Remote nursing management method based on data feature mining

Abstract

The invention relates to the technical field of data feature mining, in particular to a remote nursing management method based on data feature mining. The method comprises the following steps: acquiring specific conditions and demand data of a patient; performing data source selection and configuration according to the specific condition and demand data of the patient to obtain a data source configuration table; performing multi-dimensional data acquisition according to the data source configuration table, and performing multi-dimensional data preprocessing to obtain patient comprehensive data; carrying out individualized baseline modeling on the patient comprehensive data to obtain an individual baseline model; performing real-time data standardization on the patient comprehensive data to obtain standardized real-time data; performing deviation calculation according to the standardized real-time data and the individual baseline model to obtain a deviation value sequence; and performing index correlation analysis on the deviation value sequence to obtain an index correlation matrix. According to the invention, the utilization of data feature mining is improved, so that remote nursing management is more accurate and personalized, and finally the nursing quality and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data feature mining, and in particular to a remote nursing management method based on data feature mining. Background Art

[0002] Early remote care management methods mainly relied on simple physiological parameter monitoring, such as blood pressure, heart rate, etc., and the data analysis methods were also relatively simple. Existing methods have gradually evolved from simple threshold judgments to statistical model-based analysis, such as linear regression and logistic regression.

[0003] However, these methods still have the problem of insufficient feature utilization, which is mainly reflected in the following aspects: 1. Lack of mining of nonlinear relationships: Traditional statistical models usually assume that there is a linear relationship between features, but in reality, there are complex nonlinear relationships between many physiological indicators and health conditions. For example, the relationship between blood sugar level and insulin secretion is not a simple linear relationship. Ignoring these nonlinear relationships will lead to reduced prediction accuracy of the model.

[0004] 2. Failure to effectively utilize high-order features: High-order features refer to features that are composed of multiple basic features, which can more comprehensively reflect the patient's health status. For example, combining the three basic features of blood pressure, heart rate, and body temperature can construct a new high-order feature, which can more effectively predict the patient's health risks than using any basic feature alone. However, many existing methods do not effectively utilize high-order features. Summary of the invention

[0005] Based on this, it is necessary to provide a remote care management method based on data feature mining to solve at least one of the above technical problems.

[0006] To achieve the above purpose, a remote nursing management method based on data feature mining includes the following steps: Step S1: Obtain patient specific conditions and demand data; select and configure data sources according to the patient specific conditions and demand data to obtain a data source configuration table; perform multi-dimensional data collection and multi-dimensional data preprocessing according to the data source configuration table to obtain comprehensive patient data; Step S2: Performing individualized baseline modeling on the comprehensive data of the patient to obtain an individual baseline model; Step S3: performing real-time data standardization on the comprehensive data of the patient to obtain standardized real-time data; performing deviation calculation based on the standardized real-time data and the individual baseline model to obtain a deviation value sequence; performing indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; performing comprehensive deviation evaluation analysis based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation report; Step S4: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; perform deviation feature extraction, historical data feature processing, and environmental feature extraction on the integrated data set to obtain a deviation feature table, a historical feature table, and an environmental feature table, respectively; construct high-order combined features based on the deviation feature table, the historical feature table, and the environmental feature table to obtain a combined feature table; select predictive features for the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a predictive feature set; construct a risk prediction model based on the predictive feature set to obtain a risk prediction model; perform risk prediction processing using the risk prediction model to obtain a risk prediction report; trigger warnings and send information based on the risk prediction report to obtain warning information, so as to realize remote nursing management tasks.

[0007] The present invention can ensure that the collected data is targeted, complete and high-quality by obtaining the individualized information and needs of the patient, and selecting the appropriate data source and performing data preprocessing based on this, thereby laying a solid data foundation for subsequent personalized baseline modeling, deviation analysis and risk prediction. By performing individualized baseline modeling on the comprehensive data of the patient, the unique physiological and behavioral patterns of each patient can be learned, and a personalized baseline model can be constructed, thereby avoiding the errors caused by using the reference range based on the group average, and improving the accuracy of subsequent deviation analysis and risk prediction. By standardizing the real-time data and calculating the deviation value sequence in combination with the individual baseline model, the degree of deviation between the patient's current physiological state and the baseline state can be accurately quantified. Further, through the correlation analysis of indicators, the complex interaction relationship between different physiological indicators can be revealed, so as to more comprehensively evaluate the health status of the patient, and generate a deviation assessment report, providing a more accurate basis for subsequent risk prediction. By integrating historical health data, deviation assessment reports and environmental data, and performing feature extraction, combination and selection, a feature set containing rich information can be constructed for training risk prediction models. By utilizing multi-source data and high-order combined features, various factors that affect the health of patients can be considered more comprehensively, and the accuracy and personalization of risk prediction can be improved. Finally, risk level division and early warning triggering can be carried out according to the risk prediction results, remote real-time monitoring and early warning can be realized, and medical staff and patients' families can be reminded in time, so as to effectively prevent and control the occurrence and development of diseases and improve the quality and efficiency of nursing. Therefore, the present invention provides a remote nursing management method based on data feature mining, which more fully mines data features through indicator correlation analysis, high-order combined feature construction, deviation value sequence analysis and multi-dimensional data integration technology, thereby improving the accuracy of risk prediction and solving the problem of insufficient feature utilization of existing methods. This makes remote nursing management more accurate and personalized, and ultimately improves the quality and efficiency of nursing.

[0008] Preferably, step S1 comprises the following steps: Step S11: Obtain the patient's specific conditions and demand data; select and configure the data source according to the patient's specific conditions and demand data to obtain a data source configuration table; Step S12: collect data in real time according to the data source configuration table to obtain the original data stream; Step S13: performing data preprocessing on the original data stream to obtain a preprocessed data stream; Step S14: extracting data features from the preprocessed data stream to obtain a feature data stream; Step S15: Integrate the characteristic data stream to obtain comprehensive patient data.

[0009] The present invention collects the patient's specific situation and demand data, and selects and configures the data source based on this, generates a personalized data source configuration table, can ensure that the collected data is consistent with the patient's actual situation and needs, avoids collecting unnecessary data or missing key data, improves the efficiency and pertinence of data collection, and lays the foundation for subsequent personalized analysis and risk prediction. According to the data source configuration table, real-time data collection can ensure that the data collection process is normalized and standardized, and obtains the required multiple types of data to form an original data stream, which provides sufficient data sources for subsequent data preprocessing and feature extraction. Preprocessing the original data stream, including data cleaning, denoising, missing value filling and other operations, can effectively improve the quality of the data, remove the influence of noise and outliers on subsequent analysis, and supplement the missing data, thereby ensuring the accuracy and reliability of subsequent analysis. Extracting meaningful features from the preprocessed data stream can convert the original data into more representative and interpretable feature data, reduce data dimensions, reduce computational complexity, and highlight important physiological patterns and change trends, providing more valuable information for subsequent baseline modeling and deviation analysis. Integrating characteristic data streams from different sources according to timestamps to form comprehensive patient data (PID) can fuse multi-dimensional data into a unified data structure, facilitate subsequent analysis and processing, and provide more comprehensive patient health information, thereby improving the accuracy and comprehensiveness of risk prediction.

[0010] Preferably, step S2 comprises the following steps: Step S21: segmenting the patient's comprehensive data to obtain a training data set and a verification data set; Step S22: performing feature selection on the training data set to obtain a screening feature set; Step S23: training a baseline model according to the training data set and the screened feature set to obtain an initial baseline model; Step S24: using the validation data set to perform model validation and optimization on the initial baseline model to obtain an optimized baseline model; Step S25: updating the optimized baseline model according to the comprehensive data of the patient to obtain an individual baseline model.

[0011] The present invention can effectively avoid model overfitting, improve the generalization ability of the model, and ensure that the model can maintain good prediction performance on unseen data by dividing the patient's comprehensive data into a training data set and a verification data set, thereby more accurately reflecting the patient's true physiological baseline. Through feature selection, the key features related to the construction of the baseline model are screened out, redundant features and noise can be removed, the model structure is simplified, the training efficiency and interpretability of the model are improved, and the risk of overfitting is reduced, making the baseline model more targeted and accurate. Using the training data set and the screened feature set for model training, the model can learn the distribution law of the patient's individual data and the relationship between the features, build an initial baseline model, and provide a basis for subsequent model optimization and updating. Using the verification data set to verify and optimize the initial baseline model can evaluate the performance of the model, and adjust the model parameters or select a more suitable model according to the evaluation results, improve the prediction accuracy and stability of the model, so that it can more accurately reflect the patient's individual physiological baseline. Regularly using the latest patient comprehensive data to update the optimized baseline model can make the model adapt to the dynamic changes of the patient's physiological data, maintain the timeliness of the baseline model, and improve the model's ability to predict future changes in physiological data, thereby more accurately reflecting the patient's current physiological state and potential health risks.

[0012] Preferably, step S3 comprises the following steps: Step S31: performing real-time data standardization on the patient's comprehensive data to obtain standardized real-time data; performing baseline distribution parameter extraction on the individual baseline model to obtain baseline distribution parameters; Step S32: Calculate the real-time data probability density according to the baseline distribution parameters and the standardized real-time data to obtain the real-time data probability density; Step S33: performing probability density conversion on the real-time data probability density to obtain a deviation value; performing time series sorting on the deviation values ​​to obtain a deviation value sequence; Step S34: performing an indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; Step S35: Perform a comprehensive deviation evaluation based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation score; generate a deviation evaluation report based on the deviation evaluation score and the deviation value sequence to obtain a deviation evaluation report.

[0013] The present invention can eliminate the dimension difference between different indicators by standardizing the real-time data and extracting the baseline distribution parameters from the individual baseline model, making the data comparable, and providing the necessary input for the subsequent deviation calculation and probability density calculation, so as to more accurately evaluate the deviation between the patient's real-time data and the baseline model. By calculating the probability density of the standardized real-time data under the baseline distribution, the deviation degree of the real-time data from the baseline model can be quantified. The lower the probability density value, the greater the deviation degree, which provides a basis for the subsequent deviation value calculation, so as to more finely characterize the changes in the patient's health status. Converting the probability density into a deviation value and sorting the deviation value by time series can more intuitively reflect the deviation degree of the real-time data from the baseline model, and retain the time series information, which is convenient for the subsequent indicator correlation analysis and deviation comprehensive evaluation, so as to more clearly show the change trend of the patient's physiological data. Performing an indicator correlation analysis on the deviation value sequence can reveal the relationship between different physiological indicators, such as the correlation between heart rate and blood pressure, and help identify which indicators' deviation values ​​change synchronously, so as to more comprehensively evaluate the patient's health status and provide an important basis for the subsequent deviation comprehensive evaluation. By combining the indicator correlation matrix and the deviation value sequence to conduct a comprehensive deviation assessment and generate a deviation assessment report, the deviation information of multiple indicators can be integrated into a single deviation assessment score, which can more comprehensively reflect the patient's overall health status and present the assessment results in a visual way, making it easier for medical staff to quickly understand the patient's condition and provide important input for subsequent risk prediction.

[0014] Preferably, step S34 includes the following steps: Step S341: construct a multi-dimensional time series matrix for the deviation value sequence to obtain a deviation value matrix; Step S342: segmenting the deviation value matrix into time windows to obtain a deviation value window sequence; Step S343: extracting a single window deviation matrix from the deviation value window sequence to obtain a single window deviation value matrix; Step S344: performing linear correlation calculation between indicators on the single window deviation value matrix to obtain a linear correlation matrix sequence; performing nonlinear correlation calculation between indicators on the single window deviation value matrix to obtain a nonlinear correlation matrix sequence; Step S345: performing linear and nonlinear correlation fusion on the linear correlation matrix sequence and the nonlinear correlation matrix sequence to obtain a comprehensive correlation matrix sequence; Step S346: Perform a time window comprehensive analysis on the comprehensive correlation matrix sequence to obtain an indicator correlation matrix.

[0015] The present invention integrates the deviation value sequence of each indicator into a unified matrix structure by constructing a multidimensional time series matrix, which is convenient for subsequent time window segmentation and correlation analysis, and lays a data foundation for more effectively mining the correlation between indicators. The deviation value matrix is ​​segmented according to the time window, and the long-term deviation value sequence can be decomposed into multiple short-term data fragments, so as to analyze the correlation between the indicators in each time period respectively, thereby capturing the dynamic relationship between the indicators on different time scales. Extracting the deviation value matrix of a single time window from the deviation value window sequence can provide independent data units for subsequent linear correlation and nonlinear correlation calculations, ensuring the accuracy and independence of the calculation results. Calculating the linear and nonlinear correlations between indicators separately can capture the complex relationship between indicators more comprehensively, not only considering the simple linear relationship between indicators, but also considering the potential nonlinear association, thereby more deeply mining the interaction between indicators. Fusion of linear and nonlinear correlation matrix sequences can comprehensively consider the correlation of different types between indicators, obtain more comprehensive and accurate indicator correlation information, thereby improving the accuracy of subsequent deviation comprehensive evaluation. By conducting a time window comprehensive analysis of the comprehensive correlation matrix sequence, the correlation information of different time windows can be integrated to obtain a final indicator correlation matrix, which reflects the overall correlation between indicators over a period of time and is more stable and representative, providing a more reliable basis for subsequent risk prediction.

[0016] Preferably, step S344 is specifically: The linear correlation of each pair of indicators in the single window is calculated for the single window deviation value matrix to obtain the single window linear correlation matrix; Perform linear correlation matrix aggregation on the single window linear correlation matrix to obtain a linear correlation matrix sequence; Discretize the single window deviation value matrix to obtain a discretized deviation value matrix; The joint probability distribution is calculated based on the discretized deviation value matrix to obtain the joint probability distribution matrix; the univariate probability distribution is calculated based on the discretized deviation value matrix to obtain the univariate probability distribution data; The mutual information value is calculated according to the joint probability distribution matrix and the univariate probability distribution data to obtain the single window mutual information matrix; A nonlinear correlation matrix sequence is constructed according to the deviation value window sequence and the single window mutual information matrix to obtain a nonlinear correlation matrix sequence.

[0017] The present invention can quantify the linear relationship strength and direction between indicators in the current time window by calculating the linear correlation between the indicators in a single window, construct a single window linear correlation matrix, and provide a basis for subsequent linear correlation analysis. The linear correlation matrices of each time window are aggregated together to form a linear correlation matrix sequence, which can retain the linear correlation information of different time windows and facilitate the subsequent analysis of the dynamic changes of linear correlation between indicators. Discretizing the continuous deviation value in the single window deviation value matrix can simplify data representation, reduce calculation complexity, and provide the necessary input data format for subsequent nonlinear correlation calculations (such as mutual information). Calculating the joint probability distribution and the univariate probability distribution can provide the necessary probability basis for subsequent mutual information calculations, which is used to quantify the nonlinear correlation strength between indicators. By calculating the mutual information value between the indicators, the nonlinear correlation strength between the indicators in the current time window can be quantified, and a single window mutual information matrix can be constructed to provide a basis for subsequent nonlinear correlation analysis. Aggregating the mutual information matrix of each time window to form a nonlinear correlation matrix sequence can retain the nonlinear correlation information of different time windows, facilitate the subsequent analysis of the dynamic changes of nonlinear correlation between indicators, and provide support for a more comprehensive understanding of the interaction between indicators.

[0018] Preferably, step S4 comprises the following steps: Step S41: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; Step S42: extracting deviation features from the integrated data set to obtain a deviation feature table; Step S43: performing historical data feature processing on the integrated data set to obtain a historical feature table; Step S44: extracting environmental features from the integrated data set to obtain an environmental feature table; Step S45: constructing a high-order combined feature according to the deviation feature table, the historical feature table and the environmental feature table to obtain a combined feature table; Step S46: performing prediction feature selection on the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a prediction feature set; Step S47: constructing a risk prediction model according to the prediction feature set and the preset annotated data set to obtain a risk prediction model; using the risk prediction model to perform risk prediction on the standardized real-time data, and generating a risk prediction report to obtain a risk prediction report; Step S48: Risk levels are classified according to the risk prediction report to obtain a risk level list; warning triggering and information sending are performed according to the risk level list and a preset warning rule library to obtain warning information.

[0019] The present invention integrates the patient's historical health data and deviation assessment report to form a unified integrated data set, which can combine the patient's static health information and dynamic physiological data change trends, provide a more comprehensive data basis for subsequent feature extraction and risk prediction, and thus improve the accuracy of risk prediction. By extracting deviation features from the integrated data set, the dynamic change trends of the patient's physiological data, such as short-term fluctuations, long-term trends, etc., can be quantified, and these change trends can be converted into features that can be used for risk prediction, thereby improving the sensitivity of the risk prediction model. The patient's historical health data is structured and converted into features that can be used for risk prediction, and the patient's past medical history, medication records, allergy history and other static health information can be integrated into the risk prediction model, thereby improving the personalization and accuracy of risk prediction. By extracting environmental features from the integrated data set, the influence of the patient's environmental factors can be incorporated into the risk prediction model, such as temperature, humidity, air quality, etc., so as to more comprehensively consider the factors affecting the patient's health and improve the accuracy of risk prediction. By constructing high-order combined features, the interaction between different features can be captured, such as the interaction between deviation features and historical features, the interaction between environmental features and historical features, etc., so as to dig deeper risk factors and further improve the performance of the risk prediction model. Predictive feature selection can remove redundant and irrelevant features, reduce the complexity of the model, improve the training efficiency and generalization ability of the model, and avoid overfitting, thereby making the risk prediction model more robust and stable. The risk prediction model is trained using the predictive feature set and the annotated data set, and the trained model is used to predict the risk of the patient's real-time data to generate a risk prediction report. The extracted features can be used for actual risk prediction and provide visual prediction results. Risk levels are divided according to the risk prediction report, and the corresponding warning information is triggered according to the preset warning rules. The risk prediction results can be converted into specific risk levels and the relevant personnel can be notified in time, thereby achieving early warning.

[0020] Preferably, step S42 includes the following steps: Step S421: defining time windows of different lengths for deviation feature extraction according to the integrated data set to obtain a deviation time window list; Step S422: grouping the integrated data set according to the deviation time window list according to the deviation score of the time window grouping to obtain a time window deviation score set; Step S423: Calculate the time window deviation statistical feature on the time window deviation score set to obtain the time window deviation statistical feature set; Step S424: Calculate the time window deviation trend feature for the time window deviation score set to obtain the time window deviation trend feature set; Step S425: merging window deviation features of the time window deviation statistical feature set and the time window deviation trend feature set to obtain a deviation feature table.

[0021] By defining time windows of different lengths, the present invention can capture the deviation of the patient's physiological data from multiple time scales, such as short-term fluctuations, medium-term changes and long-term trends, and provide a more comprehensive time perspective for subsequent deviation feature extraction. By grouping the deviation scores according to the defined time windows, the deviation information of different time scales can be separated, and the statistical features and trend features of each time window can be calculated separately in preparation. The statistical features of the deviation scores in each time window, such as mean, standard deviation, maximum value, minimum value, etc., can be calculated to quantify the distribution of deviations on different time scales, and provide richer feature information for risk prediction. The trend features of the deviation scores in each time window, such as slope, rate of change, etc., can be calculated to capture the changing trend of the deviation of the patient's physiological data, such as whether the deviation is gradually increasing or decreasing, and provide more dynamic feature information for risk prediction. The statistical features and trend features of different time windows are merged into the deviation feature table, and the deviation information of multiple time scales can be integrated together to form a more comprehensive deviation feature set, thereby improving the accuracy and robustness of the risk prediction model.

[0022] Preferably, step S44 includes the following steps: Step S441: defining time windows of different lengths for environmental feature extraction according to the integrated data set to obtain an environmental time window list; Step S442: grouping the integrated data set into environmental data based on time window grouping according to the environmental time window list to obtain a time window environmental data set; Step S443: performing time window environment statistical features on the time window deviation score set to obtain a time window environment statistical feature set; Step S444: Calculate the time window environment change characteristics according to the time window environment statistical feature set and the time window environment data set to obtain the time window environment change feature set; Step S445: merging the window environment features of the time window environment statistical feature set and the time window environment change feature set to obtain an environment feature table.

[0023] The present invention can examine the impact of environmental factors on patients from multiple angles, such as short-term impact, long-term impact, etc., by defining environmental time windows of different lengths, so as to more comprehensively capture the relationship between environmental factors and patient health status. By grouping environmental data according to different time windows, environmental information of different time scales can be separated, so as to prepare for the subsequent calculation of environmental statistical characteristics and change characteristics of each time window. By calculating environmental statistical characteristics in each time window, such as average temperature, maximum temperature, minimum temperature, average humidity, etc., environmental conditions on different time scales can be quantified, and more abundant environmental information can be provided for risk prediction. By calculating environmental change characteristics in each time window, such as temperature change rate, humidity change rate, etc., the changing trend of environmental factors can be captured, such as whether the temperature is gradually increasing or gradually decreasing, so as to provide more dynamic environmental information for risk prediction. By merging environmental statistical characteristics and change characteristics of different time windows into the environmental feature table, environmental information of multiple time scales can be integrated together to form a more comprehensive set of environmental characteristics, so as to improve the accuracy of the risk prediction model and the sensitivity to environmental factors.

[0024] Preferably, step S45 includes the following steps: Step S451: preparing a feature matrix for the deviation feature table, the historical feature table, and the environmental feature table to obtain a feature matrix; Step S452: generating basic combination features for the feature matrix to obtain a basic combination feature set; Step S453: using the deep feature synthesis technology to perform deep feature synthesis on the basic combined feature set to obtain deep features; Step S454: Optimize the deep features using a genetic algorithm to obtain an optimized combined feature set; Step S455: Generate a combination feature table according to the optimized combination feature set to obtain a combination feature table.

[0025] The present invention can integrate different types of features into a unified data structure by merging the deviation feature table, the historical feature table and the environmental feature table into a feature matrix, so as to facilitate the subsequent combined feature generation and feature selection, and ensure that all features are processed consistently in the subsequent analysis. By generating basic combined features on the feature matrix, such as addition, subtraction, multiplication and division operations between features, the interaction between features can be preliminarily explored, and new features can be generated to expand the feature space, providing a basis for subsequent deep feature synthesis. Using deep feature synthesis technology, more complex and deeper combined features can be automatically generated from the basic combined feature set, so as to more comprehensively capture the nonlinear relationship and interaction between features, and significantly improve the expression and prediction capabilities of features. Using genetic algorithms to optimize deep features, feature subsets that contribute most to risk prediction can be screened out from a large number of combined features, reducing feature redundancy, improving the efficiency and generalization ability of the model, and preventing overfitting. The optimized combined feature set is organized into a combined feature table, which can provide a structured feature data set for subsequent feature selection and model training, facilitate the use and management of subsequent steps, and ensure that the features ultimately used for model training are the optimal subsets. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of the steps of a remote nursing management method based on data feature mining is provided; Figure 2 for Figure 1 Detailed implementation steps of step S3 in FIG. Figure 3 for Figure 1 Detailed implementation steps of step S4 in the flowchart;.

[0027] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0028] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0029] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0030] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0031] To achieve this, please refer to Figures 1 to 3 , a remote nursing management method based on data feature mining, comprising the following steps: Step S1: Obtain patient specific conditions and demand data; select and configure data sources according to the patient specific conditions and demand data to obtain a data source configuration table; perform multi-dimensional data collection and multi-dimensional data preprocessing according to the data source configuration table to obtain comprehensive patient data; Step S2: Performing individualized baseline modeling on the comprehensive data of the patient to obtain an individual baseline model; Step S3: performing real-time data standardization on the comprehensive data of the patient to obtain standardized real-time data; performing deviation calculation based on the standardized real-time data and the individual baseline model to obtain a deviation value sequence; performing indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; performing comprehensive deviation evaluation analysis based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation report; Step S4: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; perform deviation feature extraction, historical data feature processing, and environmental feature extraction on the integrated data set to obtain a deviation feature table, a historical feature table, and an environmental feature table, respectively; construct high-order combined features based on the deviation feature table, the historical feature table, and the environmental feature table to obtain a combined feature table; select predictive features for the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a predictive feature set; construct a risk prediction model based on the predictive feature set to obtain a risk prediction model; perform risk prediction processing using the risk prediction model to obtain a risk prediction report; trigger warnings and send information based on the risk prediction report to obtain warning information, so as to realize remote nursing management tasks.

[0032] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a flow chart of the steps of the remote nursing management method based on data feature mining of the present invention. In this example, the remote nursing management method based on data feature mining includes the following steps: Step S1: Obtain patient specific conditions and demand data; select and configure data sources according to the patient specific conditions and demand data to obtain a data source configuration table; perform multi-dimensional data collection and multi-dimensional data preprocessing according to the data source configuration table to obtain comprehensive patient data; In the embodiment of the present invention, the patient's condition information, health status, nursing needs, and expected monitoring indicators are first obtained through various channels. Then, according to these data and preset rules, appropriate data acquisition equipment and sensors are selected, and acquisition parameters are configured to generate a data source configuration table. Finally, multi-dimensional data acquisition is performed according to the configuration table, and the collected raw data is pre-processed, including data cleaning, denoising, missing value filling, and format unification, and finally the patient comprehensive data (PID) containing multi-dimensional information is obtained.

[0033] Step S2: Performing individualized baseline modeling on the comprehensive data of the patient to obtain an individual baseline model; In an embodiment of the present invention, an individualized baseline model (IBM) is constructed using patient comprehensive data (PID). First, the PID is divided into a training data set and a validation data set in chronological order. Then, feature selection is performed on the training data set to screen out key features related to the target variable. Next, a suitable machine learning algorithm (such as AR, GMM, RNN, etc.) is used to train the model on the training data set to obtain an initial baseline model. Afterwards, the initial model is verified and optimized using the validation data set, and the model parameters are adjusted or different models are tried to obtain an optimized baseline model. Finally, the optimized model is regularly updated using the latest patient comprehensive data to obtain the final individual baseline model (IBM).

[0034] Step S3: performing real-time data standardization on the comprehensive data of the patient to obtain standardized real-time data; performing deviation calculation based on the standardized real-time data and the individual baseline model to obtain a deviation value sequence; performing indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; performing comprehensive deviation evaluation analysis based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation report; In an embodiment of the present invention, the real-time data in the patient's comprehensive data is first standardized, for example, using the Z-score standardization method. Then, according to the baseline distribution parameters (such as the mean, variance and weight of the GMM) extracted from the individual baseline model (IBM), the probability density value of the standardized real-time data under the baseline distribution is calculated, and converted into a deviation value, for example, using a negative logarithmic transformation. The deviation values ​​of each indicator are arranged in chronological order to form a deviation value sequence. Next, the deviation value sequence is subjected to an indicator correlation analysis, and the linear correlation (such as the Pearson correlation coefficient) and the nonlinear correlation (such as the mutual information) between the indicators are calculated, and the two are fused to obtain an indicator correlation matrix. Finally, according to the indicator correlation matrix and the deviation value sequence, the deviation of the patient's physiological data is comprehensively evaluated, the deviation assessment score is calculated, and a deviation assessment report (DAR) is generated.

[0035] Step S4: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; perform deviation feature extraction, historical data feature processing, and environmental feature extraction on the integrated data set to obtain a deviation feature table, a historical feature table, and an environmental feature table, respectively; construct high-order combined features based on the deviation feature table, the historical feature table, and the environmental feature table to obtain a combined feature table; perform predictive feature selection on the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a predictive feature set; construct a risk prediction model based on the predictive feature set to obtain a risk prediction model; perform risk prediction processing using the risk prediction model to obtain a risk prediction report; trigger an early warning and send information based on the risk prediction report to obtain early warning information, so as to realize remote nursing management tasks; In an embodiment of the present invention, the patient's historical health data is obtained and integrated with the deviation assessment report (DAR) to form an integrated data set. Then, feature extraction is performed on the integrated data set, including extracting deviation features (such as mean, standard deviation, trend, etc.) from the deviation assessment score, processing historical health data to obtain historical features (such as disease diagnosis, medication records, etc.), and extracting environmental features (such as temperature, humidity, etc.). Next, high-order combined features are constructed based on the extracted features, such as through mathematical operations or logical operations between features. After that, all features are selected to obtain a prediction feature set. Then, a risk prediction model (such as random forest, SVM, GBDT, etc.) is trained using the labeled data set and the prediction feature set. The trained model is used to predict the risk of the patient's real-time data and generate a risk prediction report (RPR). Finally, according to the risk prediction results in the RPR, the risk level is divided, and the corresponding warning information is triggered according to the preset warning rule library, and the warning information is sent to the doctor and family members.

[0036] Preferably, step S1 comprises the following steps: Step S11: Obtain the patient's specific conditions and demand data; select and configure the data source according to the patient's specific conditions and demand data to obtain a data source configuration table; Step S12: collect data in real time according to the data source configuration table to obtain the original data stream; Step S13: performing data preprocessing on the original data stream to obtain a preprocessed data stream; Step S14: extracting data features from the preprocessed data stream to obtain a feature data stream; Step S15: Integrate the characteristic data stream to obtain comprehensive patient data.

[0037] In an embodiment of the present invention, first, the patient's specific situation and demand data, such as age, gender, past medical history, current medication, allergy history, living habits, nursing needs, and physiological indicators expected to be monitored, are obtained through electronic health record systems, patient questionnaires, and doctor interviews. Then, based on these data, combined with pre-defined disease-data source mapping rules and patient personalized needs, the appropriate data acquisition source is automatically selected and the acquisition parameters are configured. For example, for patients with hypertension and sleep apnea, the system will automatically select smart blood pressure monitors, heart rate monitors, and sleep breathing monitors, and configure the sampling frequency and data transmission method. Finally, a data source configuration table is generated to record information such as the data source type, acquisition frequency, data type, and data transmission protocol that each patient needs to use. This configuration table will guide the subsequent data acquisition process to ensure that multi-dimensional data that meets the individual needs of patients is collected.

[0038] Based on the "Data Source Configuration Table", start the corresponding wearable devices and smart home sensors, such as smart watches, smart blood pressure monitors, blood glucose meters, sleep monitors, ambient temperature and humidity sensors, etc., to collect the patient's physiological, environmental and behavioral data in real time. The data collection process follows the preset parameters in the data source configuration table, such as sampling frequency, data type and data transmission protocol. The collected data is transmitted and stored in the form of a raw data stream consisting of timestamps, data types and values. For example, the heart rate data stream contains a series of timestamps and corresponding heart rate values.

[0039] After obtaining the original data stream, a series of data preprocessing techniques are used to improve data quality. First, the data is cleaned using preset rules and thresholds to remove obviously erroneous data, such as data that exceeds the normal range of physiological indicators. Secondly, signal processing techniques such as Kalman filtering or wavelet transform are used to denoise the data to remove the influence of interference and noise. Thirdly, for missing data, linear interpolation or model-based interpolation methods are used to fill in. Finally, the data collected from different data sources are unified in format and converted in units, such as converting blood pressure data collected by different devices into mmHg units. The preprocessed data still maintains the form of a data stream, forming a preprocessed data stream, which provides a high-quality data foundation for subsequent feature extraction.

[0040] Extract meaningful features from the preprocessed data stream for subsequent modeling and analysis. For example, extract heart rate variability (HRV) features such as RMSSD, SDNN, pNN50, etc. from the heart rate data stream; extract sleep duration, deep and light sleep ratio, sleep efficiency, number of apnea, etc. from the sleep data stream; extract steps, activity duration, activity intensity, energy consumption, etc. from the activity data stream. In addition, features such as average temperature, humidity, and light intensity can also be extracted from environmental data. The extracted features constitute a feature data stream in the form of timestamps and feature values. For example, the HRV feature data stream contains a series of timestamps and corresponding feature values ​​such as RMSSD, SDNN, pNN50, etc.

[0041] Multiple feature data streams are aligned and integrated according to timestamps to form comprehensive patient data (PID). PID is stored in a structured form and contains multi-dimensional information such as the patient's physiology, environment, and behavior. For example, a record in a PID may contain information such as heart rate, blood pressure, blood oxygen saturation, sleep stage, activity level, ambient temperature and humidity at a certain point in time. The structured storage of PID facilitates subsequent baseline modeling, deviation analysis, and risk prediction. Timestamp alignment ensures the temporal consistency of data from different data sources and avoids time deviation during data fusion.

[0042] Preferably, step S2 comprises the following steps: Step S21: segmenting the patient's comprehensive data to obtain a training data set and a verification data set; Step S22: performing feature selection on the training data set to obtain a screening feature set; Step S23: training a baseline model according to the training data set and the screened feature set to obtain an initial baseline model; Step S24: using the validation data set to perform model validation and optimization on the initial baseline model to obtain an optimized baseline model; Step S25: updating the optimized baseline model according to the comprehensive data of the patient to obtain an individual baseline model.

[0043] In an embodiment of the present invention, the patient comprehensive data (PID) is segmented in chronological order. For example, the first 80% of the data is used as a training data set for model training; the last 20% of the data is used as a validation data set for model evaluation and parameter adjustment. In order to ensure that the model can adapt to changes in patient data, a sliding window mechanism is adopted. For example, the training data set and validation data set are updated once a day, the data of the latest day is added to the training data set, and the data of the oldest day is removed to keep the size of the data set unchanged.

[0044] Perform feature selection on the training data set to remove redundant and irrelevant features and improve model efficiency and generalization ability. Use a filter-based feature selection method, for example, calculate the Pearson correlation coefficient between each feature and the target variable (e.g., heart rate), and sort them according to the absolute value of the correlation coefficient. Select the top N features or the features with an absolute value of the correlation coefficient greater than a preset threshold to form a screening feature set. For example, select the 10 features with the highest correlation with heart rate, such as resting heart rate, heart rate variability, sleep duration, activity level, etc., to form the screening feature set of the heart rate baseline model.

[0045] Based on the training data set and feature set, an individualized baseline model is trained. Taking the heart rate baseline model as an example, an autoregressive model (AR) is used for training. The AR model is trained using the historical heart rate data in the training data set and other features in the screened feature set, such as sleep duration, activity level, etc., to learn the law of changes in heart rate data over time. During the training process, the least squares method is used to estimate the parameters of the AR model, and the Akaike information criterion (AIC) or the Bayesian information criterion (BIC) is used to select the optimal model order. After the training is completed, the initial baseline model is obtained, which can predict future heart rate values ​​based on the patient's historical data and related features.

[0046] Use the validation dataset to validate and optimize the initial baseline model. Calculate the mean square error (MSE) or mean absolute error (MAE) of the model on the validation dataset to evaluate the prediction accuracy of the model. If the prediction error of the model is large, adjust the model parameters, such as the order or regularization coefficient of the AR model, or try other models, such as Gaussian mixture model (GMM) or recurrent neural network (RNN), to improve the prediction accuracy of the model. Through iterative model validation and parameter adjustment, the optimized baseline model is finally obtained, which has high prediction accuracy on the validation dataset.

[0047] In order to make the baseline model adapt to the changes in the patient's physiological data, the optimized baseline model is regularly updated with the latest comprehensive patient data. For example, the model is incrementally learned every day using the newly added comprehensive patient data. Specifically, the newly added data is added to the training data set, and the model parameters are fine-tuned using the updated training data set. The model update frequency can be adjusted according to the patient's condition and data changes. The updated model can better reflect the patient's current physiological state, improve the accuracy and timeliness of the baseline model, and finally obtain an individual baseline model (IBM).

[0048] Preferably, step S3 comprises the following steps: Step S31: performing real-time data standardization on the patient's comprehensive data to obtain standardized real-time data; performing baseline distribution parameter extraction on the individual baseline model to obtain baseline distribution parameters; Step S32: Calculate the real-time data probability density according to the baseline distribution parameters and the standardized real-time data to obtain the real-time data probability density; Step S33: performing probability density conversion on the real-time data probability density to obtain a deviation value; performing time series sorting on the deviation values ​​to obtain a deviation value sequence; Step S34: performing an indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; Step S35: Perform a comprehensive deviation evaluation based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation score; generate a deviation evaluation report based on the deviation evaluation score and the deviation value sequence to obtain a deviation evaluation report.

[0049] As an example of the present invention, refer to Figure 2 As shown, in this example, step S3 includes: Step S31: performing real-time data standardization on the patient's comprehensive data to obtain standardized real-time data; performing baseline distribution parameter extraction on the individual baseline model to obtain baseline distribution parameters; In an embodiment of the present invention, first, the real-time data in the patient comprehensive data (PID) is standardized. The Z-score standardization method is used to subtract the mean of the indicator in the baseline model from the real-time data of each indicator, and then divide it by the standard deviation of the indicator in the baseline model. The standardized data obeys a standard normal distribution with a mean of 0 and a standard deviation of 1. For example, the patient's real-time heart rate value is subtracted from the average heart rate value of the heart rate baseline model, and then divided by the standard deviation of the heart rate baseline model to obtain a standardized real-time heart rate value. At the same time, the baseline distribution parameters are extracted from the individual baseline model (IBM). Assuming that IBM is constructed using a Gaussian mixture model (GMM), the parameters such as the mean, variance, and weight of the GMM corresponding to each indicator are extracted to form the baseline distribution parameters.

[0050] Step S32: Calculate the real-time data probability density according to the baseline distribution parameters and the standardized real-time data to obtain the real-time data probability density; In an embodiment of the present invention, based on the baseline distribution parameters and the standardized real-time data, the probability density value of the real-time data under the baseline distribution is calculated. Taking the heart rate index as an example, if the heart rate baseline model adopts a Gaussian mixture model (GMM) modeling, the probability density value of the standardized real-time heart rate value under the GMM is calculated according to the mean, variance and weight parameters of the GMM. The real-time data at each time point will calculate a corresponding probability density value.

[0051] Step S33: performing probability density conversion on the real-time data probability density to obtain a deviation value; performing time series sorting on the deviation values ​​to obtain a deviation value sequence; In an embodiment of the present invention, the probability density value of real-time data is converted into a deviation value. A negative logarithm conversion method is adopted to take the negative logarithm of the probability density value to obtain a deviation value. The smaller the probability density value, the greater the deviation between the real-time data and the baseline model, and the larger the corresponding deviation value. The deviation value of each indicator is sorted according to the timestamp to form a deviation value sequence. For example, the deviation values ​​corresponding to the real-time heart rate data at each time point are arranged in chronological order to form a heart rate deviation value sequence.

[0052] Step S34: performing an indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; In an embodiment of the present invention, an indicator correlation analysis is performed on the deviation value sequence. First, the deviation value sequences of all indicators are constructed into a multidimensional time series matrix. Then, the sliding window technology is used to divide the multidimensional time series matrix into multiple time windows. For the deviation value data in each time window, the Pearson correlation coefficient between the indicators is calculated to measure the linear correlation between the indicators. At the same time, the mutual information between the indicators is calculated to measure the nonlinear correlation between the indicators. The linear and nonlinear correlations are weighted averaged to obtain the final indicator correlation matrix.

[0053] Step S35: Perform a comprehensive deviation evaluation based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation score; generate a deviation evaluation report based on the deviation evaluation score and the deviation value sequence to obtain a deviation evaluation report; In an embodiment of the present invention, a comprehensive evaluation is performed on the deviation of the patient's physiological data based on the indicator correlation matrix and the deviation value sequence. The weighted summation method is adopted to multiply the deviation value of each indicator by the weight of the indicator in the correlation matrix, and then sum them to obtain the deviation evaluation score. The weight reflects the importance of the indicator and its correlation with other indicators. The higher the deviation evaluation score, the greater the degree of deviation of the patient's physiological data and the higher the potential health risk. Finally, a deviation assessment report (DAR) is generated based on the deviation evaluation score, the deviation value sequence and the indicator correlation matrix. DAR contains information such as the deviation value, deviation trend, indicator correlation and comprehensive evaluation results of each indicator, and is presented in a visual way, such as tables, charts, etc., so that medical staff can quickly understand the health status of the patient.

[0054] Preferably, step S34 includes the following steps: Step S341: construct a multi-dimensional time series matrix for the deviation value sequence to obtain a deviation value matrix; Step S342: segmenting the deviation value matrix into time windows to obtain a deviation value window sequence; Step S343: extracting a single window deviation matrix from the deviation value window sequence to obtain a single window deviation value matrix; Step S344: performing linear correlation calculation between indicators on the single window deviation value matrix to obtain a linear correlation matrix sequence; performing nonlinear correlation calculation between indicators on the single window deviation value matrix to obtain a nonlinear correlation matrix sequence; Step S345: performing linear and nonlinear correlation fusion on the linear correlation matrix sequence and the nonlinear correlation matrix sequence to obtain a comprehensive correlation matrix sequence; Step S346: Perform a time window comprehensive analysis on the comprehensive correlation matrix sequence to obtain an indicator correlation matrix.

[0055] In an embodiment of the present invention, the deviation value sequences of various indicators are integrated into a multidimensional time series matrix. Each row of the matrix represents a time point, and each column represents the deviation value of an indicator. For example, the first column is the heart rate deviation value sequence, the second column is the blood pressure deviation value sequence, the third column is the blood oxygen saturation deviation value sequence, and so on. Ensure that the deviation value sequences of all indicators are aligned in time, for example, the first row of the matrix contains the deviation values ​​of all indicators at the first time point.

[0056] The multidimensional time series matrix is ​​split according to the preset time window size. For example, the time window size is set to 5 minutes, and the sliding window is split with a step size of 1 minute. The deviation value data of every 5 minutes is used as a time window to form a series of deviation value window data. The data in each time window constitutes a submatrix, the number of rows of the submatrix is ​​the time window size (for example, 5), and the number of columns is the number of indicators. All submatrices constitute a deviation value window sequence.

[0057] Extract the deviation value data of a single time window from the deviation value window sequence one by one to form a single window deviation value matrix. Each single window deviation value matrix represents the multi-dimensional deviation information within a time window. For example, extract the deviation value data of the first time window to form a matrix of 5 rows (5 minutes) x indicator number columns.

[0058] For each single window deviation value matrix, the linear correlation and nonlinear correlation between the indicators are calculated. The linear correlation is calculated using the Pearson correlation coefficient. The Pearson correlation coefficient is calculated for each two columns of indicator data in the matrix (for example, heart rate deviation value and blood pressure deviation value), and a correlation coefficient between -1 and 1 is obtained, which reflects the degree of linear correlation between the two indicators. The Pearson correlation coefficients between all indicators are combined to form a linear correlation matrix. The nonlinear correlation is calculated using the mutual information method. First, the data in the single window deviation value matrix is ​​discretized, for example, the deviation value of each indicator is divided into three levels: high, medium, and low. Then, the mutual information value between each two indicators is calculated to measure the degree of nonlinear correlation between them. The mutual information values ​​between all indicators are combined to form a nonlinear correlation matrix. The above calculations are performed for each single window deviation value matrix, and finally a linear correlation matrix sequence and a nonlinear correlation matrix sequence are obtained.

[0059] The linear correlation matrix sequence and the nonlinear correlation matrix sequence are fused to obtain a comprehensive correlation matrix sequence. The fusion is performed using the weighted average method. For example, the linear correlation weight is set to 0.5 and the nonlinear correlation weight is set to 0.5. The linear correlation matrix and the nonlinear correlation matrix of the corresponding time window are multiplied by the corresponding weights and then added to obtain a comprehensive correlation matrix. The same operation is performed for each time window, and finally a comprehensive correlation matrix sequence is obtained.

[0060] Perform a time window comprehensive analysis on the comprehensive correlation matrix sequence to obtain the final indicator correlation matrix. Calculate the average correlation coefficient between each indicator pair in all time windows to obtain the final indicator correlation matrix. This matrix reflects the overall correlation between indicators over a period of time. For example, calculate the average correlation coefficient of heart rate and blood pressure in all time windows as the final correlation coefficient of heart rate and blood pressure in the indicator correlation matrix.

[0061] Preferably, step S344 is specifically: The linear correlation of each pair of indicators in the single window is calculated for the single window deviation value matrix to obtain the single window linear correlation matrix; Perform linear correlation matrix aggregation on the single window linear correlation matrix to obtain a linear correlation matrix sequence; Discretize the single window deviation value matrix to obtain a discretized deviation value matrix; The joint probability distribution is calculated based on the discretized deviation value matrix to obtain the joint probability distribution matrix; the univariate probability distribution is calculated based on the discretized deviation value matrix to obtain the univariate probability distribution data; The mutual information value is calculated according to the joint probability distribution matrix and the univariate probability distribution data to obtain the single window mutual information matrix; A nonlinear correlation matrix sequence is constructed according to the deviation value window sequence and the single window mutual information matrix to obtain a nonlinear correlation matrix sequence.

[0062] In an embodiment of the present invention, a single window deviation value matrix is ​​obtained. For each pair of indicators in the matrix, such as heart rate deviation value and blood pressure deviation value, their Pearson correlation coefficients are calculated. The calculation formula of the Pearson correlation coefficient is: covariance (X, Y) / (standard deviation (X) * standard deviation (Y)). Wherein, X and Y represent the deviation value vectors of the two indicators. The calculated Pearson correlation coefficient ranges from -1 to 1. The closer the value is to 1, the stronger the positive correlation is. The closer the value is to -1, the stronger the negative correlation is. The value close to 0 indicates a weak correlation. The Pearson correlation coefficients between all pairs of indicators are stored in a matrix to form a single window linear correlation matrix. The matrix is ​​a symmetric matrix with elements on the diagonal being 1.

[0063] The single-window linear correlation matrix calculated for each single-window deviation value matrix is ​​arranged in chronological order to form a linear correlation matrix sequence, which contains the linear correlation information between each indicator in each time window.

[0064] Discretize the deviation value of each indicator in the single window deviation value matrix. For example, use the equal frequency binning method to divide the deviation value of each indicator into three intervals: high, medium, and low. Specifically, sort the deviation values ​​of each indicator, and then evenly divide the deviation values ​​into three intervals. The deviation value in each interval is represented by the corresponding discrete value, for example, the high interval is represented by "2", the middle interval is represented by "1", and the low interval is represented by "0". Replace the original deviation value with the discretized value to obtain a discretized deviation value matrix.

[0065] Based on the discretized deviation value matrix, calculate the joint probability distribution between every two indicators. For example, calculate the joint probability that the heart rate deviation value and the blood pressure deviation value are in the "high", "medium" and "low" intervals respectively. For example, P(heart rate = high, blood pressure = low) represents the probability that the heart rate deviation value is in the "high" interval and the blood pressure deviation value is in the "low" interval. Store the joint probability distribution of all pairs of indicators in a matrix to form a joint probability distribution matrix. At the same time, calculate the univariate probability distribution of each indicator, that is, the probability that each indicator is in the "high", "medium" and "low" intervals. For example, P(heart rate = high) represents the probability that the heart rate deviation value is in the "high" interval.

[0066] According to the joint probability distribution matrix and univariate probability distribution data calculated in the previous step, calculate the mutual information value between each two indicators. The calculation formula of the mutual information value is: ΣP(X, Y)*log(P(X, Y) / (P(X)*P(Y))), where X and Y represent the discrete values ​​of the two indicators, and "Σ" means summing all possible X and Y values. The larger the mutual information value, the stronger the nonlinear correlation between the two indicators. The mutual information values ​​between all pairs of indicators are stored in a matrix to form a single-window mutual information matrix.

[0067] The single-window mutual information matrix calculated for each single-window deviation value matrix is ​​arranged in chronological order to form a nonlinear correlation matrix sequence, which contains the nonlinear correlation information between each indicator in each time window.

[0068] Preferably, step S4 comprises the following steps: Step S41: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; Step S42: extracting deviation features from the integrated data set to obtain a deviation feature table; Step S43: performing historical data feature processing on the integrated data set to obtain a historical feature table; Step S44: extracting environmental features from the integrated data set to obtain an environmental feature table; Step S45: constructing a high-order combined feature according to the deviation feature table, the historical feature table and the environmental feature table to obtain a combined feature table; Step S46: performing prediction feature selection on the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a prediction feature set; Step S47: constructing a risk prediction model according to the prediction feature set and the preset annotated data set to obtain a risk prediction model; using the risk prediction model to perform risk prediction on the standardized real-time data, and generating a risk prediction report to obtain a risk prediction report; Step S48: Risk levels are classified according to the risk prediction report to obtain a risk level list; warning triggering and information sending are performed according to the risk level list and a preset warning rule library to obtain warning information.

[0069] As an example of the present invention, refer to Figure 3 As shown, in this example, step S4 includes: Step S41: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; In an embodiment of the present invention, the patient's historical health data, including past medical history, diagnosis information, medication records, allergy history, family history, laboratory test results, etc., is obtained from an electronic health record (EHR) system, a hospital information system (HIS), or other medical databases. These historical health data are integrated with the data in the deviation assessment report (DAR) generated in step S3. During the integration process, the DAR data and historical health data are aligned according to the timestamp, and the data from different sources are converted into a unified format and unit. The integrated data is stored in an integrated data set, which includes the patient's real-time physiological data deviation information, historical health information, and other relevant information.

[0070] Step S42: extracting deviation features from the integrated data set to obtain a deviation feature table; In an embodiment of the present invention, features related to deviations in the integrated data set are analyzed to construct a deviation feature table. For different time windows (e.g., the past 24 hours, the past 7 days, the past 30 days), the statistical features of each indicator deviation value sequence are calculated, such as the mean, standard deviation, maximum value, minimum value, median, skewness, kurtosis, and the value of a specific quantile. In addition, the change trend of the deviation value sequence is calculated, such as the slope, rate of change, etc. These statistical features and trend features are used as deviation features and stored in the deviation feature table.

[0071] Step S43: performing historical data feature processing on the integrated data set to obtain a historical feature table; In an embodiment of the present invention, feature processing is performed on the historical health data of the patient in the integrated data set to construct a historical feature table. The patient's past medical history, medication records, allergy history and other information are converted into structured features. For example, one-hot encoding is used to convert disease diagnosis information into a binary feature vector, with 1 indicating that the patient has the disease and 0 indicating that the patient does not have the disease. The patient's medication record is converted into features of drug dosage and time of administration. The patient's allergy history is converted into a binary feature vector of allergens. The processed historical data features are stored in the historical feature table.

[0072] Step S44: extracting environmental features from the integrated data set to obtain an environmental feature table; In an embodiment of the present invention, environmental features related to the patient are extracted from the integrated data set to construct an environmental feature table. For different time windows (e.g., the past 24 hours, the past 7 days, the past 30 days), statistical features of environmental data are calculated, such as average temperature, maximum temperature, minimum temperature, average humidity, average light intensity, etc. In addition, the change trend of environmental data is calculated, such as the temperature change rate, humidity change rate, etc. These statistical features and trend features are used as environmental features and stored in the environmental feature table.

[0073] Step S45: constructing a high-order combined feature according to the deviation feature table, the historical feature table and the environmental feature table to obtain a combined feature table; In an embodiment of the present invention, a high-order combined feature is constructed based on a deviation feature table, a historical feature table, and an environmental feature table to capture the interaction between different features. For example, the mean of the heart rate deviation value is multiplied by the average temperature to construct a new combined feature. Or the ratio of the standard deviation of the blood pressure deviation value to the average activity in the past 7 days is calculated to construct another combined feature. The constructed high-order combined feature is stored in a combined feature table. The high-order combined feature can be automatically constructed using Deep Feature Synthesis technology.

[0074] Step S46: performing prediction feature selection on the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a prediction feature set; In an embodiment of the present invention, feature selection is performed on all features in the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to remove redundant features and irrelevant features, thereby improving the efficiency and generalization ability of the model. A wrapper-based feature selection method, such as recursive feature elimination (RFE), is used in combination with a machine learning model (such as a random forest) to evaluate the importance of features, and features with low importance are gradually eliminated. Finally, a subset of features with the most predictive ability is selected to form a prediction feature set.

[0075] Step S47: constructing a risk prediction model according to the prediction feature set and the preset annotated data set to obtain a risk prediction model; using the risk prediction model to perform risk prediction on the standardized real-time data, and generating a risk prediction report to obtain a risk prediction report; In an embodiment of the present invention, a preset annotated data set (including patient characteristics and corresponding risk labels) and a prediction feature set are used to train a risk prediction model. The annotated data set is obtained by retrospectively analyzing historical case data. Select a suitable machine learning model, such as a random forest, a support vector machine, or a gradient boosting decision tree (GBDT), and train the model using the annotated data set. After the training is completed, the trained risk prediction model is used to perform risk prediction on the patient's standardized real-time data to predict the probability of the patient developing a specific health risk (such as heart failure, falls, infections, etc.) in the future. The prediction results and other relevant information, such as the patient's personal information, current physiological indicators, and the type and probability of predicted risks, are integrated into a risk prediction report (RPR). RPR presents the prediction results in a clear and easy-to-understand manner, such as using visualization methods such as charts and tables, to facilitate medical staff to quickly understand the patient's risk status.

[0076] Step S48: Classify the risk levels according to the risk prediction report to obtain a risk level list; trigger the warning and send the information according to the risk level list and the preset warning rule library to obtain the warning information; In an embodiment of the present invention, according to the risk probability predicted in the risk prediction report (RPR), the risk of the patient is divided into different levels, such as low risk, medium risk and high risk. The threshold for dividing the risk level can be adjusted according to the specific application scenario and risk type. For example, patients with a heart failure risk probability lower than 0.1 are classified as low risk, patients with a probability between 0.1 and 0.5 are classified as medium risk, and patients with a probability higher than 0.5 are classified as high risk. The risk level information of the patient is stored in a risk level list. Then, according to the risk level list and the preset early warning rule library, the corresponding early warning mechanism is triggered. The early warning rule library predefines the early warning method and information content corresponding to different risk levels. For example, for high-risk patients, the system will immediately send early warning information to doctors and family members; for medium-risk patients, the system will regularly send reminder information to doctors and recommend enhanced monitoring; for low-risk patients, the system will not send early warning information. Early warning information can be sent in a variety of ways such as SMS, email, and APP notification.

[0077] Preferably, step S42 includes the following steps: Step S421: defining time windows of different lengths for deviation feature extraction according to the integrated data set to obtain a deviation time window list; Step S422: grouping the integrated data set according to the deviation time window list according to the deviation score of the time window grouping to obtain a time window deviation score set; Step S423: Calculate the time window deviation statistical feature on the time window deviation score set to obtain the time window deviation statistical feature set; Step S424: Calculate the time window deviation trend feature for the time window deviation score set to obtain the time window deviation trend feature set; Step S425: merging window deviation features of the time window deviation statistical feature set and the time window deviation trend feature set to obtain a deviation feature table.

[0078] In an embodiment of the present invention, time windows of different lengths are defined according to the data characteristics and analysis requirements in the integrated data set to extract deviation features. For example, three time windows are defined: 24 hours, 7 days, and 30 days. The length values ​​of these time windows are stored in a deviation time window list, for example, the deviation time window list = [24, 168, 720], in hours. The selection of these time windows is based on clinical experience and the needs of data analysis. For example, a 24-hour window can reflect short-term changes in patients, while a 30-day window can reflect long-term trends in patients.

[0079] According to the list of deviation time windows, the deviation assessment scores in the integrated data set are grouped according to the time windows. For example, for a 24-hour time window, the deviation assessment scores for the past 24 hours are grouped together; for a 7-day time window, the deviation assessment scores for the past 7 days are grouped together; for a 30-day time window, the deviation assessment scores for the past 30 days are grouped together. Each time window corresponds to a deviation assessment score set, and these sets constitute a time window deviation score set.

[0080] For the deviation score set of each time window, calculate the corresponding statistical features. For example, calculate the mean, standard deviation, maximum, minimum, median, skewness, kurtosis, 25% quantile and 75% quantile of the deviation evaluation score in each time window. Store the calculated statistical features in the time window deviation statistical feature set. Each time window corresponds to a set of statistical features.

[0081] For the deviation score set of each time window, the corresponding trend features are calculated. For example, linear regression is used to fit the deviation evaluation score in each time window, and the slope of the regression line is calculated to represent the change trend of the deviation evaluation score. The rate of change of the deviation evaluation score in the time window can also be calculated, such as (last value - first value) / first value. The calculated trend features are stored in the time window deviation trend feature set. Each time window corresponds to a set of trend features.

[0082] The time window deviation statistical feature set and the time window deviation trend feature set are combined to form a deviation feature table. In the deviation feature table, each row represents the deviation feature of a time window, and each column represents a feature, such as the 24-hour mean, 7-day standard deviation, 30-day slope, etc.

[0083] Preferably, step S44 includes the following steps: Step S441: defining time windows of different lengths for environmental feature extraction according to the integrated data set to obtain an environmental time window list; Step S442: grouping the integrated data set into environmental data based on time window grouping according to the environmental time window list to obtain a time window environmental data set; Step S443: performing time window environment statistical features on the time window deviation score set to obtain a time window environment statistical feature set; Step S444: Calculate the time window environment change characteristics according to the time window environment statistical feature set and the time window environment data set to obtain the time window environment change feature set; Step S445: merging the window environment features of the time window environment statistical feature set and the time window environment change feature set to obtain an environment feature table.

[0084] In an embodiment of the present invention, time windows of different lengths are defined according to the data characteristics and analysis requirements in the integrated data set to extract environmental features. The setting of the environmental time window should match the patient's physiological data change cycle and the influence cycle of environmental factors. For example, three time windows are defined: 24 hours, 7 days, and 30 days, so as to capture short-term, medium-term, and long-term environmental change trends. The length values ​​of these time windows are stored in the environmental time window list, for example, the environmental time window list = [24, 168, 720], in hours.

[0085] According to the environmental time window list, the environmental data in the integrated dataset, such as temperature, humidity, light, noise, etc., are grouped according to the time window. For example, the temperature data and humidity data of the past 24 hours are grouped together; the temperature data and humidity data of the past 7 days are grouped together; the temperature data and humidity data of the past 30 days are grouped together. Each time window corresponds to an environmental dataset, which contains all the environmental data within the time window. These datasets constitute the time window environmental dataset.

[0086] For each time window environment data set, calculate the corresponding environment statistical features. For example, calculate the mean, maximum, minimum, and standard deviation of temperature in each time window; calculate the mean, maximum, minimum, and standard deviation of humidity; calculate the mean and standard deviation of light intensity; calculate the mean and standard deviation of noise, etc. Store the calculated statistical features in the time window environment statistical feature set. Each time window corresponds to a set of environment statistical features.

[0087] In addition to statistical features, the changing trend of environmental data is further analyzed. For example, the rate of change of temperature in each time window is calculated, that is, (the last temperature value - the first temperature value) / the first temperature value; the rate of change of humidity; the rate of change of light intensity, etc. Linear regression can also be used to fit the environmental data in each time window, such as temperature, and the slope of the regression line is calculated to represent the changing trend of temperature. The calculated changing trend features are stored in the time window environmental change feature set. Each time window corresponds to a set of environmental change features.

[0088] The time window environment statistical feature set and the time window environment change feature set calculated in step S444 are combined to form an environment feature table. In the environment feature table, each row represents an environment feature of a time window, and each column represents a feature, such as 24-hour average temperature, 7-day temperature change rate, 30-day average humidity, etc.

[0089] Preferably, step S45 includes the following steps: Step S451: preparing a feature matrix for the deviation feature table, the historical feature table, and the environmental feature table to obtain a feature matrix; Step S452: generating basic combination features for the feature matrix to obtain a basic combination feature set; Step S453: using the deep feature synthesis technology to perform deep feature synthesis on the basic combined feature set to obtain deep features; Step S454: Optimize the deep features using a genetic algorithm to obtain an optimized combined feature set; Step S455: Generate a combination feature table according to the optimized combination feature set to obtain a combination feature table.

[0090] In an embodiment of the present invention, the feature data in the deviation feature table, the historical feature table, and the environmental feature table are combined into a feature matrix. Each row of the feature matrix represents a sample (for example, a time point or a time period), and each column represents a feature. For example, the first column can be the 24-hour heart rate deviation mean, the second column can be the 7-day blood pressure deviation maximum, the third column can be whether there is a history of hypertension, and the fourth column can be the 24-hour average temperature, etc.

[0091] Generate basic combination features based on the feature matrix. Basic combination features refer to new features obtained by performing simple mathematical operations (such as addition, subtraction, multiplication, division, square, square root, etc.) or logical operations (such as AND, OR, NOT) on the original features. For example, multiply the 24-hour heart rate deviation mean by the 24-hour average temperature to obtain a new combination feature that represents the interaction between heart rate deviation and temperature. All generated combination features are stored in the basic combination feature set.

[0092] Using deep feature synthesis (DFS) technology, more complex deep features are automatically generated from basic combined feature sets. DFS is a graph-based feature generation method that generates new features by iteratively applying predefined transformation operators (such as addition, subtraction, multiplication, division, average, maximum, minimum, etc.) to original features and generated features. DFS can automatically explore the feature space and generate deep features with stronger expressive power. For example, DFS can first generate the product feature of "24-hour heart rate deviation mean" and "7-day average temperature", and then perform a logical operation on this feature with "whether there is a history of hypertension" to generate a new deep feature.

[0093] Use genetic algorithm to optimize deep features and select the most effective feature subset. Genetic algorithm is an optimization algorithm that simulates the natural selection process. It gradually optimizes feature subsets by simulating operations such as crossover, mutation and selection of genes. Specifically, deep features are used as genes to construct a population of feature subsets. Then, the fitness of the feature subset is evaluated by calculating the performance of each feature subset on the risk prediction model (such as AUC value). Feature subsets with high fitness are selected for crossover and mutation operations to generate new feature subsets. The above operations are iterated until the optimal feature subset is found. The optimal feature subset is stored in the optimized combination feature set.

[0094] The features in the optimized combined feature set and the original features are stored together in the combined feature table. The combined feature table contains the original features and the combined features after deep feature synthesis and genetic algorithm optimization, which are used for subsequent risk prediction model training.

[0095] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0096] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A remote nursing management method based on data feature mining, characterized in that: The following steps are involved: Step S1: Obtain patient specific conditions and demand data; select and configure data sources according to the patient specific conditions and demand data to obtain a data source configuration table; Collect multi-dimensional data according to the data source configuration table, and perform multi-dimensional data preprocessing to obtain comprehensive patient data; Step S2: Performing individualized baseline modeling on the comprehensive data of the patient to obtain an individual baseline model; Step S3: performing real-time data standardization on the comprehensive data of the patient to obtain standardized real-time data; performing deviation calculation based on the standardized real-time data and the individual baseline model to obtain a deviation value sequence; performing indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; performing comprehensive deviation evaluation analysis based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation report; Step S4: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; perform deviation feature extraction, historical data feature processing, and environmental feature extraction on the integrated data set to obtain a deviation feature table, a historical feature table, and an environmental feature table, respectively; construct high-order combined features based on the deviation feature table, the historical feature table, and the environmental feature table to obtain a combined feature table; select predictive features for the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a predictive feature set; construct a risk prediction model based on the predictive feature set to obtain a risk prediction model; perform risk prediction processing using the risk prediction model to obtain a risk prediction report; trigger warnings and send information based on the risk prediction report to obtain warning information, so as to realize remote nursing management tasks.

2. The remote nursing management method based on data feature mining according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Obtain the patient's specific conditions and demand data; select and configure the data source according to the patient's specific conditions and demand data to obtain a data source configuration table; Step S12: collect data in real time according to the data source configuration table to obtain the original data stream; Step S13: performing data preprocessing on the original data stream to obtain a preprocessed data stream; Step S14: extracting data features from the preprocessed data stream to obtain a feature data stream; Step S15: Integrate the characteristic data stream to obtain comprehensive patient data.

3. The remote nursing management method based on data feature mining according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: segmenting the patient's comprehensive data to obtain a training data set and a verification data set; Step S22: performing feature selection on the training data set to obtain a screening feature set; Step S23: training a baseline model according to the training data set and the screened feature set to obtain an initial baseline model; Step S24: using the validation data set to perform model validation and optimization on the initial baseline model to obtain an optimized baseline model; Step S25: updating the optimized baseline model according to the comprehensive data of the patient to obtain an individual baseline model.

4. The remote nursing management method based on data feature mining according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: performing real-time data standardization on the patient's comprehensive data to obtain standardized real-time data; performing baseline distribution parameter extraction on the individual baseline model to obtain baseline distribution parameters; Step S32: Calculate the real-time data probability density according to the baseline distribution parameters and the standardized real-time data to obtain the real-time data probability density; Step S33: performing probability density conversion on the real-time data probability density to obtain a deviation value; performing time series sorting on the deviation values ​​to obtain a deviation value sequence; Step S34: performing an indicator correlation analysis on the deviation value sequence to obtain an indicator correlation matrix; Step S35: Perform a comprehensive deviation evaluation based on the indicator correlation matrix and the deviation value sequence to obtain a deviation evaluation score; generate a deviation evaluation report based on the deviation evaluation score and the deviation value sequence to obtain a deviation evaluation report.

5. The remote nursing management method based on data feature mining according to claim 4 is characterized in that: Step S34 includes the following steps: Step S341: construct a multi-dimensional time series matrix for the deviation value sequence to obtain a deviation value matrix; Step S342: segmenting the deviation value matrix into time windows to obtain a deviation value window sequence; Step S343: extracting a single window deviation matrix from the deviation value window sequence to obtain a single window deviation value matrix; Step S344: performing linear correlation calculation between indicators on the single window deviation value matrix to obtain a linear correlation matrix sequence; performing nonlinear correlation calculation between indicators on the single window deviation value matrix to obtain a nonlinear correlation matrix sequence; Step S345: performing linear and nonlinear correlation fusion on the linear correlation matrix sequence and the nonlinear correlation matrix sequence to obtain a comprehensive correlation matrix sequence; Step S346: Perform a time window comprehensive analysis on the comprehensive correlation matrix sequence to obtain an indicator correlation matrix.

6. The remote nursing management method based on data feature mining according to claim 5 is characterized in that: Step S344 is specifically as follows: The linear correlation of each pair of indicators in the single window is calculated for the single window deviation value matrix to obtain the single window linear correlation matrix; Perform linear correlation matrix aggregation on the single window linear correlation matrix to obtain a linear correlation matrix sequence; Discretize the single window deviation value matrix to obtain a discretized deviation value matrix; The joint probability distribution is calculated based on the discretized deviation value matrix to obtain the joint probability distribution matrix; the univariate probability distribution is calculated based on the discretized deviation value matrix to obtain the univariate probability distribution data; The mutual information value is calculated according to the joint probability distribution matrix and the univariate probability distribution data to obtain the single window mutual information matrix; A nonlinear correlation matrix sequence is constructed according to the deviation value window sequence and the single window mutual information matrix to obtain a nonlinear correlation matrix sequence.

7. The remote nursing management method based on data feature mining according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: Acquire the patient's historical health data; integrate the deviation assessment report and the patient's historical health data to obtain an integrated data set; Step S42: extracting deviation features from the integrated data set to obtain a deviation feature table; Step S43: performing historical data feature processing on the integrated data set to obtain a historical feature table; Step S44: extracting environmental features from the integrated data set to obtain an environmental feature table; Step S45: constructing a high-order combined feature according to the deviation feature table, the historical feature table and the environmental feature table to obtain a combined feature table; Step S46: performing prediction feature selection on the deviation feature table, the historical feature table, the environmental feature table, and the combined feature table to obtain a prediction feature set; Step S47: constructing a risk prediction model according to the prediction feature set and the preset annotated data set to obtain a risk prediction model; using the risk prediction model to perform risk prediction on the standardized real-time data, and generating a risk prediction report to obtain a risk prediction report; Step S48: Risk levels are classified according to the risk prediction report to obtain a risk level list; warning triggering and information sending are performed according to the risk level list and a preset warning rule library to obtain warning information.

8. The remote nursing management method based on data feature mining according to claim 7 is characterized in that: Step S42 includes the following steps: Step S421: defining time windows of different lengths for deviation feature extraction according to the integrated data set to obtain a deviation time window list; Step S422: grouping the integrated data set according to the deviation time window list according to the deviation score of the time window grouping to obtain a time window deviation score set; Step S423: Calculate the time window deviation statistical feature on the time window deviation score set to obtain the time window deviation statistical feature set; Step S424: Calculate the time window deviation trend feature for the time window deviation score set to obtain the time window deviation trend feature set; Step S425: merging window deviation features of the time window deviation statistical feature set and the time window deviation trend feature set to obtain a deviation feature table.

9. The remote nursing management method based on data feature mining according to claim 7 is characterized in that: Step S44 includes the following steps: Step S441: defining time windows of different lengths for environmental feature extraction according to the integrated data set to obtain an environmental time window list; Step S442: grouping the integrated data set into environmental data based on time window grouping according to the environmental time window list to obtain a time window environmental data set; Step S443: performing time window environment statistical features on the time window deviation score set to obtain a time window environment statistical feature set; Step S444: Calculate the time window environment change characteristics according to the time window environment statistical feature set and the time window environment data set to obtain the time window environment change feature set; Step S445: merging the window environment features of the time window environment statistical feature set and the time window environment change feature set to obtain an environment feature table.

10. The remote nursing management method based on data feature mining according to claim 7 is characterized in that: Step S45 includes the following steps: Step S451: preparing a feature matrix for the deviation feature table, the historical feature table, and the environmental feature table to obtain a feature matrix; Step S452: generating basic combination features for the feature matrix to obtain a basic combination feature set; Step S453: using the deep feature synthesis technology to perform deep feature synthesis on the basic combined feature set to obtain deep features; Step S454: Optimize the deep features using a genetic algorithm to obtain an optimized combined feature set; Step S455: Generate a combination feature table according to the optimized combination feature set to obtain a combination feature table.

Images