A method for identifying early behavior sequence abnormalities of hypertension and processing data

Abstract

The application discloses a hypertension early behavior sequence anomaly identification and data processing method, relates to the technical field of data processing, and comprises the following steps: acquiring blood pressure related monitoring data and behavior state data of a target patient from a preset database, wherein the blood pressure related monitoring data comprises monitoring sequence data in multiple dimensions; calculating abnormal degree values of each abnormal data segment based on the difference between the abnormal data segment in the blood pressure related monitoring data and a preset normal value range; calculating the attention range of a preset multi-head attention mechanism based on the abnormal degree values; adjusting the attention weight based on the interaction relationship between the behavior state data and the blood pressure related monitoring data to obtain a final attention weight; and identifying and processing the blood pressure related monitoring data based on the preset multi-head attention mechanism and the final attention weight to obtain a hypertension risk identification result. The application can provide accurate hypertension risk early warning.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for identifying and processing abnormal behavioral sequences in the early stages of hypertension. Background Technology

[0002] Currently, the diagnosis of hypertension mainly relies on occasional blood pressure measurements or ambulatory blood pressure monitoring. However, this method has a lag effect, and a diagnosis is usually made only when blood pressure has already risen significantly. Studies have shown that the occurrence and development of hypertension are closely related to daily behavioral patterns (such as sedentary lifestyle, sleep disorders, stress response, etc.). These behavioral abnormalities often appear months or even years before blood pressure rises significantly. However, existing methods can only detect overt abnormalities (such as a certain indicator exceeding the standard) during the monitoring of patients' characteristic data. It is difficult to identify the synchronous correlation of multiple monitoring data in the early stages of hypertension. Therefore, it is impossible to provide accurate early warning of hypertension risk. Summary of the Invention

[0003] The main purpose of this application is to provide a method for identifying and processing abnormal behavioral sequences in the early stages of hypertension. This method aims to solve the technical problem in related technologies that rely on occasional blood pressure measurements or ambulatory blood pressure monitoring, which makes it difficult to identify the synchronous correlation of multiple monitoring data in the early stages of hypertension, thus failing to provide accurate early warning of hypertension risk.

[0004] To achieve the above objectives, embodiments of this application provide a method for identifying and processing abnormal behavioral sequences in the early stages of hypertension, including:

[0005] Blood pressure-related monitoring data and behavioral status data of the target patients are obtained from a pre-set database. The blood pressure-related monitoring data includes monitoring sequence data from multiple dimensions.

[0006] Based on the difference between abnormal data segments in blood pressure-related monitoring data and the preset normal value range, the degree of abnormality of each abnormal data segment is calculated.

[0007] Based on the anomaly level value, the attention range of the preset multi-head attention mechanism is calculated;

[0008] Based on the interaction between behavioral state data and blood pressure-related monitoring data, the attention weights of the monitoring sequence data corresponding to each dimension of the attention range are adjusted to obtain the final attention weights.

[0009] Based on a pre-set multi-head attention mechanism and final attention weight, blood pressure-related monitoring data are identified and processed to obtain hypertension risk identification results.

[0010] In one possible implementation of this application, based on the difference between abnormal data segments in blood pressure-related monitoring data and a preset normal value range, the degree of abnormality for each abnormal data segment is calculated, including:

[0011] Determine the first number of monitoring data points in the abnormal data segment of blood pressure-related monitoring data;

[0012] Based on the minimum difference between the monitored data in the abnormal data segment and the preset normal value range, the first quantity, and the average difference between adjacent monitored data in the abnormal data segment.

[0013] In one possible implementation of this application, before calculating the degree of abnormality for each abnormal data segment based on the difference between the abnormal data segments in blood pressure-related monitoring data and a preset normal value range, the method further includes:

[0014] Identify multiple monitoring data points in blood pressure-related monitoring data that are outside the preset normal range;

[0015] Any consecutive segment of a preset number of monitoring data points is treated as an abnormal data segment, generating multiple abnormal data segments in the blood pressure-related monitoring data.

[0016] In one possible implementation of this application, the attention range of a preset multi-head attention mechanism is calculated based on anomaly severity values, including:

[0017] Based on the anomaly severity value and the difference in time length between different anomaly data segments, the degree of intra-sequence deviation of each anomaly data segment is calculated;

[0018] Based on the degree of intra-sequence bias, the attention range of the preset multi-head attention mechanism is calculated.

[0019] In one possible implementation of this application, the degree of intra-sequence deviation of each abnormal data segment is calculated based on the anomaly severity value and the difference in time length between different abnormal data segments, including:

[0020] For any abnormal data segment, select abnormal data segments whose time length difference with the current abnormal data segment is less than a preset similar time length, and use them as the neighboring abnormal data segments;

[0021] Calculate the sum of the time length differences between adjacent abnormal data segments and the current abnormal data segment, as well as the average difference in the abnormality level values;

[0022] Based on the sum of differences, the average difference, and the number of adjacent abnormal data segments, the degree of intra-sequence deviation for each abnormal data segment corresponding to different monitoring sequence data is calculated.

[0023] In one possible implementation of this application, before calculating the attention range of a preset multi-head attention mechanism based on the degree of intra-sequence bias, the method further includes:

[0024] Based on the degree of intra-sequence deviation, each abnormal data segment is divided to obtain the final abnormal data segment.

[0025] Based on the degree of intra-sequence bias, the attention range of the preset multi-head attention mechanism is calculated, including:

[0026] Based on the time length of the final abnormal data segment and the degree of intra-sequence deviation, the attention range of each monitoring sequence data corresponding to the preset multi-head attention mechanism is calculated.

[0027] In one possible implementation of this application, the abnormal data segments are divided based on the degree of intra-sequence deviation to obtain the final abnormal data segments, including:

[0028] The outlier data segments are sorted in ascending order of their degree of deviation within the sequence to obtain an ordered sequence.

[0029] Identify the two adjacent data points corresponding to the maximum absolute value of the difference in the degree of deviation within the ordered sequence, and divide the abnormal data segments on both sides of these two adjacent data points in the ordered sequence into the first data segment and the second data segment;

[0030] The data segment corresponding to the minimum mean of the degree of deviation within the sequence in the first and second data segments is selected as the final outlier data segment.

[0031] In one possible implementation of this application, based on the interaction between behavioral state data and blood pressure-related monitoring data, the attention weights of the monitoring sequence data corresponding to each dimension of the attention range are adjusted to obtain the final attention weights, including:

[0032] By combining behavioral state data with abnormal data segments from blood pressure-related monitoring data, multiple data pairs are obtained;

[0033] The correlation strength of each data pair is calculated based on the frequency of occurrence of the data pairs and the proportion of behavioral state data in the data pairs relative to the number of all abnormal data segments in each monitoring sequence data.

[0034] Based on the correlation strength and the degree of anomalousness of the corresponding anomalous data segment, the adjustment coefficient of the attention weight is calculated.

[0035] The attention weights are adjusted based on the adjustment coefficients to obtain the final attention weights.

[0036] In one possible implementation of this application, the adjustment coefficient for the attention weight is calculated based on the correlation strength and the degree of anomalousness of the corresponding anomalous data segment, including:

[0037] Determine the first slope of the fitted line for the degree of anomalousness of the data for the corresponding anomalous data segment;

[0038] Based on the ratio between the mean and standard deviation of the association strength, and the first slope, the adjustment coefficient of the attention weight corresponding to each monitoring sequence data for the behavioral state data is calculated.

[0039] In one possible implementation of this application, the attention weights are adjusted based on an adjustment coefficient to obtain the final attention weights, including:

[0040] Multiply the adjustment factor by the attention weight to obtain the weight adjustment value;

[0041] The final attention weight is obtained by adding the weight adjustment value to the attention weight.

[0042] This application provides a method for identifying and processing abnormal behavioral sequences in the early stages of hypertension. Compared to related technologies that rely on occasional blood pressure measurements or ambulatory blood pressure monitoring, which struggle to identify synchronous correlations among multiple monitoring data in the early stages of hypertension and thus fail to provide accurate hypertension risk warnings, this application retrieves blood pressure-related monitoring data and behavioral state data of the target patient across multiple dimensions from a pre-set database. Based on the difference between abnormal data segments in the blood pressure-related monitoring data and a pre-set normal value range, the degree of abnormality for each abnormal data segment is determined. Then, based on this degree of abnormality, the attention range of a pre-set multi-head attention mechanism is calculated. Finally, based on the relationship between the behavioral state data and the blood pressure-related monitoring data... The interaction between the attentional states and the blood pressure-related monitoring data is analyzed to adjust the attention weights of the monitoring sequence data in each dimension of the attention range, thereby obtaining the final attention weights. Finally, based on the preset multi-head attention mechanism and the final attention weights, the blood pressure-related monitoring data is identified and processed to obtain the hypertension risk identification results. By using the multi-head attention mechanism to identify and process the monitoring sequence data in different dimensions, and then adjusting the attention weights based on the interaction between the behavioral state data and the blood pressure-related monitoring data, the neural network corresponding to the attention mechanism can locate the behavioral segments that cause hypertension through the distribution of attention weights, identify the risk of hypertension in advance, and thus provide accurate early warning of hypertension risk. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the first embodiment of the method for identifying and processing abnormal behavioral sequences in the early stages of hypertension according to this application.

[0044] Figure 2 This is a schematic diagram of data changes corresponding to the small attention span involved in the method for identifying and processing abnormal behavioral sequences in the early stages of hypertension in this application.

[0045] Figure 3 This is a schematic diagram of data changes corresponding to the large attention span involved in the method for identifying and processing abnormal behavioral sequences in the early stages of hypertension in this application.

[0046] Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application. Detailed Implementation

[0047] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0048] This application provides a method for identifying and processing abnormal behavioral sequences in the early stages of hypertension. In the first embodiment of this method, refer to... Figure 1 ,include:

[0049] Step S10: Obtain blood pressure-related monitoring data and behavioral status data of the target patient from a preset database. The blood pressure-related monitoring data includes monitoring sequence data from multiple dimensions.

[0050] Step S20: Based on the difference between the abnormal data segments in the blood pressure-related monitoring data and the preset normal value range, calculate the abnormality value of each abnormal data segment.

[0051] Step S30: Calculate the attention range of the preset multi-head attention mechanism based on the anomaly level value;

[0052] Step S40: Based on the interaction between behavioral state data and blood pressure-related monitoring data, adjust the attention weights of the monitoring sequence data of each dimension corresponding to the attention range to obtain the final attention weights;

[0053] Step S50: Based on the preset multi-head attention mechanism and the final attention weight, the blood pressure-related monitoring data is identified and processed to obtain the hypertension risk identification result.

[0054] This embodiment aims to: use a multi-head attention mechanism to identify and process monitoring sequence data under different dimensions, and then adjust the attention weights according to the interaction between behavioral state data and blood pressure-related monitoring data, so that the neural network corresponding to the attention mechanism can locate the behavioral segments that cause hypertension through the distribution of attention weights, identify the risk of hypertension in advance, and thus provide accurate early warning of hypertension risk.

[0055] The specific steps are as follows:

[0056] Step S10: Obtain blood pressure-related monitoring data and behavioral status data of the target patient from a preset database. The blood pressure-related monitoring data includes monitoring sequence data from multiple dimensions.

[0057] As an example, the method for identifying and processing abnormal behavioral sequences in the early stages of hypertension can be applied to a device for identifying and processing abnormal behavioral sequences in the early stages of hypertension. This device belongs to a system for identifying and processing abnormal behavioral sequences in the early stages of hypertension, and the system belongs to a device for identifying and processing abnormal behavioral sequences in the early stages of hypertension.

[0058] As an example, the preset database can be a database in a hospital used to store patients' medical records. When determining whether a target patient has hypertension, blood pressure-related monitoring data and behavioral status data from historical time periods in the preset database can be extracted for subsequent judgment processes.

[0059] As an example, blood pressure-related monitoring data can include blood pressure data, heart rate data, and heart rate variability data extracted from electrocardiogram signals. Each type of data corresponds to a dimension. For example, blood pressure data and heart rate data are monitoring sequence data of different dimensions.

[0060] As an example, behavioral state data can be the patient's behavioral state at different points in time, such as: sleeping, resting, active, exercising, emotional fluctuations, drinking alcohol, etc. Among them, blood pressure-related monitoring data can reflect early characteristic indicators of suspected hypertension patients. Before use, a sliding window algorithm is used to identify fluctuation outliers to ensure data reliability.

[0061] Step S20: Based on the difference between the abnormal data segments in the blood pressure-related monitoring data and the preset normal value range, calculate the abnormality value of each abnormal data segment.

[0062] As an example, the abnormality value represents the degree of deviation between the data in different data segments of blood pressure-related monitoring data and the normal data. The abnormality value of each abnormal data segment is calculated by determining the difference between the abnormal data segment and the preset normal value range.

[0063] As an example, the preset normal value range is a set of values, which can be a reasonable range of values ​​determined based on the age and basic physical condition of the current target patient for each blood pressure-related monitoring data, without any specific limitation.

[0064] Prior to step S20, which involves identifying and processing abnormal behavioral sequences in the early stages of hypertension, the following steps are also included:

[0065] Identify multiple monitoring data points in blood pressure-related monitoring data that are outside the preset normal range.

[0066] Any consecutive segment of a preset number of monitoring data points is treated as an abnormal data segment, generating multiple abnormal data segments in the blood pressure-related monitoring data.

[0067] As an example, before calculating the degree of abnormality of different abnormal data segments, it is necessary to identify the abnormal data segments in the blood pressure-related monitoring data. First, based on the comparison between the data in the blood pressure-related monitoring data and the preset normal value range, identify multiple monitoring data that are not in the preset normal value range. The more the monitoring data deviates from the preset normal value range, and the more obvious the fluctuation of the value of a certain dimension of data, the more abnormal it is.

[0068] As an example, any continuous sequence of multiple monitoring data points of a predetermined number can be considered as an abnormal data segment. Thus, for each dimension of the monitoring sequence data, multiple abnormal data segments can be obtained.

[0069] Step S20, which involves identifying and processing abnormal behavioral sequences in the early stages of hypertension, includes:

[0070] Step S21: Determine the first number of monitoring data in the abnormal data segment of the blood pressure-related monitoring data;

[0071] Step S22: Based on the minimum difference between the monitored data in the abnormal data segment and the preset normal value range, the first quantity, and the average difference between adjacent monitored data in the abnormal data segment.

[0072] As an example, each abnormal data segment may include at least two monitoring data points, where the first number can be 3, 5, 7, etc., without any specific limitation.

[0073] As an example, taking an anomalous data segment t as an example, the anomalousness value... The calculation method can be:

[0074] ;

[0075] In the formula, Indicates abnormal data segment The degree of abnormality; Indicates abnormal data segment The first quantity; Indicates monitoring data The minimum difference from the preset normal value range; These represent two adjacent monitoring data points.

[0076] Formula logic: This reflects the difference between all monitored data and the preset normal value range in an abnormal data segment. When the monitored data is greater than the upper limit of the preset normal value range, the difference refers to the difference between the monitored data and the upper limit. When the monitored data is less than the lower limit of the preset normal value range, the difference refers to the difference between the lower limit and the monitored data. The larger the difference and the more data in the abnormal data segment, the greater the abnormality of the abnormal data segment. This represents the mean of the absolute values ​​of the differences between all two adjacent data points within an outlier data segment. It reflects the degree of fluctuation of the data within the outlier data segment. The larger the value, the greater the degree of outlier the outlier data segment. Similarly, the degree of outlier values ​​for each outlier data segment can be obtained.

[0077] As an example, abnormal data segments in monitoring sequence data under a certain dimension reflect the deviation data of that dimension. When the abnormality value is larger, the deviation of the corresponding abnormal data segment from the normal range is also larger. However, directly identifying hypertension early based on abnormal data segments and their abnormality values ​​will be somewhat arbitrary. Therefore, it is necessary to combine the abnormal performance of monitoring sequence data from different dimensions to achieve early abnormal identification of hypertensive patients.

[0078] Step S30: Calculate the attention range of the preset multi-head attention mechanism based on the anomaly level value.

[0079] As an example, based on the degree of anomalousness of each anomalous data segment, the degree of intra-sequence deviation of each anomalous data segment is determined, and then the attention range of the preset multi-head attention mechanism is calculated based on the degree of intra-sequence deviation.

[0080] As an example, the pre-defined multi-head attention mechanism introduces multiple parallel attention heads, each performing attention operations independently and processing different information, thereby capturing feature information more comprehensively from different subspaces of the sequence. This helps the model detect long-distance dependencies or short-term fluctuations based on changes in different monitored sequence data, capturing richer features. The multi-head self-attention mechanism allows the model to focus on different parts of the input sequence in different representation subspaces, with each head computing independently and finally concatenating the output, improving feature representation capabilities.

[0081] Specifically, for the monitoring sequence data of each dimension of the target patient, when the patient's body gradually develops into hypertension, there may be related abnormalities in some dimensions of data in the early stage. For example, abnormalities in heart rate variability indicate a decline in the patient's autonomic regulation function, which may increase the risk of future cardiovascular events and lead to hypertension. Therefore, a multi-head attention mechanism is applied to analyze the data of each dimension to obtain its predictive support for the patient's hypertension. Each attention head processes a different subspace, thereby capturing the correlation between its corresponding dimensional features and blood pressure.

[0082] During the processing, the corresponding subspace is divided according to the distribution characteristics of the monitoring sequence data of each dimension. When the data changes are more unstable, the magnitude of each change is usually larger. In this case, a high-frequency subspace is divided by a smaller attention range to capture short-term fluctuations. When the data changes are stable and the fluctuations are more gradual, a low-frequency subspace is divided by a larger attention range to capture long-term dependencies.

[0083] The step S30, which calculates the attention range of the preset multi-head attention mechanism based on the anomaly level value, includes:

[0084] Step S31: Based on the anomaly severity value and the difference in time length between different anomaly data segments, calculate the degree of intra-sequence deviation for each anomaly data segment.

[0085] As an example, the degree of intra-sequence bias is used to represent the degree of deviation of the abnormal data segment from the monitoring sequence data of the current dimension. The greater the degree of intra-sequence bias, the more interference the corresponding abnormal data segment will cause in the identification of hypertension, which will affect the accuracy of the determined attention range.

[0086] The step S31, which calculates the degree of intra-sequence deviation for each abnormal data segment based on the anomaly severity value and the difference in time length between different abnormal data segments, includes:

[0087] For any abnormal data segment, select abnormal data segments whose time length difference with the current abnormal data segment is less than a preset similar time length, and use them as the neighboring abnormal data segments;

[0088] Calculate the sum of the time length differences between adjacent abnormal data segments and the current abnormal data segment, as well as the average difference in the abnormality level values;

[0089] As an example, the preset similar time length can be the minimum time length among all the time lengths of the abnormal data segments.

[0090] As an example, for any abnormal data segment, the time length between other abnormal data segments and the current abnormal data segment is compared. When the time length difference between the two is less than a preset similar time length, the abnormal data segment is regarded as the neighboring abnormal data segment of the current abnormal data segment.

[0091] As an example, the difference sum represents the sum of the absolute values ​​of the differences in time length between all adjacent outlier data segments and the current outlier data segment, while the difference mean is the average of the absolute values ​​of the differences in the outlier severity values ​​between all adjacent outlier data segments and the current outlier data segment.

[0092] Based on the sum of differences, the average difference, and the number of adjacent abnormal data segments, the degree of intra-sequence deviation for each abnormal data segment corresponding to different monitoring sequence data is calculated.

[0093] As an example, consider the outlier segment t of a monitoring sequence data x of arbitrary dimensions, and the degree of intra-sequence bias. The calculation method can be:

[0094] ;

[0095] In the formula, This indicates the number of all adjacent abnormal data segments of the abnormal data segment; This represents the sum of the time length differences between all adjacent outlier data segments and the outlier data segment t; This represents the average difference in the degree of abnormality between adjacent abnormal data segments and abnormal data segment t.

[0096] As an example, the fewer the number of adjacent abnormal data segments in the corresponding monitoring sequence data and the greater the difference in time length, the greater the deviation of the abnormal data segment. This means that the abnormal data segment will interfere with the identification of hypertension abnormalities. Furthermore, among all adjacent abnormal data segments, the greater the difference in the abnormality value between the abnormal data segment and the other adjacent abnormal data segments, the more the change trend of the current abnormal data segment t deviates from the overall abnormal performance, which will affect the final determined attention range.

[0097] Step S32: Calculate the attention range of the preset multi-head attention mechanism based on the degree of intra-sequence bias.

[0098] As an example, the shorter the time length of the abnormal data segment in the monitored sequence data and the smaller the degree of deviation within the sequence, the more detailed the subspace should be divided. The sensitivity of the attention head to the changing trend of the data is enhanced to short-term anomalies. Therefore, the attention range of the preset multi-head attention mechanism is calculated based on the degree of deviation within the sequence.

[0099] Before step S32, which calculates the attention range of the preset multi-head attention mechanism based on the degree of intra-sequence bias, the following steps are also included:

[0100] Based on the degree of deviation within the sequence, each abnormal data segment is divided to obtain the final abnormal data segment.

[0101] As an example, in the monitoring sequence data, there may be some abnormal data segments that are minor disturbances. Their abnormal performance over the entire time scale has a weak correlation with early abnormalities of hypertension. Therefore, in the process of determining the attention range of different subspaces of the multi-head attention mechanism, it is necessary to screen out abnormal data segments that can effectively reflect the data in this dimension as data of early behavioral abnormalities of hypertension. When the time length of the abnormal data segment and the deviation between the data performance of the abnormal data segment and the overall monitoring sequence data are too large, such abnormal data segments need to be removed.

[0102] As an example, the first step is to divide the outlier data into segments based on the degree of intra-sequence deviation of different outlier data segments. The segments with smaller intra-sequence deviation are selected as the final outlier data segments for focused analysis. This allows us to determine the scope of attention based on the degree of intra-sequence deviation of the final outlier data segments.

[0103] The step of dividing the outlier data segments based on the degree of intra-sequence deviation to obtain the final outlier data segments includes:

[0104] The outlier data segments are sorted in ascending order of their degree of deviation within the sequence to obtain an ordered sequence.

[0105] As an example, in the process of dividing abnormal data segments, each abnormal data segment is sorted according to the degree of deviation within the sequence, in ascending order, to obtain an ordered sequence.

[0106] Identify the two adjacent data points corresponding to the maximum absolute value of the difference in the degree of deviation within the ordered sequence, and divide the abnormal data segments on both sides of these two adjacent data points in the ordered sequence into the first data segment and the second data segment.

[0107] The data segment corresponding to the minimum mean of the degree of deviation within the sequence in the first and second data segments is selected as the final outlier data segment.

[0108] As an example, in all the data in the abnormal data segment, the difference in the degree of deviation between two adjacent data (the difference between the degree of deviation between the next data and the degree of deviation between the previous data) is calculated, and the two adjacent data corresponding to the maximum value of these differences are determined. The positions of these two adjacent data are located, and the two sides of the two adjacent data are divided into the first data segment and the second data segment respectively. The two adjacent data are then assigned to the corresponding first data segment or the second data segment respectively.

[0109] As an example, when selecting the final outlier data segment, the data segment corresponding to the minimum mean of the degree of deviation within the sequence between the two outlier data segments is selected as the final outlier data segment. The final outlier data segment has strong reference value and can be used to calculate a relatively accurate attention range.

[0110] As an example, for all the final abnormal data segments, they may induce the co-occurrence of abnormal data in patients and hypertension. Therefore, analyzing the duration of the abnormal data segments reveals that the longer the duration of the abnormal data segments, the longer the corresponding attention span. This allows for the effective identification of possible fluctuations in the abnormal data segments of this feature sequence data, such as... Figure 2 The image shows a schematic diagram of monitoring sequence data with a small attention range, such as... Figure 3 The diagram shows a monitoring sequence data with a large attention range.

[0111] Step S32, which calculates the attention range of the preset multi-head attention mechanism based on the degree of intra-sequence bias, includes:

[0112] Based on the time length of the final abnormal data segment and the degree of intra-sequence deviation, the attention range of each monitoring sequence data corresponding to the preset multi-head attention mechanism is calculated.

[0113] As an example, the attention range of the subspace corresponding to the monitored sequence data x. The calculation method can be:

[0114] ;

[0115] In the formula, Indicates the time length of the final abnormal data segment t; This indicates the degree of intra-sequence deviation in the final outlier data segment t. It represents the sum of the intra-sequence deviations of all final abnormal data segments t in the monitored sequence data x.

[0116] As an example, the attention range of the subspace corresponding to the monitored sequence data is obtained by weighting the time length of all final abnormal data segments with the degree of deviation within the sequence, thereby capturing data fluctuations of an effective length. The shorter the time length of the abnormal data segments in the data sequence and the greater the degree of deviation within the sequence, the more detailed the subspace division should be. In other words, the attention head is more sensitive to short-term anomalies in capturing the changing trends of the data.

[0117] Step S40: Based on the interaction between behavioral state data and blood pressure-related monitoring data, the attention weights of the monitoring sequence data corresponding to each dimension of the attention range are adjusted to obtain the final attention weights.

[0118] As an example, after determining the attention range of each monitoring feature data subspace, a multi-head attention mechanism is used in parallel to employ several attention heads. Each attention head corresponds to a monitoring sequence data, and based on its attention range, it captures the corresponding degree of short-term fluctuations or long-term trends (for example, one head may focus on the instantaneous fluctuations of heart rate, while another head focuses on the long-term trend of a certain monitoring indicator of the patient's body, thereby identifying abnormal behaviors at different stages), thus obtaining the attention weight of each monitoring sequence data x. These attention weights reflect the correlation between input features and blood pressure.

[0119] As an example, before the monitoring sequence data appears, there may be abnormal data segments in the monitoring sequence data caused by a specific behavioral state, such as elevated blood pressure after exercise or drinking. In this case, it is necessary to identify the combination pattern of this column, bind its weight through the attention mechanism, and enhance the model's sensitivity to the combination pattern. Therefore, based on the interaction between behavioral state data and blood pressure-related monitoring data, the attention weights of different dimensions of data are adjusted to obtain the final attention weights, making the weight values ​​more accurate.

[0120] Step S40 includes:

[0121] Step S41: Combine the behavioral state data with the abnormal data segments in the blood pressure-related monitoring data to obtain multiple data pairs.

[0122] As an example, behavioral state data can be combined with abnormal data segments in blood pressure-related monitoring data. The combination method can be to obtain the behavioral state data of the target patient before the occurrence of abnormal data segments in the monitoring sequence data of the time axis, and integrate these behavioral state data with the corresponding abnormal data segments into a data pair.

[0123] Step S42: Based on the occurrence frequency of data pairs and the proportion of behavioral state data in a data pair relative to the number of all abnormal data segments in each monitoring sequence data, the correlation strength of each data pair is calculated.

[0124] As an example, a data pair includes (behavioral state data, abnormal data segments of a certain monitoring sequence data). After obtaining multiple data pairs, if a certain combination of data pairs occurs more frequently and stably, it is considered an associated behavioral sequence. When repeated abnormalities occur repeatedly, it is necessary to merge the combinations of behavioral changes and bind corresponding attention weights to all associated behavioral sequences within the combination.

[0125] As an example, the occurrence frequency of a data pair is the occurrence frequency of the current data pair among all data pairs. For example, if there are a total of 5 data pairs and there is only 1 current data pair, then the occurrence frequency is 1 / 5.

[0126] As an example, the quantity proportion represents the proportion of behavioral state data to all abnormal data segments in the monitoring sequence data. For example, if there are 5 data pairs containing a certain behavioral state data in 10 data pairs, then the quantity proportion is 1 / 2. The correlation strength of each data pair is determined based on the frequency of occurrence and the quantity proportion.

[0127] As an example, consider data pair u; the correlation strength of data pair u. The calculation method can be:

[0128] ;

[0129] In the formula, This indicates the frequency of occurrence of the current data pair u among all data pairs; This indicates the percentage of all abnormal data segments in the monitored sequence data relative to the behavior status data of u.

[0130] As an example, the higher the frequency of occurrence of data pair u in all data pairs in historical data, and the higher the proportion of current behavioral state data in all data pairs of characteristic abnormal data segments at a specific time, the stronger the correlation of the current data pair.

[0131] Step S43: Based on the correlation strength and the degree of anomalousness of the corresponding anomalous data segment, calculate the adjustment coefficient of the attention weight.

[0132] As an example, when the correlation strength of a data pair shows a trend of larger and more stable values, the current leading behavior state has a stronger predictive power for the monitored feature data. Furthermore, when the abnormality level of the abnormal data segment in the data pair shows a trend of gradually increasing values, the current behavior state has a more significant effect on the corresponding monitoring sequence data, and the adjustment of its attention weight is greater. Therefore, the adjustment coefficient of the attention weight is calculated based on the changing trend of the correlation strength and the abnormality level of the corresponding abnormal data segment.

[0133] Step S43, which calculates the adjustment coefficient of the attention weight based on the correlation strength and the degree of anomalousness of the corresponding anomalous data segment, includes:

[0134] Determine the first slope of the fitted line for the degree of anomalousness of the data for the corresponding anomalous data segment.

[0135] Based on the ratio between the mean and standard deviation of the association strength, and the first slope, the adjustment coefficient of the attention weight corresponding to each monitoring sequence data for the behavioral state data is calculated.

[0136] As an example, for any dimension of monitoring sequence data, based on the correlation strength and trend of all data pairs of its corresponding behavioral state data, the attention weights of the monitoring sequence data are adjusted when different behavioral states occur, and the adjustment coefficient is determined. The method can be:

[0137] ;

[0138] In the formula, This represents the adjustment factor of the behavioral state data y to the monitoring sequence data x. , Let represent the mean and standard deviation of the association strength of all data pairs that conform to this combination (data combination is (x, y)); denoted by , represents the first slope of the fitted line for all data to the corresponding outlier data segment, tanh represents the hyperbolic tangent function used for normalization calculation, its range is [-1,1], and exp is the natural constant.

[0139] Step S44: Adjust the attention weights based on the adjustment coefficients to obtain the final attention weights.

[0140] As an example, after determining the adjustment coefficients, attention weights are applied to each monitored sequence data. The adjustment coefficient is increased to different degrees to obtain the final attention weight of each behavioral state data to each monitoring sequence data. By allocating attention weights to each dimension of data, the contribution of key monitoring sequence data is highlighted.

[0141] Step S44 further includes:

[0142] Multiply the adjustment factor by the attention weight to obtain the weight adjustment value;

[0143] The final attention weight is obtained by adding the weight adjustment value to the attention weight.

[0144] As an example, the adjustment of attention weights by adjusting the coefficient can be done by multiplying the adjustment coefficient by the attention weight to obtain the weight adjustment value. The weight adjustment value is less than the attention weight, indicating the degree to which the attention weight needs to be adjusted. Then, the weight adjustment value is added to the attention weight to obtain the final attention weight.

[0145] Step S50: Based on the preset multi-head attention mechanism and the final attention weight, the blood pressure-related monitoring data is identified and processed to obtain the hypertension risk identification result.

[0146] As an example, when processing blood pressure-related monitoring data through the multi-head attention mechanism in a deep learning model, the acquired monitoring sequence data is first processed through the initial attention mechanism in the model to obtain the original attention weights. Then, the original attention weights are combined with the attention weights obtained from the analysis to obtain the weighted attention weights. At this point, the attention weights are not adjusted based on the behavioral state data.

[0147] Furthermore, the aforementioned attention module is inserted into an appropriate layer (such as the decoder part in an encoder-decoder structure) to enhance the model's ability to identify anomalies in monitoring sequence data of hypertensive patients.

[0148] When corresponding behavioral state data of the target patient exists, the attention weight of the associated monitoring sequence data is adjusted to obtain the final attention weight. Then, through the preset multi-head attention mechanism and the final attention weight, the blood pressure-related monitoring data is identified and processed. The model outputs the hypertension risk identification result, which increases the accuracy of identifying possible abnormalities in early hypertension patients. Finally, a mechanism for identifying early behavioral sequence abnormalities in suspected hypertension patients under continuous monitoring is obtained, and a hypertension risk warning is issued in a timely manner.

[0149] This application provides a method for identifying and processing abnormal behavioral sequences in the early stages of hypertension. Compared to related technologies that rely on occasional blood pressure measurements or ambulatory blood pressure monitoring, which struggle to identify synchronous correlations among multiple monitoring data in the early stages of hypertension and thus fail to provide accurate hypertension risk warnings, this application retrieves blood pressure-related monitoring data and behavioral state data of the target patient across multiple dimensions from a pre-set database. Based on the difference between abnormal data segments in the blood pressure-related monitoring data and a pre-set normal value range, the degree of abnormality for each abnormal data segment is determined. Then, based on this degree of abnormality, the attention range of a pre-set multi-head attention mechanism is calculated. Finally, based on the relationship between the behavioral state data and the blood pressure-related monitoring data... The interaction between the attentional states and the blood pressure-related monitoring data is analyzed to adjust the attention weights of the monitoring sequence data in each dimension of the attention range, thereby obtaining the final attention weights. Finally, based on the preset multi-head attention mechanism and the final attention weights, the blood pressure-related monitoring data is identified and processed to obtain the hypertension risk identification results. By using the multi-head attention mechanism to identify and process the monitoring sequence data in different dimensions, and then adjusting the attention weights based on the interaction between the behavioral state data and the blood pressure-related monitoring data, the neural network corresponding to the attention mechanism can locate the behavioral segments that cause hypertension through the distribution of attention weights, identify the risk of hypertension in advance, and thus provide accurate early warning of hypertension risk.

[0150] Reference Figure 4 , Figure 4 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.

[0151] like Figure 4 As shown, the device for identifying and processing abnormal early behavioral sequences of hypertension may include: a processor 1001, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005.

[0152] Optionally, the device for identifying and processing abnormal early behavioral sequences of hypertension may also include a user interface, a network interface, a camera, RF (Radio Frequency) circuitry, sensors, a WiFi module, etc. The user interface may include a display screen and an input submodule such as a keyboard; optional user interfaces may also include standard wired or wireless interfaces. The network interface may include standard wired or wireless interfaces (such as a Wi-Fi interface).

[0153] Those skilled in the art will understand that Figure 4The structure of the hypertension early behavioral sequence abnormality identification and data processing device shown does not constitute a limitation on the hypertension early behavioral sequence abnormality identification and data processing device. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0154] like Figure 4 As shown, the memory 1005, serving as a storage medium, may include an operating system, a network communication module, and a program for identifying and processing early behavioral sequences of hypertension. The operating system is a program that manages and controls the hardware and software resources of the device for identifying and processing early behavioral sequences of hypertension, supporting the operation of the program and other software and / or programs. The network communication module is used to enable communication between the various components within the memory 1005, as well as communication with other hardware and software in the system for identifying and processing early behavioral sequences of hypertension.

[0155] exist Figure 4 In the hypertension early behavioral sequence anomaly identification and data processing device shown, the processor 1001 is used to execute the hypertension early behavioral sequence anomaly identification and data processing program stored in the memory 1005 to implement the steps of the hypertension early behavioral sequence anomaly identification and data processing method described above.

[0156] The specific implementation of the device for identifying and processing abnormal early behavioral sequences of hypertension in this application is basically the same as the embodiments of the method for identifying and processing abnormal early behavioral sequences of hypertension described above, and will not be repeated here.

[0157] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0158] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0159] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0160] The above are merely preferred embodiments of this application and do not limit the scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the scope of protection of this application.

[0161] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0162] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for identifying and processing abnormal behavioral sequences in the early stages of hypertension, characterized in that, The method includes: Blood pressure-related monitoring data and behavioral status data of the target patient are obtained from a preset database. The blood pressure-related monitoring data includes monitoring sequence data in multiple dimensions. Based on the difference between the abnormal data segments in the blood pressure-related monitoring data and the preset normal value range, the abnormality degree value of each abnormal data segment is calculated. The calculation of the degree of abnormality for each abnormal data segment based on the difference between the abnormal data segments in the blood pressure-related monitoring data and the preset normal value range includes: Determine the first number of monitoring data in the abnormal data segment of the blood pressure-related monitoring data; Based on the minimum difference between the monitored data in the abnormal data segment and the preset normal value range, the first quantity, and the average difference between adjacent monitored data in the abnormal data segment, the abnormality degree value of each abnormal data segment is calculated. Based on the anomaly level value, the attention range of the preset multi-head attention mechanism is calculated; The calculation of the attention range of the preset multi-head attention mechanism based on the anomaly level value includes: For any abnormal data segment, select abnormal data segments whose time length difference with the current abnormal data segment is less than a preset similar time length, and use them as the neighboring abnormal data segments; Calculate the sum of the time length differences between the adjacent abnormal data segments and the current abnormal data segment, and the average difference of the abnormality values; Based on the sum of differences, the average of differences, and the number of adjacent abnormal data segments, the degree of intra-sequence deviation of each of the abnormal data segments corresponding to different monitoring sequence data is calculated; The abnormal data segments are sorted in ascending order of the degree of deviation within the sequence to obtain an ordered sequence; Determine the two adjacent data points corresponding to the maximum absolute value of the difference in the degree of deviation within the ordered sequence, and divide the abnormal data segments on both sides of these two adjacent data points in the ordered sequence into a first data segment and a second data segment; The data segment corresponding to the minimum mean of the degree of deviation within the sequence in the first data segment and the second data segment is selected as the final abnormal data segment; Based on the time length of the final abnormal data segment and the degree of intra-sequence deviation, the attention range of each monitoring sequence data corresponding to the preset multi-head attention mechanism is calculated. Based on the interaction between the behavioral state data and the blood pressure-related monitoring data, the attention weights of the monitoring sequence data corresponding to each dimension of the attention range are adjusted to obtain the final attention weights. The attention weights are adjusted based on the interaction between the behavioral state data and the blood pressure-related monitoring data to obtain the final attention weights, including: The behavioral state data is combined with the abnormal data segments in the blood pressure-related monitoring data to obtain multiple data pairs; Based on the occurrence frequency of the data pairs and the proportion of behavioral state data in the data pairs relative to the number of all abnormal data segments in each monitoring sequence data, the correlation strength of each data pair is calculated. Based on the correlation strength and the degree of anomalousness of the corresponding anomalous data segment, the adjustment coefficient of the attention weight is calculated. Based on the adjustment coefficient, the attention weights are adjusted to obtain the final attention weights; Based on the preset multi-head attention mechanism and the final attention weight, the blood pressure-related monitoring data is identified and processed to obtain the hypertension risk identification result.

2. The method for identifying and processing abnormal behavioral sequences in the early stages of hypertension as described in claim 1, characterized in that, Before calculating the degree of abnormality for each abnormal data segment based on the difference between the abnormal data segments in the blood pressure-related monitoring data and the preset normal value range, the method further includes: Identify multiple monitoring data points in the blood pressure-related monitoring data that are not within a preset normal value range; Any consecutive segment of a predetermined number of monitoring data points is treated as an abnormal data segment, generating multiple abnormal data segments in the blood pressure-related monitoring data.

3. The method for identifying and processing abnormal behavioral sequences in the early stages of hypertension as described in claim 1, characterized in that, The adjustment coefficient for the attention weight is calculated based on the correlation strength and the degree of anomalousness of the corresponding anomalous data segment, including: Determine the first slope of the fitted straight line for the degree of anomalousness of the data corresponding to the anomalous data segment; Based on the ratio between the mean and standard deviation of the association strength, and the first slope, the adjustment coefficient of the attention weight corresponding to each monitoring sequence data for the behavioral state data is calculated.

4. The method for identifying and processing abnormal behavioral sequences in the early stages of hypertension as described in claim 1, characterized in that, The step of adjusting the attention weights based on the adjustment coefficient to obtain the final attention weights includes: Multiply the adjustment coefficient by the attention weight to obtain the weight adjustment value; The final attention weight is obtained by adding the weight adjustment value to the attention weight.

Images