Multi-device linkage old-age care data monitoring and early warning method, system and device

Abstract

The invention provides a multi-device linkage old-age care data monitoring and early warning method, system and device, and relates to the technical field of data processing, and the method comprises the steps: carrying out the linkage monitoring of home-based old-age care users, obtaining an old-age care monitoring data set, and carrying out the feature recognition of an activity scene, carrying out multi-dimensional abnormal analysis attention optimization according to the user activity scene label, carrying out risk accident type fitting according to the user activity scene label, carrying out multi-dimensional risk accident deduction on the monitoring abnormal attention coupling matrix according to a dynamic disturbance distillation mechanism, and obtaining a risk accident deduction map; and optimizing the risk accident deduction map based on the accompanying user monitoring data to obtain a risk deduction optimization map, and synchronously generating a pension monitoring early warning signal. The technical problem that the accuracy and reliability of pension accident monitoring and prediction are insufficient due to one-sidedness of pension monitoring in the prior art is solved. The technical effect of improving the accuracy and reliability of accident monitoring and early warning of home-based care users is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a multi-device linkage pension data monitoring and early warning method, system and device. BACKGROUND

[0002] Traditional home-based pension monitoring usually relies on a single data source, such as monitoring physiological monitoring data of the elderly through wearable devices for risk identification, or only using indoor cameras to collect user behavior information. Although a single data source can provide basic monitoring functions, it cannot comprehensively evaluate the overall health status of the elderly. In addition, the traditional monitoring method usually ignores the influence and intervention effect accompanying the user, resulting in a lack of comprehensiveness and accuracy in risk identification and early warning, thereby affecting the accuracy and reliability of the pension accident monitoring and early warning.

[0003] The existing technology has the technical problem of one-sidedness in pension monitoring, which leads to insufficient accuracy and reliability of pension accident monitoring and prediction. SUMMARY

[0004] The purpose of the present application is to provide a multi-device linkage pension data monitoring and early warning method, system and device, which solves the technical problem of one-sidedness in the prior art pension monitoring, which leads to insufficient accuracy and reliability of pension accident monitoring and prediction.

[0005] In view of the above problems, the present application provides a multi-device linkage pension data monitoring and early warning method, system and device.

[0006] The first aspect of the present application provides a multi-device linkage pension data monitoring and early warning method, the method comprising: monitoring a home-based pension user through a multi-device monitoring array, obtaining a pension monitoring data set, and identifying activity scene features based on the pension monitoring data set, generating user activity scene labels; according to the user activity scene label, multi-dimensional abnormality attention optimization is carried out on the pension monitoring data set, and a monitoring abnormality attention coupling matrix is established; according to the user activity scene label, risk accident type fitting is carried out, N fitting risk accident types are obtained, N is a positive integer greater than 1; based on the N fitting risk accident types, according to the dynamic disturbance distillation mechanism, the monitoring abnormality attention coupling matrix is subjected to multi-dimensional risk accident deduction, and a risk accident deduction graph is obtained; read the corresponding accompanying user monitoring data of the real-time accompanying user, and based on the accompanying user monitoring data, the risk accident deduction graph is subjected to accompanying accident intervention analysis optimization, and a risk deduction optimization graph is obtained, and a pension monitoring early warning signal is generated synchronously.

[0007] Optionally, data cleaning and feature identification are performed according to the elderly care monitoring data set to obtain a physiological monitoring feature sequence, an environmental monitoring feature sequence, and a behavior monitoring feature sequence; the physiological monitoring feature sequence is subjected to abnormal analysis attention optimization according to the user activity scene label to generate a physiological abnormality attention coupling matrix; the environmental monitoring feature sequence is subjected to abnormal analysis attention optimization according to the user activity scene label to generate an environmental abnormality attention coupling matrix; the behavior monitoring feature sequence is subjected to abnormal analysis attention optimization according to the user activity scene label to generate a behavior abnormality attention coupling matrix; and the physiological abnormality attention coupling matrix, the environmental abnormality attention coupling matrix, and the behavior abnormality attention coupling matrix are aggregated to generate the monitoring abnormality attention coupling matrix.

[0008] Optionally, a normal physiological monitoring sample of the home-based elderly care user is retrieved based on the user activity scene label to obtain a scene normal physiological sample set; multi-dimensional physiological feature set trend analysis is performed according to the scene normal physiological sample set to construct a scene normal physiological monitoring space; abnormality identification is performed on the physiological monitoring feature sequence according to the scene normal physiological monitoring space to obtain a physiological abnormality identification result; abnormality degree evaluation is performed on the physiological abnormality identification result according to the scene normal physiological monitoring space to obtain a physiological abnormality evaluation result; and abnormality attention coupling optimization is performed on the physiological monitoring feature sequence according to the physiological abnormality identification result and the physiological abnormality evaluation result to obtain the physiological abnormality attention coupling matrix.

[0009] Optionally, a risk accident is retrieved according to the user activity scene label to obtain a scene matching historical risk accident set; accident type identification is performed according to the scene matching historical risk accident set to obtain an accident type sample set; support degree calculation is performed on each accident type sample in the accident type sample set according to the scene matching historical risk accident set to obtain accident type support degrees; confidence degree calculation is performed on each accident type sample according to the accident type support degrees to obtain accident type confidence degrees; and the accident type sample set is subjected to optimization identification according to the accident type confidence degrees based on a predetermined type confidence degree to generate the N fitted risk accident types.

[0010] Optionally, risk accident deduction record retrieval is performed according to the N fitted risk accident types to obtain an N risk accident deduction record set; accident tree training is performed according to the N risk accident deduction record set to obtain N risk accident deduction models; dynamic perturbation distillation reinforcement is performed on the N risk accident deduction models according to the dynamic perturbation distillation mechanism to establish N risk accident deduction channels; the monitoring abnormality attention coupling matrix is input into the N risk accident deduction channels to obtain N risk accident deduction paths; and the N risk accident deduction paths are sorted to generate the risk accident deduction graph.

[0011] Optionally, based on the N risk incident simulation models, an nth risk incident simulation model is extracted, where n is a positive integer, 1 ≤ n ≤ N; based on the N risk incident simulation record sets, an nth risk incident simulation record set corresponding to the nth risk incident simulation model is extracted; the dynamic perturbation distillation mechanism is activated, which includes a multimodal perturbation factor, comprising temporal perturbation, feature perturbation, and semantic perturbation; the nth risk incident simulation record set is perturbed multiple times according to the multimodal perturbation factor to obtain a temporal perturbation risk simulation. The system comprises a record set, a feature-based perturbation risk inference record set, and a semantic-based perturbation risk inference record set. Incremental learning is performed on the nth risk event inference model based on these three sets, respectively, to establish a first enhanced risk inference model, a second enhanced risk inference model, and a third enhanced risk inference model. Hierarchical distillation enhancement is then performed on these three enhanced models to generate the nth risk event inference channel.

[0012] Optionally, based on the accompanying user monitoring data, accident prediction is performed on the real-time accompanying users to obtain accompanying accident prediction results; based on the accompanying accident prediction results, accident triggering interference analysis is performed on the risk accident projection map to obtain accident triggering interference analysis results; based on the accompanying accident prediction results, accident propagation interference analysis is performed on the risk accident projection map to obtain accident propagation interference analysis results; based on the accompanying accident prediction results, accident accumulation interference analysis is performed on the risk accident projection map to obtain accident accumulation interference analysis results; based on the accident triggering interference analysis results, the accident propagation interference analysis results, and the accident accumulation interference analysis results, the risk accident projection map is corrected to obtain the optimized risk projection map.

[0013] Optionally, an accompanying early warning signal is generated based on the accompanying accident prediction results.

[0014] In a second aspect of the present application, a multi-device joint elderly data monitoring and early warning system is provided, comprising: a scene label generation module, configured to perform joint monitoring on a home-based elderly user through a multi-device monitoring array, obtain an elderly monitoring data set, and perform activity scene feature recognition based on the elderly monitoring data set to generate a user activity scene label; a matrix establishment module, configured to perform multi-dimensional abnormality analysis attention optimization on the elderly monitoring data set according to the user activity scene label, and establish a monitoring abnormality attention coupling matrix; a risk accident type fitting module, configured to perform risk accident type fitting according to the user activity scene label, and obtain N fitting risk accident types, N being a positive integer greater than 1; a risk accident deduction module, configured to perform multi-dimensional risk accident deduction on the monitoring abnormality attention coupling matrix according to a dynamic disturbance distillation mechanism based on the N fitting risk accident types, and obtain a risk accident deduction graph; and a monitoring and early warning signal generation module, configured to read corresponding accompanying user monitoring data of a real-time accompanying user, and perform accompanying accident interference analysis optimization on the risk accident deduction graph based on the accompanying user monitoring data to obtain a risk deduction optimization graph, and synchronously generate an elderly monitoring and early warning signal.

[0015] In a third aspect of the present application, an electronic device is provided, comprising: at least one processor; 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-mentioned multi-device joint elderly data monitoring and early warning method.

[0016] One or more technical solutions provided in the present application have at least the following technical effects or advantages: The method provided by the embodiment of the application can obtain a pension monitoring data set by monitoring the home pension user through a multi-device monitoring array, identify activity scene features based on the pension monitoring data set, generate a user activity scene label, perform multi-dimensional abnormality analysis attention optimization on the pension monitoring data set according to the user activity scene label, establish a monitoring abnormality attention coupling matrix, fit a risk accident type according to the user activity scene label, obtain N fitted risk accident types, N is a positive integer greater than 1, perform multi-dimensional risk accident deduction on the monitoring abnormality attention coupling matrix according to a dynamic disturbance distillation mechanism based on the N fitted risk accident types, obtain a risk accident deduction graph, read corresponding accompanying user monitoring data of a real-time accompanying user, and perform accompanying accident intervention analysis optimization on the risk accident deduction graph based on the accompanying user monitoring data, obtain a risk deduction optimization graph, and synchronously generate a pension monitoring early warning signal. Through the multi-device linkage monitoring and multi-dimensional risk deduction, the influence of the accompanying user is considered, and the technical effects of improving the accuracy and reliability of the accident monitoring and early warning of the home pension user are achieved.

[0017] The above description is only a summary of the technical solutions of the application. In order to more clearly understand the technical means of the application, the specific embodiments of the application can be implemented according to the content of the description, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the application, nor is it intended to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0019] Figure 1 A flowchart of a multi-device linkage pension data monitoring and early warning method provided by the application.

[0020] Figure 2 A structure diagram of a multi-device linkage pension data monitoring and early warning system provided by the application.

[0021] Figure 3 A structure diagram of an exemplary electronic device provided by the application.

[0022] Description of附图标记: Scene label generation module 11, matrix establishment module 12, risk accident type fitting module 13, risk accident deduction module 14, monitoring and early warning signal generation module 15, bus 300, receiver 301, processor 302, transmitter 303, memory 304, bus interface 305. Detailed implementation

[0023] This application provides a multi-device linkage-based elderly care data monitoring and early warning method, system and device, which are used to solve the technical problem that the existing elderly care monitoring is one-sided, resulting in insufficient accuracy and reliability of the monitoring and prediction of elderly care accidents. Through the linkage monitoring of multiple devices and multi-dimensional risk deduction, and considering the influence of accompanying users, the technical effect of improving the accuracy and reliability of accident monitoring and early warning for home-based elderly care users is achieved.

[0024] Next, the technical solutions in the present invention will be described clearly and completely with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited by the example embodiments described here. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention. Additionally, it should be noted that for the sake of description, only parts related to the present invention are shown in the accompanying drawings rather than all.

[0025] Embodiment 1, as Figure 1 shown, this application provides a multi-device linkage-based elderly care data monitoring and early warning method, and the multi-device linkage-based elderly care data monitoring and early warning method includes: Carry out linkage monitoring on home-based elderly care users through a multi-device monitoring array, obtain an elderly care monitoring data set, and generate user activity scene labels based on the elderly care monitoring data set for activity scene feature recognition.

[0026] Specifically, a multi-device monitoring array is constructed, which is composed of multiple different types of monitoring devices, including but not limited to wearable physiological monitoring devices, environmental sensors, and monitoring cameras, wherein the wearable physiological monitoring devices, such as smart bracelets, can collect physiological indicators such as heart rate, blood pressure, body temperature, and blood oxygen saturation in real time, the environmental sensors, such as temperature and humidity sensors and illumination intensity sensors, are used to collect information such as temperature, humidity, and illumination intensity of the home environment, and the monitoring cameras are used to collect user daily behavior actions, reflecting user life activity state, such as activity trajectory, sitting, standing, walking, lying, and other action types and activity frequency, etc. The multiple monitoring devices are connected to each other through wireless communication technology, such as wifi. The multi-device monitoring array is used to realize all-around and linkage monitoring of home-based elderly users, collect multiple data information including physiological monitoring data, environmental monitoring data, and behavior monitoring data, and form an elderly monitoring data set.

[0027] The elderly monitoring data set is analyzed to identify user activity scenes, and user activity scene labels are generated according to the identified user activity scene types. For example, a machine learning algorithm is used to analyze and identify the data set, a large amount of home-based elderly monitoring data is collected, including physiological monitoring data, environmental monitoring data, and behavior monitoring data under different activity scenes, noise and outliers in the data are removed, such as physiological data exceeding the range of human body and obviously unreasonable data in environmental data, and each data sample is labeled with a corresponding activity scene label, such as resting, eating, and exercising. The behavior data is processed to convert the dynamic information obtained by the camera into data features, including frequency, amplitude, and action type. The preprocessed data set is divided into a training set and a test set according to 7:3 or 8:2, and the training set is used to train a machine learning algorithm, such as a decision tree algorithm. In the training process, the split features and split points of the nodes are determined by calculating the information gain or Gini coefficient. The test set is used to evaluate the trained decision tree model, which can be evaluated by accuracy, recall rate, and F1 value. When the evaluation decision tree model meets the performance requirement, a well-constructed decision tree model is obtained, and the elderly monitoring data set is input into the decision tree model. The decision tree model classifies the input data according to the learned rules and patterns, identifies user activity scene features, and outputs corresponding user activity scene labels. Activity scene features refer to different types of activities performed by the elderly in a home environment, such as resting, preparing food, eating, and exercising. After identifying the activity scene features, each activity scene is labeled with a corresponding scene label, such as preparing food in the kitchen and resting in the bedroom.

[0028] In different activity scenarios, the physiological indicators, environmental needs and behavior patterns of users are different. For example, in the rest scenario, the physiological indicators such as heart rate and blood pressure of the user should be at a relatively stable level. If there is an abnormal fluctuation, there may be a health problem. In the exercise scenario, the physiological indicators may be elevated, which may be a normal phenomenon. By generating an accurate user activity scenario label, a more accurate user actual state is realized, the accuracy and reliability of the pension data monitoring and early warning are improved, and the safety and health of the home-based pension users are ensured.

[0029] According to the user activity scenario label, the multi-dimensional anomaly analysis attention optimization of the pension monitoring data set is performed, and a monitoring anomaly attention coupling matrix is established.

[0030] Further, according to the user activity scenario label, the multi-dimensional anomaly analysis attention optimization of the pension monitoring data set is performed, and a monitoring anomaly attention coupling matrix is established, including: according to the pension monitoring data set, data cleaning and feature recognition are performed to obtain a physiological monitoring feature sequence, an environmental monitoring feature sequence and a behavior monitoring feature sequence; according to the user activity scenario label, the physiological monitoring feature sequence is subjected to anomaly analysis attention optimization to generate a physiological anomaly attention coupling matrix; according to the user activity scenario label, the environmental monitoring feature sequence is subjected to anomaly analysis attention optimization to generate an environmental anomaly attention coupling matrix; according to the user activity scenario label, the behavior monitoring feature sequence is subjected to anomaly analysis attention optimization to generate a behavior anomaly attention coupling matrix; the physiological anomaly attention coupling matrix, the environmental anomaly attention coupling matrix and the behavior anomaly attention coupling matrix are aggregated to generate the monitoring anomaly attention coupling matrix.

[0031] Further, according to the user activity scenario label, the physiological monitoring feature sequence is subjected to anomaly analysis attention optimization to generate a physiological anomaly attention coupling matrix, including: based on the user activity scenario label, a normal physiological monitoring sample of the home-based pension user is searched to obtain a scenario normal physiological sample set; according to the scenario normal physiological sample set, multi-dimensional physiological feature set trend analysis is performed to construct a scenario normal physiological monitoring space; according to the scenario normal physiological monitoring space, the physiological monitoring feature sequence is subjected to anomaly identification to obtain a physiological anomaly identification result; according to the scenario normal physiological monitoring space, the physiological anomaly identification result is subjected to anomaly degree evaluation to obtain a physiological anomaly evaluation result; according to the physiological anomaly identification result and the physiological anomaly evaluation result, the physiological monitoring feature sequence is subjected to anomaly attention coupling optimization to obtain the physiological anomaly attention coupling matrix.

[0032] Specifically, the data cleaning is performed on the elderly monitoring dataset to remove noise and abnormal data, such as error values exceeding the normal physiological limit of the human body in the physiological monitoring data, and obviously unreasonable data in the environmental monitoring data. Through data cleaning, the data quality is ensured. Then, feature recognition is performed on the cleaned data based on time series to form physiological monitoring feature sequences, environmental monitoring feature sequences and behavior monitoring feature sequences. The physiological monitoring feature sequences are sequences formed by sorting the physiological data of the user collected in real time in time sequence, including heart rate, blood pressure, body temperature, blood oxygen saturation and other physiological indicators reflecting the health status of the user. The environmental monitoring feature sequences include temperature, humidity, light intensity and other environmental indicators reflecting the living environment of the user. The behavior monitoring feature sequences include action type, activity frequency and activity trajectory and other behavior information reflecting the activity mode and life regularity of the user.

[0033] According to the user activity scene label, the physiological monitoring feature sequence, the environmental monitoring feature sequence and the behavior monitoring feature sequence are respectively subjected to abnormal analysis attention optimization. In the process of abnormal analysis attention optimization of the physiological monitoring feature sequence according to the user activity scene label, first, the physiological data of the home-based elderly user is searched according to the user activity scene label. In the historical data, the normal physiological monitoring samples conforming to the user activity scene label are searched out in the activity scenes which have been labeled, such as rest, meal, exercise and the like. For example, when searching for normal physiological samples in the rest scene, the data records of the physiological indicators of the user in the rest scene, such as heart rate, blood pressure, body temperature and the like, are extracted, which are in the normal range and relatively stable, and a scene normal physiological sample set is formed.

[0034] The central tendency of the multi-dimensional physiological features of the scene normal physiological sample set is analyzed, and the mean, median and other statistical quantities of the multi-dimensional physiological samples are calculated. Taking the heart rate in the physiological sample set as an example, the mean heart rate of all normal samples under the user activity scene label is calculated as the central value of the heart rate in this scene. By analyzing the central tendency of each physiological feature in the scene normal physiological sample set, the normal fluctuation range of the user index in the scene can be reflected. The central values of the multi-dimensional physiological features are integrated to construct a scene normal physiological monitoring space, which is the integration of the normal range of multi-dimensional physiological features. For example, in a three-dimensional scene normal physiological monitoring space, the three dimensions represent heart rate, blood pressure and body temperature respectively, and the data points in the space correspond to the normal combination of physiological features in different scenes. Then the real-time collected physiological monitoring feature sequence is compared with the scene normal physiological monitoring space. If the physiological feature at a certain time point exceeds the range of the normal monitoring space in the corresponding scene, it is determined to be abnormal, and the physiological abnormality recognition result is obtained. For example, in the rest scene, the normal blood pressure range of the user is 90-120 mmHg for systolic pressure and 60-80 mmHg for diastolic pressure, and at a certain time, the systolic pressure is monitored to be 130 mmHg and the diastolic pressure is 85 mmHg, which exceeds the normal range, and is identified as abnormal.

[0035] According to the scene normal physiological monitoring space, the abnormal deviation degree index of the physiological abnormality recognition result is obtained. For each physiological feature identified as abnormal, the deviation value of its actual value from the normal range in the scene normal physiological monitoring space is calculated. Combined with the abnormal duration, the abnormal degree evaluation of the physiological abnormality recognition result is weighted. If the abnormal duration is long, it means that the abnormality has a greater impact on the user's health, and a time weight coefficient is set. For example, in the rest scene, the normal blood pressure range of the user is 90-120 mmHg for systolic pressure and 60-80 mmHg for diastolic pressure, and at a certain time, the systolic pressure is monitored to be 130 mmHg and the diastolic pressure is 85 mmHg. The systolic pressure deviates from the normal upper limit by 10 mmHg, and the diastolic pressure deviates from the normal upper limit by 5 mmHg. The blood pressure is abnormal for 5 minutes, and the time weight coefficient is 0.1. The weighted deviation degree of systolic pressure is 10x(1+0.1x5)=15, and the weighted deviation degree of diastolic pressure is 5x(1+0.1x5)=7.5. When the abnormal duration increases, the time weight coefficient also increases accordingly. Different physiological features have different impacts on health, for example, heart rate abnormality is more urgent than body temperature abnormality. According to medical knowledge and clinical experience, important weights are assigned to each physiological feature, for example, heart rate weight 0.4, blood pressure weight 0.4, and body temperature weight 0.2. The weighted deviation degree and the importance weight are multiplied to obtain the physiological abnormality evaluation result corresponding to the physiological abnormality recognition result, for example, for systolic pressure, the abnormality evaluation degree score=15x0.4=6, and the diastolic pressure abnormality degree score=7.5x0.4=3.

[0036] According to the physiological abnormality recognition result and the physiological abnormality evaluation result, the physiological monitoring sequence is subjected to abnormal attention coupling optimization, which refers to assigning different attention weights to different physiological characteristics in the physiological monitoring feature sequence according to the type and severity of the abnormality. For abnormal features with high severity, higher attention weights are assigned. The attention mechanism is used to dynamically adjust the attention degree of each abnormality according to the recognition result and the evaluation result of each abnormality, and to obtain abnormal attention weights. The abnormal attention weights corresponding to each physiological feature are sorted according to their order in the physiological monitoring feature sequence to form a physiological abnormality attention coupling matrix. Each element in the physiological abnormality attention coupling matrix represents the attention weight of the corresponding physiological feature in the abnormal monitoring. The higher the weight, the higher the attention degree of the feature.

[0037] Similarly, according to the user activity scene label, the environment monitoring feature sequence and the behavior monitoring feature sequence are subjected to normal environment monitoring sample retrieval and normal behavior monitoring sample retrieval, respectively, and centralized trend analysis is performed to construct a scene normal environment monitoring space and a scene normal behavior monitoring space. Abnormality recognition and abnormality degree evaluation are performed on the environment monitoring feature sequence and the behavior monitoring feature sequence, and abnormal attention coupling optimization is performed to obtain an environment abnormal attention coupling matrix and a behavior abnormal attention coupling matrix, respectively.

[0038] According to historical data statistics or expert experience, a weight is assigned to each abnormal attention coupling matrix to reflect the importance of different types of abnormalities in monitoring. For example, physiological abnormalities have a greater impact on the health of elderly users, so the weight of the physiological abnormality attention coupling matrix is set higher, such as 0.4, and the weights of the environment abnormality and behavior abnormality are set to 0.3, respectively. According to the assigned weights, the physiological abnormality attention coupling matrix, the environment abnormality attention coupling matrix, and the behavior abnormality attention coupling matrix are subjected to weighted summation to generate a monitoring abnormality attention coupling matrix. The monitoring abnormality attention coupling matrix integrates abnormal information in the physiological, environmental, and behavioral dimensions, and can comprehensively and accurately reflect the abnormal conditions of the user in different scenes.

[0039] Through multi-dimensional abnormality analysis and integration, false positives or false negatives caused by single-dimensional analysis are avoided, more accurate identification of abnormal conditions of the user in different activity scenes is achieved, and the comprehensiveness and accuracy of the evaluation of the user's health and safety status are improved to ensure the health and safety of the home-based elderly users.

[0040] According to the user activity scene label, risk accident type fitting is performed to obtain N fitted risk accident types, where N is a positive integer greater than 1.

[0041] Furthermore, risk incident types are fitted based on the user activity scenario tags to obtain N fitted risk incident types, including: risk incident retrieval based on the user activity scenario tags to obtain a scenario-matching historical risk incident set; incident type identification based on the scenario-matching historical risk incident set to obtain an incident type sample set; support calculation for each incident type sample in the incident type sample set based on the scenario-matching historical risk incident set to obtain the support of each incident type; confidence calculation for each incident type sample based on the support of each incident type to obtain the confidence of each incident type; and optimization identification of the incident type sample set based on the predetermined type confidence and the confidence of each incident type to generate the N fitted risk incident types.

[0042] Specifically, based on user activity scenario tags, database retrieval technology is used to search for historical risk accident records that match the user activity scenario tags in a pre-built database containing a large amount of historical risk accident data, thus obtaining a scenario-matched historical risk accident set. For example, in a kitchen scenario, accident data related to falls and burns is retrieved. The scenario-matched historical risk accident set is then categorized to extract different accident types, such as falls, strokes, and heart attacks. These extracted accident types are then classified to generate an accident type sample set, where each sample represents a specific type of accident.

[0043] Support is calculated for each accident type sample in the historical risk accident set for scene matching. Support refers to the number of times a specific accident type appears in the historical risk accident set. The support for each accident type is obtained by statistically analyzing the occurrence counts of multiple accident types in the historical risk accident set for scene matching. For example, in the historical risk accident set for the cooking scene, a fall accident occurred 5 times, so the support for the fall accident type is 5. The confidence score is calculated using the formula: Accident Type Confidence Score = Accident Type Support Score ÷ Sum of Supports for All Accident Types. The confidence score for each accident type sample is calculated based on the support scores of each accident type. For example, in the historical risk accident set for the cooking scene, which includes fall accidents and burn accidents, the support for fall accidents is 5, and the support for burn accidents is 3. Therefore, the confidence score for fall accidents is 5 ÷ (5 + 3) = 0.625, and the confidence score for burn accidents is 3 ÷ (5 + 3) = 0.375.

[0044] The support of each type of accident in the historical accident data is counted, the probability of each type of accident occurring in different scenarios is calculated, the predetermined type confidence is set based on the probability and the requirement of safety risk tolerance, which is used to distinguish high-risk and low-risk accident types. According to the predetermined type confidence, the accident type sample set is screened and optimized to identify, that is, the confidence of each accident type is compared with the predetermined type confidence, the accident types with confidence greater than or equal to the predetermined type confidence in the accident type sample set are selected as N fitting risk accident types, and N is a positive integer greater than 1. The N fitting risk accident types refer to the most likely accidents in a specific user activity scenario.

[0045] Through support calculation, confidence calculation and optimization, high-risk accident types related to a specific activity scenario are extracted from historical data, and then accurate identification of potential risk factors in different user activity scenarios is realized, real-time and accurate safety warnings are provided for users, and more accurate safety protection is provided for home-based elderly users.

[0046] Based on the N fitting risk accident types, the multi-dimensional risk accident deduction of the monitoring anomaly attention coupling matrix is performed according to the dynamic disturbance distillation mechanism, and a risk accident deduction graph is obtained.

[0047] Further, based on the N fitting risk accident types, the multi-dimensional risk accident deduction of the monitoring anomaly attention coupling matrix is performed according to the dynamic disturbance distillation mechanism, and a risk accident deduction graph is obtained, including: according to the N fitting risk accident types, risk accident deduction record retrieval is performed to obtain an N risk accident deduction record set; according to the N risk accident deduction record set, an accident tree training is performed to obtain N risk accident deduction models; according to the dynamic disturbance distillation mechanism, the N risk accident deduction models are dynamically disturbed and distilled to strengthen, and N risk accident deduction channels are established; the monitoring anomaly attention coupling matrix is input into the N risk accident deduction channels to obtain N risk accident deduction paths; the N risk accident deduction paths are sorted to generate the risk accident deduction graph.

[0048] Specifically, according to N fitted risk accident types, risk accident deduction record retrieval is performed in a large amount of pre-stored risk accident deduction record data, and N risk accident deduction record sets are obtained, which contain detailed information about the occurrence of historical accident types related to the fitted risk accident types, including occurrence time, environment, and user health status, etc. The N risk accident deduction records are used as a training set, and the fault tree analysis method is used for fault tree training. The fault tree analysis method is a method for analyzing the possibility and consequences of various events through logical relationships, which can identify the root cause and trigger condition of an accident in a specific scenario. During the training process, the key information in each risk accident deduction record set is first extracted, including accident type, occurrence condition, trigger factor, and result, etc., to construct a preliminary fault tree framework. By using the fault tree analysis (FTA) method, the high-risk nodes of each accident occurrence are analyzed by logically associating the conditions and events of the accident. Based on the N risk accident deduction record sets, machine learning algorithms such as decision trees or random forests are used to train the N accident tree models, adjust the weights and connection relationships of each node, and form N risk accident deduction models. The N risk accident deduction models are used to simulate the occurrence path of different accident types in a specific scenario and environment.

[0049] According to the dynamic disturbance distillation mechanism, the N risk accident deduction models are dynamically disturbed and distilled to strengthen them. The dynamic disturbance distillation mechanism refers to introducing random disturbances to optimize the N risk accident deduction models and improve their generalization ability. During the strengthening process, different intensity and type of random disturbances are added to each risk accident deduction model, and the disturbed risk accident deduction models are fused through distillation technology to obtain more stable and accurate deduction models as N risk accident deduction channels. The monitoring anomaly attention coupling matrix is input into the N risk accident deduction channels for analysis, and N risk accident deduction paths are output, each representing a possible development process of a certain risk accident under specific monitoring data. For example, if the user's heart rate is abnormal and is in the kitchen scenario, a risk path of falling may be deduced. The obtained N risk accident deduction paths are sorted according to their importance and relevance, and a graphical tool is used to generate a risk accident deduction map. Each path in the risk accident deduction map represents a potential risk evolution process, which comprehensively and intuitively reflects the possible risk accidents, their occurrence order and relevance of the user under different monitoring data and scenarios, providing comprehensive and effective data support for user risk warning and intervention, and thus ensuring the safety of home-based elderly users.

[0050] Further, the N risk accident deduction models are dynamically disturbed and distilled for reinforcement according to the dynamic disturbance distillation mechanism, to establish N risk accident deduction channels, including: according to the N risk accident deduction models, extracting an nth risk accident deduction model, n being a positive integer, 1≤n≤N; according to the N risk accident deduction record sets, extracting an nth risk accident deduction record set corresponding to the nth risk accident deduction model; activating the dynamic disturbance distillation mechanism, the dynamic disturbance distillation mechanism including multi-modal disturbance factors, the multi-modal disturbance factors including time sequence disturbance, feature disturbance and semantic disturbance; performing multi-round disturbance on the nth risk accident deduction record set according to the multi-modal disturbance factors, to obtain a time sequence disturbance risk deduction record set, a feature disturbance risk deduction record set and a semantic disturbance risk deduction record set; respectively performing incremental learning on the nth risk accident deduction model according to the time sequence disturbance risk deduction record set, the feature disturbance risk deduction record set and the semantic disturbance risk deduction record set, to establish a risk deduction first enhanced model, a risk deduction second enhanced model and a risk deduction third enhanced model; performing hierarchical distillation reinforcement according to the risk deduction first enhanced model, the risk deduction second enhanced model and the risk deduction third enhanced model, to generate an nth risk accident deduction channel.

[0051] Specifically, an nth risk accident deduction model is sequentially extracted from the N risk accident deduction models, where n is a positive integer, 1≤n≤N. According to the extracted nth risk accident deduction model, a corresponding nth risk accident deduction record set is extracted from the N risk accident deduction record sets. The dynamic disturbance distillation mechanism is activated, and the dynamic disturbance distillation mechanism contains multi-modal disturbance factors, including time sequence disturbance, feature disturbance and semantic disturbance. The time sequence disturbance refers to changing the time sequence information in the risk accident deduction record, and the model's risk changes in different time periods can be disturbed by a preset time change rule. The feature disturbance refers to changing the monitoring features in the risk accident deduction record, such as physiological data, environmental data or behavior data, simulating the influence of physiological, environmental or behavior changes on the occurrence of accidents, and modifying the features according to a preset feature adjustment strategy. The semantic disturbance refers to interfering with the semantic information in the risk accident deduction record, for example, modifying the semantic expression of the accident description to improve the deduction ability of the deduction model under different semantic understandings, and transforming the semantics using natural language processing technology.

[0052] According to the time sequence disturbance, feature disturbance and semantic disturbance in the multi-modal disturbance factor, the n-th risk accident deduction record set is disturbed for multiple rounds, and the disturbed data is arranged to form a time sequence disturbance risk deduction record set, a feature disturbance risk deduction record set and a semantic disturbance risk deduction record set. Through an incremental learning algorithm, the n-th risk accident deduction model is incrementally learned by using the time sequence disturbance risk deduction record set, the feature disturbance risk deduction record set and the semantic disturbance risk deduction record set. Incremental learning refers to a method of inputting new data to update and optimize the model on the basis of the n-th risk accident deduction model. In the incremental learning process, the n-th risk accident deduction model processes new information in the time sequence disturbance risk deduction record set, the feature disturbance risk deduction record set and the semantic disturbance risk deduction record set after disturbance, realizes optimization of the n-th risk accident deduction model, and establishes a risk deduction first enhanced model, a risk deduction second enhanced model and a risk deduction third enhanced model. The risk deduction first enhanced model, the risk deduction second enhanced model and the risk deduction third enhanced model correspond to optimization of the n-th risk accident deduction model by the time sequence disturbance risk deduction record set, the feature disturbance risk deduction record set and the semantic disturbance risk deduction record set, respectively.

[0053] The knowledge distillation technology is used to strengthen the hierarchical distillation of the risk deduction first enhanced model, the risk deduction second enhanced model and the risk deduction third enhanced model. The knowledge distillation technology is a technology of fusing the knowledge of multiple models, extracts useful deduction knowledge from the risk deduction first enhanced model, the risk deduction second enhanced model and the risk deduction third enhanced model, including the path of independent deduction of each model and the output result under different disturbance factors, and defines soft labels. Through the teacher and student model mode, the risk deduction first enhanced model, the risk deduction second enhanced model and the risk deduction third enhanced model are used as teacher models, and a target deduction model is used as a learning model. The learning model learns the soft labels of the teacher model, minimizes the difference between the soft labels of the teacher model output by the student model, and improves the deduction ability. Through hierarchical distillation strengthening, the knowledge of the risk deduction first enhanced model, the risk deduction second enhanced model and the risk deduction third enhanced model is compressed and integrated to generate an n-th risk accident deduction channel with higher deduction efficiency and accuracy. The above steps are repeated to obtain N risk accident deduction channels.

[0054] Through dynamic disturbance and knowledge distillation strengthening, the adaptability of the risk accident deduction channel to changing environment and risk scene is improved, the risk prediction accuracy is improved, and then through processing of the old-age monitoring data, the risk is quickly and accurately evaluated and early warning is provided, various potential safety risks are effectively prevented and coped with, and the home safety of the old users is ensured.

[0055] read the real-time accompanying user corresponding accompanying user monitoring data, and based on the accompanying user monitoring data, the risk accident deduction graph is intervened and analyzed and optimized, the risk deduction optimization graph is obtained, and the old-age monitoring early warning signal is generated synchronously.

[0056] Further, based on the accompanying user monitoring data, the risk accident deduction graph is intervened and analyzed and optimized to obtain a risk deduction optimization graph, including: according to the accompanying user monitoring data, the real-time accompanying user is accident predicted to obtain an accompanying accident prediction result; according to the accompanying accident prediction result, the risk accident deduction graph is intervened and analyzed for accident triggering to obtain an accident triggering intervention analysis result; according to the accompanying accident prediction result, the risk accident deduction graph is intervened and analyzed for accident propagation to obtain an accident propagation intervention analysis result; according to the accompanying accident prediction result, the risk accident deduction graph is intervened and analyzed for accident accumulation to obtain an accident accumulation intervention analysis result; according to the accident triggering intervention analysis result, the accident propagation intervention analysis result and the accident accumulation intervention analysis result, the risk accident deduction graph is corrected to obtain the risk deduction optimization graph.

[0057] Specifically, the monitoring data of the accompanying user is read in real time, and the accompanying user refers to a population living together with the home-based elderly user or generated, and the monitoring data of the accompanying user includes physiological index data, environment data and behavior activity data. According to the accompanying user monitoring data, the real-time accompanying user is accident predicted to obtain an accompanying accident prediction result. For example, the historical monitoring data of the accompanying user and the corresponding accident label are obtained, and the historical monitoring data is divided into a training set and a validation set. The neural network model is trained through the training set and the corresponding accident label. In the training process, the weights and biases in the network are continuously adjusted through the back propagation algorithm to minimize the error between the accompanying accident prediction result and the true label, and the trained neural network is evaluated and adjusted using the validation set. When the performance of the neural network model on the validation set no longer improves, the training is stopped. The accompanying user monitoring data is input into the trained neural network model for accident prediction to obtain an accompanying accident prediction result. The accompanying accident prediction result is the type and probability of abnormal events that the accompanying user may occur in the future, such as falling.

[0058] According to the accident triggering intervention analysis result of the risk accident deduction graph, it is analyzed whether the behavior or health change of the accompanying user triggers the accident risk of the home-based elderly user. For example, when the accompanying user falls, the heart rate of the home-based elderly user fluctuates sharply, affecting the health status of the home-based elderly user. By using time series analysis, the risk degree and time relationship of the accompanying accident prediction are calculated by matching the accompanying accident prediction result with the risk accident deduction graph, and it is evaluated whether the accident of the home-based elderly user will be triggered, and the accident triggering intervention analysis result is obtained. At the same time, by analyzing the historical accident data and the risk accident deduction graph, the propagation path of the accompanying accident to the accident of the home-based elderly user is established, which means that the abnormal behavior of the accompanying user gradually expands to the health or behavior state of the home-based elderly user. Based on the accompanying accident prediction result, the accident propagation intervention analysis of the risk accident deduction graph is performed, the possibility of different risk propagation paths is calculated, and the influence of the accompanying accident on the propagation of the accident of the home-based elderly user is evaluated. For example, the accompanying accident may increase the risk of falling of the home-based elderly user. Further analysis of the risk accumulation effect of the accompanying accident on the home-based elderly user, long-term or multiple accompanying accidents may have a cumulative effect, increasing the risk of the home-based elderly user. According to the accident accumulation intervention analysis result of the risk accident deduction graph, the cumulative weighting of multiple related nodes in the deduction graph is performed, the cumulative effect of the accompanying accident is identified, and the accident accumulation intervention analysis result is obtained.

[0059] According to the accident triggering intervention analysis result, the accident propagation intervention analysis result and the accident accumulation intervention analysis result, the risk accident deduction graph is corrected. The accident triggering intervention analysis result identifies which accompanying accidents may directly or indirectly trigger the risk of the home-based elderly user, and the risk nodes in the risk accident deduction graph are adjusted according to the accident triggering intervention analysis result. According to the propagation path in the accident propagation intervention analysis result, the propagation weight of the related nodes in the risk accident deduction graph is updated. The accident accumulation intervention analysis result analyzes the superposition effect of the accompanying accident, and the risk weight of the related risk nodes is increased according to the cumulative effect. Through the three kinds of triggering intervention analysis results, the accident nodes, paths and weights in the risk accident deduction graph are corrected, and the risk deduction optimization graph is generated. When the risk deduction optimization graph shows that there is an accident risk exceeding the threshold value, the related home-based monitoring warning signals are generated, and the related personnel are notified by means of short message, APP push, email, etc.

[0060] By analyzing the monitoring data of the accompanying user and optimizing the risk accident deduction graph, the risk accident deduction graph is more in line with the actual situation, and the comprehensiveness, accuracy and timeliness of the risk accident prediction in the home-based process are improved. The generation of the home-based monitoring warning signal in synchronization can timely notify the relevant personnel to take measures, effectively reduce the harm caused by the risk accident to the user, ensure the safety and health of the user, and improve the quality and safety of the home-based service.

[0061] Further, according to the predicted result of the accompanying accident, an accompanying warning signal is generated.

[0062] Specifically, the accident prediction of the real-time accompanying user is carried out according to the accompanying user monitoring data. After obtaining the predicted result of the accompanying accident, the occurrence probabilities of multiple risk events in the predicted result of the accompanying accident are compared with a preset threshold. When the occurrence probability of the risk event is greater than or equal to the prediction threshold, an accompanying warning signal is generated, and preventive measures are taken before the risk occurs to improve the safety of the home care environment, reduce potential accidents, and ensure the safety and health of the elderly.

[0063] Embodiment 2, based on the same inventive concept as a multi-device linkage-based elderly care data monitoring and warning method in the foregoing embodiment, as Figure 2 shown, the present application provides a multi-device linkage-based elderly care data monitoring and warning system, wherein the multi-device linkage-based elderly care data monitoring and warning system includes: A scenario label generation module 11, configured to perform linkage monitoring on home care users through a multi-device monitoring array to obtain an elderly care monitoring data set, and perform activity scenario feature recognition based on the elderly care monitoring data set to generate user activity scenario labels; a matrix establishment module 12, configured to perform multi-dimensional abnormal analysis attention optimization on the elderly care monitoring data set according to the user activity scenario labels to establish a monitoring abnormal attention coupling matrix; a risk accident type fitting module 13, configured to perform risk accident type fitting according to the user activity scenario labels to obtain N fitting risk accident types, where N is a positive integer greater than 1; a risk accident deduction module 14, configured to perform multi-dimensional risk accident deduction on the monitoring abnormal attention coupling matrix based on the N fitting risk accident types according to a dynamic perturbation distillation mechanism to obtain a risk accident deduction map; a monitoring warning signal generation module 15, configured to read the accompanying user monitoring data corresponding to the real-time accompanying user, and perform accompanying accident interference analysis optimization on the risk accident deduction map based on the accompanying user monitoring data to obtain a risk deduction optimization map, and synchronously generate an elderly care monitoring warning signal.

[0064] Furthermore, the matrix building module 12 is also used for: performing data cleaning and feature recognition based on the elderly care monitoring dataset to obtain physiological monitoring feature sequences, environmental monitoring feature sequences, and behavioral monitoring feature sequences; performing anomaly analysis and attention optimization on the physiological monitoring feature sequences based on the user activity scene labels to generate a physiological anomaly attention coupling matrix; performing anomaly analysis and attention optimization on the environmental monitoring feature sequences based on the user activity scene labels to generate an environmental anomaly attention coupling matrix; performing anomaly analysis and attention optimization on the behavioral monitoring feature sequences based on the user activity scene labels to generate a behavioral anomaly attention coupling matrix; and aggregating the physiological anomaly attention coupling matrix, the environmental anomaly attention coupling matrix, and the behavioral anomaly attention coupling matrix to generate the monitoring anomaly attention coupling matrix.

[0065] Furthermore, the matrix establishment module 12 is also used for: retrieving normal physiological monitoring samples of the home-based elderly care user based on the user activity scene tags to obtain a scene normal physiological sample set; performing multi-dimensional physiological feature central trend analysis based on the scene normal physiological sample set to construct a scene normal physiological monitoring space; performing anomaly identification on the physiological monitoring feature sequence based on the scene normal physiological monitoring space to obtain physiological anomaly identification results; evaluating the degree of anomaly based on the physiological anomaly identification results based on the scene normal physiological monitoring space to obtain physiological anomaly evaluation results; and performing anomaly attention coupling optimization on the physiological monitoring feature sequence based on the physiological anomaly identification results and the physiological anomaly evaluation results to obtain the physiological anomaly attention coupling matrix.

[0066] Furthermore, the risk incident type fitting module 13 is also used for: performing risk incident retrieval based on the user activity scenario tags to obtain a scenario-matching historical risk incident set; performing incident type identification based on the scenario-matching historical risk incident set to obtain an incident type sample set; calculating the support of each incident type sample in the incident type sample set based on the scenario-matching historical risk incident set to obtain the support of each incident type; calculating the confidence of each incident type sample based on the support of each incident type to obtain the confidence of each incident type; and performing optimization identification on the incident type sample set based on the predetermined type confidence and the confidence of each incident type to generate the N fitted risk incident types.

[0067] Further, the risk accident deduction module 14 is further configured to: perform risk accident deduction record retrieval according to the N fitted risk accident types, to obtain N risk accident deduction record sets; perform fault tree training according to the N risk accident deduction record sets, to obtain N risk accident deduction models; perform dynamic perturbation distillation reinforcement on the N risk accident deduction models according to the dynamic perturbation distillation mechanism, to establish N risk accident deduction channels; input the monitoring abnormality attention coupling matrix into the N risk accident deduction channels, to obtain N risk accident deduction paths; and collate the N risk accident deduction paths, to generate the risk accident deduction graph.

[0068] Further, the risk accident deduction module 14 is further configured to: extract an nth risk accident deduction model according to the N risk accident deduction models, n being a positive integer and 1≤n≤N; extract an nth risk accident deduction record set corresponding to the nth risk accident deduction model according to the N risk accident deduction record sets; activate the dynamic perturbation distillation mechanism, the dynamic perturbation distillation mechanism including a multi-modal perturbation factor, the multi-modal perturbation factor including a time sequence perturbation, a feature perturbation and a semantic perturbation; perform multi-round perturbation on the nth risk accident deduction record set according to the multi-modal perturbation factor, to obtain a time sequence perturbation risk deduction record set, a feature perturbation risk deduction record set and a semantic perturbation risk deduction record set; perform incremental learning on the nth risk accident deduction model according to the time sequence perturbation risk deduction record set, the feature perturbation risk deduction record set and the semantic perturbation risk deduction record set respectively, to establish a risk deduction first enhanced model, a risk deduction second enhanced model and a risk deduction third enhanced model; and perform hierarchical distillation reinforcement according to the risk deduction first enhanced model, the risk deduction second enhanced model and the risk deduction third enhanced model, to generate an nth risk accident deduction channel.

[0069] Further, the monitoring and early warning signal generation module 15 is further configured to: perform accident prediction on the real-time accompanying user according to the accompanying user monitoring data, to obtain an accompanying accident prediction result; perform accident trigger interference analysis on the risk accident deduction graph according to the accompanying accident prediction result, to obtain an accident trigger interference analysis result; perform accident propagation interference analysis on the risk accident deduction graph according to the accompanying accident prediction result, to obtain an accident propagation interference analysis result; perform accident accumulation interference analysis on the risk accident deduction graph according to the accompanying accident prediction result, to obtain an accident accumulation interference analysis result; and correct the risk accident deduction graph according to the accident trigger interference analysis result, the accident propagation interference analysis result and the accident accumulation interference analysis result, to obtain the risk deduction optimized graph.

[0070] Further, the monitoring and early warning signal generation module 15 is further configured to generate a companion early warning signal according to the companion accident prediction result.

[0071] In the third embodiment, based on the same inventive concept as the method for monitoring and early warning of the elderly data in the preceding embodiments, the present application further provides an electronic device, comprising: at least one processor; a memory in communication connection with 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 method for monitoring and early warning of the elderly data in any one of the first embodiment.

[0072] The accompanying drawings are included to provide a further understanding of the present application, and are incorporated herein and constitute a part of the detailed description. The embodiments illustrated in the drawings are presented using only for explanatory purposes. It should be apparent that the application can be practiced without such specific details. Figure 3 The structure of the exemplary electronic device of the present application is shown in the accompanying drawings. In the drawings: Figure 3 The bus architecture is represented by a bus 300, which can include any number of interconnecting buses and bridges, the bus 300 connecting various circuitry including one or more processors represented by processor 302 and memory represented by memory 304. The bus 300 can also connect various other circuitry, such as peripheral devices, voltage stabilizers, and power management circuitry, which are well known in the art and will not be described further. A bus interface 305 provides an interface between the bus 300 and a receiver 301 and a transmitter 303. The receiver 301 and the transmitter 303 can be the same element, i.e., a transceiver, which provides a means for communicating with various other apparatuses over a transmission medium. The processor 302 is responsible for managing the bus 300 and general processing, while the memory 304 can be used for storing data used by the processor 302 in executing operations.

[0073] The above description of disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0074] Obviously, for those skilled in the art, without departing from the principles of the present application, the present application can be improved and modified in several ways, and these improvements and modifications also fall within the scope of protection of the present application.

Claims

1. A method for monitoring and early warning of elderly care data involving multiple devices, characterized in that, The method includes: By using a multi-device monitoring array to conduct joint monitoring of home-based elderly care users, an elderly care monitoring dataset is obtained, and activity scene features are identified based on the elderly care monitoring dataset to generate user activity scene tags. Based on the user activity scenario tags, multidimensional anomaly analysis and attention optimization are performed on the elderly care monitoring dataset to establish a monitoring anomaly attention coupling matrix. Based on the user activity scenario tags, risk incident types are fitted to obtain N fitted risk incident types, where N is a positive integer greater than 1; Based on the N fitted risk accident types, a multi-dimensional risk accident deduction is performed on the monitoring anomaly attention coupling matrix according to the dynamic perturbation distillation mechanism to obtain a risk accident deduction map. Read the real-time accompanying user monitoring data corresponding to the accompanying user, and perform accompanying accident interference analysis and optimization on the risk accident projection map based on the accompanying user monitoring data to obtain the risk projection optimization map, and simultaneously generate elderly care monitoring early warning signals.

2. The method for monitoring and early warning of elderly care data through multi-device linkage as described in claim 1, characterized in that, Based on the user activity scenario tags, multidimensional anomaly analysis and attention optimization are performed on the elderly care monitoring dataset to establish a monitoring anomaly attention coupling matrix, including: Data cleaning and feature recognition were performed on the elderly care monitoring dataset to obtain physiological monitoring feature sequences, environmental monitoring feature sequences, and behavioral monitoring feature sequences. Based on the user activity scenario tags, the physiological monitoring feature sequence is subjected to anomaly parsing and attention optimization to generate a physiological anomaly attention coupling matrix. Based on the user activity scene tags, the environmental monitoring feature sequence is subjected to anomaly parsing and attention optimization to generate an environmental anomaly attention coupling matrix. Based on the user activity scenario tags, the behavior monitoring feature sequence is subjected to anomaly parsing and attention optimization to generate a behavior anomaly attention coupling matrix. The physiological abnormality attention coupling matrix, the environmental abnormality attention coupling matrix, and the behavioral abnormality attention coupling matrix are aggregated to generate the monitoring abnormality attention coupling matrix.

3. The method for monitoring and early warning of elderly care data through multi-device linkage as described in claim 2, characterized in that, Based on the user activity scenario tags, anomaly analysis and attention optimization are performed on the physiological monitoring feature sequence to generate a physiological anomaly attention coupling matrix, including: Based on the user activity scenario tags, normal physiological monitoring samples of the home-based elderly care users are retrieved to obtain a scenario normal physiological sample set. Based on the normal physiological sample set of the scenario, perform multidimensional physiological feature central trend analysis to construct a normal physiological monitoring space for the scenario; Based on the normal physiological monitoring space of the scenario, anomaly identification is performed on the physiological monitoring feature sequence to obtain physiological anomaly identification results; The degree of abnormality of the physiological abnormality identification result is evaluated based on the normal physiological monitoring space of the scenario to obtain the physiological abnormality evaluation result. Based on the physiological abnormality identification results and the physiological abnormality evaluation results, the physiological monitoring feature sequence is optimized for abnormal attention coupling to obtain the physiological abnormality attention coupling matrix.

4. The method for monitoring and early warning of elderly care data through multi-device linkage as described in claim 1, characterized in that, Based on the user activity scenario tags, risk incident types are fitted to obtain N fitted risk incident types, including: Based on the user activity scenario tags, risk incident retrieval is performed to obtain a set of historical risk incidents matching the scenario. Based on the scenario, historical risk incident sets are matched to identify incident types and obtain an incident type sample set. Based on the historical risk accident set matched to the scenario, the support of each accident type sample in the accident type sample set is calculated to obtain the support of each accident type. Calculate the confidence score for each accident type sample based on the support score for each accident type to obtain the confidence score for each accident type; Based on the predetermined type confidence level, the accident type sample set is optimized and identified according to the confidence level of each accident type to generate the N fitted risk accident types.

5. The method for monitoring and early warning of elderly care data through multi-device linkage as described in claim 1, characterized in that, Based on the N fitted risk incident types, a multi-dimensional risk incident extrapolation is performed on the monitoring anomaly attention coupling matrix according to the dynamic perturbation distillation mechanism to obtain a risk incident extrapolation map, including: Based on the N fitted risk accident types, risk accident simulation records are retrieved to obtain N risk accident simulation record sets. Based on the N risk incident simulation record sets, an incident tree is trained to obtain N risk incident simulation models; The N risk incident simulation models are enhanced by dynamic perturbation distillation based on the dynamic perturbation distillation mechanism to establish N risk incident simulation channels. Input the monitoring anomaly attention coupling matrix into the N risk incident deduction channels to obtain N risk incident deduction paths; Organize the N risk incident simulation paths to generate the risk incident simulation map.

6. The method for monitoring and early warning of elderly care data through multi-device linkage as described in claim 5, characterized in that, Based on the aforementioned dynamic perturbation distillation mechanism, the N risk incident simulation models are enhanced by dynamic perturbation distillation to establish N risk incident simulation channels, including: Based on the N risk incident simulation models, extract the nth risk incident simulation model, where n is a positive integer, 1≤n≤N; Based on the N risk incident simulation record sets, extract the nth risk incident simulation record set corresponding to the nth risk incident simulation model; The dynamic perturbation distillation mechanism is activated, which includes a multimodal perturbation factor, comprising temporal perturbation, feature perturbation, and semantic perturbation. The nth risk event simulation record set is perturbed multiple times according to the multimodal perturbation factor to obtain the temporal perturbation risk simulation record set, the feature perturbation risk simulation record set, and the semantic perturbation risk simulation record set; Based on the time-series perturbation risk simulation record set, the feature perturbation risk simulation record set, and the semantic perturbation risk simulation record set, incremental learning is performed on the nth risk accident simulation model to establish a first enhanced risk simulation model, a second enhanced risk simulation model, and a third enhanced risk simulation model. Based on the first enhanced risk simulation model, the second enhanced risk simulation model, and the third enhanced risk simulation model, hierarchical distillation enhancement is performed to generate the nth risk accident simulation channel.

7. The method for monitoring and early warning of elderly care data through multi-device linkage as described in claim 1, characterized in that, Based on the accompanying user monitoring data, the risk incident projection map is optimized by accompanying incident interference analysis to obtain an optimized risk projection map, including: Based on the accompanying user monitoring data, an accident prediction is performed on the real-time accompanying user to obtain the accompanying accident prediction result. Based on the accompanying accident prediction results, the risk accident projection map is subjected to accident triggering interference analysis to obtain the accident triggering interference analysis results. Based on the accompanying accident prediction results, the accident propagation interference analysis is performed on the risk accident projection map to obtain the accident propagation interference analysis results; Based on the accompanying accident prediction results, the risk accident projection map is subjected to accident cumulative interference analysis to obtain the accident cumulative interference analysis results; Based on the accident triggering interference analysis results, the accident propagation interference analysis results, and the accident accumulation interference analysis results, the risk accident projection map is revised to obtain the optimized risk projection map.

8. The method for monitoring and early warning of elderly care data through multi-device linkage as described in claim 7, characterized in that, Based on the predicted results of the accompanying accidents, an accompanying early warning signal is generated.

9. A pension data monitoring and early warning system with multi-device linkage, characterized in that, The steps for implementing the multi-device linkage elderly care data monitoring and early warning method according to any one of claims 1 to 8 include: The scene tag generation module is used to conduct joint monitoring of home-based elderly care users through a multi-device monitoring array, obtain elderly care monitoring dataset, and generate user activity scene tags based on the elderly care monitoring dataset to identify activity scene features. The matrix building module is used to perform multi-dimensional anomaly analysis and attention optimization on the elderly care monitoring dataset based on the user activity scene labels, and to build a monitoring anomaly attention coupling matrix. The risk incident type fitting module is used to fit risk incident types based on the user activity scenario tags to obtain N fitted risk incident types, where N is a positive integer greater than 1. The risk incident simulation module is used to perform multi-dimensional risk incident simulation on the monitoring anomaly attention coupling matrix based on the N fitted risk incident types and according to the dynamic perturbation distillation mechanism, so as to obtain a risk incident simulation map. The monitoring and early warning signal generation module is used to read the real-time monitoring data of the accompanying users, and perform accompanying accident interference analysis and optimization on the risk accident projection map based on the accompanying user monitoring data to obtain the risk projection optimization map, and simultaneously generate elderly care monitoring and early warning signals.

10. An electronic device, characterized in that, include: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the steps of the multi-device linkage elderly care data monitoring and early warning method according to any one of claims 1 to 8.

Images