Intelligent prediction method and system for abnormal blood pressure during pregnancy

Abstract

The present application provides a method and system for intelligent prediction of abnormal blood pressure during pregnancy, which relates to the technical field of smart medical systems, wherein the method includes: constructing a continuous blood pressure monitoring sequence for a target pregnant woman based on real-time blood pressure sampling parameters and a historical blood pressure sampling parameter set; inputting the continuous blood pressure monitoring sequence into an equipment monitoring fault identification model to determine whether the blood pressure monitoring equipment has a transient sensitivity loss fault; when there is no equipment fault, querying the target user attribute data and target daily activity plan of the target pregnant woman; inputting the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan into a preset blood pressure prediction model to determine the predicted blood pressure parameter interval corresponding to the second preset time period. Thus, the comprehensive use of real-time blood pressure monitoring, deep learning prediction models and equipment fault identification mechanisms provides a comprehensive, accurate and personalized solution for blood pressure management during pregnancy.

Description

Technical Field

[0001] The present application relates to the technical field of smart medical systems, and in particular to a method and system for intelligently predicting abnormal blood pressure during pregnancy. Background Art

[0002] In the field of modern medical health monitoring, blood pressure management during pregnancy is a key link, which is directly related to the health of pregnant women and fetuses. During pregnancy, fluctuations in pregnant women's blood pressure may lead to a variety of complications, such as gestational hypertension, preeclampsia and eclampsia. These complications not only threaten the health of pregnant women, but may also affect the normal development of the fetus. Therefore, effective blood pressure monitoring and management during pregnancy is crucial.

[0003] Traditional methods of monitoring blood pressure during pregnancy mainly rely on regular medical examinations, such as home blood pressure monitoring or blood pressure measurements in medical institutions. However, on the one hand, traditional methods cannot continuously and real-time monitor the blood pressure of pregnant women. Temporary fluctuations or long-term trends in blood pressure may occur during non-measurement periods, resulting in the loss of key health status information, such as the possibility of missing important information such as sudden changes in blood pressure. On the other hand, traditional methods mostly focus on blood pressure problems that have already occurred, which limits the possibility of taking measures before abnormal blood pressure occurs, increasing the risk of pregnant women facing serious health problems.

[0004] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the invention

[0005] The embodiments of the present application provide a method, system, electronic device and storage medium for intelligently predicting abnormal blood pressure during pregnancy, which are used to at least solve the problem in the prior art that it is impossible to continuously monitor blood pressure in real time and to predict the risk of future abnormal blood pressure in advance.

[0006] In a first aspect, an embodiment of the present application provides an intelligent prediction method for abnormal blood pressure during pregnancy, comprising: when real-time blood pressure sampling parameters for a target pregnant woman received from a blood pressure monitoring device are detected, a historical blood pressure sampling parameter set is queried from a blood pressure database according to a user identifier of the target pregnant woman; based on the real-time blood pressure sampling parameters and the historical blood pressure sampling parameter set, a continuous blood pressure monitoring sequence for the target pregnant woman is constructed; the continuous blood pressure monitoring sequence is input into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a transient sensitivity loss fault; when it is determined that the blood pressure monitoring device does not have a transient sensitivity loss fault, the target user attribute data and target daily activity plan of the target pregnant woman are queried from a preset user database based on the user identifier of the target pregnant woman; the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan are input into a preset blood pressure prediction model to determine the predicted blood pressure parameter interval corresponding to the second preset time period; the blood pressure prediction model adopts a deep learning model.

[0007] Optionally, the equipment monitoring fault identification model comprises a typical variable analysis module, a vector autoregression module and an equipment abnormality identification module, and is used to determine whether there is a transient sensitivity loss fault of the equipment by performing the following operations:

[0008] Based on the continuous blood pressure monitoring sequence, a history vector matrix is ​​constructed:

[0009] a new,t =[a new (t),a new (t-1),…,a new (t-p+1)] T

[0010] In the formula, a new,t represents the history vector matrix, a new (t) represents the blood pressure sampling parameters at time point t;

[0011] The history vector matrix a new,t Projected into the typical variable analysis module to obtain the corresponding typical variable vector z new,t :

[0012] z new,t =J G a new,t

[0013] In the formula, J G represents a transfer matrix, which is determined by training the canonical variable analysis module based on a historical working sample data set of the blood pressure monitoring device;

[0014] Based on the vector autoregression module, the time series relationship matrix of the typical variable vector is extracted

[0015]

[0016] In the formula, z G,t-1 Represents the typical variable vector at time t-1; A1, A2, …, A d represents the model coefficient, β represents the constant term deviation, ε(t) represents the fitting error, and d represents the model order;

[0017] According to the difference between the historical vector matrix and the time series relationship matrix, the residual matrix e is determined new；t :

[0018]

[0019] According to the residual matrix e new；t , calculate the online monitoring statistical index S for time series correlation analysis new :

[0020]

[0021] In the formula, Represents the residual matrix e new,t The transpose of

[0022] The device abnormality identification module is used to perform the following operations:

[0023] If S new ≤S E , it is determined that there is no equipment transient sensitivity loss fault, and if S new >S E , it is determined that there is a transient sensitivity loss fault of the equipment, S E Indicates the preset threshold.

[0024] Optionally, after inputting the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan into a preset blood pressure prediction model to determine a predicted blood pressure parameter interval corresponding to a second preset time period in the future, the method further includes: comparing the predicted blood pressure parameter interval with the normal blood pressure range for pregnant women; and when it is determined that the predicted blood pressure parameter interval does not belong to a subset of the normal blood pressure range for pregnant women, performing an early warning operation on the daily activity plan.

[0025] Optionally, after inputting the continuous blood pressure monitoring sequence into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a device transient sensitivity loss fault, the method further includes: sending an equipment maintenance notification to the blood pressure monitoring device if it is determined that the blood pressure monitoring device has a device transient sensitivity loss fault.

[0026] Optionally, the blood pressure prediction model adopts a MIMO deep learning model, and the blood pressure prediction model includes a cascaded multi-input module, a feature fusion module and a multi-output module; the multi-input module includes a first input layer, a second input layer and a third input layer in parallel; the first input layer adopts an attention-based LSTM to extract the blood pressure time series features corresponding to the continuous blood pressure monitoring sequence:

[0027] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0028] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0029] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0030] C t =f t ×C t-1 +i t ×tanh(W C ·[h t-1 ,x t ]+b C )

[0031] h t =o t ×tanh(C t )

[0032] α t =softmax(W a ×h t +b a )

[0033]

[0034] In the formula, i t 、ft , o t , C t and h t Respectively represent the input gate, forget gate, output gate, unit state and initial output features of the LSTM unit at time step t; W i ,W f ,W o ,W C Represents the weight matrices of the input gate, forget gate, output gate, and cell state in the LSTM unit, and b i ,b f ,b o ,b C Respectively represent the bias term of the corresponding gate; σ represents the Sigmoid activation function; W a ,b a Represent the weight and bias of the attention layer, α t represents the attention weight, h' t Represents the weighted output feature; tanh() represents the hyperbolic tangent function.

[0035] Optionally, the second input layer adopts a fully connected layer to extract user attribute features corresponding to the target user attribute data:

[0036] y user =ReLU(W user ·x user +b user )

[0037] In the formula, x user Represents the input user attribute data, W user and b user Represent the weight matrix and bias term of the second input layer respectively; y user Represents the output user attribute features; ReLU() represents the RELU activation function;

[0038] The third input layer uses a fully connected layer to extract the schedule activity features corresponding to the target schedule activity plan:

[0039] x encoded =TimeEncode(x schedule )

[0040] y schedule =ReLU(W schedule *x encoded +b schedule )

[0041] In the formula, x schedule represents the input schedule activity plan; TimeEncode represents the time encoding function to encode time information into the feature; xencoded Represents the encoded schedule activity data; W schedule and b schedule Respectively represent the weight matrix bias items of the third input layer; y schedule Represents the output schedule activity characteristics;

[0042] Optionally, the feature fusion module includes a first feature cross layer and a second feature cross layer; the first feature cross layer includes a feature attention sublayer and a first nonlinear transformation sublayer; through the first feature cross layer, the blood pressure time series feature is fused with the user attribute feature to determine a corresponding first fusion feature:

[0043] γ=softmax(W att1 ·[h' t ;y user ]+b att1 )

[0044]

[0045] m1=ReLU(W relu1 ·m 1,att +b relu1 )

[0046] Where γ represents the feature attention weight; W att1 ,b att1 Represent the weight matrix and bias term of the feature attention sublayer respectively; m 1,att W represents the feature after feature attention weighting; relu1 ,b relu1 Respectively represent the weight and bias of the first nonlinear transformation sublayer, m1 represents the first fusion feature of the output, h' t Represents the weighted output features;

[0047] The second feature cross layer includes a second deep cross network sublayer and a second nonlinear transformation sublayer; the first fused feature is fused with the schedule activity feature through the second feature cross layer to determine a corresponding second fused feature:

[0048]

[0049] m2=ReLU(W relu2 ·m 2,cross +b relu2 )

[0050] In the formula, y schedule represents the feature vector of the schedule activity, W cross2 ,b cross2 represents the weight matrix and bias term of the second deep cross network sublayer; W relu2 ,brelu2 represents the weight matrix and bias term of the second nonlinear transformation sublayer; m2 represents the second fusion feature of the output;

[0051] The multi-output module comprises a first output layer and a second output layer;

[0052] The first output layer includes a plurality of nonlinear processing sublayers; through the first output layer, the upper limit value BP of the blood pressure in the predicted blood pressure parameter interval is determined according to the second fusion feature. upper :

[0053] BP upper =ReLU(W upper1 (ReLU(W upper2 ·m2+b upper2 ))+b upper1 )

[0054] Where W upper1 ,W upper2 and b upper1 ,b upper2 Respectively represent the weight matrix and bias term of the first nonlinear processing sublayer and the second nonlinear processing sublayer;

[0055] The second output layer is used to determine the lower limit of blood pressure BP in the predicted blood pressure parameter interval according to the second fusion feature. lower :

[0056] BP lower =Sigmoid(W lower ·m2+b lower )

[0057] Where W lower ,b lower Represents the weight matrix and bias term of the second output layer.

[0058] Optionally, the time encoding function is used to encode the schedule hour information hour in the schedule activity plan in 24-hour format to obtain the corresponding schedule activity feature:

[0059] x encoded_sin =sin(2π×hour / 24)

[0060] x encoded_cos =cos(2π×hour / 24)

[0061] x encoded =[x schedule ,x encoded_sin ,x encoded_cos ].

[0062] Optionally, the loss function Total Loss of the blood pressure prediction model is:

[0063] Total Loss=Loss+Penalty

[0064] In the formula, Loss represents the weighted combination loss, Penal ty represents the penalty term;

[0065] The weighted combination loss Loss is expressed as follows:

[0066] Loss = λ × Loss upper +(1-λ)×Loss lower

[0067]

[0068] In the formula, Loss upper and Loss lower They represent the loss terms for the upper and lower limits of blood pressure, respectively, and λ represents the weight factor used to balance different loss terms; and They represent the predicted values ​​of the upper and lower limits of blood pressure for the i-th sample respectively; and They represent the actual upper and lower limits of blood pressure of the i-th sample respectively; N represents the total number of samples in the training sample set;

[0069] The penalty term Penalty is expressed as follows:

[0070] Penalty=η×max(0,|BP pred -BP actual |-threshold) 2

[0071] In the formula, BP pred and BP actual They represent the predicted blood pressure value and the actual blood pressure value respectively; η represents the penalty coefficient, and threshold represents the preset blood pressure prediction tolerance error threshold.

[0072] In a second aspect, an embodiment of the present application provides an intelligent prediction system for abnormal blood pressure during pregnancy, including: a historical parameter query unit, configured to query a historical blood pressure sampling parameter set from a blood pressure database according to a user identifier of the target pregnant woman when real-time blood pressure sampling parameters for the target pregnant woman are detected from a blood pressure monitoring device; the historical blood pressure sampling parameter set includes a plurality of historical blood pressure sampling parameters of the target pregnant woman corresponding to a first preset time period in the past; a time series construction unit, configured to construct a continuous blood pressure monitoring sequence for the target pregnant woman based on the real-time blood pressure sampling parameters and the historical blood pressure sampling parameter set; an equipment fault identification unit, configured to input the continuous blood pressure monitoring sequence into an equipment monitoring fault identification model to determine whether the blood pressure monitoring device has a transient sensitivity loss fault; a schedule query unit, configured to, upon determining that the blood pressure monitoring device does not have a transient sensitivity loss fault, , based on the user identifier of the target pregnant woman, query the target user attribute data and target daily activity plan of the target pregnant woman from a preset user database; the target user attribute data includes the user's age, past medical history information and the number of weeks of pregnancy; the target daily activity plan includes the diet plan data and exercise activity plan data of the target pregnant woman in the second preset time period in the future; the blood pressure interval prediction unit is configured to input the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan into a preset blood pressure prediction model to determine the predicted blood pressure parameter interval corresponding to the second preset time period; the blood pressure prediction model adopts a deep learning model; the equipment monitoring fault identification model includes a typical variable analysis module, a vector autoregression module and an equipment abnormality identification module, and is used to determine whether there is a transient sensitivity loss fault of the equipment by performing the following operations: based on the continuous blood pressure monitoring sequence, construct a historical vector matrix:

[0073] a new,t =[a new (t),a new (t-1),…,a new (t-p+1)] T

[0074] In the formula, a new,t represents the history vector matrix, a new (t) represents the blood pressure sampling parameters at time point t;

[0075] The history vector matrix a new,t Projected into the typical variable analysis module to obtain the corresponding typical variable vector z new,t :

[0076] z new,t =J G a new,t

[0077] In the formula, J G represents a transfer matrix, which is determined by training the canonical variable analysis module based on a historical working sample data set of the blood pressure monitoring device;

[0078] Based on the vector autoregression module, the time series relationship matrix of the typical variable vector is extracted

[0079]

[0080] In the formula, z G (t-1) represents the typical variable vector at time t-1; A1, A2, …, A d represents the model coefficient, β represents the constant term deviation, ε(t) represents the fitting error, and d represents the model order;

[0081] According to the difference between the historical vector matrix and the time series relationship matrix, the residual matrix e is determined new；t :

[0082]

[0083] According to the residual matrix e new；t , calculate the online monitoring statistical index S for time series correlation analysis new :

[0084]

[0085] In the formula, Represents the residual matrix e new,t The transpose of

[0086] The device abnormality identification module is used to perform the following operations:

[0087] If S new ≤S E , it is determined that there is no equipment transient sensitivity loss fault, and

[0088] If S new >S E , it is determined that there is a transient sensitivity loss fault of the equipment, S E Indicates the preset threshold.

[0089] In a third aspect, an embodiment of the present application provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the above method.

[0090] In a fourth aspect, an embodiment of the present application provides a storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of the above-mentioned method of the present application.

[0091] In a fifth aspect, an embodiment of the present application further provides a computer program product, wherein the computer program product comprises a computer program stored on a storage medium, wherein the computer program comprises program instructions, and when the program instructions are executed by a computer, the computer executes the steps of the above method.

[0092] The intelligent prediction method, system, electronic device and non-transitory computer-readable storage medium for abnormal blood pressure during pregnancy provided by the present application can produce at least the following technical effects:

[0093] (1) By collecting the real-time blood pressure sampling parameters of the target pregnant women and combining them with their historical blood pressure data, a continuous blood pressure monitoring sequence can be constructed to achieve real-time and continuous monitoring of the pregnant women's blood pressure. This not only provides more comprehensive blood pressure data than traditional regular tests, but also captures blood pressure fluctuations that may occur between regular checks, reducing the omission of key health information.

[0094] (2) The deep learning model can be used to combine continuous blood pressure monitoring data, user attribute data, and daily activity plans to more accurately predict blood pressure fluctuations in the future. It can learn patterns and relationships from complex data and improve the accuracy and reliability of blood pressure abnormality prediction.

[0095] (3) Through the equipment monitoring fault identification model, the typical variable analysis module, the vector autoregression module and the equipment abnormality identification module are integrated to effectively identify the transient sensitivity loss failure of the blood pressure monitoring equipment. This ensures the accuracy and reliability of the monitoring data and reduces the risk of misdiagnosis caused by equipment failure.

[0096] (4) By combining personal attribute data (such as age, medical history, and gestational age) with daily activity plans (including diet and exercise plans), more personalized blood pressure management advice can be provided to pregnant women, which is more helpful in formulating effective personalized blood pressure control plans.

[0097] (5) By predicting the future range of blood pressure parameters, it helps to detect potential blood pressure problems early, allowing doctors and pregnant women to take preventive measures, such as avoiding certain daily activities, thereby reducing the risk of complications such as gestational hypertension and preeclampsia, and significantly improving the health and safety of pregnant women and fetuses.

[0098] Through the embodiments of the present application, the comprehensive use of real-time blood pressure monitoring, deep learning prediction models and equipment fault identification mechanisms can provide a comprehensive, accurate and personalized solution for blood pressure management during pregnancy, providing pregnant women with more comprehensive and reliable health management and monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0100] Figure 1 A flowchart showing an example of an intelligent prediction method for abnormal blood pressure during pregnancy according to an embodiment of the present application is shown;

[0101] Figure 2 A schematic diagram showing a structure of an example of a device monitoring fault identification model according to an embodiment of the present application is shown;

[0102] Figure 3 A flowchart showing an example of an intelligent prediction method for abnormal blood pressure during pregnancy according to an embodiment of the present application is shown;

[0103] Figure 4 A schematic structural diagram of an example of a blood pressure prediction model according to an embodiment of the present application is shown;

[0104] Figure 5 A structural block diagram of an example of an intelligent prediction system for abnormal blood pressure during pregnancy according to an embodiment of the present application is shown;

[0105] Figure 6 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0106] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be clearly and completely described below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0107] Unless otherwise defined, the technical terms or scientific terms used in this application should be understood by people with ordinary skills in the field to which this application belongs. "First", "second" and similar words used in this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, similar words such as "one", "one" or "the" do not indicate quantitative restrictions, but indicate the existence of at least one. "Include" or "comprises" and other similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and other similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0108] It should be noted that the terms "up", "down", "left", "right", "front", "back", etc. used in this application are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0109] Figure 1 A flowchart showing an example of a method for intelligently predicting abnormal blood pressure during pregnancy according to an embodiment of the present application is shown.

[0110] Regarding the execution subject of the method of the embodiment of the present application, it can be any controller or processor with computing or processing capabilities, so as to realize the deep learning technology and personalized data analysis, greatly improve the accuracy and timeliness of prediction, and provide a more comprehensive blood pressure health management solution for pregnant women. In some examples, it can be integrated and configured in the client or server through software, hardware, or a combination of software and hardware, which should not be limited here.

[0111] The following will take the pregnancy smart service platform as an exemplary implementation subject to introduce the details of the technical solution involved in this application. However, it should be understood that one or more steps involved in the following process can be implemented by one or more controllers or software installed and deployed in the client or server.

[0112] like Figure 1 As shown, in step S110, when the real-time blood pressure sampling parameters for the target pregnant woman received from the blood pressure monitoring device are detected, a historical blood pressure sampling parameter set is queried from the blood pressure database according to the user identifier of the target pregnant woman.

[0113] In some embodiments, the blood pressure monitoring device may be a smart wearable device, such as a smart bracelet, a smart chest strap, and a health monitoring patch, etc. For example, with the user's authorization, the smart wearable device can collect various physiological activity data of the user in real time, and upload the various physiological activity data to the pregnancy smart service platform in real time, so as to realize real-time blood pressure monitoring and intelligently predict future abnormal blood pressure during pregnancy.

[0114] In addition, the historical blood pressure sampling parameter set includes multiple historical blood pressure sampling parameters of the target pregnant woman corresponding to the first preset time period in the past. Here, the length of the first preset time period can be non-restrictive and can be set according to business or accuracy requirements. For example, if system resources permit, it can be set to a longer period, such as 1 month, so that the subsequent model can fully learn the blood pressure change pattern of the pregnant mother in daily life.

[0115] In step S120, a continuous blood pressure monitoring sequence for the target pregnant woman is constructed based on the real-time blood pressure sampling parameters and the historical blood pressure sampling parameter set.

[0116] Exemplarily, abnormal mutation data in the historical blood pressure sampling parameter set is screened out and standardized to obtain a continuous blood pressure monitoring sequence.

[0117] In step S130 , the continuous blood pressure monitoring sequence is input into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a device transient sensitivity loss fault.

[0118] It should be noted that transient loss of sensitivity failure is a common type of failure in sensor equipment and is also a type of failure that is difficult to distinguish. Generally, it is caused by an increase in pressure on the sensor. The occurrence of this failure is highly random and uncertain, and it will return to normal when the user resets the device.

[0119] In step S140, when it is determined that the blood pressure monitoring device does not have a transient sensitivity loss failure, the target user attribute data and target daily activity plan of the target pregnant woman are queried from a preset user database based on the user identification of the target pregnant woman.

[0120] Here, the target user attribute data includes the user's age, medical history information, and number of weeks of pregnancy, and the target daily activity plan includes the target pregnant woman's diet plan data and exercise activity plan data for the second preset time period in the future. User attribute data can better describe the character portraits of different pregnant women, which helps to assist in guiding the system's prediction direction. For example, when a pregnant mother has a history of hypertension, she may pay more attention to the risk monitoring of extremely high blood pressure values, and when a pregnant mother has a history of malnutrition, she may pay more attention to the risk monitoring of extremely low blood pressure values. Through the daily activity plan, the system is able to learn and predict the impact of different diet plans or exercise plans on blood pressure changes.

[0121] In step S150, the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan are input into a preset blood pressure prediction model to determine a predicted blood pressure parameter interval corresponding to a second preset time period.

[0122] Here, the blood pressure prediction model can adopt various types of deep learning models, such as CNN or RNN, which should not be limited here. In some preferred embodiments, the blood pressure prediction model can adopt a MIMO (Multiple-Input Multiple-Output) deep learning model, and more details are described below in conjunction with other examples. In addition, the length of the second preset time period can be diverse, but should preferably be lower than the first preset time period. For example, when the time unit of the first preset time period is a month or a week, the time unit of the second preset time period can be a day.

[0123] Figure 2 A structural schematic diagram of an example of an equipment monitoring fault identification model according to an embodiment of the present application is shown.

[0124] like Figure 2 As shown, the equipment monitoring fault identification model 200 includes a canonical variate analysis (CVA) module 210, a vector autoregressive model (VAR) module 220 and an equipment abnormality identification module 230. Through the coordination of signal data of the above-mentioned cascaded modules, the identification and monitoring of equipment transient sensitivity loss faults are achieved.

[0125] Therefore, the reliability of blood pressure measurement data is ensured by continuously monitoring the performance of the device. In the management of blood pressure during pregnancy, the accuracy of the data is crucial. Based on this monitoring, data deviations caused by equipment failure can be discovered and corrected in a timely manner, ensuring that medical decisions are based on reliable data.

[0126] More specifically, based on the continuous blood pressure monitoring sequence, a history vector matrix is ​​constructed:

[0127] a new,t =[a new (t),a new (t-1),…,a new (t-p+1)] T Formula (1)

[0128] In the formula, a new,t represents the history vector matrix, a new (t) represents the blood pressure sampling parameters at time point t.

[0129] The history vector matrix a new,t Projected into the canonical variable analysis module 210 to obtain the corresponding canonical variable vector z new,t :

[0130] z new,t =J G a new,t Formula (2)

[0131] In the formula, J G represents the transfer matrix, which is determined by training the canonical variable analysis module based on the historical working sample data set of the blood pressure monitoring equipment.

[0132] Based on the vector autoregression module 220, the time series relationship matrix of the typical variable vector is extracted

[0133] In the formula, z G,t-1 Represents the typical variable vector at time t-1; A1, A2, …, A d represents the model coefficient, β represents the constant term deviation, ε(t) represents the fitting error, and d represents the model order.

[0134] According to the difference between the historical vector matrix and the time series relationship matrix, the residual matrix e is determined new；t :

[0135]

[0136] According to the residual matrix e new；t , calculate the online monitoring statistical index S for time series correlation analysis new :

[0137]

[0138] In the formula, Represents the residual matrix e new,t The transpose of .

[0139] The device abnormality identification module 230 is used to perform the following operations:

[0140] If S new ≤S E , it is determined that there is no equipment transient sensitivity loss fault, and

[0141] If S new >S E , it is determined that there is a transient sensitivity loss fault of the equipment, S E Indicates the preset threshold.

[0142] Through the embodiments of the present application, combined with the design of the CVA module and the VAR module, real-time monitoring and analysis of the functional status of the equipment can be achieved, and the nature and possible causes of the failure can also be analyzed. When the performance of the equipment is detected to be degraded or abnormal, the system can immediately issue an early warning to reduce the risk of monitoring interruption or erroneous data caused by equipment failure.

[0143] It should be noted that the design of this fusion CVA module and VAR module enables the model to self-calibrate based on a continuous data stream, ensuring long-term monitoring accuracy and helping to maintain the long-term stability and reliability of the equipment.

[0144] Furthermore, the deployment details for the CVA module and the VAR module will be introduced below.

[0145] In the vector autoregression model (VAR), determine the model parameters A1, A2, ..., A d , the constant term deviation β, and the model order d usually follow the following steps:

[0146] First, in the process of determining the model order, the optimal order is selected through information criteria (such as Akaike Information Criterion AIC, Bayesian Information Criterion BIC) to balance the goodness of fit and complexity of the model (higher order models may lead to overfitting). It should be noted that the use of these criteria requires estimating multiple VAR models of different orders, and then selecting the model that minimizes the selected information criterion.

[0147] Then, once the model order is determined, the model coefficients A1, A2, ..., A need to be estimated. d and a constant term bias β. This is usually done by minimizing the sum of squares of the prediction errors, a so-called least squares estimation. More specifically, it can be done with statistical packages such as the "vars" package or Python's "statsmodels" library, which provide functions and tools to automatically estimate these coefficients.

[0148] Then, the fitting error ε(t) is generated during the model fitting process, which represents the difference between the actual observed value and the model predicted value at a specific time point. In the VAR model, it is usually assumed that the error term is white noise, that is, the error terms are uncorrelated and have a constant variance.

[0149] Finally, residual tests and other statistical tests can be used to assess the suitability of the model. For example, check whether the residuals are close to white noise and whether the model is stable (by checking the characteristic roots). In some cases, model optimization may be required, such as using techniques such as rolling forecasts to improve the forecast accuracy of the VAR model.

[0150] The deployment of the CVA module is done by training it using a dataset of historical working samples of blood pressure monitoring devices.

[0151] More specifically, first collect and prepare two sets of data: one set is the data of the blood pressure monitoring device in normal operation, and the other set is the data of the device in a possible fault state. These data should include blood pressure and related variables that may affect fault identification (such as environmental factors, equipment performance indicators, etc.). Clean the data, handle missing values ​​and outliers, and ensure data quality.

[0152] Then, for each set of data, calculate the mean and covariance matrix of each variable. This will be used for subsequent statistical analysis.

[0153] Next, the within-class scatter matrix and the between-class scatter matrix are calculated. In the within-class scatter matrix (Within-ClassScatter Matrix), the scatter within each group of data is mainly calculated. In the between-class scatter matrix (Between-ClassScatterMatrix), the scatter between classes is mainly calculated based on the mean vectors of different groups of data.

[0154] Afterwards, the within-class scatter matrix and the between-class scatter matrix are used to solve the generalized eigenvalue problem to find the eigenvector that maximizes the ratio of between-class scatter to within-class scatter.

[0155] Finally, the most important eigenvectors are selected to construct the transfer matrix J G These eigenvectors define how the original data is projected into a new feature space that is the one that best distinguishes between different groups of data.

[0156] It should be noted that when applying CVA, the key is to ensure that there is sufficient and representative data to support the training and validation of the model. Especially in the scenario of blood pressure (medical data) monitoring failure, accuracy and reliability are crucial. Through precise training and rigorous validation, the CVA model can be effectively used to identify whether the device has a transient loss of sensitivity failure.

[0157] Figure 3 A flowchart showing an example of a method for intelligently predicting abnormal blood pressure during pregnancy according to an embodiment of the present application is shown.

[0158] like Figure 3As shown, in step S310, when the real-time blood pressure sampling parameters for the target pregnant woman received from the blood pressure monitoring device are detected, a historical blood pressure sampling parameter set is queried from the blood pressure database according to the user identifier of the target pregnant woman.

[0159] In step S320, a continuous blood pressure monitoring sequence for the target pregnant woman is constructed based on the real-time blood pressure sampling parameters and the historical blood pressure sampling parameter set.

[0160] In step S330, the continuous blood pressure monitoring sequence is input into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a device transient sensitivity loss fault.

[0161] In step S341, when it is determined that the blood pressure monitoring device does not have a transient sensitivity loss failure, the target user attribute data and target daily activity plan of the target pregnant woman are queried from a preset user database based on the user identification of the target pregnant woman.

[0162] In step S343, when it is determined that the blood pressure monitoring device has a transient sensitivity loss failure, a device maintenance notification is sent to the blood pressure monitoring device.

[0163] At this time, the pregnancy smart service platform uses big data AI analysis to determine that the blood pressure monitoring device has a fault, and sends a device maintenance notification to the blood pressure detection device to remind the pregnant mother to restart or repair the device to ensure the continuous reliability of the blood pressure monitoring process. In addition, the real-time blood pressure sampling parameters corresponding to the transient sensitivity loss failure of the device will not be retained in the database, and will not trigger the subsequent system analysis process, saving system resource consumption on the platform.

[0164] In step S350, the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan are input into a preset blood pressure prediction model to determine a predicted blood pressure parameter interval corresponding to a second preset time period.

[0165] In step S360, the predicted blood pressure parameter interval is compared with the normal blood pressure range of pregnant women.

[0166] Generally speaking, the normal blood pressure range for pregnant women is: 60-139 mmHg. Specifically, the systolic pressure is 90~139 mmHg and the diastolic pressure is 60~89 mmHg.

[0167] In step S370, when it is determined that the predicted blood pressure parameter interval does not belong to a subset of the normal blood pressure range for pregnant women, an early warning operation is performed on the daily activity plan.

[0168] Through the embodiments of the present application, when it is predicted that the blood pressure parameter interval exceeds the normal blood pressure range of pregnant women, the pregnancy smart service platform promptly performs early warning operations on the target pregnant woman's daily activity plan, such as pushing messages or interacting with inquiries through AI robots, to prompt that the corresponding food plan or exercise plan will have potential risks of hypertension or hypotension.

[0169] Figure 4 A structural schematic diagram of an example of a blood pressure prediction model according to an embodiment of the present application is shown.

[0170] like Figure 4 As shown, the blood pressure prediction model 400 adopts a MIMO deep learning model, which includes a cascaded multi-input module 410, a feature fusion module 420 and a multi-output module 430.

[0171] The multi-input module 410 includes a first input layer 411 , a second input layer 413 , and a third input layer 415 connected in parallel.

[0172] The first input layer 411 uses an attention-based LSTM to extract the blood pressure time series features corresponding to the continuous blood pressure monitoring sequence:

[0173] i t =σ(W i ·[h t-1 ,x t ]+b i ) Formula (6)

[0174] f t =σ(W f ·[h t-1 ,x t ]+b f ) Formula (7)

[0175] o t =σ(W o ·[h t-1 ,x t ]+b o ) Formula (8)

[0176] C t =f t ×C t-1 +i t ×tanh(W C ·[h t-1 ,x t ]+b C ) Formula (9)

[0177] h t =o t ×tanh(C t ) Formula (10)

[0178] α t =softmax(W a ×h t +b a ) Formula (11)

[0179] h' t =∑ t (α t ×h t ) Formula (12)

[0180] In the formula, i t 、f t , o t , C t and h t Respectively represent the input gate, forget gate, output gate, unit state and initial output features of the LSTM unit at time step t; W i ,W f ,W o ,W C Represents the weight matrices of the input gate, forget gate, output gate, and cell state in the LSTM unit, and b i ,b f ,b o ,b C Respectively represent the bias term of the corresponding gate; σ represents the Sigmoid activation function; W a ,b a Represent the weight and bias of the attention layer, α t represents the attention weight, h' t Represents the weighted output features.

[0181] Therefore, in the first input layer 411, by using an LSTM network with an attention mechanism to process time series data, for example, using an LSTM layer containing 128 or more neurons, the time dependency in the blood pressure data can be fully captured, especially the most important part of the blood pressure time series data, thereby improving the sensitivity to the blood pressure change trend and the prediction accuracy.

[0182] The second input layer 413 uses a fully connected layer to extract user attribute features corresponding to the target user attribute data:

[0183] y user =ReLU(W user ·x user +b user ) Formula (13)

[0184] In the formula, x user Represents the input user attribute data, W user and b userRepresent the weight matrix and bias term of the second input layer respectively; y user Represents the output user attribute features.

[0185] It should be noted that user attribute data is usually static and structured, such as age, gestational age, etc. This type of data does not involve time series and may contain categorical properties (such as gender, past medical history). Therefore, the second input layer 413 can use one or more fully connected layers to process these attribute data. Since this type of data may not require complex feature extraction, a smaller number of neurons can be considered to reduce model complexity.

[0186] The third input layer 415 uses a fully connected layer to extract the schedule activity features corresponding to the target schedule activity plan:

[0187] x encoded =TimeEncode(x schedule ) Formula (14)

[0188] y schedule =ReLU(W schedule ·x encoded +b schedule ) Formula (15)

[0189] In the formula, x schedule represents the input schedule activity plan; TimeEncode represents the time encoding function to encode time information into the feature; x encoded Represents the encoded schedule activity data; W schedule and b schedule Respectively represent the weight matrix bias items of the third input layer; y schedule Represents the output calendar activity characteristics.

[0190] It should be noted that the time-related activity data in the target schedule activity plan can effectively improve the efficiency of data processing by introducing time-coded time series processing methods, and can also help the model better understand and utilize schedule activity data to make accurate blood pressure predictions.

[0191] Here, the time encoding function is a method of converting time information into useful features. It is particularly suitable for processing time-related data, such as daily activity plans, and it is more convenient for the model to understand complex activity information at different times. In deep learning, time encoding helps the model understand and capture the impact of time variables on target variables (such as blood pressure).

[0192] In some examples of the embodiments of the present application, the time encoding function is used to encode the schedule hour information "hour" in the schedule activity plan in 24-hour format to obtain the corresponding schedule activity features:

[0193] x encoded_sin =sin(2π×hour / 24) Formula (16)

[0194] x encoded_cos =cos(2π×hour / 24) Formula (17)

[0195] x encoded =[x schedule ,x encoded_sin ,x encoded_cos ] Formula (18)

[0196] In the formula, x encoded_sin The sine function encoding value representing the schedule hour information, x encoded_cos The cosine function encoding value representing the schedule hour information.

[0197] In this way, time-encoded data can be directly input into the neural network, which helps the neural network capture and understand the periodicity and trends of time, allowing the model to more effectively process daily activity data containing time information, thereby improving the accuracy of predicting blood pressure changes during pregnancy.

[0198] Through the first input layer, the second input layer and the third input layer designed for different structures of the input module, a differentiated input structure design is achieved, ensuring that the model can effectively extract features according to different types of data characteristics.

[0199] The feature fusion module 420 includes a first feature cross layer 421 and a second feature cross layer 423 .

[0200] The first feature cross layer 421 includes a feature attention sublayer (not shown) and a first nonlinear transformation sublayer (not shown). Through the first feature cross layer 421, the blood pressure time series feature is fused with the user attribute feature to determine the corresponding first fusion feature:

[0201] γ=softmax(W att1 ·[h' t ;y user ]+b att1 ) Formula (19)

[0202]

[0203] m1=ReLU(W relu1 ·m 1,att +b relu1 ) Formula (21)

[0204] Where γ represents the feature attention weight; W att1 ,b att1Represent the weight matrix and bias term of the feature attention sublayer respectively; m 1,att W represents the feature after feature attention weighting; relu1 ,b relu1 They represent the weight and bias of the first nonlinear transformation sublayer respectively, and m1 represents the first fusion feature of the output.

[0205] Therefore, in the first feature cross-layer 411, automatic feature selection is achieved through the attention mechanism to highlight the most important feature combinations. The model can focus more on those feature combinations that contribute most to the prediction, which helps to enhance the model's ability to process complex data, thereby improving the accuracy of the prediction.

[0206] The second feature cross layer 423 includes a second deep cross network sublayer (not shown) and a second nonlinear transformation sublayer (not shown). The first fusion feature is fused with the schedule activity feature through the second feature cross layer 423 to determine the corresponding second fusion feature:

[0207]

[0208] m2=ReLU(W relu2 ·m 2,cross +b relu2 ) Formula (23)

[0209] In the formula, y schedule represents the feature vector of the schedule activity, W cross2 ,b cross2 represents the weight matrix and bias term of the second deep cross network sublayer; W relu2 ,b relu2 represents the weight matrix and bias term of the second nonlinear transformation sublayer; m2 represents the second fusion feature of the output.

[0210] Therefore, by introducing a deep cross network and a nonlinear transformation layer in the second feature cross layer 423, the model can more effectively learn and characterize the complex relationship between different data sources, enhance the effect of feature fusion, and thus improve the accuracy and robustness of the prediction.

[0211] The multi-output module 430 includes a first output layer 431 and a second output layer 433 .

[0212] The first output layer 431 includes a plurality of nonlinear processing sublayers (not shown). The first output layer 431 is used to determine the upper limit of the blood pressure in the predicted blood pressure parameter interval according to the second fusion feature. upper :

[0213] BP upper =ReLU(W upper1 (ReLU(W upper2 ·m2+bupper2 ))+b upper1 ) Formula (24)

[0214] Where W upper1 ,W upper2 and b upper1 ,b upper2 Represent the weight matrix and bias term of the first nonlinear processing sublayer and the second nonlinear processing sublayer respectively.

[0215] It should be noted that the risk probability of high blood pressure is much greater than the risk of low blood pressure. Therefore, during the model working process, we should pay special attention to the prediction and monitoring of the upper limit of blood pressure, and introduce more complex structures in the first output layer, such as additional nonlinear processing layers or more layers of network structures.

[0216] The second output layer 433 is used to determine the lower limit of blood pressure BP in the predicted blood pressure parameter interval according to the second fusion feature. lower :

[0217] BP lower =Sigmoid(W lower ·m2+b lower ) Formula (25)

[0218] Where W lower ,b lower Represents the weight matrix and bias term of the second output layer.

[0219] Here, for the prediction of the lower limit of blood pressure, the structure can be relatively simplified to simplify the structural design of the model.

[0220] The MIMO deep learning model provided by the embodiment of the present application adopts a multi-input module, so that the model can integrate multiple data sources (such as real-time blood pressure data, user personal health records and daily activity information), providing a more comprehensive data basis for prediction. When processing blood pressure time series data, the attention-based LSTM can dynamically focus on the most critical time points, improve the ability to capture sudden changes or trend changes in blood pressure, and predict potential blood pressure problems at an early stage. The independent processing of user personal attributes and daily activity plans ensures that the model can extract the most suitable features for different types of data, thereby accurately capturing the impact of individual differences on blood pressure. In the feature fusion module, through advanced feature crossover and attention mechanisms, not only the ability to extract key information from complex data is improved, but also the deep interaction and fusion between different data sources is ensured, so that the model can better capture the complex relationship between blood pressure and various factors. In the output module, through the output layer of the differential design, the model can perform specialized predictions on the upper and lower limits of blood pressure respectively, especially, for the key prediction of the upper limit of blood pressure, the model improves the recognition ability of dangerous blood pressure levels through a complex network structure, and realizes more targeted early detection and intervention of hypertension during pregnancy.

[0221] It should be noted that in the intelligent prediction model of abnormal blood pressure during pregnancy, the design of the loss function is a key difficult project, because it not only needs to accurately reflect the prediction error, but also should take into account the particularity and clinical importance of blood pressure prediction.

[0222] In view of this, in some examples of the embodiments of the present application, the loss function T of the blood pressure prediction model is ota lLoss is:

[0223] Total Loss=Loss+Penalty Formula (26)

[0224] In the formula, Loss represents the weighted combination loss, and Penalty represents the penalty term. In this way, not only the importance difference between the upper and lower limit predictions of blood pressure is taken into account, but also a penalty mechanism is introduced to ensure the prediction accuracy which is clinically critical.

[0225] The weighted combination loss Loss is expressed as follows:

[0226] Loss = λ × Loss upper +(1-λ)×Loss lower Formula (27)

[0227]

[0228] In the formula, Loss upper and Loss lowerThey represent the loss terms for the upper and lower limits of blood pressure, respectively, and λ represents the weight factor used to balance different loss terms; and They represent the predicted values ​​of the upper and lower limits of blood pressure for the i-th sample respectively; and They represent the actual upper and lower limits of blood pressure of the ith sample respectively; N represents the total number of samples in the training sample set.

[0229] Here, since blood pressure prediction involves two different objectives (the upper and lower limits of blood pressure), a weighted combination loss function can be designed, which combines the losses of the two different outputs and assigns a weight λ according to their clinical importance. For example, if the accurate prediction of the upper limit of blood pressure is more clinically important, λ can be set to a value greater than 0.5.

[0230] The penalty term Penalty is expressed as follows:

[0231] Penalty=η×max(0,|BP pred -BP actual |-threshold) 2 Formula (30)

[0232] In the formula, BP pred and BP actual Represent the predicted blood pressure value and the actual blood pressure value respectively; η represents the penalty coefficient, and threshold represents the preset blood pressure prediction tolerance error threshold. In this way, when the prediction error exceeds the threshold, the loss function will add additional penalties, prompting the model to pay more attention to reducing larger prediction errors.

[0233] Therefore, by adding a penalty term to the loss function, the model is punished more severely when the prediction error is large, which encourages the model to focus on reducing the occurrence of large errors. Because large blood pressure deviations may lead to serious health risks, it is necessary to ensure the reliability of the prediction and management of abnormal blood pressure in pregnant women.

[0234] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of actions combined, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0235] like Figure 5The figure shows a schematic diagram of the structure of an intelligent prediction system for abnormal blood pressure during pregnancy provided in one embodiment of the present application. The system can execute the intelligent prediction method for abnormal blood pressure during pregnancy described in any of the above embodiments and be configured in a terminal.

[0236] The present embodiment provides an intelligent prediction system 500 for abnormal blood pressure during pregnancy, which includes a historical parameter query unit 510 , a time series construction unit 520 , an equipment fault identification unit 530 , a schedule query unit 540 , and a blood pressure interval prediction unit 550 .

[0237] The historical parameter query unit 510 is configured to query a historical blood pressure sampling parameter set from the blood pressure database according to the user identifier of the target pregnant woman when the real-time blood pressure sampling parameters for the target pregnant woman are detected from the blood pressure monitoring device; the historical blood pressure sampling parameter set includes multiple historical blood pressure sampling parameters of the target pregnant woman corresponding to a first preset time period in the past.

[0238] The time series construction unit 520 is configured to construct a continuous blood pressure monitoring sequence for the target pregnant woman based on the real-time blood pressure sampling parameters and the historical blood pressure sampling parameter set.

[0239] The device fault identification unit 530 is configured to input the continuous blood pressure monitoring sequence into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a device transient sensitivity loss fault.

[0240] The schedule query unit 540 is configured to query the target user attribute data and target schedule activity plan of the target pregnant woman from a preset user database based on the user identification of the target pregnant woman when it is determined that the blood pressure monitoring device does not have a transient sensitivity loss failure of the device; the target user attribute data includes the user's age, past medical history information and number of weeks of pregnancy; the target schedule activity plan includes the diet plan data and exercise activity plan data of the target pregnant woman in a second preset time period in the future.

[0241] The blood pressure interval prediction unit 550 is configured to input the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan into a preset blood pressure prediction model to determine the predicted blood pressure parameter interval corresponding to the second preset time period; the blood pressure prediction model adopts a deep learning model.

[0242] The equipment monitoring fault identification model comprises a typical variable analysis module, a vector autoregression module and an equipment abnormality identification module, and is used to determine whether there is a transient sensitivity loss fault of the equipment by performing the following operations:

[0243] Based on the continuous blood pressure monitoring sequence, a history vector matrix is ​​constructed:

[0244] a new,t =[a new (t),a new (t-1),…,a new (t-p+1)] T

[0245] In the formula, a new,t represents the history vector matrix, a new (t) represents the blood pressure sampling parameters at time point t;

[0246] The history vector matrix a new,t Projected into the typical variable analysis module to obtain the corresponding typical variable vector z new,t :

[0247] z new,t =J G a new,t

[0248] In the formula, J G represents a transfer matrix, which is determined by training the canonical variable analysis module based on a historical working sample data set of the blood pressure monitoring device;

[0249] Based on the vector autoregression module, the time series relationship matrix of the typical variable vector is extracted

[0250]

[0251] In the formula, z G (t-1) represents the typical variable vector at time t-1; A1, A2, …, A d represents the model coefficient, β represents the constant term deviation, ε(t) represents the fitting error, and d represents the model order;

[0252] According to the difference between the historical vector matrix and the time series relationship matrix, the residual matrix e is determined new；t :

[0253]

[0254] According to the residual matrix e new；t , calculate the online monitoring statistical index S for time series correlation analysis new :

[0255]

[0256] In the formula, Represents the residual matrix e new,t The transpose of

[0257] The device abnormality identification module is used to perform the following operations:

[0258] If S new ≤S E , it is determined that there is no equipment transient sensitivity loss fault, and

[0259] If S new >S E , it is determined that there is a transient sensitivity loss fault of the equipment, S E Indicates the preset threshold.

[0260] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the above-mentioned intelligent prediction method for abnormal blood pressure during pregnancy in the present application.

[0261] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the above-mentioned intelligent prediction method for abnormal blood pressure during pregnancy.

[0262] In some embodiments, the embodiments of the present application also provide an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute an intelligent prediction method for abnormal blood pressure during pregnancy.

[0263] Figure 6 FIG. 1 is a schematic diagram of the hardware structure of an electronic device for executing an intelligent prediction method for abnormal blood pressure during pregnancy provided by another embodiment of the present application. Figure 6 As shown, the device includes:

[0264] One or more processors 610 and memory 620, Figure 6 A processor 610 is taken as an example.

[0265] The device for executing the intelligent prediction method for abnormal blood pressure during pregnancy may further include: an input device 630 and an output device 640 .

[0266] The processor 610, the memory 620, the input device 630 and the output device 640 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0267] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the method for intelligent prediction of abnormal blood pressure during pregnancy in the embodiment of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 620, that is, the method for intelligent prediction of abnormal blood pressure during pregnancy in the above method embodiment is implemented.

[0268] The memory 620 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 620 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include a memory remotely arranged relative to the processor 610, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0269] The input device 630 can receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 640 can include a display device such as a display screen.

[0270] The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, the intelligent prediction method for abnormal blood pressure during pregnancy in any of the above method embodiments is executed.

[0271] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0272] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0273] (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0274] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, etc.

[0275] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0276] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

[0277] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0278] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0279] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent prediction method for abnormal blood pressure during pregnancy, characterized in that: include: When the real-time blood pressure sampling parameters for the target pregnant woman received from the blood pressure monitoring device are detected, querying a historical blood pressure sampling parameter set from a blood pressure database according to a user identifier of the target pregnant woman; constructing a continuous blood pressure monitoring sequence for the target pregnant woman based on the real-time blood pressure sampling parameters and the historical blood pressure sampling parameter set; Inputting the continuous blood pressure monitoring sequence into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a device transient sensitivity loss fault; In the case where it is determined that the blood pressure monitoring device does not have a transient sensitivity loss failure of the device, querying the target user attribute data and the target daily activity plan of the target pregnant woman from a preset user database based on the user identification of the target pregnant woman; The target user attribute data includes the user's age, medical history information, and the number of weeks of pregnancy, and the target schedule activity plan includes the target pregnant woman's diet plan data and exercise activity plan data in the second preset time period in the future; Inputting the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan into a preset blood pressure prediction model to determine a predicted blood pressure parameter interval corresponding to the second preset time period; the blood pressure prediction model adopts a deep learning model; The equipment monitoring fault identification model includes a typical variable analysis module, a vector autoregression module and an equipment abnormality identification module, and is used to determine whether there is a transient sensitivity loss fault of the equipment by performing the following operations: Based on the continuous blood pressure monitoring sequence, a history vector matrix is ​​constructed: a new,t =[a new (t),a new (t-1),…,a new (t-p+1)] T In the formula, a new,t represents the history vector matrix, a new (t) represents the blood pressure sampling parameters at time point t; The history vector matrix a new,t Projected into the canonical variable analysis module to obtain the corresponding canonical variable vector z new,t : from new,t =J G and new,t In the formula, J G represents a transfer matrix, which is determined by training the canonical variable analysis module based on a historical working sample data set of the blood pressure monitoring device; Based on the vector autoregression module, the time series relationship matrix of the typical variable vector is extracted In the formula, z G,t-1 Represents the typical variable vector at time t-1; A1, A2, …, A d represents the model coefficient, β represents the constant term deviation, ε(t) represents the fitting error, and d represents the model order; According to the difference between the historical vector matrix and the time series relationship matrix, the residual matrix e is determined new；t : According to the residual matrix e new；t , calculate the online monitoring statistical index S for time series correlation analysis new : In the formula, Represents the residual matrix e new,t The transpose of The device abnormality identification module is used to perform the following operations: If S new ≤S E , it is determined that there is no equipment transient sensitivity loss fault, and If S new >S E , it is determined that there is a transient sensitivity loss fault of the equipment, S E Indicates the preset threshold.

2. The method according to claim 1, characterized in that After inputting the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan into a preset blood pressure prediction model to determine a predicted blood pressure parameter interval corresponding to a second preset time period in the future, the method further includes: Comparing the predicted blood pressure parameter interval with the normal blood pressure range of pregnant women; When it is determined that the predicted blood pressure parameter interval does not belong to a subset of the normal blood pressure range of the pregnant woman, an early warning operation is performed on the daily activity plan.

3. The method according to claim 1, characterized in that After inputting the continuous blood pressure monitoring sequence into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a device transient sensitivity loss fault, the method further includes: When it is determined that the blood pressure monitoring device has a transient sensitivity loss failure of the device, a device maintenance notification is sent to the blood pressure monitoring device.

4. The method according to claim 1, characterized in that The blood pressure prediction model adopts a MIMO deep learning model, and the blood pressure prediction model includes a cascaded multi-input module, a feature fusion module and a multi-output module; The multi-input module comprises a first input layer, a second input layer and a third input layer connected in parallel; The first input layer uses an attention-based LSTM to extract the blood pressure time series features corresponding to the continuous blood pressure monitoring sequence: i t =σ(W i ·[h t-1 ,x t ]+b i ) f t =σ(W f ·[h t-1 ,x t ]+b f ) the t =σ(W o ·[h t-1 ,x t ]+b o ) C t =f t ×C t-1 +i t ×tanh(W C ·[h t-1 ,x t ]+b C ) h t =o t ×tanh(C t ) a t =softmax(W a ×h t +b a ) In the formula, i t 、f t , o t , C t and h t Respectively represent the input gate, forget gate, output gate, unit state and initial output features of the LSTM unit at time step t; W i ,W f ,W o ,W C Represents the weight matrices of the input gate, forget gate, output gate, and cell state in the LSTM unit, and b i ,b f ,b o ,b C Respectively represent the bias term of the corresponding gate; σ represents the Sigmoid activation function; W a ,b a Represent the weight and bias of the attention layer, α t represents the attention weight, h' t Represents the weighted output features; tanh() represents the hyperbolic tangent function.

5. The method according to claim 4, characterized in that The second input layer uses a fully connected layer to extract user attribute features corresponding to the target user attribute data: y user =ReLU(W user ·x user +b user ) In the formula, x user Represents the input user attribute data, W user and b user Represent the weight matrix and bias term of the second input layer respectively; y user Represents the output user attribute features; ReLU() represents the RELU activation function; The third input layer uses a fully connected layer to extract the schedule activity features corresponding to the target schedule activity plan: x encoded =TimeEncode(x schedule ) y schedule =ReLU(W schedule ·x encoded +b schedule ) In the formula, x schedule Represents the input schedule activity plan; TimeEncode represents the time encoding function to encode time information into the feature; x encoded Represents the encoded schedule activity data; W schedule and b schedule Respectively represent the weight matrix bias items of the third input layer; y schedule Represents the output calendar activity characteristics.

6. The method according to claim 5, characterized in that The feature fusion module includes a first feature cross layer and a second feature cross layer; The first feature cross layer includes a feature attention sublayer and a first nonlinear transformation sublayer; through the first feature cross layer, the blood pressure time series feature is fused with the user attribute feature to determine a corresponding first fusion feature: γ=softmax(W att1 ·[h' t ;y user ]+b att1 ) m1=ReLU(W relu1 m 1,att +b relu1 ) In the formula, γ represents the feature attention weight; W att1 ,b att1 Represent the weight matrix and bias term of the feature attention sublayer respectively; m 1,att Represents the features weighted by feature attention; W relu1 ,b relu1 Respectively represent the weight and bias of the first nonlinear transformation sublayer, m1 represents the first fusion feature of the output, h' t Represents the weighted output features; The second feature cross layer includes a second deep cross network sublayer and a second nonlinear transformation sublayer; the first fused feature is fused with the schedule activity feature through the second feature cross layer to determine a corresponding second fused feature: <h2 style=";text-align:left;direction:ltr">m2 = ReLU(W)<h2 style=";text-align:left;direction:ltr"> relu2 <h2 style=";text-align:left;direction:ltr"> ·m<h2 style=";text-align:left;direction:ltr"> 2,cross <h2 style=";text-align:left;direction:ltr"> +b<h2 style=";text-align:left;direction:ltr"> relu2 <h2 style=";text-align:left;direction:ltr"> ) In the formula, y schedule represents the feature vector of the schedule activity, W cross2 ,b cross2 represents the weight matrix and bias term of the second deep cross network sublayer; W relu2 ,b relu2 represents the weight matrix and bias term of the second nonlinear transformation sublayer; m2 represents the second fusion feature of the output; The multi-output module comprises a first output layer and a second output layer; The first output layer includes a plurality of nonlinear processing sublayers; through the first output layer, the upper limit value BP of the blood pressure in the predicted blood pressure parameter interval is determined according to the second fusion feature. upper : BP upper =ReLU(W upper1 ·(ReLU(W upper2 ·m2+b upper2 ))+b upper1 ) Where W upper1 ,W upper2 and b upper1 ,b upper2 Respectively represent the weight matrix and bias term of the first nonlinear processing sublayer and the second nonlinear processing sublayer; The second output layer is used to determine the lower limit of blood pressure BP in the predicted blood pressure parameter interval according to the second fusion feature. lower : BP lower =Sigmoid(W lower ·m2+b lower ) Where W lower ,b lower Represents the weight matrix and bias term of the second output layer.

7. The method according to claim 6, characterized in that The time encoding function is used to encode the schedule hour information hour in the schedule activity plan in 24-hour format to obtain the corresponding schedule activity characteristics: x encoded_sin =sin(2π×hour / 24) x encoded_cos =cos(2π×hour / 24) x encoded =[x schedule ,x encoded_sin ,x encoded_cos ] Among them, x encoded_sin The sine function encoding value representing the schedule hour information, x encoded_cos The cosine function encoding value representing the schedule hour information.

8. The method according to claim 1, characterized in that: The loss function T of the blood pressure prediction model is ota oeLh oss for: Total Loss=Loss+Penalty Where, L oss represents the weighted combination loss, Penalty represents the penalty term; The weighted combination loss Loss is expressed as follows: Loss=λ×Loss upper +(1-λ)×Loss lower In the formula, Loss upper and Loss lower They represent the loss terms for the upper and lower limits of blood pressure, respectively, and λ represents the weight factor used to balance different loss terms; and They represent the predicted values ​​of the upper and lower limits of blood pressure for the i-th sample respectively; and They represent the actual upper and lower limits of blood pressure of the i-th sample respectively; N represents the total number of samples in the training sample set; The penalty term Penalty is expressed as follows: Penalty=η×max(0,|BP pred -BP actual |-threshold) 2 In the formula, BP pred and BP actual They represent the predicted blood pressure value and the actual blood pressure value respectively; η represents the penalty coefficient, and threshold represents the preset blood pressure prediction tolerance error threshold.

9. An intelligent prediction system for abnormal blood pressure during pregnancy, the system being used to implement the method according to any one of claims 1 to 8, characterized in that: The system comprises: a historical parameter query unit configured to query a historical blood pressure sampling parameter set from a blood pressure database according to a user identifier of a target pregnant woman when detecting real-time blood pressure sampling parameters for the target pregnant woman received from a blood pressure monitoring device; a time series construction unit configured to construct a continuous blood pressure monitoring sequence for the target pregnant woman based on the real-time blood pressure sampling parameters and the historical blood pressure sampling parameter set; a device fault identification unit configured to input the continuous blood pressure monitoring sequence into a device monitoring fault identification model to determine whether the blood pressure monitoring device has a device transient sensitivity loss fault; A schedule query unit is configured to query the target user attribute data and target schedule activity plan of the target pregnant woman from a preset user database based on the user identification of the target pregnant woman when it is determined that the blood pressure monitoring device does not have the transient sensitivity loss failure of the device; the target schedule activity plan includes the diet plan data and exercise activity plan data of the target pregnant woman in a second preset time period in the future; The blood pressure interval prediction unit is configured to input the continuous blood pressure monitoring sequence, the target user attribute data and the target daily activity plan into a preset blood pressure prediction model to determine the predicted blood pressure parameter interval corresponding to the second preset time period; the blood pressure prediction model adopts a deep learning model.

Images