Multi-parameter real-time wireless monitoring integrated device and method

Abstract

The invention relates to the field of real-time wireless monitoring, in particular to a multi-parameter real-time wireless monitoring integrated device and method, and the device comprises a multi-sensor data collection module, a multi-modal fusion processing module, an anomaly detection analysis module and a multi-stage alarm module. The multi-sensor data acquisition module eliminates motion artifacts by using a generative adversarial network and acquires various physiological signals; the multi-modal fusion processing module realizes space-time alignment and correlation modeling by means of linear interpolation and a graph convolution network to generate a joint feature vector; the anomaly detection and analysis module presets a causal relationship based on a causal graph, optimizes a model through a weight matrix, and judges a root cause through anti-fact reasoning; the multi-level alarm module distinguishes alarm levels according to the risk scoring model and processes the alarm levels; the device and the method solve the problems of low precision, poor collaboration and the like of a traditional monitoring technology, realize multi-parameter accurate monitoring, intelligent analysis and efficient alarm, and can be widely applied to scenes such as remote medical monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the field of multi-parameter real-time wireless monitoring integration, and specifically to a multi-parameter real-time wireless monitoring integration device and method. Background Art

[0002] In today's medical field, multi-parameter real-time wireless monitoring technology is crucial for disease prevention, diagnosis, and patient health management; traditional multi-parameter monitoring methods have many drawbacks and are difficult to meet the needs of modern medicine;

[0003] Traditional monitoring devices often rely on a single sensor or a simple combination, and are unable to accurately capture complex physiological signal changes; various physiological parameters are interrelated, but traditional devices lack effective fusion and analysis means, and the data collected by different sensors are often processed independently, and the potential relationships between parameters cannot be explored. For example, there is a close connection between respiratory rate and blood oxygen saturation. When breathing is abnormal, blood oxygen saturation usually changes, but traditional monitoring is difficult to use this correlation for comprehensive judgment and is prone to missing important pathological information.

[0004] Lacking accurate root cause analysis, traditional monitoring mainly judges abnormalities based on preset thresholds and is unable to deeply analyze the causes of abnormalities; when blood oxygen saturation is below the normal range, it is difficult to determine whether it is caused by respiratory problems, cardiovascular problems, or other factors; to solve the above defects, a technical solution is provided now. Summary of the Invention

[0005] To solve the technical problems raised in the above background art, the present invention provides a multi-parameter real-time wireless monitoring integration device and method.

[0006] The object of the present invention can be achieved by the following technical solutions:

[0007] The first aspect of the present invention provides a multi-parameter real-time wireless monitoring integration device, including a multi-sensor data acquisition module, a multi-modal fusion processing module, an anomaly detection and analysis module, a multi-level alarm module, and a database.

[0008] The multi-sensor data acquisition module collects the user's state according to multiple sensors and eliminates motion artifacts by establishing a generative adversarial network. The specific process is as follows:

[0009] The multi-sensors include a photoelectric sensor, a heart rate piezoelectric sensor, an impedance sensor, a temperature sensor, and a MEMS accelerometer; the photoelectric sensor is used to collect blood oxygen saturation signals in real time, the heart rate piezoelectric sensor is used to collect electrocardiogram signals in real time, the temperature sensor is used to collect body temperature signals in real time, the impedance sensor is used to collect respiratory rate signals in real time, and the MEMS accelerometer collects three-axis dynamic acceleration signals in real time when the user is in a moving state;

[0010] The generative adversarial network architecture includes a noise generator, a signal generator, and a discriminator. The dynamic acceleration signal is input into the noise generator, and the time-domain features of the dynamic acceleration signal are extracted through the feature extraction layer of a one-dimensional convolutional neural network, which contains 3 convolutional layers. The high-frequency vibration and low-frequency displacement features of the motion signal are extracted layer by layer, and then the convolutional features are mapped into a motion feature matrix through a fully connected layer. The electrocardiogram signal, respiratory frequency signal, and motion feature matrix are input into the signal generator, and the artifact-free signal after denoising is output through a residual convolutional network. The artifact-free signal is input into the discriminator to extract the global features of the signal. The global features include waveform periodicity, amplitude distribution, and spectral energy. The output layer outputs the discrimination probability p ∈ [0, 1] through the Sigmoid activation function. If p ≈ 1, it means that the input is a real signal and the motion artifact has been removed. If p ≈ 0, it means that the input is a generated signal, and the above steps are repeated until it is a real signal;

[0011] The multi-modal fusion processing module realizes spatio-temporal alignment and correlation modeling according to linear interpolation and the graph convolutional network GCN, and generates joint features through a feed-forward network to improve the accuracy of anomaly detection. The specific steps are as follows:

[0012] Synchronize the blood oxygen saturation signal, electrocardiogram signal, body temperature signal, respiratory frequency signal, and dynamic acceleration signal to a unified time axis according to linear interpolation;

[0013] After synchronization, model the GCN. Specifically: set the node set V = {X1, X2......X5}, corresponding to 5 types of nodes. Each node includes each feature dimension. The X1 blood oxygen saturation node includes amplitude, waveform rising edge time, and pulse conduction time. The X2 electrocardiogram node includes RR interval coefficient of variation and QRS amplitude. The X3 body temperature signal includes real-time temperature and temperature change rate. The X4 respiratory frequency signal includes tidal volume and respiratory cycle. The X5 includes acceleration amplitude, acceleration, and motion direction; the edge set δ is initialized to the causal relationship preset by medical guidelines, and multiple directed edges are set to form the initial adjacency matrix as Indicates the number of directed edges;

[0014] Learn the edge weights through 3 layers of GCN, convert the initial adjacency matrix into a learnable weight matrix, and then obtain the spatial correlation features of the nodes. Specifically: the first layer of GCN normalizes the adjacency matrix through the symmetric normalization method, and its calculation logic is: where D is the degree matrix, ensuring that the node features are normalized according to the number of neighbors, and perform feature transformation and activation on it. Its calculation logic is: where the input H (0) Feature vector, W (0) Is the weight matrix, σ is the activation function, and the output H (1)Capture the feature interactions of first-order neighbors; the second-layer GCN repeats the adjacency matrix normalization step and introduces residual connections to avoid gradient vanishing. Its calculation logic is as follows: Output H (2) Model the feature coupling of second-order neighbors, such as heart rate variability → cardiac output change → blood oxygen transport efficiency; the third-layer GCN outputs an edge weight matrix according to a linear transformation, and the causal strength between nodes is characterized as: Optimize the adjacency matrix through the objective function to output the updated learnable weight matrix A θ , and the edge weight range is (0, 1);

[0015] Integrate the spatio-temporal features of the features of each processed node through a feed-forward network, and output a joint feature vector with a dimension of 128. The joint feature vector includes the time-domain indicators and frequency-domain indicators of heart rate variability in the time series features, and the respiratory-heart rate coupling phase difference and the mapping relationship between conduction time and blood pressure in the spatial correlation features;

[0016] The anomaly detection and analysis module presets the causal relationships of physiological parameters based on the causal graph, optimizes the causal graph model through the weight matrix, introduces the root cause analysis algorithm for counterfactual reasoning to determine the root cause, and outputs the anomaly type, probability, and trend graph. The specific process is as follows: Define a directed acyclic graph Map the joint feature vector with a dimension of 128 output by the feed-forward network to a node state vector X ∈ R T×5 , where T is the number of time steps;

[0017] Input the node state vector into the objective function Among them, y t represents the physiological parameter vector actually observed at time t, represents the parameter vector predicted by the causal graph model. Calculate the causal transmission between nodes through the graph neural network. λ = 0.01 represents the L2 regularization coefficient;

[0018] Calculate the output of the causal graph model according to the learnable weight matrix. Its calculation logic is as follows:

[0019] Calculate the prediction error MES through backpropagation. Its calculation logic is as follows: Use the Adam optimizer to accelerate convergence. When MSR is less than 10 for 50 consecutive rounds -5 , stop to obtain the optimized causal graph model;

[0020] Introduce the root cause analysis algorithm's counterfactual reasoning to determine the root cause, and output the abnormal type and probability value. Specifically: when the blood oxygen saturation < 90% or the coefficient of variation of the electrocardiogram interval > 0.15 or the body temperature > 37.5°C is detected for 3 consecutive time steps, trigger the root cause analysis. Taking blood oxygen saturation as an example, mark the blood oxygen saturation, electrocardiogram, body temperature, respiratory rate, and dynamic acceleration as TY, TR, TE, TW, and TQ respectively. Sort the intervention variables of blood oxygen saturation. If the first sorting is the respiratory rate, calculate the normal value of the user's respiratory rate to get TW normal Perform an intervention operation on the current respiratory rate and set it to TW normal Use the trained causal graph model, input the intervened respiratory rate value into the model, and calculate the theoretical value of blood oxygen saturation through the forward propagation of the model Calculate the expected value E[TW normal |TY] of the intervened blood oxygen saturation and take the difference from the currently observed blood oxygen saturation value E[TW|TY], and divide the obtained difference by the historical standard deviation TW of blood oxygen saturation σ to obtain the standardized causal effect SCE, and its calculation logic is:

[0021]

[0022] Obtain the difference between the theoretical value of the intervened blood oxygen saturation and the actual observed value E[TW|TY], and perform standardization processing. Its calculation logic is: Extract the corresponding preset threshold in the database to determine the strength of the causal association: when ZS > 2, it is determined that there is a strong causal association; when 1.5 ≤ ZS < 2, it is determined that there is a weak causal association;

[0023] Integrate multi-signal evidence through a logistic regression model, and calculate the root cause probability P of the respiratory rate (r) Its calculation logic is:

[0024] where 2, 1.5, and 1 are fixed weight coefficients, and TR 偏离度 is the deviation degree of the current electrocardiogram from the normal range, and TE 偏差 is the difference between the current body temperature and the normal body temperature range; and so on. For the root cause probabilities of electrocardiogram, body temperature, and dynamic acceleration, select the maximum root cause probability as the abnormal type and send it to the multi-level alarm module.

[0025] The multi-level alarm module differentiates and processes the alarm level through risk scoring based on the abnormal detection and root cause analysis results, combined with the abnormal degree of physiological parameters and the clinical risk level. The specific process is:

[0026] Obtain the root cause probability value P of the current abnormal type (r)and the absolute value deviation of the parameter P 偏差 , substitute into the risk score calculation model: SW = P (r) 0.7+P 偏差 The risk score is calculated based on the value of 0.03, where 0.7 and 0.3 are fixed weight coefficients for the deviation between the root cause probability value and the absolute value of the parameter. The risk score is divided into three levels, including blue warning, yellow alert, and red alert. Specifically, 30≤S<40 is a blue warning, which pushes health advice via SMS and records parameter fluctuations; 40≤S<70 is a yellow alert, which sends a flashing warning light and voice reminder to the user's mobile terminal and generates a parameter trend report; S≥70 is a red alert, which notifies the emergency contact by phone, triggers an audible and visual alarm, and initiates location tracking and sends it to the corresponding doctor's mobile terminal.

[0027] The second aspect of the present invention provides a multi-parameter real-time wireless monitoring integrated method, the specific steps of which are as follows:

[0028] Step 1: Multi-sensor data collection: collect user status based on multiple sensors and establish a generative adversarial network to eliminate motion artifacts;

[0029] Step 2: Multimodal fusion processing: Using linear interpolation and GCN to achieve spatiotemporal alignment and correlation modeling, and generating joint features through a feedforward network to improve the accuracy of anomaly detection analysis;

[0030] Step 3: Anomaly detection and analysis: Based on the causal graph, the causal relationship of physiological parameters is preset, the causal graph model is optimized through the weight matrix, and the root cause analysis algorithm is introduced to determine the root cause through counterfactual reasoning. The anomaly type, probability and trend graph are output;

[0031] Step 4. Multi-level alarm: Calculate the risk scoring function based on the risk scoring calculation model, which comprehensively considers the root cause probability, parameter deviation degree and trend risk. Differentiate and process the alarm levels based on the risk score.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] Enhanced multi-parameter fusion analysis capabilities: This approach leverages linear interpolation and graph convolutional networks to achieve spatiotemporal alignment and correlation modeling of multiple physiological signals, exploring potential relationships between parameters. The joint feature vector generated by the feedforward network incorporates rich temporal and spatial correlation information, such as the phase difference of respiratory-heart rate coupling, significantly improving anomaly detection accuracy.

[0034] Precise anomaly detection and root cause analysis: Preset the causal relationships of physiological parameters based on causal graphs, optimize the model through a weight matrix, and combine counterfactual reasoning to determine the root cause. When abnormal blood oxygen saturation is detected, it can accurately determine whether it is caused by respiratory problems, cardiovascular problems, or other factors, helping doctors quickly formulate targeted treatment plans, avoid misdiagnosis and missed diagnosis, and improve the treatment effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. The following drawings are not deliberately drawn to scale in actual size, and the focus is on showing the gist of the present invention.

[0036] Figure 1 It is a block diagram of the module connection of the present invention.

[0037] Figure 2 It is a flowchart of the method steps of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts also belong to the scope of protection of the present invention.

[0039] Please refer to Figure 1 As shown, the first aspect of the present invention provides a multi-parameter real-time wireless monitoring integration device, including a multi-sensor data acquisition module, a multi-modal fusion processing module, an anomaly detection and analysis module, a multi-level alarm module, and a database.

[0040] The multi-sensor data acquisition module collects the user's state according to multiple sensors, and establishes a generative adversarial network to eliminate motion artifacts. The specific process is as follows:

[0041] The multiple sensors include a photoelectric sensor, a heart rate piezoelectric sensor, an impedance sensor, a temperature sensor, and a MEMS accelerometer; the photoelectric sensor is used to collect blood oxygen saturation signals in real time, the heart rate piezoelectric sensor is used to collect electrocardiogram signals in real time, the temperature sensor is used to collect body temperature signals in real time, the impedance sensor is used to collect respiratory frequency signals in real time, and the MEMS accelerometer collects three-axis dynamic acceleration signals in real time when the user is in a moving state;

[0042] The generative adversarial network architecture includes a noise generator, a signal generator, and a discriminator. The dynamic acceleration signal is input into the noise generator, and the time-domain features of the dynamic acceleration signal are extracted through the feature extraction layer of a one-dimensional convolutional neural network, which includes 3 convolutional layers. The high-frequency vibration and low-frequency displacement features of the motion signal are extracted layer by layer, and then the convolutional features are mapped into a motion feature matrix through a fully connected layer. The electrocardiogram signal, respiratory frequency signal, and motion feature matrix are input into the signal generator, and the artifact-free signal after denoising is output through a residual convolutional network. The artifact-free signal is input into the discriminator to extract the global features of the signal. The global features include waveform periodicity, amplitude distribution, and spectral energy. The output layer outputs the discrimination probability p ∈ [0, 1] through the Sigmoid activation function. If p ≈ 1, it means that the input is a real signal and the motion artifact has been removed. If p ≈ 0, it means that the input is a generated signal, and the above steps are repeated until it is a real signal;

[0043] The multi-modal fusion processing module realizes spatio-temporal alignment and correlation modeling according to linear interpolation and the graph convolutional network GCN, and generates joint features through a feed-forward network to improve the accuracy of anomaly detection. The specific steps are as follows:

[0044] Synchronize the blood oxygen saturation signal, electrocardiogram signal, body temperature signal, respiratory frequency signal, and dynamic acceleration signal to a unified time axis according to linear interpolation;

[0045] After synchronization, model the GCN. Specifically: Set the node set V = {X1, X2......X5}, corresponding to 5 types of nodes. Each node includes each feature dimension. The X1 blood oxygen saturation node includes amplitude, waveform rising edge time, and pulse conduction time. The X2 electrocardiogram node includes the coefficient of variation of the RR interval and the QRS amplitude. The X3 body temperature signal includes the real-time temperature and the temperature change rate. The X4 respiratory frequency signal includes tidal volume and respiratory cycle. The X5 acceleration amplitude, acceleration, and motion direction; The edge set δ is initially set to the causal relationship preset by medical guidelines, such as abnormal respiration → decreased blood oxygen or increased heart rate → change in stroke volume. Set multiple directed edges to form the initial adjacency matrix as Indicates the number of directed edges;

[0046] Learn the edge weights through 3 layers of GCN, convert the initial adjacency matrix into a learnable weight matrix, and then obtain the spatial correlation features of the nodes. Specifically: The first layer of GCN normalizes the adjacency matrix through the symmetric normalization method, and its calculation logic is: Where D is the degree matrix, which ensures that the node features are normalized according to the number of neighbors, and performs feature transformation and activation on it. Its calculation logic is: Where the input H (0) Feature vector, W (0) Is the weight matrix, σ is the activation function, and the output H (1)Capture the feature interactions of first-order neighbors; the second-layer GCN repeats the adjacency matrix normalization step and introduces residual connections to avoid gradient vanishing. Its calculation logic is as follows: Output H (2) Model the feature coupling of second-order neighbors, such as heart rate variability → cardiac output change → blood oxygen transport efficiency; the third-layer GCN outputs an edge weight matrix according to a linear transformation, and the causal strength between nodes is characterized as: Optimize the adjacency matrix through the objective function to output the updated learnable weight matrix A θ , and the edge weight range is (0, 1); for example, the edge weight of heart rate variability → respiratory rate change is optimized from the initial 1 to 0.85, indicating a strong causal association, and capturing the spatial association features of physiological couplings such as electrocardiogram-blood oxygen and respiration-heart rate; integrate the spatio-temporal features of the features of each processed node through a feed-forward network, and output a joint feature vector with a dimension of 128, including the time-domain index and frequency-domain index of heart rate variability in the time-series features, the respiratory-heart rate coupling phase difference in the spatial association features, and the mapping relationship between the conduction time and blood pressure;

[0047] The anomaly detection and analysis module presets the causal relationships of physiological parameters based on the causal graph, optimizes the causal graph model through the weight matrix, introduces the root cause analysis algorithm for counterfactual reasoning to determine the root cause, and outputs the anomaly type, probability, and trend graph. The specific process is as follows: Define a directed acyclic graph Map the joint feature vector with a dimension of 128 output by the feed-forward network to a node state vector X ∈ R T×5 , where T is the number of time steps;

[0048] Input the node state vector into the objective function where, y t represents the vector of physiological parameters actually observed at time t, represents the parameter vector predicted by the causal graph model. Calculate the causal transmission between nodes through the graph neural network. λ = 0.01 represents the L2 regularization coefficient, which is used to avoid overfitting;

[0049] Calculate the output of the causal graph model according to the learnable weight matrix, and its calculation logic is:

[0050] Calculate the prediction error MES through backpropagation, and its calculation logic is: Use the Adam optimizer to accelerate convergence. Stop when MSR is less than 10 for 50 consecutive rounds -5 , and finally obtain the optimized causal graph model;

[0051] Introduce the counterfactual reasoning of the root cause analysis algorithm to determine the root cause, and output the abnormal type, probability, and trend chart. Specifically: when the blood oxygen saturation < 90% or the coefficient of variation of the electrocardiogram interval > 0.15 or the body temperature > 37.5°C is detected for three consecutive time steps, trigger the root cause analysis. Taking blood oxygen saturation as an example, mark blood oxygen saturation, electrocardiogram, body temperature, respiratory rate, and dynamic acceleration as TY, TR, TE, TW, and TQ respectively. Sort the intervention variables of blood oxygen saturation. If the first sorting is the respiratory rate, calculate the normal value of the user's respiratory rate to obtain TW normal It should be noted that by collecting the respiratory rate data of this individual in the past week and calculating its average value as the normal value; perform an intervention operation on the current respiratory rate and set it to TW normal Use the trained causal graph model, input the intervened respiratory rate value into the model, and calculate the theoretical value of blood oxygen saturation through the forward propagation of the model Calculate the expected value E[TW normal |TY] of the intervened blood oxygen saturation and subtract it from the currently observed blood oxygen saturation value E[TW|TY], and divide the obtained difference by the historical standard deviation TW of blood oxygen saturation σ to obtain the standardized causal effect SCE, and its calculation logic is:

[0052]

[0053] Detect the signals of electrocardiogram and body temperature. After observing the intervened respiratory rate, check whether the relevant characteristics of the electrocardiogram signal have changed, and check whether the ST segment depression amplitude of the electrocardiogram signal has decreased. If the ST segment depression amplitude does decrease, this may indicate that the change in respiratory rate has an impact on the electrical activity of the heart, thus supporting the inference of a causal relationship between respiratory rate and blood oxygen saturation; check whether the correlation between body temperature and heat metabolic rate returns to normal. If the correlation returns to normal, this also provides additional supporting evidence for the respiratory rate as the root cause;

[0054] Obtain the theoretical value of the intervened blood oxygen saturation and the difference between the actual observed value E[TWTY], and perform standardization processing. Its calculation logic is: Extract the corresponding preset threshold in the database to determine the strength of the causal association: when ZS > 2, it is determined that there is a strong causal association; when 1.5 ≤ ZS < 2, it is determined that there is a weak causal association;

[0055] Integrate multi-signal evidence through a logistic regression model and calculate the root cause probability P of the respiratory rate (r) Its calculation logic is: Among them, 2, 1.5, and 1 are fixed weight coefficients, and TR 偏离度 is the deviation degree of the current electrocardiogram from the normal range, and TE 偏差is the difference between the current body temperature and the normal body temperature range; and so on. For the root cause probabilities of electrocardiogram, body temperature, and dynamic acceleration, select the maximum root cause probability as the abnormal type and send it to the multi-level alarm module.

[0056] Based on the results of anomaly detection and root cause analysis, the multi-level alarm module combines the degree of physiological parameter abnormality and the clinical risk level, and differentiates and processes the alarm levels through risk scoring. The specific process is as follows:

[0057] Obtain the root cause probability value P of the current abnormal type (r) and the absolute value deviation P of the parameter 偏差 , and substitute them into the risk scoring calculation model: SW = P (r) ·0.7 + P 偏差 ·0.3 to obtain the risk scoring value, where 0.7 and 0.3 are the fixed weight coefficients of the root cause probability value and the absolute value deviation of the parameter; divide the risk scoring value into three levels, including blue warning, yellow alarm, and red alarm. Specifically: 30 ≤ S < 40 is a blue warning, and health advice is pushed via text message and parameter fluctuations are recorded; 40 ≤ S < 70 is a yellow alarm, a flashing warning light + voice reminder is sent to the user's mobile terminal, and a parameter trend report is generated; S ≥ 70 is a red alarm, an emergency contact is notified by phone, a sound and light alarm is triggered, and location tracking is started and sent to the mobile terminal of the corresponding doctor.

[0058] As Figure 2 shown, the second aspect of the present invention provides a multi-parameter real-time wireless monitoring integration method, and its specific steps are as follows:

[0059] Step 1: Multi-sensor data collection: Collect the user's status according to multiple sensors, and establish a generative adversarial network to eliminate motion artifacts.

[0060] Step 2: Multi-modal fusion processing: Achieve spatio-temporal alignment and correlation modeling according to linear interpolation and GCN, generate joint features through a feedforward network, and improve the accuracy of anomaly detection and analysis.

[0061] Step 3: Anomaly detection and analysis: Preset the causal relationship of physiological parameters based on a causal graph, optimize the causal graph model through a weight matrix, introduce a root cause analysis algorithm for counterfactual reasoning to determine the root cause, and output the abnormal type, probability, and trend graph.

[0062] Step 4: Multi-level alarm: Calculate the risk scoring function based on the risk scoring calculation model by integrating the root cause probability, parameter deviation degree, and trend risk, and differentiate and process the alarm levels through risk scoring.

[0063] The foregoing is a description of the invention and should not be construed as limiting thereof. Although several exemplary embodiments of the invention have been described, those skilled in the art will readily appreciate that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the invention. Accordingly, all such modifications are intended to be included within the scope of the invention as defined by the claims. It should be understood that the foregoing is a description of the invention and should not be considered limited to the particular embodiments disclosed, and modifications to the disclosed embodiments as well as other embodiments are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.

Claims

1. A multi-parameter real-time wireless monitoring integrated device, comprising a multi-sensor data acquisition module, a multi-modal fusion processing module, an anomaly detection and analysis module, a multi-level alarm module, and a database, characterized by: The multimodal fusion processing module implements spatiotemporal alignment and association modeling based on linear interpolation and GCN, and generates joint features through a feedforward network. The multi-level alarm module calculates a risk scoring function based on the risk scoring calculation model, comprehensively considering the root cause probability, parameter deviation degree, and trend risk, and distinguishes and processes the alarm level based on the risk score. The abnormality detection and analysis module presets the causal relationship of physiological parameters based on the causal graph, and optimizes the causal graph model through the weight matrix. Specifically, it defines a directed acyclic graph The joint feature vector with a dimension of 128 output by the feedforward network is mapped to the node state vector X∈R through the fully connected layer T×5 , where T is the number of time steps; the node state vector is input into the objective function Among them, y t Represented as the physiological parameter vector actually observed at time t, It is represented as the parameter vector predicted by the causal graph model, and the causal transmission between nodes is calculated through the graph neural network. λ is represented as the L2 regularization coefficient; The causal graph model output is calculated based on the learnable weight matrix. The calculation logic is: Calculate the prediction error MES through backpropagation, and its calculation logic is as follows: Use the Adam optimizer to accelerate convergence. Stop when MSR is less than 10 for 50 consecutive rounds -5 to obtain the optimized causal graph model.

2. A multi-parameter real-time wireless monitoring integrated device according to claim 1, characterized in that: The abnormality detection and analysis module is specifically as follows: when the blood oxygen saturation is detected to be less than 90% or the ECG interval variation coefficient is greater than 0.15 or the body temperature is greater than 37.5℃ for three consecutive time steps, the root cause analysis is triggered, the blood oxygen saturation, ECG, body temperature, respiratory rate and dynamic acceleration are marked as TY, TR, TE, TW and TQ respectively, the intervention variables of blood oxygen saturation are sorted, if the first sort is respiratory rate, the normal value of the user's respiratory rate is calculated to obtain TW normal , intervene in the current respiratory rate and set it to TW normal , using the trained causal graph model, input the respiratory rate value after intervention into the model, and calculate the theoretical value of blood oxygen saturation through the forward propagation of the model Calculate the expected value of blood oxygen saturation after intervention E[TW normal The difference between the current observed blood oxygen saturation value E[TW|TY] and the historical standard deviation of blood oxygen saturation TW is divided by the difference. σ , we get the standardized causal effect SCE, whose calculation logic is: Obtain theoretical blood oxygen saturation value after intervention The difference between the actual observed value E[TW|TY] and the normalized value is calculated as follows: The corresponding preset threshold in the database is extracted to determine the strength of the causal relationship: when ZS>2, it is determined that there is a strong causal relationship; when 1.5≤ZS<2, it is determined that there is a weak causal relationship.

3. The multi-parameter real-time wireless monitoring integration device according to claim 2, characterized in that, The anomaly detection and analysis module introduces a root cause analysis algorithm to determine the root cause through counterfactual reasoning and outputs the anomaly type and probability value, specifically: The logistic regression model was used to integrate multiple signal evidence and calculate the root cause probability P of respiratory rate. (r) , its calculation logic is: Among them, 2, 1.5 and 1 are fixed weight coefficients, TR 偏离度 TE is the degree of deviation between the current ECG and the normal range. 偏差 is the difference between the current body temperature and the normal body temperature range; similarly, for the root cause probabilities of ECG, body temperature, and dynamic acceleration, the maximum root cause probability is selected as the abnormality type and sent to the multi-level alarm module.

4. A multi-parameter real-time wireless monitoring integrated device according to claim 1, characterized in that The multimodal fusion processing module implements spatiotemporal alignment and association modeling based on linear interpolation and graph convolutional network (GCN). The specific steps are as follows: Synchronize the blood oxygen saturation signal, electrocardiogram signal, body temperature signal, respiratory rate signal and dynamic acceleration signal to a unified time axis according to linear interpolation; Model the GCN after synchronization, specifically: Set the node set V = {X1, X2......X5}, corresponding to 5 types of nodes. Each node includes various feature dimensions. The X1 blood oxygen saturation node includes amplitude, waveform rising edge time, and pulse transit time. The X2 electrocardiogram node includes the coefficient of variation of the RR interval and the QRS amplitude. The X3 body temperature signal includes the real-time temperature and the temperature change rate. The X4 respiratory rate signal includes tidal volume and respiratory cycle. The X5 includes acceleration amplitude, acceleration, and movement direction; The edge set δ is initialized to the causal relationship preset by the medical guidelines, and multiple directed edges are set to form the initial adjacency matrix as Indicated as the number of directed edges; Learn the edge weights through three layers of GCN, convert the initial adjacency matrix into a learnable weight matrix, and then obtain the spatial correlation features of the nodes. Specifically: The first layer of GCN normalizes the adjacency matrix through the symmetric normalization method, and its calculation logic is: where D is the degree matrix, ensuring that the node features are normalized by the number of neighbors, and perform feature transformation and activation on it. Its calculation logic is: where the input H (0) feature vector, W (0) is the weight matrix, σ is the activation function, and the output H (1) captures the feature interactions of first-order neighbors; The second layer of GCN repeats the adjacency matrix normalization step and introduces residual connections to avoid gradient disappearance. Its calculation logic is: Output H (2) Modeling the feature coupling of second-order neighbors, such as heart rate variability → cardiac output change → blood oxygen transport efficiency; the third layer GCN outputs the edge weight matrix based on the linear transformation, representing the causal strength between nodes as: The adjacency matrix is optimized by the objective function to output the updated learnable weight matrix A θ , whose edge weight range is (0,1).

5. The multi-parameter real-time wireless monitoring integration device according to claim 4, characterized in that, The multimodal fusion processing module generates joint features through a feedforward network, specifically: The features of each node after processing are integrated with spatiotemporal features through a feedforward network, and a joint feature vector with a dimension of 128 is output. The joint feature vector includes the time domain index and frequency domain index of heart rate variability in the time series features, the mapping relationship between the respiratory-heart rate coupling phase difference and conduction time and blood pressure in the spatial correlation features.

6. The multi-parameter real-time wireless monitoring integrated device according to claim 1, wherein The multi-level alarm module of the abnormality detection and analysis module distinguishes and processes the alarm level through risk scoring based on the abnormality detection and root cause analysis results, combined with the abnormality degree of physiological parameters and clinical risk level. The specific process is as follows: Obtain the root cause probability value P of the current exception type (r) and the absolute value deviation P of the parameter 偏差 , substitute them into the risk scoring calculation model: SW = P (r) ·0.7 + P 偏差 ·0.3 to obtain the risk score value, where 0.7 and 0.3 are the fixed weight coefficients of the root cause probability value and the absolute value deviation of the parameter; divide the risk score value into three levels, including blue warning, yellow alert and red alert, specifically: 30 ≤ S < 40 is the blue warning, push health advice through text messages and record parameter fluctuations 40≤S<70 is a yellow alarm, which will be alerted to the user's mobile terminal through flashing warning lights and voice reminders, and a parameter trend report will be generated; S≥70 is a red alert, and the emergency contact will be notified by phone, an audible and visual alarm will be triggered, and location tracking will be initiated and sent to the corresponding doctor's mobile terminal.

7. The multi-parameter real-time wireless monitoring integrated device according to claim 1, characterized in that, The multi-sensor data acquisition module collects user status based on multiple sensors, which is specifically: The multiple sensors include photoelectric sensors, heart rate piezoelectric sensors, impedance sensors, temperature sensors and MEMS accelerometers; the photoelectric sensors are used to collect blood oxygen saturation signals in real time, the heart rate piezoelectric sensors are used to collect electrocardiogram signals in real time, the temperature sensors are used to collect body temperature signals in real time, the impedance sensors are used to collect respiratory rate signals in real time, and the MEMS accelerometers collect three-axis dynamic acceleration signals in real time when the user is in motion.

8. The multi-parameter real-time wireless monitoring integration device according to claim 7, characterized in that The multi-sensor data acquisition module establishes a generative adversarial network to eliminate motion artifacts. The specific process is as follows: The generative adversarial network architecture includes a noise generator, a signal generator, and a discriminator. The dynamic acceleration signal is input into the noise generator. The dynamic acceleration signal is extracted from the time domain through the feature extraction layer of the one-dimensional convolutional neural network. It contains three convolutional layers, which extract the high-frequency vibration and low-frequency displacement features of the motion signal layer by layer. The convolution features are then mapped into a motion feature matrix through a fully connected layer. The electrocardiogram signal, respiratory rate signal, and motion feature matrix are input into the signal generator, and the residual convolutional network outputs a denoised, artifact-free signal. The artifact-free signal is input into the discriminator to extract the global features of the signal, including waveform periodicity, amplitude distribution and spectral energy. The output layer outputs the discrimination probability p∈[0,1] through the Sigmoid activation function. If p≈1, it means that the input is a real signal and the motion artifact has been removed. If p≈0, it means that the input is a generated signal. The above steps are repeated until it is a real signal.

9. A multi-parameter real-time wireless monitoring integrated method, used to implement a multi-parameter real-time wireless monitoring integrated device according to any one of claims 1 to 8, characterized in that: The specific steps are as follows: Step 1: Multi-sensor data collection: collect user status based on multiple sensors and establish a generative adversarial network to eliminate motion artifacts; Step 2: Multimodal fusion processing: Using linear interpolation and GCN to achieve spatiotemporal alignment and correlation modeling, and generating joint features through a feedforward network to improve the accuracy of anomaly detection analysis; Step 3: Anomaly detection and analysis: Based on the causal graph, the causal relationship of physiological parameters is preset, the causal graph model is optimized through the weight matrix, and the root cause analysis algorithm is introduced to determine the root cause through counterfactual reasoning. The anomaly type, probability and trend graph are output; Step 4. Multi-level alarm: Calculate the risk scoring function based on the risk scoring calculation model, which comprehensively considers the root cause probability, parameter deviation degree and trend risk. Differentiate and process the alarm levels based on the risk score.

Images