A nursing system and a monitoring method for the nursing system

Abstract

The present invention discloses a nursing system and a monitoring method thereof for the nursing system, comprising the following steps: collecting physiological data of a patient through intelligent sensors to form a dynamic health data stream, analyzing the data in combination with historical health data, extracting key health features based on a temporal attention mechanism, predicting the health status of the patient using a machine learning model, calculating the health patterns between the patient and similar patients to generate a personalized nursing plan, combining a temporal collaborative multi-task learning framework to jointly optimize the health status prediction and the nursing recommendation plan, and realizing dynamic adjustment of the nursing plan. The system further includes an adaptive nursing intervention module, which optimizes the nursing intervention strategy according to the latest health data of the patient and continuously iteratively optimizes the nursing plan based on the feedback of nursing execution. The present invention constructs a complete nursing closed-loop from data collection, analysis, prediction, recommendation to optimization and intervention, realizing precise monitoring of the patient's health status and personalized nursing management.

Description

Technical Field

[0001] The present invention relates to the technical field of nursing monitoring, and particularly to a nursing system and a monitoring method for the nursing system. Background Art

[0002] With the development of intelligent medical technology, the role of the nursing system in medical services has become increasingly important. However, in the face of increasingly complex nursing needs and the dynamic changes in the health status of patients, traditional nursing systems still have many deficiencies and are difficult to meet the personalized, intelligent, and efficient nursing needs. The core goal of the nursing system is to provide real-time and accurate health status prediction and nursing plan recommendation based on the physiological data of patients and historical nursing records. However, the existing nursing systems have obvious defects in the following aspects:

[0003] 1. Low nursing work efficiency: Traditional nursing systems rely on manual recording and analysis of patient data. Nursing staff need to manually track the health status of patients and adjust nursing plans, resulting in cumbersome nursing processes and low efficiency, and being unable to meet the high-frequency health monitoring needs.

[0004] 2. Difficulty in real-time monitoring of nursing quality: Existing nursing systems usually rely on regular health checks or subjective nursing evaluations, lacking automatic monitoring of the real-time health status of patients and data-driven optimization of nursing plans, resulting in lagged nursing interventions and being unable to timely adjust nursing measures to adapt to the dynamic changes in the health status of patients.

[0005] 3. Lack of personalized health management: Most traditional nursing plans adopt standardized nursing processes, failing to fully consider individual differences and being difficult to formulate accurate nursing plans based on the historical health data of patients, nursing experience of similar patients, and individual health characteristics, resulting in limited nursing effects.

[0006] 4. Low data utilization rate and insufficient intelligence in nursing decision-making: Existing nursing systems do not make full use of the historical health data and real-time physiological data of patients, lacking efficient data analysis methods, making it difficult to dynamically optimize nursing plans and affecting the scientificity and accuracy of nursing plans.

[0007] 5. Lack of adaptability in nursing intervention strategies: Traditional nursing systems mainly rely on static nursing plans and are difficult to dynamically adjust nursing intervention measures according to the real-time changes in the health status of patients, resulting in a lack of flexible response capabilities of nursing plans in the face of sudden health events or changes in the patient's state.

[0008] Therefore, how to provide a nursing system and a monitoring method for the nursing system is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0009] An object of the present invention is to provide a nursing system and a monitoring method thereof. The present invention utilizes a temporal attention mechanism, a collaborative filtering algorithm, and a temporal collaborative multi-task learning framework, and details a closed-loop optimization process from physiological data collection, health status prediction, personalized nursing recommendation to adaptive nursing intervention, with the advantages of strong real-time performance, accurate personalized nursing, and high ability to adaptively adjust nursing intervention.

[0010] A nursing system and a monitoring method thereof according to an embodiment of the present invention include the following steps:

[0011] S1. A physiological data collection module, which collects multi-dimensional physiological data of a patient in real time through intelligent sensors to form a continuously updated dynamic health data stream;

[0012] S2. A temporal data analysis and feature extraction module, which analyzes the dynamic health data stream by applying a temporal attention mechanism and automatically extracts physiological features at key time nodes and historical health data;

[0013] S3. A health status prediction module, which predicts the health status of the patient based on the extracted physiological features at key time nodes in combination with a machine learning algorithm;

[0014] S4. A personalized nursing recommendation module, which analyzes the health patterns between the patient and similar patients through a collaborative filtering algorithm based on the patient's health status prediction and historical health data to generate a nursing recommendation plan;

[0015] S5. A nursing plan optimization and adjustment module, which constructs a temporal collaborative multi-task learning framework, jointly optimizes the patient's health status prediction and nursing recommendation plan, determines the patient's nursing needs, and adjusts the nursing recommendation plan in real time according to the patient's health status changes and nursing needs and generates a historical nursing record;

[0016] S6. An adaptive nursing intervention and feedback module, which automatically generates a nursing intervention strategy and optimizes the nursing recommendation plan according to the dynamic health data stream, historical nursing record, and adjusted nursing recommendation plan.

[0017] Optionally, the S2 specifically includes:

[0018] S21. A data preprocessing module, which standardizes and temporally aligns the dynamic health data stream, eliminates abnormal data by using an outlier detection method, and constructs a time series feature matrix :

[0019] ;

[0020] Wherein, represents a physiological feature and is used to identify the physiological measurement index at the time step and represents the total number of physiological features, represents the time step of the physiological feature at;

[0021] S22, the time series feature extraction module, based on the time series feature matrix , applies the time series attention mechanism to calculate the weight matrix of multi-dimensional physiological data at different time steps , and uses weighted summation to generate the time series feature vector at time step ; ;

[0022] S23, the key time step identification module, based on the time series feature vector at time step ; , uses the dynamic time warping method to calculate the feature change amplitude between time steps and determines the key time step :

[0023] ;

[0024] wherein, is the th key time step, is the sliding window size, automatically identifies the key change moment of the health state based on the feature change trend, is the time step ;

[0025] S24, the historical health data matching module, based on the key time step , extracts the historical health data of the corresponding time window from the time series feature matrix , and calculates the matching degree between the feature component of the key time step and the historical health data : ;

[0026] ;

[0027] wherein, reflects the matching degree between the health state at the key time step and the historical health data , is the feature component of the key time step , is the corresponding feature component of the historical health data, represents the physiological feature, is the total number of physiological features;

[0028] S25. The timing feature optimization module dynamically corrects the timing feature vector at the key time step based on the matching degree between the health state at the key time step and the historical health data: , for the key time step :

[0029] ;

[0030] Wherein, is the optimized timing feature vector, is the timing feature vector extracted at the key time step , is the historical health data, is the feature fusion adjustment factor, and the value range is .

[0031] Optionally, the S22 specifically includes:

[0032] S221. The timing attention initialization module initializes the initial weight matrix of the timing attention mechanism based on the time series feature matrix , calculates the forward calculation weight matrix and the backward calculation weight matrix and jointly generates the initial weight matrix: :

[0033] ;

[0034] Wherein, represents the initial weight matrix at time step , is the normalization function, and are the forward calculation weight matrix and the backward calculation weight matrix respectively, is the time series feature matrix at the current time step , is the time series feature matrix at the previous time step ;

[0035] S222. The time step attention distribution calculation module calculates the attention distribution weight of the multi-dimensional physiological data at the current time step based on the initial weight matrix by using the normalization mechanism:

[0036] ;

[0037] Wherein, represents the attention distribution weight of the physiological feature at time step , is the time step of the physiological characteristics for the initial weight matrix, is the total number of physiological characteristics, is the exponential function;

[0038] S223. A timing feature calculation module, based on the attention distribution weight , performs weighted calculation on the physiological characteristics at the current time step to generate a timing feature vector :

[0039] ;

[0040] wherein, represents the timing feature vector at time step , is the time step when the physiological characteristic is the observed value, is the physiological characteristic at time step of the attention distribution weight.

[0041] Optionally, the S4 specifically includes:

[0042] S41. A patient health feature construction module, based on patient health status prediction and historical health data, extracts multi-dimensional physiological parameters, medical history records, and nursing feedback information to construct a patient health feature vector ;

[0043] S42. A similar patient matching module, based on the patient health feature vector , calculates the health feature similarity with similar patients, and uses the Mahalanobis distance to calculate the health feature similarity between different patients:

[0044] ;

[0045] wherein, represents the health feature similarity between patient and similar patient , and respectively represent the patient health feature vectors of patient and similar patient , is the inverse matrix of the covariance matrix of the health features;

[0046] S43. A similar group construction module, based on the health feature similarity between the patient and similar patients, screens a set of similar patients whose similarity meets the set threshold ​ ;

[0047] S44. Nursing plan scoring calculation module, based on the set of similar patients and their historical health data, apply collaborative filtering algorithm to calculate the recommended score of the candidate nursing plan :

[0048] ;

[0049] Among them, represents the recommended score of the candidate nursing plan for patient , is the historical score of the similar patient for the candidate nursing plan ;

[0050] S45. Personalized nursing plan generation module, based on the recommended scores of the candidate nursing plans , sort all candidate nursing plans, select the candidate nursing plan with the highest score, and form the final nursing recommendation plan .

[0051] Optionally, the S42 specifically includes:

[0052] S421. Health feature normalization module, normalize the multi-dimensional physiological parameters in the patient health feature vector to generate the normalized feature values on the health feature components of the patient;

[0053] S422. Health feature correlation calculation module, based on the normalized feature values on the health feature components of the patient, calculate the covariance matrix of the health feature components to measure the correlation between each health feature variable:

[0054] ;

[0055] Among them, represents the covariance between the health feature component and the health feature component , is the mean value of the health feature component , is the mean value of the health feature component , is the total number of patient data, is the patient on the health feature component of the normalized feature value, is the patient On the health feature component The normalized eigenvalue;

[0056] S423. The individual health feature similarity calculation module, based on the normalized eigenvalue on the health feature component of the patient and the covariance matrix of the health features, calculates the health feature similarity between the patient and the similar patient using the Mahalanobis distance.

[0057] Optionally, the S5 specifically includes:

[0058] S51. The nursing demand modeling module, based on the prediction of the patient's health status and the nursing recommendation plan, extracts the health change trend of the patient and the historical nursing records, and constructs a nursing demand vector ;

[0059] S52. The time series collaborative learning module, constructs a time series collaborative multi-task learning framework, and jointly optimizes the prediction of the patient's health status and the nursing recommendation plan:

[0060] ;

[0061] Among them, is the total loss function, is the health status prediction loss, is the nursing recommendation loss, and are the task weight factors;

[0062] S53. The personalized nursing optimization module, based on the total loss function optimizes the nursing demand vector to generate an optimized nursing demand vector , and updates the nursing recommendation plan :

[0063] ;

[0064] Among them, is the updated nursing recommendation plan, represents the matching degree between the nursing demand vector and the candidate nursing plan;

[0065] S54. The historical nursing record update module, based on the updated nursing recommendation plan , updates the patient's historical nursing records.

[0066] Optionally, a monitoring method for a nursing system, the method includes the following steps:

[0067] S71. Real-time collect the multi-dimensional physiological data of the patient through intelligent sensors to form a dynamic health data stream;

[0068] S72. Based on the dynamic health data stream, use the temporal attention mechanism to extract physiological features at key time steps and construct a time series feature matrix;

[0069] S73. Based on the time series feature matrix and key time steps, use a machine learning model to predict the patient's future health status and output the health risk assessment result of the patient;

[0070] S74. Based on the patient's health status prediction result and historical nursing records, use the Mahalanobis distance to calculate the health feature similarity between the patient and the set of similar patients, and use the collaborative filtering algorithm to calculate the nursing plan score to generate a nursing recommendation plan;

[0071] S75. Based on the nursing recommendation plan, construct a temporal collaborative multi-task learning framework, jointly optimize the patient's health status prediction and nursing recommendation plan, and dynamically adjust the nursing recommendation plan;

[0072] S76. Based on the dynamically adjusted nursing recommendation plan and the dynamic health data stream, calculate the nursing intervention state vector and use the optimization function to calculate the optimal nursing intervention decision vector;

[0073] S77. Based on the execution feedback of the nursing intervention decision vector, update the patient's historical nursing records, and synchronously adjust the nursing intervention state vector and the nursing recommendation plan.

[0074] The beneficial effects of the present invention are as follows:

[0075] (1) By combining the temporal attention mechanism, collaborative filtering algorithm and temporal collaborative multi-task learning framework, the present invention realizes the accurate prediction of the patient's health status and personalized nursing recommendation by the nursing system, enabling the system to dynamically adapt to the patient's health changes and optimize the nursing plan. Based on the dynamic health data stream collected by intelligent sensors, the system can automatically identify key health features and generate the optimal nursing plan by combining the analysis of the health patterns of similar patients, ensuring the accuracy and adaptability of nursing interventions.

[0076] (2) Through the temporal collaborative optimization mechanism, the present invention jointly optimizes the generation of health status prediction and nursing recommendation plan, realizes the dynamic adjustment and personalized matching of the nursing plan. Based on the nursing intervention state modeling, the system can calculate the optimal decision vector of the nursing intervention strategy according to the patient's historical nursing records, current health features and recommended nursing plan, and dynamically adjust the nursing recommendation plan, enabling the nursing plan to be continuously optimized and adapted to the patient's health changes.

[0077] (3) By constructing a closed-loop nursing monitoring process, which covers data collection, feature extraction, health prediction, nursing recommendation, nursing optimization, nursing intervention, and feedback update, etc., the nursing system of the present invention is enabled with an adaptive learning ability. The system can iteratively optimize the nursing recommendation plan based on the feedback information of the nursing intervention after the nursing execution, and update the nursing intervention status, reducing the dependence on manual intervention, improving the accuracy and dynamic adaptability of the nursing intervention, and thus effectively improving the quality and efficiency of nursing services. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0079] Figure 1 is an overall architecture diagram of a nursing system proposed by the present invention;

[0080] Figure 2 is a flowchart of a monitoring method of a nursing system proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0081] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0082] Refer to Figure 1-2 , a nursing system and a monitoring method of the nursing system.

[0083] Specifically, a nursing system includes the following steps:

[0084] S1. A physiological data collection module, which collects multi-dimensional physiological data of patients in real time through intelligent sensors to form a continuously updated dynamic health data stream;

[0085] In this embodiment, multi-dimensional physiological data of patients are collected in real time through intelligent sensors to form a continuously updated dynamic health data stream, providing high-precision and timely data support for subsequent health status prediction and personalized nursing recommendation. This method can effectively reduce the lag and error of manual data recording, improve the real-time monitoring ability of the nursing system, and ensure the accuracy and continuity of nursing intervention.

[0086] S2. A time series data analysis and feature extraction module, which analyzes the dynamic health data stream by applying a time series attention mechanism to automatically extract physiological features at key time nodes and historical health data;

[0087] S3. A health status prediction module, which predicts the health status of patients based on the physiological features at the extracted key time nodes in combination with machine learning algorithms;

[0088] This embodiment uses a machine learning algorithm that combines a long short-term memory network with a gradient boosting decision tree to predict the patient's health status. The long short-term memory network can effectively model the temporal changes in the patient's physiological parameters, capture short-term fluctuations and long-term trends, and improve the timeliness and accuracy of health status prediction. The gradient boosting decision tree uses historical health data to explore the health evolution patterns of different patient groups and optimize health risk assessment through multi-level feature extraction. Combining the temporal modeling capabilities of the long short-term memory network and the efficient feature learning capabilities of the gradient boosting decision tree, this embodiment can accurately predict the patient's future health status and provide reliable data support for personalized care plans, while ensuring the robustness and generalization capabilities of the prediction model, and improving the decision-making accuracy and adaptability of the nursing system.

[0089] S4, personalized nursing recommendation module, through collaborative filtering algorithm, based on the patient's health status prediction and historical health data, analyzes the health pattern between the patient and similar patients, and generates nursing recommendation plans;

[0090] S5, Nursing plan optimization and adjustment module, builds a time-series collaborative multi-task learning framework, jointly optimizes the patient's health status prediction and nursing recommendation plan, determines the patient's nursing needs, adjusts the nursing recommendation plan in real time according to the patient's health status changes and nursing needs, and generates historical nursing records;

[0091] S6, the adaptive nursing intervention and feedback module, automatically generates nursing intervention strategies and optimizes nursing recommendation plans based on dynamic health data streams, historical nursing records, and adjusted nursing recommendation plans.

[0092] This implementation method builds an adaptive nursing intervention strategy based on dynamic health data streams, historical nursing records, and optimized nursing recommendation plans, which can adjust the nursing plan in real time to adapt to changes in the patient's health status. Through the automated strategy generation and optimization mechanism, the accuracy and response efficiency of nursing decisions are improved, and the dynamic adaptation of personalized nursing interventions is achieved, ensuring the continuous optimization of nursing plans and the efficient implementation of patient health management.

[0093] In this implementation manner, S2 specifically includes:

[0094] S21, data preprocessing module, standardizes and aligns the dynamic health data stream, uses outlier detection method to remove abnormal data, and constructs the time series feature matrix :

[0095] ;

[0096] in, Represents physiological characteristics and is used to identify time steps Physiological measurement indicators at represent the total number of physiological characteristics, represent the time step physiological characteristics at the observed value of;

[0097] S22. The time series feature extraction module, based on the time series feature matrix , applies the time series attention mechanism to calculate the weight matrix of multi-dimensional physiological data at different time steps , and uses weighted summation to generate the time series feature vector at time step ; ;

[0098] S23. The key time step identification module, based on the time series feature vector at time step , uses the dynamic time warping method to calculate the feature change amplitude between time steps and determines the key time step : :

[0099] ;

[0100] Among them, is the th key time step, is the sliding window size, automatically identifies the key change moment of the health state based on the feature change trend, is the time step the time series feature vector at;

[0101] S24. The historical health data matching module, based on the key time step , extracts the historical health data of the corresponding time window from the time series feature matrix , calculates the matching degree between the feature components of the key time step and the historical health data : :

[0102] ;

[0103] Among them, reflects the matching degree between the health state at the key time step and the historical health data , is the feature component of the key time step , is the corresponding feature component of the historical health data, represents the physiological characteristic, is the total number of physiological characteristics;

[0104] S25. The timing feature optimization module dynamically corrects the timing feature vector at the key time step based on the matching degree between the health state at the key time step and the historical health data. , for the key time step :

[0105] ;

[0106] Among them, is the optimized timing feature vector, is the timing feature vector extracted at the key time step , is the historical health data, is the feature fusion adjustment factor, and the value range is .

[0107] The present invention realizes the accurate analysis of the dynamic health data stream through data preprocessing, timing feature extraction, and key time step identification. The health feature similarity at the key time step is calculated by combining the historical health data matching module, and the key features are dynamically corrected in the timing feature optimization module to ensure the accuracy of the physiological data and the stability of the timing features. This method improves the accuracy of health state prediction, enabling the nursing plan to be personalized adjusted and optimized based on high-quality health feature data.

[0108] In this embodiment, the S22 specifically includes:

[0109] S221. The timing attention initialization module initializes the initial weight matrix of the timing attention mechanism based on the time series feature matrix , calculates the forward calculation weight matrix and the backward calculation weight matrix and jointly generates the initial weight matrix: ;

[0110] ;

[0111] Among them, represents the initial weight matrix at time step , is the normalization function, and are the forward calculation weight matrix and the backward calculation weight matrix respectively, is the time series feature matrix at the current time step , is the time series feature matrix at the previous time step ;

[0112] S222. The time step attention distribution calculation module, based on the initial weight matrix ​​​​, using a normalization mechanism to calculate the attention distribution weights of multi-dimensional physiological data at the current time step of :

[0113] ;

[0114] wherein, represents the attention distribution weight of the physiological feature at the time step , is the initial weight matrix of the physiological feature at the time step , is the total number of physiological features, is the exponential function;

[0115] S223. A timing feature calculation module, based on the attention distribution weight , performs weighted calculation on the physiological features at the current time step to generate a timing feature vector :

[0116] ;

[0117] wherein, represents the timing feature vector at the time step , is the observed value of the physiological feature at the time step , is the attention distribution weight of the physiological feature at the time step .

[0118] In this embodiment, through the timing attention mechanism, the weights of the patient's physiological features at different time steps are dynamically calculated to ensure that key health information is intensively processed in the care system. The combined application of the forward and backward calculation weight matrices enables the model to combine historical and current health data and optimize the accuracy of timing feature extraction. Through attention distribution calculation and weighted feature fusion, the present invention can accurately capture the changing trend of the patient's health status, improve the prediction accuracy of the care system for the health status, and provide more reliable data support for personalized care plans.

[0119] In this embodiment, the S4 specifically includes:

[0120] S41. A patient health feature construction module, based on the patient's health status prediction and historical health data, extracts multi-dimensional physiological parameters, medical history records, and nursing feedback information to construct a patient health feature vector ;

[0121] S42. Similar patient matching module, based on the patient's health feature vector , calculate the health feature similarity with similar patients, and use Mahalanobis distance to calculate the health feature similarity between different patients:

[0122] ;

[0123] Among them, represents the health feature similarity between patient and similar patient , and respectively represent the patient health feature vectors of patient and similar patient , is the inverse matrix of the covariance matrix of health features;

[0124] S43. Similar group construction module, based on the health feature similarity between patients and similar patients , screen the set of similar patients whose similarity meets the set threshold ;

[0125] S44. Nursing plan score calculation module, based on the set of similar patients and their historical health data, apply collaborative filtering algorithm to calculate the recommended score of the candidate nursing plan :

[0126] ;

[0127] Among them, represents the recommended score of the candidate nursing plan for patient , is the historical score of similar patient for the candidate nursing plan ;

[0128] S45. Personalized nursing plan generation module, based on the recommended scores of the candidate nursing plans, sort all candidate nursing plans, screen the candidate nursing plan with the highest score, and form the final nursing recommendation plan .

[0129] The present invention combines Mahalanobis distance similarity calculation and collaborative filtering algorithm to accurately match similar patient groups, and optimizes the nursing plan recommendation based on the historical nursing data of similar patients. This method can not only dynamically adapt to the health status of individual patients, but also optimize nursing decisions through learning from historical data, improve the personalized matching degree and recommendation accuracy of nursing plans, enhance the scientificity and effectiveness of nursing interventions, and achieve the reasonable allocation and intelligent optimization of nursing resources.

[0130] In this embodiment, step S42 specifically includes:

[0131] S421. A health feature normalization module normalizes the multi-dimensional physiological parameters in the patient's health feature vector to generate a normalized eigenvalue on the health feature component of the patient;

[0132] S422. A health feature correlation calculation module calculates the covariance matrix of the health feature components based on the normalized eigenvalues on the health feature components of the patient to measure the correlation between each health feature variable:

[0133] ;

[0134] Wherein, represents the covariance between the health feature component and the health feature component , is the mean value of the health feature component , is the mean value of the health feature component , is the total number of data of the patient, is the patient on the health feature component of the normalized eigenvalue, is the patient on the health feature component of the normalized eigenvalue;

[0135] S423. An individual health feature similarity calculation module calculates the health feature similarity between the patient and the similar patient using the Mahalanobis distance based on the normalized eigenvalues on the health feature components of the patient and the covariance matrix of the health features.

[0136] The present invention normalizes the patient's health feature vector and calculates the covariance matrix of the health features to ensure that data in different feature dimensions have a unified scale in similarity calculation. The Mahalanobis distance is used to measure the similarity of the patient's health status, enabling the system to accurately match patients with similar health features, improving the accuracy and effectiveness of personalized nursing recommendations, thereby optimizing the matching of nursing plans and enhancing the adaptability of nursing interventions.

[0137] In this embodiment, step S5 specifically includes:

[0138] S51. A nursing demand modeling module extracts the patient's health change trend and historical nursing records based on the patient's health status prediction and nursing recommendation plan, and constructs a nursing demand vector ;

[0139] S52. A time-series collaborative learning module constructs a time-series collaborative multi-task learning framework to jointly optimize the patient's health status prediction and nursing recommendation plan:

[0140] ;

[0141] Among them, is the total loss function, is the health status prediction loss, is the nursing recommendation loss, and are task weight factors;

[0142] S53. A personalized nursing optimization module optimizes the nursing demand vector based on the total loss function to generate an optimized nursing demand vector , and updates the nursing recommendation plan :

[0143] ;

[0144] Among them, is the updated nursing recommendation plan, represents the matching degree between the nursing demand vector and the candidate nursing plan;

[0145] S54. A historical nursing record update module updates the patient's historical nursing records based on the updated nursing recommendation plan .

[0146] In this embodiment, a time-series collaborative multi-task learning framework is constructed to jointly optimize the prediction of the patient's health status and the nursing recommendation plan, achieving accurate modeling and optimization of nursing needs. Based on the optimized nursing need vector, the nursing recommendation plan is dynamically adjusted, and the historical nursing records are iteratively updated in combination with the nursing execution feedback to ensure that the nursing plan can continuously adapt to the changes in the patient's health status and improve the accuracy and intelligence level of nursing intervention.

[0147] In this embodiment, a monitoring method for a nursing system, the method comprising the following steps:

[0148] S71. Real-time collect multi-dimensional physiological data of the patient through intelligent sensors to form a dynamic health data stream;

[0149] S72. Based on the dynamic health data stream, use a time-series attention mechanism to extract physiological features at key time steps and construct a time-series feature matrix;

[0150] S73. Based on the time-series feature matrix and key time steps, use a machine learning model to predict the patient's future health status and output the patient's health risk assessment result;

[0151] S74. Based on the patient's health status prediction result and historical nursing records, use the Mahalanobis distance to calculate the health feature similarity between the patient and the set of similar patients, and use the collaborative filtering algorithm to calculate the nursing plan score to generate a nursing recommendation plan;

[0152] S75. Based on the nursing recommendation plan, construct a time-series collaborative multi-task learning framework to jointly optimize the prediction of the patient's health status and the nursing recommendation plan, and dynamically adjust the nursing recommendation plan;

[0153] S76. Based on the dynamically adjusted nursing recommendation plan and the dynamic health data stream, calculate the nursing intervention state vector and use the optimization function to calculate the optimal nursing intervention decision vector;

[0154] S77. Based on the execution feedback of the nursing intervention decision vector, update the patient's historical nursing records, and synchronously adjust the nursing intervention state vector and the nursing recommendation plan.

[0155] Example 1:

[0156] To verify the feasibility of the present invention, the present invention is applied to the intelligent nursing system of a large general hospital. The hospital receives a large number of inpatients and outpatients every day, with a huge nursing workload. Moreover, the health conditions of the patients are complex and changeable. Nursing staff need to monitor the physiological states of the patients in real time and adjust the nursing plans in a timely manner. The existing nursing methods mainly rely on manual records and empirical judgments, making it difficult to achieve precision and personalization. To improve the nursing efficiency and optimize the nursing plan, the hospital decides to introduce the intelligent nursing system of the present invention, and utilize technologies such as physiological data collection, health status prediction, personalized nursing recommendation, nursing optimization and intervention strategies to achieve full-process intelligent nursing monitoring.

[0157] The hospital installs intelligent sensors in each ward to collect multi-dimensional physiological data of patients, including heart rate, blood pressure, blood oxygen saturation, body temperature, blood glucose level, etc., and transmits these data to the nursing database in real time. The nursing system automatically extracts the physiological characteristics at key time steps, combines with the historical health data of the patients, constructs a time series feature matrix, and analyzes the data using a time series attention mechanism to screen out the key factors affecting the health status of the patients. On the basis of data collection, the system uses a machine learning model to predict the health risks of the patients, and calculates the health patterns of the patients and similar patients through a collaborative filtering algorithm, so as to generate a personalized nursing recommendation plan for each patient.

[0158] To ensure the dynamic adaptability of the nursing plan, the system is based on a time series collaborative multi-task learning framework, jointly optimizes health status prediction and nursing plan generation, and adjusts the plan in combination with nursing execution feedback. For example, the nursing plan for a diabetic inpatient initially recommends monitoring blood glucose levels every 6 hours, but the system finds that the blood glucose of this patient fluctuates greatly after nursing execution. Therefore, the plan is dynamically adjusted to increase the monitoring frequency and recommend a more precise diet management plan. In addition, the system can also automatically generate nursing intervention strategies to optimize the implementation process of the nursing plan. For example, when the blood glucose level of a patient abnormally increases, the system immediately reminds the nursing staff to adjust the insulin dose and recommends personalized diet intervention measures in combination with the patient's historical nursing records.

[0159] To verify the effectiveness of the nursing system of the present invention, the hospital selects 500 inpatients for a comparative experiment. Among them, 250 patients use the traditional nursing method, and 250 patients adopt the intelligent nursing system of the present invention. The experimental period is three months. The performances of the two nursing methods in terms of nursing efficiency, nursing quality, health risk control, etc. are compared. The experimental data are shown in the following table:

[0160] Table 1 Comparison of the effects of the traditional nursing method and the intelligent nursing system

[0161]

[0162] As can be seen from the experimental results, the intelligent nursing system of the present invention has significantly improved the nursing efficiency, increasing the number of patients cared for by each nurse per day by 50%. At the same time, since the nursing plan is optimized based on data-driven, the matching degree of the nursing plan has increased from 78.5% to 92.3%, and patients can receive a nursing plan that better suits their own health conditions. In addition, since the intelligent system can timely identify the health risks of patients and dynamically adjust the nursing plan, the nursing intervention response time has been shortened from 45 minutes to 15 minutes, and the incidence of health emergencies has decreased by 61%. In terms of nursing satisfaction, the satisfaction of patients using the intelligent nursing system has increased by 10.8%, indicating that the system can better meet the nursing needs of patients.

[0163] In the specific implementation process, the system has also optimized the personalized care for certain special patient groups. For example, in the ICU ward, the nursing system automatically detects the changes in the patient's condition based on the real-time health monitoring data of the patient and pushes nursing adjustment suggestions to the nursing staff. A certain postoperative recovery patient had a drop in blood pressure at night. The system combined his historical health data and the nursing records of similar patients and automatically recommended that the nursing staff adjust the intravenous infusion rate and take additional nursing measures, successfully avoiding the deterioration of the condition.

[0164] In summary, the intelligent nursing system of the present invention effectively improves the nursing efficiency, nursing quality and health management level through real-time data collection, time series analysis, personalized recommendation, dynamic optimization and nursing intervention feedback. The system can dynamically adapt to the health changes of patients, provide precise and personalized nursing plans, and reduce the work burden of nursing staff, providing an efficient and feasible solution for the intelligent nursing management of medical institutions.

[0165] The above is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A nursing system, characterized in that: Includes the following modules: The physiological data collection module collects the patient's multi-dimensional physiological data in real time through intelligent sensors, forming a continuously updated dynamic health data stream; The time series data analysis and feature extraction module uses the time series attention mechanism to analyze the dynamic health data stream and automatically extract the physiological characteristics and historical health data of key time nodes; The health status prediction module predicts the patient's health status based on the extracted physiological characteristics of key time nodes combined with machine learning algorithms; The personalized nursing recommendation module uses collaborative filtering algorithms to analyze the health patterns of patients and similar patients based on the patient's health status prediction and historical health data, and generates nursing recommendations; Nursing plan optimization and adjustment module, builds a time-series collaborative multi-task learning framework, jointly optimizes patient health status prediction and nursing recommendation plans, determines patient nursing needs, adjusts nursing recommendation plans in real time according to patient health status changes and nursing needs, and generates historical nursing records; Adaptive nursing intervention and feedback module, which automatically generates nursing intervention strategies and optimizes nursing recommendation plans based on dynamic health data streams, historical nursing records, and adjusted nursing recommendation plans; The personalized care recommendation module specifically includes: The patient health feature construction module extracts multi-dimensional physiological parameters, medical history records and nursing feedback information based on the patient's health status prediction and historical health data to construct the patient's health feature vector ; Similar patient matching module, based on patient health feature vector , calculate the similarity of health characteristics with similar patients, and use Mahalanobis distance to calculate the similarity of health characteristics between different patients: ； in, Indicates patient Similar patients The similarity of health characteristics between and Respectively, patients and similar patients The patient health feature vector, is the inverse matrix of the covariance matrix of the health characteristics; Similarity group building module, based on the similarity of health characteristics between patients and similar patients , filter similarity to meet the set threshold Similar patient collections ; Nursing plan scoring calculation module, based on similar patient sets and historical health data, and apply collaborative filtering algorithms to calculate candidate care plans Recommended rating : ； in, Indicates for patients Candidate Nursing Programs Recommended rating, For similar patients Nursing Program Candidates Historical ratings; Personalized nursing plan generation module, based on the recommendation score of candidate nursing plans , sort all candidate nursing plans, select the candidate nursing plans with the highest scores, and form the final nursing recommendation plan ; The nursing plan optimization and adjustment module specifically includes: Nursing demand modeling module, based on patient health status prediction and nursing recommendation plan, extracts patient health change trends and historical nursing records, and constructs nursing demand vector ; The time series collaborative learning module builds a time series collaborative multi-task learning framework to jointly optimize the patient health status prediction and nursing recommendation plan: ； in, is the total loss function, Predicting loss for health status, Recommend loss for nursing, and is the task weight factor; Personalized care optimization module, based on the total loss function Optimize the nursing demand vector to generate the optimized nursing demand vector , updated nursing recommendations: ； in, For updated care recommendations, Indicates the matching degree between the nursing demand vector and the nursing recommendation plan; Historical nursing record update module, based on updated nursing recommendations , update patient historical care records.

2. A nursing system according to claim 1, characterized in that: The time series data analysis and feature extraction module specifically includes: The data preprocessing module standardizes and aligns the dynamic health data stream, uses the outlier detection method to remove abnormal data, and constructs the time series feature matrix : ； in, Represents physiological characteristics and is used to identify time steps Physiological measurements of Represents the total number of physiological characteristics, Represents the time step Physiological characteristics of Observed value of Time series feature extraction module, based on the time series feature matrix , applying the temporal attention mechanism to calculate the weight matrix of multi-dimensional physiological data at different time steps , and use weighted summation to generate time steps The time series feature vector at ; Critical time step identification module, based on time step The time series feature vector at , dynamic time warping method is used to calculate the feature change amplitude between time steps and determine the key time step : ； in, For the A key time step, is the sliding window size, automatically identifying the key change moments of health status based on the feature change trend. is the time step The time series feature vector at ; Historical health data matching module, based on key time steps , from the time series feature matrix Extract historical health data of the corresponding time window , calculate the matching degree between the characteristic components of the key time step and the historical health data : ； in, Reflects the key time step Health status and historical health data The matching degree between The key time step The characteristic component of is the corresponding feature component of historical health data, Represents physiological characteristics, is the total number of physiological characteristics; The time series feature optimization module uses the matching degree between the health status at key time steps and the historical health data , for the key time step Dynamic correction of the time series feature vector at: ； in, is the optimized time series feature vector, The key time step The extracted time series feature vector, For historical health data, is the feature fusion adjustment factor, with a value range of .

3. A nursing system according to claim 2, characterized in that: The temporal feature extraction module specifically includes: Temporal attention initialization module, based on the time series feature matrix , initialize the initial weight matrix of the temporal attention mechanism , calculate the forward calculation weight matrix And the backward calculation weight matrix And jointly generate the initial weight matrix: ； in, Represents the time step The initial weight matrix, is the normalization function, and They are the forward calculation weight matrix and the backward calculation weight matrix, is the current time step The time series feature matrix, is the previous time step The time series feature matrix of Time step attention distribution calculation module, based on the initial weight matrix , using the normalization mechanism to calculate the multi-dimensional physiological data at the current time step The attention distribution weight : ； in, Indicates physiological characteristics At time step The attention distribution weight, is the time step Physiological characteristics The initial weight matrix, is the total number of physiological characteristics, is the exponential operation function; Time series feature calculation module, based on attention distribution weight , for the current time step The physiological characteristics are weighted to generate a time series feature vector : ； in, Represents the time step The time series feature vector of is the time step Physiological characteristics The observed value of For physiological characteristics At time step The attention distribution weight.

4. A nursing system according to claim 1, characterized in that: The similar patient matching module specifically includes: Health feature normalization module, which normalizes the patient's health feature vector Normalizing the multi-dimensional physiological parameters in the patient data to generate normalized characteristic values ​​on the patient's health characteristic components; The health feature correlation calculation module calculates the covariance matrix of the health feature components based on the normalized eigenvalues ​​of the patient's health feature components. , to measure the correlation between various health characteristic variables: ； in, Health characteristic component Health characteristics The covariance between Health characteristic component The mean of Health characteristic component The mean of is the total number of patient data, For patients In the health characteristics The normalized eigenvalue on , For patients In the health characteristics The normalized eigenvalues ​​on ; The individual health feature similarity calculation module uses the Mahalanobis distance to calculate the patient's health feature similarity based on the normalized eigenvalues ​​of the patient's health feature components and the covariance matrix of the health features. Similar patients The similarity of health characteristics between them.

Images