset period, generate long-term statistical features based on morning blood pressure data within a second preset period, and dynamically adjust the reference baselines of the short-term and long-term statistical features through an individual adaptive learning model; The deviation of the current morning blood pressure peak is compared with the long-term reference baseline, and the blood pressure rise rate is dynamically compared with the short-term reference baseline. If both exceed the preset threshold, it is determined to be an abnormal morning blood pressure event; The process of presetting and adjusting the morning blood pressure analysis window includes: Using the inferred user's wake-up time as a reference base point, extend the first preset time forward to determine the measurement interval before waking up, and extend the second preset time backward to determine the measurement interval after waking up; Merge the measurement intervals before and after waking up to form a morning blood pressure analysis window, and dynamically adjust the window boundaries based on blood pressure measurement data, behavioral data, and blood pressure measurement density; The process of dynamically fine-tuning window boundaries includes: constructing an input feature vector for predicting the optimal analysis window boundary based on blood pressure measurement data, behavioral data, and blood pressure measurement density; Input the input feature vector into the pre-trained LightGBM regression model and output the optimal adjustment amount of the window start and end points; Finely modifying the boundary of the preliminary analysis window according to the adjustment amount to form a final morning blood pressure analysis window; The process of extracting recursive quantitative analysis features includes: Based on the continuous blood pressure time series of the morning blood pressure segment, phase space reconstruction is performed to generate a trajectory matrix; A recurrence graph is calculated based on the trajectory matrix to describe the similarity between blood pressure states; Extract recursive quantitative analysis features from the recursive graph as a supplementary indicator of the dynamic characteristics of morning blood pressure; The dynamic adjustment process includes: In the second sliding time window, the statistical features of the user's daily morning blood pressure segment are collected; The feature distribution of the current window is compared with that of the historical window. When the distribution difference is detected to be greater than the difference threshold, it is determined that a concept drift event has occurred, and the short-term and long-term reference baseline parameters are updated to make the abnormality judgment rules adapt to the long-term change trend of the individual's physiological state.

2. The method for analyzing data of an intelligent blood pressure monitor based on an adaptive algorithm according to claim 1, characterized in that: The process of inferring the user's wake-up time includes: Based on historical behavior data, an SVM classification model is trained to determine the characteristics of waking behavior patterns; Receive user behavior data in real time and extract behavior feature vectors, and input the behavior feature vectors into the SVM classification model to determine the waking behavior and output the user's waking time.

3. The method for analyzing data of an intelligent blood pressure monitor based on an adaptive algorithm according to claim 1, characterized in that: The process of cleaning the morning blood pressure data includes: Using a first sliding window to traverse the blood pressure time series, the blood pressure change rate within the first sliding window is calculated, and measurement points whose change rate exceeds a preset threshold are identified as outliers and removed; For the series after removing outliers, we check whether there are missing measurement data, and based on the timestamps and blood pressure values ​​of the previous and next valid measurement points, we use linear interpolation to estimate and fill in the missing values ​​to form a complete and continuous blood pressure time series.

4. The method for analyzing data of an intelligent blood pressure monitor based on an adaptive algorithm according to claim 3, characterized in that: The criteria for determining abnormal morning blood pressure events are: the peak blood pressure deviation exceeds 10% of the long-term reference baseline, and the blood pressure rise rate exceeds 2 times the standard deviation of the short-term reference baseline.

Images