Home safety monitoring system for retired old people based on Internet of Things

Abstract

The invention belongs to the technical field of intelligent health monitoring, and discloses a retired old people home safety monitoring system based on Internet of Things, which comprises the following steps: deploying multiple types of sensors for the home environment of old people to carry out all-weather data acquisition and analysis to obtain behavior data with activity feature tags; performing correlation analysis to obtain a personalized security risk assessment model; performing real-time monitoring and mode recognition to obtain a behavior track analysis report; performing security level division and intelligent configuration to obtain a virtual security guard area and a self-adaptive reminding rule; performing grade classification processing and early warning signal generation to obtain graded early warning information and a response authority strategy; performing integration analysis to obtain a dynamic intervention strategy, and updating the personalized security risk assessment model according to the dynamic intervention strategy to obtain a target security monitoring model; blind spots and defects of traditional manual monitoring are effectively overcome, and a safe home environment is provided for the old.

Description

Technical Field

[0001] The present invention relates to the field of intelligent health monitoring technology, and more specifically, to a home safety monitoring system for retired elderly people based on the Internet of Things. Background Art

[0002] As the trend of population aging intensifies, the issue of home safety for retired elderly people has become increasingly prominent. At the same time, the rapid development of Internet of Things technology has provided new possibilities for solving this problem. Traditional home monitoring solutions for the elderly mainly rely on simple calling devices or passive camera monitoring, which lack intelligence and foresight, and cannot effectively prevent dangerous events such as falls, getting lost or sudden illnesses. The highly individualized health status and behavioral habits of the elderly make one-size-fits-all monitoring solutions limited in effect and easily cause privacy and excessive intervention issues.

[0003] The existing technologies for home safety monitoring of the elderly have the following main problems: the means of perception are single and lack semantic understanding ability, making it impossible to accurately identify and analyze the complex behavior patterns of the elderly; the risk assessment model is too static and general, making it difficult to adapt to the dynamic changes in the health status and behavioral habits of the elderly; the living space safety monitoring strategy is rigid and cannot provide differentiated protection based on the risk characteristics of different areas; the early warning mechanism is not smart enough, either the sensitivity is too high leading to frequent false alarms, or the response is slow leading to missed reporting of key risk events; the intervention response lacks systematicity and coordination, making it difficult to efficiently dispatch family and medical resources for timely assistance.

[0004] In view of this, the present invention proposes a home safety monitoring system for retired elderly people based on the Internet of Things to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: Collection and processing module: deploy multiple types of sensors in the elderly’s home environment to collect and analyze data around the clock, and obtain behavioral data with activity feature labels; Model building module: performing correlation analysis based on the behavioral data and the health record information of the elderly to obtain a personalized safety risk assessment model; Analysis module: Based on the personalized safety risk assessment model, the daily activities of the elderly are monitored in real time and pattern recognized to obtain a behavior trajectory analysis report; Division module: Based on the behavior trajectory analysis report, the living area of ​​the elderly is divided into security levels and intelligently configured to obtain virtual security guard areas and adaptive reminder rules; Classification module: classifies the behavior data and generates warning signals according to the activity feature labels, the virtual security guard area and the adaptive reminder rules, and obtains graded warning information and response authority strategies; Optimization module: Based on the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information, an integrated analysis is performed to obtain a dynamic intervention strategy, and the personalized safety risk assessment model is updated according to the dynamic intervention strategy to obtain a target safety monitoring model.

[0006] Furthermore, the deployment of multiple types of sensors in the elderly's home environment to collect and analyze data around the clock, and obtain behavioral data with activity feature labels, including: Conduct spatial planning and sensor demand analysis on the elderly’s home environment, obtain a sensor distribution plan, and deploy multiple types of sensor devices based on the sensor distribution plan to build an environmental perception network; Continuously collecting environmental parameters and activity signals through the environmental perception network to obtain an original perception data stream, and performing signal filtering and timing alignment processing on the original perception data stream to obtain cleaned structured data; Based on the structured data, spatiotemporal feature calculation and behavior sequence segmentation are performed to obtain a basic behavior unit set, and a pattern recognition algorithm and context association analysis are applied to the basic behavior unit set to obtain behavior semantic features; Activity classification and attribute labeling are performed according to the behavior semantic features to obtain preliminary behavior labels, and the preliminary behavior labels are verified and enhanced in combination with the historical behavior pattern library to obtain behavior data with activity feature labels.

[0007] Furthermore, the correlation analysis based on the behavior data and the health file information of the elderly is performed to obtain a personalized safety risk assessment model, including: Performing structured analysis and feature extraction on the health record information of the elderly to obtain a health status feature vector, and identifying health risk factors based on the health status feature vector to obtain an initial health risk rating; Based on the initial health risk rating and the behavior data, a correlation analysis is performed on the relationship between the health status and the behavior pattern to obtain a risk association rule set and a health behavior characteristic model, and according to the risk association rule set and the health behavior characteristic model, the health status and the behavior pattern are cross-validated and weighted to obtain a health behavior risk association map; Calculating the importance and prioritizing the risk factors based on the health behavior risk association map to obtain a risk factor weight matrix, and quantifying the levels of potential risk scenarios based on the risk factor weight matrix and the activity feature labels to obtain an initial risk assessment matrix; Based on the initial risk assessment matrix, a time series pattern analysis is performed on the risk prediction to obtain a set of risk evolution rules, and according to the set of risk evolution rules, a Markov chain modeling and risk transfer probability calculation are performed on the initial risk assessment matrix to obtain a dynamic risk probability distribution; Based on the dynamic risk probability distribution, risk prediction accuracy assessment and threshold optimization are performed to obtain a risk model adjustment parameter set, and based on the risk model adjustment parameter set, risk assessment multi-feature fusion and model training verification are performed to obtain a personalized safety risk assessment model.

[0008] Furthermore, according to the personalized safety risk assessment model, the daily activities of the elderly are monitored in real time and pattern recognized to obtain a behavior trajectory analysis report, including: Real-time collection and signal processing of elderly activity data are performed to obtain an original behavior data stream, and normal behavior definition and abnormality detection are performed on the elderly activities based on the original behavior data stream and the personalized safety risk assessment model to obtain an activity status marking result; Based on the activity state marking results, time and space feature extraction and activity type recognition are performed on the behavior events to obtain a multi-dimensional behavior feature set, and based on the multi-dimensional behavior feature set, the activity patterns of the elderly are analyzed and conventional behavior modeling is performed to obtain a behavior pattern library; Based on the behavior pattern library and historical activity data, a behavior time series model is constructed and trend analysis is performed to obtain a behavior change trajectory diagram, and behavior deviation calculation and risk scoring are performed based on the behavior change trajectory diagram and preset health and safety indicators to obtain a behavior risk measurement scale; Based on the behavior risk measurement table and the personalized safety risk assessment model, risk level classification and urgency ranking are performed to obtain a risk event priority queue, and monitoring tasks are intelligently allocated and monitored in parallel through the risk event priority queue to obtain a distributed monitoring task network; Based on the distributed monitoring task network, behavior data statistics aggregation and pattern extraction are performed to obtain structured behavior records, and activity association analysis and safety trend prediction are performed based on the structured behavior records to obtain a behavior trajectory analysis report.

[0009] Furthermore, based on the behavior trajectory analysis report, the living area of ​​the elderly is divided into security levels and intelligently configured to obtain a virtual security guard area and adaptive reminder rules, including: Performing spatiotemporal distribution statistics and frequency analysis on the behavior trajectory analysis report to obtain a behavior pattern feature matrix, and performing living area correlation calculation and classification based on the behavior pattern feature matrix and environmental factor data to obtain an initial area function set; Based on the initial regional function set, heat map modeling and regional clustering analysis are performed on the correlation between the activity trajectories of the elderly to obtain a behavior density distribution map, and safety risk assessment and level division are performed based on the behavior density distribution map to obtain a preliminary safety area division result; Based on the preliminary safety area division results, high-risk area identification and abnormal behavior analysis are performed to obtain a risk area association matrix, and daily activity pattern modeling and scenario deduction are performed based on the risk area association matrix to obtain a safety monitoring process map; Based on the safety monitoring process map and the health status indicators of the elderly, regional safety warning thresholds are set and monitoring points are deployed to obtain a regional monitoring strategy set, and based on the regional monitoring strategy set, a virtual safety guarding environment is constructed and sensors are configured to obtain a virtual safety guarding area; Based on the virtual safety guard area and the caregiver authority setting, the safety warning behavior is extracted and parameterized to obtain adaptive reminder rules.

[0010] Furthermore, the behavior data is graded and processed and a warning signal is generated according to the activity feature label, the virtual security guard area and the adaptive reminder rule to obtain graded warning information and a response authority strategy, including: Performing scenario analysis and feature extraction on the behavior data to obtain a behavior feature vector set, and performing abnormal behavior severity assessment and priority sorting according to the behavior feature vector set and the activity feature labels to obtain a graded abnormal behavior task list; Automatically selecting and configuring the early warning algorithm and parameters based on the hierarchical abnormal behavior task list and the virtual security guard area to obtain a differentiated early warning scheme, and performing parallel analysis and processing of abnormal behaviors and accuracy verification according to the differentiated early warning scheme to obtain an initial early warning data set; Based on the initial warning data set and the adaptive reminder rules, the warning information is hierarchically organized and notified to control the marking, so as to obtain a warning information package with multi-level protection, and the response authority is finely divided and the strategy is generated according to the multi-level protection warning information package to obtain an initial response authority rule set; Based on the initial response authority rule set and monitoring scenario requirements, response process state machine modeling and conversion rule definition are performed to obtain a dynamic response state diagram, and response operation risk assessment and threshold setting are performed according to the dynamic response state diagram to obtain response risk control parameters; Based on the response risk control parameters, early warning signals are generated and distributedly pushed to obtain a distributed early warning instruction set, and protocol design and performance verification of the response process are performed according to the distributed early warning instruction set to obtain hierarchical early warning information and response authority strategy.

[0011] Furthermore, the integrated analysis based on the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information to obtain a dynamic intervention strategy includes: Performing time series correlation and pattern extraction on the behavior trajectory analysis report and the processing record of the graded warning information to obtain a health risk feature matrix, and performing abnormal state correlation calculation and classification according to the health risk feature matrix and the response authority strategy to obtain an initial risk scenario set; Based on the initial risk scenario set, trend chart modeling and type cluster analysis are performed on the health status change trend of the elderly to obtain a health status change chart, and medical needs assessment and urgency classification are performed based on the health status change chart to obtain a preliminary intervention needs classification result; Based on the preliminary intervention demand classification results, high-risk state identification and intervention effect analysis are performed to obtain an intervention demand association matrix, and emergency response mode modeling and plan deduction are performed according to the intervention demand association matrix to obtain an intervention process map; Based on the intervention process map and external medical resource information, medical resource allocation thresholds are set and rescue path planning is performed to obtain a resource allocation strategy set, and based on the resource allocation strategy set, a dynamic intervention environment is constructed and a rescue team is configured to obtain a dynamic intervention strategy.

[0012] Furthermore, the personalized safety risk assessment model is updated according to the dynamic intervention strategy to obtain a target safety monitoring model, including: The dynamic intervention strategy and historical intervention records are associated and pattern mined to obtain an intervention effect sequence diagram, and the effectiveness and adaptability of the intervention measures are evaluated and tested based on the intervention effect sequence diagram and changes in the health status of the elderly to obtain an intervention effect measurement table; Perform risk factor association analysis based on the intervention effect measurement table and the behavioral characteristics of the elderly to obtain a risk factor association network, and perform potential health risk reasoning and probability estimation based on the risk factor association network to obtain a health risk probability tree; Based on the health risk probability tree, optimal allocation of monitoring resources and monitoring planning are performed to obtain a preliminary monitoring strategy plan, and feasibility assessment and priority sorting of monitoring measures are performed according to the preliminary monitoring strategy plan to obtain a hierarchical monitoring strategy set; Dynamically generate and check the consistency of monitoring rules based on the hierarchical monitoring strategy set to obtain a target monitoring rule base, and automatically arrange the daily monitoring process and build a decision tree according to the target monitoring rule base to obtain an intelligent monitoring solution; Based on the intelligent monitoring solution, risk assessment parameters are dynamically adjusted and monitoring conditions are updated to obtain a temporary safety assessment strategy. According to the temporary safety assessment strategy, the personalized safety risk assessment model is globally optimized and balanced to obtain a target safety monitoring model.

[0013] The home safety monitoring device for retired elderly people based on the Internet of Things includes: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory to enable the home safety monitoring device for retired elderly people based on the Internet of Things to execute the home safety monitoring system for retired elderly people based on the Internet of Things.

[0014] A computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute a method for home safety monitoring of retired elderly people based on the Internet of Things.

[0015] The technical effects and advantages of the home safety monitoring system for retired elderly people based on the Internet of Things of the present invention are as follows: The present invention achieves precise monitoring by constructing a risk assessment model, thus avoiding the one-size-fits-all problem of traditional monitoring systems. Through an environmental perception network constructed by multiple types of sensors, intelligent identification of abnormal behaviors and risk classification warnings are realized, greatly improving the accuracy and timeliness of warnings. According to the analysis of behavioral trajectories, safe areas are automatically divided and intelligently configured to form virtual safe guard areas, realizing key monitoring of high-risk areas. Self-optimization is carried out according to the intervention effect and changes in health status, and the risk assessment model is continuously updated to improve the long-term monitoring effect. Dynamic intervention strategies are formulated in combination with external medical resource information to achieve efficient allocation of medical resources in emergency situations. A hierarchical response authority strategy is adopted to protect the privacy and dignity of the elderly to the greatest extent while ensuring safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the home safety monitoring system for retired elderly people based on the Internet of Things of the present invention; Figure 2 It is a schematic diagram of the method for home safety monitoring of retired elderly people based on the Internet of Things of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Embodiment 1; See also Figure 1 As shown, the home safety monitoring system for retired elderly people based on the Internet of Things described in this embodiment includes: Collection and processing module: deploy multiple types of sensors in the elderly’s home environment to collect and analyze data around the clock, and obtain behavioral data with activity feature labels; Model building module: performing correlation analysis based on the behavioral data and the health record information of the elderly to obtain a personalized safety risk assessment model; Analysis module: Based on the personalized safety risk assessment model, the daily activities of the elderly are monitored in real time and pattern recognized to obtain a behavior trajectory analysis report; Division module: Based on the behavior trajectory analysis report, the living area of ​​the elderly is divided into security levels and intelligently configured to obtain virtual security guard areas and adaptive reminder rules; Classification module: classifies the behavior data and generates warning signals according to the activity feature labels, the virtual security guard area and the adaptive reminder rules, and obtains graded warning information and response authority strategies; Optimization module: based on the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information, an integrated analysis is performed to obtain a dynamic intervention strategy, and the personalized safety risk assessment model is updated according to the dynamic intervention strategy to obtain a target safety monitoring model; The modules are connected to each other via wired and / or wireless means to achieve data transmission between modules.

[0019] Deploy multiple types of sensors in the elderly’s home environment to collect and analyze data around the clock, and obtain behavioral data with activity feature labels; Specifically, we will conduct spatial planning and sensor demand analysis on the home environment of the elderly, and determine the key monitoring areas and monitoring parameters by dividing the functional areas of the living space and studying the characteristics of the elderly's daily activities. Functional areas include main living spaces such as bedrooms, living rooms, kitchens, and bathrooms, as well as transition areas such as corridors and stairs. Based on the analysis results, we will formulate a sensor distribution plan to ensure that the sensor network covers all key areas, and configure appropriate types and quantities of sensor equipment according to the characteristics and risk levels of different areas. Based on the sensor distribution plan, we will deploy multiple types of sensor equipment in the home environment, including motion sensors, door and window sensors, pressure sensors, temperature and humidity sensors, smoke sensors, and smart cameras, to build a comprehensive environmental perception network.

[0020] Through the environmental perception network, environmental parameters and activity signals are continuously collected to achieve all-weather monitoring of the living environment and behavioral activities of the elderly. Environmental parameters include environmental indicators such as temperature, humidity, light, and air quality, and activity signals include behavioral-related data such as motion detection, door and window status, and mattress pressure distribution. These data together constitute the original perception data stream, providing basic data support for subsequent behavioral analysis. The collected original perception data stream is subjected to signal filtering and timing alignment processing to eliminate sensor noise and interference factors in the data collection process, and the data of different sensors are synchronized and aligned according to a unified time standard to obtain cleaned structured data, providing a high-quality data foundation for subsequent behavioral analysis.

[0021] Based on structured data, spatiotemporal feature calculation and behavior sequence segmentation are performed. Through time window analysis and behavior feature extraction, the start and end points of the elderly's activities are identified, and the continuous perception data stream is divided into discrete behavior fragments. These behavior fragments form a basic behavior unit set, each unit represents a relatively independent and complete behavior action or activity process. Deep learning pattern recognition algorithms and context association analysis are applied to the basic behavior unit set to identify the intrinsic characteristics and semantic meaning of each behavior unit, while considering the environmental conditions, time background and logical relationship between the previous and next behaviors, and obtaining behavioral semantic features with a semantic understanding level.

[0022] Activities are classified and attributes are annotated based on the semantic features of the behavior. The identified behavior patterns are classified into daily activity types such as sleeping, eating, washing, resting, and exercising, and key attributes such as duration, intensity, and regularity are annotated to obtain preliminary behavior labels. To improve the accuracy of the labels, the system will verify and enhance the preliminary behavior labels in combination with the historical behavior pattern library, correct possible recognition errors through comparison and similarity analysis with the elderly's past behavior patterns, and supplement detailed attributes related to personal habits, ultimately obtaining accurate and comprehensive behavior data with activity feature labels.

[0023] In a specific embodiment, the deployment of multiple types of sensors in the home environment of the elderly to collect and analyze data around the clock to obtain behavioral data with activity feature tags may specifically include the following steps: Conduct spatial planning and sensor demand analysis on the elderly’s home environment, obtain a sensor distribution plan, and deploy multiple types of sensor devices based on the sensor distribution plan to build an environmental perception network; The environmental parameters and activity signals are continuously collected through the environmental perception network to obtain the original perception data stream, and the original perception data stream is subjected to signal filtering and time sequence alignment processing to obtain cleaned structured data; Based on structured data, spatiotemporal feature calculation and behavior sequence segmentation are performed to obtain a basic behavior unit set, and pattern recognition algorithms and context association analysis are applied to the basic behavior unit set to obtain behavior semantic features; Activities are classified and attributes are labeled according to behavioral semantic features to obtain preliminary behavior labels, which are then verified and enhanced in combination with the historical behavior pattern library to obtain behavioral data with activity feature labels.

[0024] Specifically, we conduct spatial planning and sensor demand analysis on the home environment of the elderly, and use a system engineering approach to divide the home environment into main functional areas, such as bedrooms, living rooms, kitchens, bathrooms, corridors, and entrance areas. For each area, we determine the monitoring demand level through evaluation by gerontologists and activity risk analysis. The monitoring demand level can be expressed as a fusion function of functional importance, accident risk probability, and consequence severity: ,in , and is the weight coefficient, Represents the monitoring requirement level, Represents the importance of function, represents the probability of accident risk, Represents the severity of the consequence and satisfies the weight condition: For example, the bathroom has a high risk of slipping, so the accident risk probability and consequence severity are high; the bedroom has a high functional importance because it involves sleep monitoring. Based on the regional monitoring demand level, a sensor distribution plan is formulated, including sensor type selection, quantity determination, and installation location optimization. Sensor coverage density Level of monitoring needs Proportional to: ,in is the proportional coefficient. Based on the sensor distribution solution, PIR motion sensors (bedrooms, living rooms, corridors), door magnetic sensors (entrances, kitchens, bathrooms), pressure sensors (mattresses, seats), temperature and humidity sensors (each room), smoke and gas sensors (kitchens), smart cameras (main activity areas, equipped with privacy protection technology), and water flow sensors (bathrooms, kitchens) are deployed in the home environment. These sensors form a mesh topology environment perception network through wireless communication protocols such as ZigBee, Z-Wave or Wi-Fi, realizing seamless data collection and transmission.

[0025] Environmental parameters and activity signals are continuously collected through the environmental perception network. The sampling frequency is dynamically adjusted according to the parameter change characteristics. Environmental parameters (temperature, humidity, light, etc.) use a lower sampling rate (such as 5 minutes / time), and activity signals (exercise, pressure, etc.) use a higher sampling rate (such as 10 seconds / time or event trigger). The data collection process forms a multi-dimensional time series: ,in Indicates that the first sensor is at time The readings, Indicates Sensors at time The original perception data stream is subjected to signal filtering and timing alignment, and a Kalman filter is used to eliminate random noise. The filtering process can be expressed as: ,in For time The estimated state at the moment, For time The observed value at time, is the state transfer matrix, is the observation matrix, is the Kalman gain. For outlier detection, the modified Z-score method is used: , is the normalized value, represents correction data (such as environmental parameters), where and are the mean and standard deviation in the moving window respectively. When the preset threshold (usually set to 3.5) is exceeded, it is determined as an outlier and replaced or interpolated. The dynamic time warping (DTW) algorithm is used for time series alignment to unify the data of different sensors onto a normalized time axis to ensure the temporal consistency of the event sequence. After processing, cleaned structured data is formed, where each structured data contains a timestamp, location information, and sensor readings.

[0026] Based on structured data, spatiotemporal feature calculation and behavior sequence segmentation are performed. The sliding time window method is used for preliminary segmentation. The window size WID is dynamically adjusted according to the behavior type, and the typical value is 30 seconds to 5 minutes. Statistical features (mean, variance, peak, etc.) and frequency domain features (main frequency components are extracted by fast Fourier transform) are calculated for the data in each window. Change point detection algorithms such as CUSUM (cumulative sum control chart) are used to identify behavior transition points. Through the identified behavior transition points, the data stream is segmented into a series of basic behavior unit sets. A deep learning model is applied to the basic behavior unit set for pattern recognition, and the long short-term memory network (LSTM) or time convolution network (TCN) is used to process the time series features. The model structure includes an input layer (feature dimension p), 1 to 2 hidden layers (64-128 neurons per layer), and an output layer (number of behavior categories). The probability of each behavior pattern is calculated by the softmax function, and context association analysis is performed at the same time. The conditional random field (CRF) model is used to calculate the conditional probability of the behavior sequence. Through pattern recognition and context analysis, behavioral semantic features with rich semantic information are obtained.

[0027] Activities are classified and attributed according to the semantic features of behaviors, and a behavior ontology model is established to map the identified behavior patterns to a predefined activity category system. Activity categories include personal life activities (such as sleeping, eating, washing), instrumental daily activities (such as cooking, cleaning, going out), and social activities (such as talking on the phone, meeting guests), etc. Attributes are annotated for each activity, including basic attributes (duration, start and end time points, location) and extended attributes (activity intensity, completion quality, regularity). Activity intensity is calculated by the amplitude and change rate of the relevant sensor signal: ,in Represents the intensity of activity, For the The activity index of a signal, For the The weight coefficient of each signal. The completion quality is obtained by evaluating the similarity with the standard pattern, and the value range is [0,1]. These attribute assignments form the initial behavior label set T_initial. To improve the accuracy of the label, the system maintains a personalized historical behavior pattern library HIST, which contains the statistical characteristics and time distribution model of the elderly's past behavior patterns. The preliminary label is compared with the historical pattern to obtain the label similarity. The comparison method uses a similarity comparison function. Common similarity comparison functions include cosine similarity. When the label similarity is lower than the preset similarity threshold, the anomaly detection and label correction process is triggered, and the Bayesian reasoning or decision tree method is used for label optimization. Finally, accurate and comprehensive behavior data with activity feature labels are generated, and each record contains original data and rich behavior label information.

[0028] Based on the association analysis of the behavioral data and the health record information of the elderly, a personalized safety risk assessment model is obtained; Specifically, the health record information of the elderly is structured and analyzed and features are extracted to obtain the health status feature vector, which comprehensively captures the key health indicators of the elderly, such as basic health status, chronic disease status, and physical function status. Natural language processing technology is applied to the unstructured text data in the health record to extract key health descriptions and medical advice information. At the same time, the structured data is standardized to unify the dimensions and metrics. The extracted health information is converted into a numerical feature vector through feature engineering methods for subsequent analysis and processing.

[0029] Health risk factors are identified based on the health status feature vectors. The feature vectors are analyzed using medical knowledge graphs and expert rule systems to identify health risk factors related to home safety, such as cardiovascular disease risk, cognitive decline risk, impaired balance risk, etc. A comprehensive assessment is then conducted based on the identified risk factors to obtain an initial health risk rating.

[0030] Based on the initial health risk rating and behavior data, the correlation analysis of the relationship between health status and behavior patterns is carried out to explore the intrinsic connection between health status and daily behavior patterns. The correlation coefficient matrix between health characteristics and behavior characteristics is calculated by statistical methods, and significantly correlated feature pairs are identified. The association rule mining algorithm is used to extract the association rules between health status and specific behavior patterns to form a risk association rule set. At the same time, a behavior pattern prediction model based on health characteristics is constructed to capture the behavioral characteristics under different health conditions and obtain a health behavior characteristic model.

[0031] Based on the risk association rule set and health behavior characteristic model, health status and behavior patterns are cross-validated and weighted, the reliability and importance of each risk association rule are evaluated, weight values ​​are assigned to different rules, and a multi-layer association network between health status nodes and behavior pattern nodes is constructed to form a health behavior risk association map, which can intuitively display the complex relationship between health status and behavioral risk.

[0032] Based on the health behavior risk association map, the importance of risk factors is calculated and prioritized. The centrality index of each node in the map is calculated using the graph structure analysis method to evaluate the influence and importance of different risk factors in the overall risk network. In combination with the knowledge of medical experts, the risk factors are prioritized by importance to obtain the risk factor weight matrix. Based on the risk factor weight matrix and activity feature labels, the potential risk scenarios in daily life are systematically evaluated and quantified, and a risk scenario scoring system is established to obtain the initial risk assessment matrix.

[0033] Based on the initial risk assessment matrix, the time series pattern analysis of risk prediction is carried out to study the evolution law of risk factors over time. Through the time series analysis method, the periodic changes, trend changes and mutation characteristics of risk indicators are explored, the time series dependencies and trigger mechanisms of risk indicators are identified, and a set of risk evolution laws is summarized. According to the set of risk evolution laws, the initial risk assessment matrix is ​​modeled by Markov chain, and the risk state transition probability matrix is ​​constructed to describe the transition probability and evolution path between different risk states, calculate the steady-state distribution and expected duration of risk states, and obtain the dynamic risk probability distribution.

[0034] Based on the dynamic risk probability distribution, the risk prediction accuracy is evaluated and the threshold is optimized. The accuracy and timeliness of the risk prediction model are verified through historical data. The prediction accuracy of different risk types is evaluated. For risk types with large prediction errors, the prediction algorithm parameters and risk threshold settings are adjusted to obtain the risk model adjustment parameter set. Based on the risk model adjustment parameter set, multi-feature fusion of risk assessment is performed, integrating health features, behavioral features and environmental features, building a multimodal feature fusion framework, applying ensemble learning methods to train personalized risk assessment models, evaluating model performance through cross-validation and optimizing parameters, and finally obtaining a personalized safety risk assessment model that adapts to individual characteristics.

[0035] In a specific embodiment, the association analysis based on the behavior data and the health file information of the elderly to obtain a personalized safety risk assessment model may specifically include the following steps: Performing structured analysis and feature extraction on the health record information of the elderly to obtain a health status feature vector, and identifying health risk factors based on the health status feature vector to obtain an initial health risk rating; Based on the initial health risk rating and the behavior data, a correlation analysis is performed on the relationship between the health status and the behavior pattern to obtain a risk association rule set and a health behavior characteristic model, and according to the risk association rule set and the health behavior characteristic model, the health status and the behavior pattern are cross-validated and weighted to obtain a health behavior risk association map; Calculating the importance and prioritizing the risk factors based on the health behavior risk association map to obtain a risk factor weight matrix, and quantifying the levels of potential risk scenarios based on the risk factor weight matrix and the activity feature labels to obtain an initial risk assessment matrix; Based on the initial risk assessment matrix, a time series pattern analysis is performed on the risk prediction to obtain a set of risk evolution rules, and according to the set of risk evolution rules, a Markov chain modeling and risk transfer probability calculation are performed on the initial risk assessment matrix to obtain a dynamic risk probability distribution; Based on the dynamic risk probability distribution, risk prediction accuracy assessment and threshold optimization are performed to obtain a risk model adjustment parameter set, and based on the risk model adjustment parameter set, risk assessment multi-feature fusion and model training verification are performed to obtain a personalized safety risk assessment model.

[0036] Specifically, the health record information of the elderly is structured and analyzed and features are extracted. Natural language processing technology is used to process text information such as diagnostic reports, medical advice and medical history in the health records. First, text segmentation and keyword extraction are performed. The text is segmented using a special word segmenter in the medical field to extract medical terms and key descriptions. Then, named entity recognition technology is used to identify medical entities such as disease names, symptom descriptions, and drug names, with an accuracy rate of more than 95%. Finally, semantic analysis is used to determine the relationship between entities and construct a structured representation of the patient's health status. For structured data such as vital signs and laboratory test results, data cleaning and standardization are performed to convert indicators of different dimensions into unified standard scores. The most representative health indicators are selected through feature selection algorithms, and health status feature vectors are constructed. The vector dimensions are usually 50 to 100, which fully describe the health status of the elderly.

[0037] Health risk factors are identified based on health status feature vectors, using a reasoning method based on medical knowledge graphs. The medical knowledge graph contains a ternary relationship network of disease-symptom-risk. The health feature vectors of the elderly are mapped to the knowledge graph through a graph matching algorithm to identify potential risk factors. The risk identification process can be expressed as: ,in is a set of risk factors, represents the knowledge graph reasoning function, is a knowledge graph. For each identified risk factor, calculate its risk intensity: ,in represents the risk intensity, is the risk weight coefficient, is the indicator function, Represents the indicator value, is the risk threshold. When the indicator value is greater than the risk threshold, 1 is returned. When the indicator value is less than the risk threshold, 0 is returned. Based on the risk factor set and the corresponding strength, a multi-level scoring model is applied to conduct a comprehensive risk assessment on the elderly to obtain an initial health risk rating, which is divided into five levels: very low risk (level 1), low risk (level 2), medium risk (level 3), high risk (level 4) and very high risk (level 5).

[0038] Based on the initial health risk rating and behavior data, the correlation analysis of the relationship between health status and behavior pattern was performed. First, the Pearson correlation coefficient matrix between the health feature vector and the behavior feature vector was calculated. The calculation formula for each element in the correlation coefficient matrix is: ,in, For health characteristics, For behavioral characteristics, is a correlation calculation function. Common correlation calculation functions include cosine similarity. Screening significantly correlated feature pairs ( greater than 0.6) as candidates for in-depth analysis. Then the Apriori algorithm is used to mine association rules and calculate the support ( ) and confidence ( ): , ,in represents a combination of health characteristics, Indicates a combination of behavioral features. Greater than 0.1 and The rules with a value greater than 0.7 constitute the risk association rule set. At the same time, a health behavior feature model was constructed, and the random forest algorithm was used to establish the mapping relationship between health features and behavior features. The model used a forest composed of 500 decision trees, the Gini index was used for feature selection, and 10-fold cross validation was used for training. The model accuracy reached more than 98%.

[0039] According to the risk association rule set and health behavior characteristic model, cross-validation and weight allocation are performed on health status and behavior patterns, and the lift degree of each association rule is calculated ( ): , evaluate the effectiveness of the rules. Sort the rules based on the lift value and assign weights: . Construct a health behavior risk association graph G=(V,E,W), where V is a node set (including health status nodes and behavior pattern nodes), E is an associated edge set, and W is an edge weight set. The graph adopts a multi-layer structure: the bottom-level nodes represent specific health indicators and behavior characteristics, the middle-level nodes represent risk factors and behavior pattern categories, and the top-level nodes represent comprehensive risk types. The connection weights between nodes are determined by the weights of the association rules. The edge weights are: . Through visualization technology, the map is presented as an intuitive network structure, clearly showing the complex relationship between health status and behavioral risks.

[0040] Based on the health behavior risk association map, the importance of risk factors is calculated and prioritized, and the centrality index of each node in the map is calculated, including degree centrality, betweenness centrality, and eigenvector centrality. Combining these three centrality indexes, the weight formula is used to calculate the comprehensive importance score of the risk factor. The risk factors are ranked according to the importance score to construct a risk factor weight matrix ,in Indicates risk factors Risk scenarios Combined with the activity feature label set, the risk score is calculated for each potential risk scenario: ,in Indicates activity characteristics In the scene According to the risk score, the scenarios are divided into different risk levels to form an initial risk assessment matrix. Each element in the initial risk assessment matrix represents the risk level of the elderly in the scenario.

[0041] Based on the initial risk assessment matrix, the time series pattern analysis of risk prediction is carried out, and the time series analysis method is used to process the risk indicator data of continuous monitoring. First, the Fourier transform is applied to identify the periodic pattern of risk indicators: ,in is the risk indicator time function, is the frequency domain representation, represents the angular frequency, Represents an imaginary unit. Then the ARIMA (Autoregressive Integrated Moving Average) model is used to capture the trend changes of the time series, and the change point detection algorithm is used to identify the mutation points of the risk indicators. By analyzing these time series characteristics, a set of risk evolution rules is summarized to describe the evolution pattern of risk factors over time.

[0042] According to the risk evolution law set, the initial risk assessment matrix is ​​modeled by Markov chain to construct the risk state transition probability matrix , each element in the risk state transition probability matrix represents the probability of transitioning from a risk state, and the calculation method is: , Risk status Move to risk state The number of observations, Status The total number of observations. Calculate the steady-state distribution of risk states based on the transition probability matrix , satisfying the distribution conditions: and , represents the probability distribution of the long-term risk level. Calculate the expected duration of the risk state: ,in, Indicates risk status The expected duration of a risk is the average duration of a specific risk state. Combining the time series pattern and the Markov model, a dynamic risk probability distribution is generated to describe the probability distribution of each risk level at different time points.

[0043] Based on the dynamic risk probability distribution, the risk prediction accuracy is evaluated and the threshold is optimized, and the performance of the risk prediction model is verified using historical data. The prediction accuracy is calculated ( )、Accuracy( )、Recall( ) and validation score , the calculation formula for prediction accuracy is: , the calculation formula of precision is: , the calculation formula of recall is: ; The calculation formula for the verification score is: ; in , , , They represent the true positive, true negative, false positive and false negative prediction results respectively. For risk types with poor prediction performance, the risk threshold setting is optimized through ROC curve analysis, and the threshold point that maximizes the Youden index is selected. The risk prediction algorithm parameters, including the time window size, feature weight and model complexity, are adjusted to form a risk model adjustment parameter set.

[0044] Based on the risk model adjustment parameter set, multi-feature fusion and model training verification of risk assessment are performed, and an integrated learning framework is constructed to integrate multi-source feature information. A method combining feature-level fusion and decision-level fusion is adopted: feature-level fusion maps heterogeneous features to a unified feature space through feature transformation and standardization; decision-level fusion integrates the prediction results of different models through multi-model voting or weighted averaging. The gradient boosting decision tree (GBDT) algorithm is used to train the personalized risk assessment model, and the model expression is: ,in is a basic decision tree, is the tree weight. The model training adopts the objective function: ,in is the mean square error loss function, represents the actual term, represents the prediction term, and Ω is the regularization term. The model performance was evaluated by 5-fold cross validation, and the grid search method was used to optimize the hyperparameters, including tree depth, learning rate, and regularization strength. Finally, a personalized safety risk assessment model was obtained, which can assess potential safety risks in real time according to the health status, behavior patterns, and environmental factors of the elderly, and provide a scientific basis for preventive measures.

[0045] According to the personalized safety risk assessment model, the daily activities of the elderly are monitored in real time and pattern recognized to obtain a behavior trajectory analysis report; Specifically, the activity data of the elderly are collected and processed in real time to obtain the original behavior data stream. A distributed sensor network is used for multi-source data collection, including motion sensors, pressure sensors, sound sensors, and wearable devices, to fully capture the activity information of the elderly. The collected original signals are denoised, filtered, and feature enhanced to eliminate environmental interference and equipment errors and improve data reliability. Through signal processing, the heterogeneous data of different sensors are converted into a time series data stream in a unified format, providing a basis for subsequent analysis. Based on the original behavior data stream and the personalized safety risk assessment model, the normal behavior of the elderly activities is defined and anomalies are detected to obtain the activity status marking results.

[0046] Based on the activity status labeling results, the temporal and spatial features of the behavioral events are extracted and the activity types are identified to obtain a multi-dimensional behavioral feature set. By applying feature engineering methods to the labeled behavioral events, temporal features (such as duration, frequency of occurrence, and time distribution) and spatial features (such as activity area, movement trajectory, and position change) are extracted to construct a multi-dimensional feature vector. A hierarchical classification model is used to identify the activity types of behavioral events and distinguish different activity categories such as daily life activities such as eating, sleeping, washing, and watching TV. Based on the multi-dimensional behavioral feature set, the activity patterns of the elderly are analyzed and conventional behaviors are modeled to obtain a behavioral pattern library.

[0047] Based on the behavior pattern library and historical activity data, the behavior time series model is constructed and trend analysis is performed to obtain the behavior change trajectory diagram. Through the time series analysis method, a behavior time series prediction model is established to capture the seasonal changes, periodic laws and long-term trends of the behavior pattern. Combined with historical data, the change rate and deviation degree of the behavior indicators are calculated to generate a trajectory diagram reflecting the behavior evolution process. According to the behavior change trajectory diagram and the preset health and safety indicators, the behavior deviation calculation and risk scoring are performed to obtain the behavior risk measurement scale.

[0048] Based on the behavioral risk measurement table and personalized safety risk assessment model, the risk level is divided and the urgency is sorted to obtain the risk event priority queue. Through hierarchical analysis and risk weight calculation, the detected abnormal events are divided into risk levels and prioritized to ensure that high-risk events can be responded to in a timely manner. Through the risk event priority queue, the monitoring tasks are intelligently allocated and monitored in parallel to obtain a distributed monitoring task network.

[0049] Based on the distributed monitoring task network, the behavior data is statistically aggregated and the pattern is extracted to obtain structured behavior records. Through the data aggregation algorithm, the behavior data collected by the distributed monitoring nodes is summarized and statistically analyzed to extract key behavior features and pattern features. Activity association analysis and security trend prediction are performed based on the structured behavior records to obtain a behavior trajectory analysis report.

[0050] In a specific embodiment, the daily activities of the elderly are monitored and pattern recognized in real time according to the personalized safety risk assessment model to obtain a behavior trajectory analysis report, which may specifically include the following steps:

[0051] Real-time data collection and signal processing of elderly people's activities are performed to obtain the original behavior data stream. Based on the original behavior data stream and the personalized safety risk assessment model, normal behavior definition and abnormality detection of elderly people's activities are performed to obtain activity status marking results. Based on the activity status labeling results, the temporal and spatial features of the behavior events are extracted and the activity types are identified to obtain a multi-dimensional behavior feature set. Based on the multi-dimensional behavior feature set, the activity patterns of the elderly are analyzed and conventional behaviors are modeled to obtain a behavior pattern library. Based on the behavior pattern library and historical activity data, the behavior time series model is constructed and trend analysis is performed to obtain a behavior change trajectory diagram. Based on the behavior change trajectory diagram and preset health and safety indicators, behavior deviation calculation and risk scoring are performed to obtain a behavior risk measurement scale. Based on the behavioral risk measurement table and personalized safety risk assessment model, risk levels are divided and urgency is sorted to obtain a risk event priority queue. Through the risk event priority queue, monitoring tasks are intelligently allocated and monitored in parallel to obtain a distributed monitoring task network. Based on the distributed monitoring task network, behavioral data statistical aggregation and pattern extraction are performed to obtain structured behavior records. Activity correlation analysis and security trend prediction are performed based on the structured behavior records to obtain a behavior trajectory analysis report.

[0052] Specifically, the activity data of the elderly are collected and processed in real time. An IoT sensor system consisting of PIR motion sensors, pressure sensors, smart bracelets, etc. is used to collect the activity data of the elderly at a sampling frequency of 5Hz-20Hz. The PIR sensor detects human activities in the room, the pressure sensor monitors the occupancy of beds, sofas and other locations, and the smart bracelet records physiological parameters such as heart rate, walking steps, and wrist movements. The collected raw signals are processed, including median filtering to remove spike noise, wavelet transform to eliminate baseline drift, and Kalman filtering to fuse multi-source data. The signal processing process can be expressed as: ,in is the original signal, represents the filter function combination, is the processed signal. The processed sensor data are aligned by timestamp to form a unified raw behavior data stream , where each data point Include timestamp and Sensor readings. Normal behavior definition and anomaly detection based on raw behavior data streams and personalized security risk assessment models, using an adaptive threshold method based on individual historical behavior. Define normal behavior boundaries ,in is the mean of historical behavior at a specific time point, is the standard deviation of the historical behavior at a specific time point, It is an adjustable parameter (usually 2-3). , it is marked as a potential anomaly. The potential anomaly is verified based on the spatiotemporal context and the anomaly score is calculated: ,,in, is the abnormality score, is the time deviation component, is the spatial deviation component, is the context bias component, , and is the weight coefficient and satisfies the weight condition: .when When the preset abnormal threshold is exceeded, it is confirmed as abnormal behavior. Through this process, the activity status marking result is obtained: ,in is the timestamp, Indicates normal or abnormal status.

[0053] Based on the activity state labeling results, the temporal and spatial features of the behavioral events are extracted. The sliding window method is used to divide the continuous data stream into event segments of fixed length. For each event segment, the temporal feature vector is extracted. ,in Indicates duration, Indicates the start time, Indicates the frequency of occurrence within 24 hours. The coefficient of variation that represents temporal regularity. At the same time, the spatial feature vector is extracted ,in represents the activity position code, represents the activity coverage area, represents the moving path length, represents the moving speed. The principal component analysis (PCA) is applied to the extracted spatiotemporal features for dimensionality reduction to obtain the spatiotemporal feature vector after dimensionality reduction. The integrated classifier composed of support vector machine (SVM) and random forest (RF) is used to identify the activity type. The activity category set is defined, including common daily activities such as getting up, washing, eating, watching TV, taking medicine, etc. The integrated classifier outputs a probability vector, which represents the probability that the event belongs to each activity category. The final recognition result is the category corresponding to the highest probability. A multi-dimensional behavioral feature set is formed through time features, spatial features and activity types. According to the multi-dimensional behavioral feature set, the activity patterns of the elderly are analyzed and the conventional behavior modeling is carried out. The Gaussian mixture model (GMM) is used to represent the multi-peak distribution characteristics of the behavior. For each type of activity, a GMM model is established. The GMM parameters are learned from historical data through the expectation maximization algorithm. The set of GMM models of all activity categories constitutes a behavioral pattern library.

[0054] The behavior time series model is constructed based on the behavior pattern library and historical activity data, and the long short-term memory network (LSTM) is used to capture the time dependency of the behavior sequence. The input of the LSTM model is the feature vector at time t and the hidden state at the previous time, and the output is the current hidden state and the predicted value. Through the trained LSTM model, the future behavior sequence is predicted and compared with the actual observation value to calculate the change trend of the behavior trajectory. Behavior change trajectory diagram: Indicates at different time points The degree of behavioral deviation Calculate behavior deviation based on behavior change trajectory diagram and preset health and safety indicators, and define behavior deviation index ,in is the actual observed behavior indicator, is the expected normal value, is the standard deviation of the indicator, is the weight coefficient, For the Health and safety indicators include activity frequency, dietary regularity, sleep quality, and medication compliance. Risk score is calculated based on the deviation index. ,in is the adjustment coefficient to control the sensitivity of the score. The risk score range is 0-100, and the higher the score, the greater the risk. Based on the risk scores of multiple indicators, a behavioral risk measurement table is generated: ,in Indicates the indicator name. Indicates the corresponding risk score.

[0055] Risk levels are divided based on behavioral risk measurement tables and personalized safety risk assessment models, and fuzzy logic methods are used to deal with uncertainty in risk assessment. Define a set of risk levels, including: low risk, medium risk, high risk, and emergency risk. The corresponding membership function represents the degree to which the risk score belongs to the risk level. Calculate the priority score based on the risk level and urgency: , where RL is the risk level value (1-4), U is the urgency (time sensitivity), For the difficulty of intervention, , and is the weight coefficient. Risk events are sorted according to the priority scores to form a risk event priority queue: ,in . Intelligently allocate monitoring tasks through risk event priority queues, and adopt a task allocation algorithm based on resource constraints. Define a set of monitoring resources, including family caregivers, community service personnel, telemedicine resources, etc. The resource allocation model takes resource capacity into account , Resource availability and task-resource fit , the objective function is to maximize the risk response efficiency: ,in ∈{0,1} indicates whether the event Assign to resources , and satisfy resource constraints . Through the parallel monitoring mechanism, a distributed monitoring task network is constructed ,in Indicates the event assigned to The best resource.

[0056] Based on the distributed monitoring task network, the behavior data is statistically aggregated, and the hierarchical time aggregation method is used to aggregate the behavior data from different time scales such as hours, days, weeks, and months. Define the aggregation function ,in is the original data set, For time point, is the aggregation level. Through the aggregation function, statistical features such as mean, median, variance, trend, etc. are extracted to form a structured behavior record. According to the structured behavior record, activity association analysis is performed, and the association rule mining algorithm is applied to discover the association relationship between behavior patterns. Define the association rule rule: X→Y, where X and Y are sets of behavior events, support supp(r)=P(X∪Y) represents the probability of X and Y appearing at the same time, and confidence conf(r)=P(Y|X) represents the probability of Y appearing under the condition that X appears. Extract significant association rules whose support and confidence exceed the threshold to form a behavior association network. Based on historical data and discovered behavior patterns, apply the time series prediction model to predict safety trends. Construct an ARIMA(p,d,q) model for key safety indicators, where p is the number of autoregressive terms, d is the difference order, and q is the number of moving average terms. Use the model to predict the trend of safety indicators in the next 7 days and 30 days, and identify potential risk rising points. Integrate activity correlation analysis and safety trend prediction results to generate a behavior trajectory analysis report, including a summary of behavior changes, risk assessment results, early warning prompts, and intervention recommendations.

[0057] Based on the behavior trajectory analysis report, the living area of ​​the elderly is divided into security levels and intelligently configured to obtain a virtual security guard area and adaptive reminder rules; Specifically, the behavior trajectory analysis report is subjected to spatiotemporal distribution statistics and frequency analysis, and the activities of the elderly in different time periods and different spatial locations are quantified and counted to extract temporal regularity and spatial distribution characteristics. Through time series analysis of the daily behaviors of the elderly, regular activity patterns and irregular activity patterns are identified, and the frequency, duration and temporal distribution characteristics of various behaviors are calculated. At the same time, density analysis and path tracking are performed on spatial activities to identify high-frequency activity areas and regular movement paths. These spatiotemporal distribution and frequency data are integrated into a behavior pattern feature matrix, which comprehensively describes the behavioral habits and life patterns of the elderly, providing a data basis for subsequent safety analysis.

[0058] The correlation between living areas is calculated and classified based on the behavior pattern feature matrix and environmental factor data. The environmental factor data includes environmental parameters that may affect the safety of the elderly, such as indoor lighting conditions, temperature and humidity changes, noise levels, and ground materials. By calculating the correlation coefficient between behavioral characteristics and environmental factors, the degree and mode of influence of environmental factors on the behavior of the elderly can be identified. Based on the results of the correlation analysis, the living areas are functionally classified into different functional areas such as bedroom rest area, kitchen activity area, bathroom area, and living room communication area to form an initial regional function set. This functional classification is not only based on the physical layout of the room, but also takes into account the actual usage and behavioral characteristics of the elderly, which is more in line with personalized living habits.

[0059] Based on the initial regional function set, the correlation between the activity trajectories of the elderly is modeled by heat map and regional cluster analysis is performed. By mapping the activity data of the elderly into the spatial coordinate system and taking the activity intensity as the thermal value, an intuitive activity heat map is generated. The high temperature area in the heat map indicates the area with frequent activities, and the low temperature area indicates the area with less activities. The temperature gradient reflects the gradual relationship of the activity frequency. Based on the heat map, the density clustering algorithm is used to perform cluster analysis on the activity area, identify regional clusters with similar activity characteristics, and calculate the center point, range and density characteristics of each cluster to form a behavior density distribution map. This distribution map not only shows the concentration of spatial activities, but also reflects the activity connection and conversion frequency between different areas.

[0060] Safety risk assessment and grading are carried out based on the behavior density distribution map. The potential safety risks of each area are assessed according to the activity density, environmental factors and the health status of the elderly. The risk assessment takes into account a variety of safety factors such as fall risk, electrical safety, scalding risk, slip risk, etc., and conducts a comprehensive assessment based on the physical condition and cognitive ability of the elderly. According to the risk assessment results, the living area is divided into three levels: high-risk area, medium-risk area and low-risk area, forming a preliminary safety area division result. High-risk areas usually include bathrooms, kitchens and other areas with greater potential dangers; medium-risk areas include stairs, corridors and other areas where there may be certain risks; low-risk areas are areas where the elderly rest and carry out safe activities on a daily basis.

[0061] Based on the preliminary safety area division results, high-risk areas are identified and abnormal behavior is analyzed. The characteristics of high-risk areas are analyzed in depth to identify specific risk factors and possible dangerous situations. At the same time, abnormal behavior patterns of the elderly in high-risk areas are analyzed, such as staying in the bathroom for too long, getting up frequently at night, irregular activities in the kitchen area, and other behavioral characteristics that may indicate safety issues. The correlation between risk area characteristics and abnormal behavior patterns is quantified into a risk area correlation matrix, which describes the correlation strength between specific areas, specific behaviors and safety risks, providing a precise positioning basis for safety monitoring.

[0062] According to the risk area association matrix, daily activity patterns are modeled and scenarios are deduced. A state transition model of daily activities for the elderly is constructed to describe the transition rules and conditions between different activity states. Based on this model, multiple safety risk scenarios are deduced and predicted to simulate dangerous situations and response measures that may occur under different conditions. Through scenario deduction, key monitoring points and decision nodes are identified to construct a complete safety monitoring process map. This map describes the complete monitoring process from risk monitoring, abnormality identification to early warning response in the form of a directed graph, and clarifies the triggering conditions and processing logic of each link.

[0063] Based on the safety monitoring process map and the health status indicators of the elderly, regional safety warning thresholds are set and monitoring points are deployed. According to the health indicators of the elderly, such as age, physical condition, and medical history, personalized warning thresholds are set for different safety parameters. For example, for elderly people with limited mobility, the sensitivity threshold of fall detection is lowered; for patients with heart disease, the monitoring frequency of abnormal heart rate is increased. On the basis of determining the warning threshold, the sensor deployment location and density are scientifically planned to form a comprehensive and focused monitoring network, and finally form a regional monitoring strategy set. This strategy set includes the monitoring parameters, warning thresholds, response strategies, and upgrade mechanisms of each region, providing a decision-making basis for the safety guarding system.

[0064] Based on the regional monitoring strategy set, a virtual security guard environment is constructed and sensors are configured, and the physical space is mapped into a virtual monitoring environment to achieve real-time status monitoring and risk warning. In the virtual environment, various sensors are deployed according to the monitoring strategy, including human activity sensors, environmental parameter sensors, image recognition equipment, etc., to build a multi-dimensional perception network. At the same time, a fusion processing mechanism for sensor data is established to improve the accuracy and reliability of risk identification through comprehensive analysis of multi-source data. The final virtual security guard area is an organic combination of physical space and digital monitoring, which not only maintains the natural comfort of the living environment, but also realizes imperceptible security protection.

[0065] Based on the virtual security guard area and caregiver authority settings, the security warning behavior is extracted and parameterized, and the optimal response strategy and notification mechanism under different security risk situations are analyzed. According to the risk level and urgency, hierarchical response rules are formulated to clarify when to notify family members, when to automatically alarm, and when the system handles autonomously. At the same time, considering the different roles and permissions of caregivers, personalized notification content and methods are designed to ensure the accuracy and timeliness of information transmission. The final adaptive reminder rules can dynamically adjust the response strategy according to the actual situation, avoid excessive intervention while ensuring safety, and balance the relationship between security protection and life privacy.

[0066] In a specific embodiment, the security level division and intelligent configuration of the elderly living area based on the behavior trajectory analysis report to obtain a virtual security guard area and adaptive reminder rules may specifically include the following steps: Performing spatiotemporal distribution statistics and frequency analysis on the behavior trajectory analysis report to obtain a behavior pattern feature matrix, and performing living area correlation calculation and classification based on the behavior pattern feature matrix and environmental factor data to obtain an initial area function set; Based on the initial regional function set, heat map modeling and regional clustering analysis are performed on the correlation between the activity trajectories of the elderly to obtain a behavior density distribution map, and safety risk assessment and level division are performed based on the behavior density distribution map to obtain a preliminary safety area division result; Based on the preliminary safety area division results, high-risk area identification and abnormal behavior analysis are performed to obtain a risk area association matrix, and daily activity pattern modeling and scenario deduction are performed based on the risk area association matrix to obtain a safety monitoring process map; Based on the safety monitoring process map and the health status indicators of the elderly, regional safety warning thresholds are set and monitoring points are deployed to obtain a regional monitoring strategy set, and based on the regional monitoring strategy set, a virtual safety guarding environment is constructed and sensors are configured to obtain a virtual safety guarding area; Based on the virtual safety guard area and the caregiver authority setting, the safety warning behavior is extracted and parameterized to obtain adaptive reminder rules.

[0067] Specifically, the spatiotemporal distribution statistics and frequency analysis of the behavior trajectory analysis report are conducted using time series analysis and spatial statistical methods. In terms of time dimension, 24 hours are divided into multiple time periods (such as a 2-hour period), and the activity frequency and type of the elderly in each time period are counted. Indicates the activity type In time period The occurrence frequency of the time distribution feature can be expressed as a time series { In terms of spatial dimension, the living space is divided into grid units, and the proportion of the elderly’s stay time and activity intensity in each grid unit are calculated. Representing coordinates The activity intensity at each location can be used to construct a spatial distribution feature map. Combining temporal features with spatial features forms a behavior pattern feature matrix. , where the matrix elements Indicates the activity type In time period In Location The occurrence intensity of the behavior pattern was calculated based on the behavior pattern feature matrix and the environmental factor data, which included lighting brightness, temperature, humidity and noise level. The correlation coefficient between the behavior intensity and the environmental factors was calculated. ,in Represents the environmental factor vector, which is composed of lighting brightness, temperature, humidity and noise level. The correlation coefficient is calculated by the similarity measurement formula, such as the cosine similarity measurement formula. Based on the correlation analysis results, the K-means or fuzzy C-means clustering algorithm is used to functionally classify the living areas and obtain the initial regional function set. Each element in the set refers to a functional area.

[0068] Based on the initial regional function set, a heat map model is constructed to identify the correlations between the activity trajectories of the elderly. The calculation formula is: ,in Indicates Class Activities at Location The intensity of activity, The weight coefficient representing the activity type reflects the importance of the activity to safety. After the heat map is generated, the DBSCAN density clustering algorithm is used to perform regional clustering analysis. The DBSCAN algorithm identifies high-density areas in the heat map as activity clusters based on two parameters: neighborhood radius and minimum number of points. For a point Point, if the number of points within its neighborhood radius is greater than or equal to the minimum number of points, then Point is a core point; if point Po is within the neighborhood radius of the core point Point, then Point and Po belong to the same cluster. In this way, the natural clustering of elderly activities is identified to form a behavior density distribution map. ,in Indicates location The activity cluster type and density value. Security risk assessment is performed based on the behavior density distribution map. The risk score is obtained by weighted summation of activity density, environmental risk factors, and personal risk factors (such as age, physical condition, etc.). Based on the risk score, the area is divided into high-risk areas based on the preset regional risk threshold ( ), medium risk areas ( ) and low risk areas ( ), and obtain the preliminary safety area division results, where represents the risk score, represents a high risk threshold, Represents the medium risk threshold.

[0069] Based on the preliminary safety zone division results, high-risk areas are identified and abnormal behavior analysis is performed, and the behavior patterns in high-risk areas are analyzed in depth. Define abnormal behavior indicators , indicating activity In time In Location Abnormality: ,in represents the historical average activity intensity, represents the standard deviation. If the value is greater than or equal to the preset activity abnormality threshold, it is considered abnormal behavior. Based on the analysis results of high-risk areas and abnormal behaviors, a risk area association matrix is ​​constructed. , the matrix elements Indicates area and Region The risk correlation strength between them is calculated as follows: ,in Indicates from the area Activities transferred to the region The frequency, and Respectively represent the area and The risk score of the elderly is obtained. The daily activity pattern is modeled according to the risk area association matrix, and the hidden Markov model (HMM) is used to describe the activity state transition of the elderly. The HMM model includes a state set, an observation set, a state transition probability matrix, an observation probability matrix, and an initial state distribution. Among them, the state set corresponds to different activity states, and the observation set corresponds to the observation results of the sensor. The HMM parameters are trained by the Baum-Welch algorithm to obtain a state transition model that can describe the daily activity patterns of the elderly. Based on the HMM model, the Monte-Carlo method is used for scenario deduction to simulate and generate possible activity sequences and corresponding risk situations to form a safety monitoring process map. The safety monitoring process map is a directed graph, in which nodes represent activity states, edges represent state transitions, node attributes include risk levels and monitoring strategies, and edge attributes include transition probabilities and trigger conditions.

[0070] The regional safety warning threshold is set based on the safety monitoring process map and the health status indicators of the elderly. The health status indicators include age index, physical function index and disease risk index. The safety warning threshold is calculated using a weighted formula based on the health status indicators. According to the set safety warning threshold, the deployment location and density of monitoring points are planned. , monitoring point density The calculation formula is: ,in represents the basic monitoring density, Indicates area The risk score of Indicates the risk sensitivity coefficient. The specific location of the monitoring point is determined by the maximum coverage algorithm, and the goal is to use the least number of monitoring points to achieve maximum coverage of key areas. The regional monitoring strategy set is formed by combining the early warning threshold and the monitoring point deployment plan. ,in Indicates the area, Indicates monitoring parameters, Indicates the warning threshold, Represents the response strategy. A virtual security guard environment is constructed based on the regional monitoring strategy set. The virtual environment uses 3D modeling and digital twin technology to map the physical space into a visual monitoring model. In the virtual environment, a sensor network is configured according to the monitoring strategy set, including motion sensors, pressure sensors, temperature and humidity sensors, cameras, etc. The data of various sensors are preliminarily processed by the edge computing node and then uploaded to the cloud for comprehensive analysis. Multi-source sensor data is integrated through data fusion algorithms to improve the accuracy of event recognition, and finally form a virtual security guard area.

[0071] Based on the virtual safety guard area and caregiver authority settings, rules are extracted and parameterized for safety warning behaviors. Warning rules consist of trigger conditions, response strategies, and notification objects. Trigger conditions are logical expressions based on sensor data and thresholds, such as "if (stay time is greater than stay threshold) AND (location∈high-risk area)". The response strategy is determined based on the risk level and urgency, including local reminders, remote notifications, automatic intervention and other levels. The notification object is determined based on the caregiver's authority and relationship with the elderly to ensure that the information is delivered to the appropriate caregiver. Through machine learning methods, warning rule patterns are extracted from historical data and parameterized. Rule effectiveness scoring Calculated by the following formula: ,in Indicates accurate warning, Indicates a false positive. Indicates missed reports. Based on the scoring results, the rules are optimized and adjusted to eventually form an adaptive reminder rule set. The adaptive reminder rules can dynamically adjust parameters according to the behavioral changes and environmental conditions of the elderly to achieve personalized safety protection.

[0072] According to the activity feature label, the virtual security guard area and the adaptive reminder rule, the behavior data is graded and processed and a warning signal is generated to obtain graded warning information and a response authority strategy; Specifically, the behavior data is subjected to scene analysis and feature extraction to establish a multidimensional feature vector containing behavior, location, time and environmental factors, so as to comprehensively capture the contextual information and characteristic attributes of abnormal behaviors of the elderly. Based on the extracted feature vectors and pre-labeled activity feature labels, a multi-index hierarchical analysis method is used to evaluate the severity of abnormal behaviors, a priority sorting model is constructed, and a list of abnormal behavior tasks graded by urgency is generated. Through this list, the system can reasonably allocate computing resources and give priority to high-risk abnormal behaviors.

[0073] According to the hierarchical abnormal behavior task list and the preset virtual security guard area, the system automatically selects the most suitable warning algorithm for the current scenario, optimizes the parameters according to the scenario characteristics, and generates a differentiated warning plan. Then parallel analysis and processing are carried out, and the accuracy of abnormal judgment is cross-validated by multi-model fusion method to generate an initial warning data set with credibility indicators. This data set contains key information such as abnormal type, location, time and severity, which provides a basis for the subsequent organization of warning information.

[0074] Based on the initial warning data set and adaptive reminder rules, the system organizes the warning information in layers, adds notification tags according to the preset notification control logic, and builds a warning information package with a multi-level protection mechanism. For different levels of warning information, the system divides the response permissions in a fine-grained manner, assigns differentiated response permissions to different roles (such as family members, community caregivers, and medical personnel), and forms an initial response permission rule set.

[0075] Combining the initial response authority rule set and specific monitoring scenario requirements, the system establishes a response process state machine model, defines state transition rules, and forms a dynamic response state diagram. Based on this state diagram, the system evaluates the potential risks of each response operation, sets corresponding security thresholds, and generates response risk control parameters. These parameters ensure that the system can maintain a safe boundary when performing response operations and avoid excessive intervention or insufficient response.

[0076] Finally, the system generates graded warning signals based on the response risk control parameters, and pushes the warning instructions to relevant terminal devices in a distributed manner. The system also designs a standardized response process protocol and performs performance verification, ultimately forming a complete graded warning information and response authority strategy to provide all-round protection for the safety of the elderly at home.

[0077] In a specific embodiment, the abnormal behaviors in different scenarios are classified and processed according to the activity feature labels, the virtual security guard area and the adaptive reminder rules, and the warning signals are generated to obtain the graded warning information and the response authority strategy, which may specifically include the following steps: Performing scenario analysis and feature extraction on the behavior data to obtain a behavior feature vector set, and performing abnormal behavior severity assessment and priority sorting according to the behavior feature vector set and the activity feature labels to obtain a graded abnormal behavior task list; Automatically selecting and configuring the early warning algorithm and parameters based on the hierarchical abnormal behavior task list and the virtual security guard area to obtain a differentiated early warning scheme, and performing parallel analysis and processing of abnormal behaviors and accuracy verification according to the differentiated early warning scheme to obtain an initial early warning data set; Based on the initial warning data set and the adaptive reminder rules, the warning information is hierarchically organized and notified to control the marking, so as to obtain a warning information package with multi-level protection, and the response authority is finely divided and the strategy is generated according to the multi-level protection warning information package to obtain an initial response authority rule set; Based on the initial response authority rule set and monitoring scenario requirements, response process state machine modeling and conversion rule definition are performed to obtain a dynamic response state diagram, and response operation risk assessment and threshold setting are performed according to the dynamic response state diagram to obtain response risk control parameters; Based on the response risk control parameters, early warning signals are generated and distributedly pushed to obtain a distributed early warning instruction set, and protocol design and performance verification of the response process are performed according to the distributed early warning instruction set to obtain hierarchical early warning information and response authority strategy.

[0078] Specifically, the behavior data is analyzed and features are extracted. The behavior data is processed through a deep learning model to extract feature vectors containing four dimensions: time, space, behavior, and environment. Among them, time features (such as behavior duration and occurrence time), spatial features (such as location coordinates and movement trajectory), behavioral features (such as posture changes and movement speed), and environmental features (such as indoor temperature and lighting conditions, etc.) together constitute a behavioral feature vector set, which comprehensively describes the behavioral state of the elderly. The severity of abnormal behavior is evaluated based on the behavioral feature vector set and activity feature labels, and the severity score is calculated using a multi-indicator hierarchical analysis method. The indicators include the degree of deviation from the normal behavior pattern, duration, location risk coefficient, and abnormal degree of physiological parameters. Priority is sorted based on the severity score to form a hierarchical abnormal behavior task list, which includes three levels: high risk, medium risk, and low risk.

[0079] The early warning algorithm is automatically selected based on the hierarchical abnormal behavior task list and the virtual security guard area. The system selects the most suitable early warning algorithm from the algorithm library according to the regional characteristics and risk level of the abnormal behavior. The algorithm selection process can be expressed as: ,in is the optimal algorithm selected, is a set of candidate algorithms, It is the virtual security guard area parameter. is the risk level, Is the evaluation function used to calculate the algorithm In the area Risk level For example, for fall detection in high-risk areas such as bathrooms, the system may choose a posture analysis algorithm based on deep learning; while for abnormal behavior detection in open areas such as living rooms, it may choose an abnormal behavior recognition algorithm based on trajectory analysis. Parameter configuration optimization is performed for the selected algorithm using the Bayesian optimization method: ,in is the optimal parameter set, is the candidate parameter set, is the expected value of parameter performance, which is calculated through cross-validation. Through parameter optimization configuration, a differentiated early warning scheme that adapts to the current scenario is obtained. Based on the differentiated early warning scheme, abnormal behavior is analyzed and processed in parallel, and a multi-model parallel computing framework is adopted to run different models at the same time to evaluate abnormal behavior. Parallel processing can be expressed as: ,in Indicates The analysis results of each model are the evaluation scores. The accuracy of the parallel processing results is checked using a weighted voting mechanism: ,in is the credibility coefficient of the model. When the preset threshold is exceeded, the system confirms the existence of abnormal behavior and generates an initial warning data set. Each element in the initial warning data set contains key information such as abnormality type, location, time, severity and credibility.

[0080] Based on the initial warning data set and adaptive reminder rules, the warning information is hierarchically organized, using a three-layer structure: the core layer contains key warning information, the extended layer contains detailed context information, and the additional layer contains auxiliary information and recommended measures. Combined with the adaptive reminder rules, the warning information is marked for notification control, adding receiver authority tags, timeliness tags, and privacy protection tags. The marking process can be expressed as: ,in Indicates information Added Through the marking process, an early warning information package with a multi-level protection mechanism is obtained.

[0081] . Based on the warning information package, the response authority is divided into fine-grained categories, and the authority matrix is ​​defined for different roles. Each element in the authority matrix represents the authority level of the role to the warning information, and the element value set is {0,1,2,3} (0 means no authority, 1-3 means different levels of access and operation authority). According to the authority matrix and the warning level, the policy rules are generated, and each rule can be expressed as a conditional statement in the form of "if (condition) then (action)". These rules together constitute the initial response authority rule set.

[0082] Based on the initial response authority rule set and monitoring scenario requirements, the response process state machine is modeled, and the state set Q={q0,q1,...,qn} is defined, where q0 is the initial state and other states represent different stages in the response process. The state transition function δ(q,e)=q' is defined, which means that after receiving event e in state q, it transitions to state q'. The state transition rule can be expressed as a condition set C={c1,c2,...,cm}, where each condition cm defines the trigger condition for a specific state transition. Through the state set Q, transition function δ and condition set C, a dynamic response state diagram G=(Q,End,δ,q0,Fs) is constructed, where End is the event set and Fs is the terminal state set. Based on the dynamic response state diagram, the response operation risk assessment is performed, the fault tree analysis method is used to identify potential risk points, and the operation risk score is calculated: ,in Indicates operation, Indicates operation The probability of failure, Indicates the severity of the impact of a failure. Set safety thresholds for high-risk operations: ,in is the minimum threshold, These thresholds together constitute the response risk control parameter set.

[0083] Based on the response risk control parameters, the warning signal is graded and generated, and a five-level warning signal system is constructed: emergency, high warning, medium warning, low warning and prompt information. The warning level determination function is: ,in Represents the warning level. For early warning data, In response to the risk control parameter set, is an evaluation function, the evaluation result is 0 or 1, and the evaluation function contains multiple evaluation indicators. The evaluation results of different evaluation indicators are summed. When each evaluation indicator is greater than the corresponding parameter in the response risk control parameter set, the value of the evaluation function is 1. For different levels of early warning signals, a distributed push strategy is adopted: the emergency level is pushed to all relevant terminals at the same time, the high warning level is pushed to the close caregiver terminal first, and other levels are pushed in a targeted manner according to the role and location of the receiver. Through this distributed push mechanism, a distributed early warning instruction set is formed. Based on the early warning instruction set, a standardized response process protocol is designed, including three stages: confirmation of receipt, response action, and result feedback. The protocol is validated for effectiveness, and simulation tests and statistical analysis methods are used to calculate key performance indicators: average response time, success rate, false alarm rate, etc. Through comprehensive evaluation, a complete hierarchical early warning information and response authority strategy are finally formed to provide comprehensive protection for the safe monitoring of the elderly at home.

[0084] Based on the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information, an integrated analysis is performed to obtain a dynamic intervention strategy; Specifically, the behavior trajectory analysis report and the processing records of the graded warning information are time-series associated and pattern extracted to explore the intrinsic correlation between the daily behavior and health status of the elderly, and obtain a feature matrix that fully reflects the health risks of the elderly. This matrix contains multi-dimensional features such as time, space, behavior, and physiology, and can characterize the abnormal patterns of the health status of the elderly. Based on the health risk feature matrix and the response authority strategy, the correlation between each abnormal state is calculated through a multi-factor association analysis method, and a systematic classification is performed to form an initial risk scenario set containing multiple risk scenarios.

[0085] Based on the initial risk scenario set, a trend chart model of the health status of the elderly is constructed, and cluster analysis of different types of health status changes is performed to form a health status change chart. This change chart clearly shows the evolution path and trend characteristics of the health status of the elderly, providing a scientific basis for intervention decisions. Based on the health status change chart, a systematic medical needs assessment is conducted, the urgency of different situations is accurately divided, and the preliminary intervention needs division results are formed, laying the foundation for subsequent precise intervention.

[0086] Based on the preliminary intervention needs classification results, a deep learning method is used to identify high-risk states, and the intervention effects are analyzed in combination with historical intervention records to construct an intervention needs association matrix to comprehensively characterize the correlation and priority between different intervention needs. Based on the intervention needs association matrix, an emergency response model is constructed, and multi-scenario plan deduction is performed to form an intervention process map, providing a clear intervention decision path and execution process.

[0087] Based on the intervention process map and external medical resource information, the threshold of medical resource allocation is set, the optimal rescue path is planned, and a resource allocation strategy set is formed. This strategy set contains resource allocation plans under different levels of urgency to ensure efficient use of medical resources. Based on the resource allocation strategy set, a dynamic intervention environment is constructed, the configuration of the rescue team is optimized, and a comprehensive, accurate, and efficient dynamic intervention strategy is finally formed to provide timely and appropriate safety protection for the elderly.

[0088] In a specific embodiment, the integrated analysis based on the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information to obtain a dynamic intervention strategy may specifically include the following steps: Performing time series correlation and pattern extraction on the behavior trajectory analysis report and the processing record of the graded warning information to obtain a health risk feature matrix, and performing abnormal state correlation calculation and classification according to the health risk feature matrix and the response authority strategy to obtain an initial risk scenario set; Based on the initial risk scenario set, trend chart modeling and type cluster analysis are performed on the health status change trend of the elderly to obtain a health status change chart, and medical needs assessment and urgency classification are performed based on the health status change chart to obtain a preliminary intervention needs classification result; Based on the preliminary intervention demand classification results, high-risk state identification and intervention effect analysis are performed to obtain an intervention demand association matrix, and emergency response mode modeling and plan deduction are performed according to the intervention demand association matrix to obtain an intervention process map; Based on the intervention process map and external medical resource information, medical resource allocation thresholds are set and rescue path planning is performed to obtain a resource allocation strategy set, and based on the resource allocation strategy set, a dynamic intervention environment is constructed and a rescue team is configured to obtain a dynamic intervention strategy.

[0089] Specifically, the behavior trajectory analysis report and the processing record of the graded warning information are temporally associated and pattern extracted, and the behavior trajectory analysis report and the warning processing record are associated and analyzed using a time series analysis method. The system first serializes the behavior trajectory analysis report by timestamp to form a temporal behavior sequence. Each element in the temporal behavior sequence represents a behavior feature vector, including information such as location, action type, and duration. At the same time, the processing record of the graded warning information is formed into a temporal warning sequence. Each element in the temporal warning sequence represents a warning event feature vector, including information such as warning type, level, and processing method. The temporal correlation of the two sequences is calculated using a dynamic time warping algorithm. The algorithm calculates the similarity of the two time series by finding the best alignment path between them, and can accurately capture their correlation even if the sequence lengths are different or there is a time delay.

[0090] Based on the results of time series association analysis, the non-negative matrix decomposition method is used to extract health risk patterns and construct a health risk feature matrix. Assume that the comprehensive feature data matrix is , the dimension is ,in represents the number of samples, Represents the feature dimension. The non-negative matrix factorization method decomposes Xty into the product of two non-negative matrices: ,in for dimensional basis matrix, for dimensional coefficient matrix, is the number of patterns extracted, usually much smaller than and Each column of the basis matrix represents a health risk pattern, and the coefficient matrix represents the weight distribution of each pattern. Through this decomposition, a clear health risk pattern can be extracted from complex behavior and early warning data to form a health risk feature matrix.

[0091] According to the health risk feature matrix and the response authority strategy, the abnormal state correlation is calculated and classified, and the correlation between abnormal states is calculated using the association rule mining algorithm. Strong association rules are identified by calculating support and confidence: , Indicates that an abnormal state occurs at the same time and probability; ,in Indicates abnormal state Under the conditions that occur, The conditional probability of occurrence is also. For association rules that meet the threshold conditions: Greater than or equal to the preset correlation threshold and If the confidence level is greater than or equal to the preset threshold, they will be included in the abnormal state association network.

[0092] Combined with the response permission strategy matrix, the elements in the permission strategy matrix Indicates abnormal status Response permission level The applicability of . By calculating the weighted association: , further optimize the abnormal state association network. The spectral clustering algorithm is used to classify the abnormal state and obtain the abnormal state classification results. Finally, according to the classification results and association strength, the initial risk scenario set is constructed. Each risk scenario contains a set of highly associated abnormal states and corresponding description information.

[0093] Based on the initial risk scenario set, trend graph modeling and type clustering analysis are performed on the trend of changes in the health status of the elderly, and a health status trend graph model is constructed using a time series graph neural network (TGNN). Each scenario in the risk scenario set is regarded as a node in the graph, and a directed edge is constructed through the transition probability within the time span to form an initial trend graph G0=(V,E,A), where V is a node set, E is an edge set, A is an adjacency matrix, and a_ij represents the probability of transitioning from scenario s_i to scenario s_j.

[0094] Use graph convolutional network (GCN) for feature extraction. The function formula of graph convolutional network is: ,in For the The node feature matrix of the layer, is the degree matrix, is the learnable weight matrix, is the activation function. By stacking multiple layers of GCN, high-order feature representations of nodes are extracted to capture the complex change patterns of health status. Combined with the temporal convolutional network (TCN), which captures the temporal features of the last layer of GCN output features, TCN captures the change features of different time scales through one-dimensional convolution and dilated convolution. Finally, the fusion features are obtained through multi-layer perceptron fusion.

[0095] Based on the fusion features, the spectral clustering algorithm is used to cluster the health status change patterns. A similarity matrix is ​​constructed, and the k-medoids clustering algorithm is applied to classify the health status change patterns into multiple categories. Through visualization technology, the clustering results are mapped to a two-dimensional plane to construct a health status change graph, which intuitively shows the evolution path and transfer relationship between different health states.

[0096] Based on the health status change graph, medical needs assessment and urgency division are performed, and a multi-factor scoring model is used to assess medical needs. An evaluation index system is designed, including multiple dimensions such as physiological condition, degree of behavioral abnormality, and warning frequency. For each state node in the health status change graph, a weight formula is used to calculate the comprehensive urgency, that is, the weighted sum of the scores of each state in the corresponding indicator. Threshold division is performed based on the comprehensive urgency, and the health status is divided into multiple urgency levels, for example, it can be divided into "normal", "need attention", "need intervention", "emergency intervention" and other levels. Combined with the urgency level and the health status change graph, a preliminary intervention needs division result is formed to provide a scientific basis for the formulation of subsequent intervention strategies.

[0097] Based on the preliminary intervention needs division results, high-risk state identification and intervention effect analysis are performed, and deep reinforcement learning methods are used to identify high-risk states. The health state space is modeled as a Markov decision process (MDP), the state space is the set of nodes in the health status change graph, and the action space is the set of possible intervention methods. Design a reward function based on health risk: ,in represents the health status before intervention, represents the health status after intervention, Indicates status The urgency of Indicates intervention action The cost, and The optimal intervention strategy is learned through the deep Q network, and the high-risk state set is identified through Q value analysis. The intervention effect is analyzed in combination with historical intervention records, and the intervention effect score is calculated based on the preset weights using the weight formula. The intervention demand association matrix is ​​constructed, and the matrix elements represent the association strength between different intervention demands, which is calculated through the cosine similarity of the eigenvectors.

[0098] According to the intervention demand association matrix, emergency response mode modeling and scheme deduction are carried out, and the emergency response mode is modeled by graph neural network. The intervention demand association matrix is ​​regarded as the adjacency matrix of the weighted graph, and an intervention demand graph is constructed, in which the nodes represent the intervention demand and the edges represent the association between the demands. The graph attention network is used to extract node features, and the features extracted by the graph attention network are used to construct a decision tree model for response scheme deduction. For each intervention demand node, a series of possible intervention schemes are derived based on its feature vector. Each scheme contains elements such as intervention timing, intervention means, and execution role. Through the Monte Carlo tree search algorithm, the execution process of multiple intervention schemes is simulated on the decision tree to evaluate the expected effect and risk of each scheme. Through multiple rounds of deduction and evaluation, the optimal set of intervention schemes is selected to construct an intervention process map, in which the nodes represent intervention decision points or actions, and the edges represent the decision process. The intervention process map adopts a directed acyclic graph structure to clearly show the complete intervention process from anomaly detection to emergency response, providing executable guidance for actual operations.

[0099] Based on the intervention process map and external medical resource information, medical resource allocation threshold setting and rescue path planning are performed, and the resource allocation threshold is set by a multi-objective optimization method. Medical resources are divided into multiple categories, including medical staff, emergency equipment, medicines, etc. For each node in the intervention process map, a resource demand vector is set according to its intervention needs. Combined with external medical resource availability information, a resource allocation constraint condition is constructed: the demand is less than or equal to the external medical resource availability information. A multi-objective optimization problem is designed, taking into account resource utilization efficiency and intervention effect at the same time. Through Pareto optimal solution analysis, the medical resource allocation threshold vector is determined, and when the demand exceeds the threshold, the corresponding level of resource allocation is triggered.

[0100] A heuristic algorithm is used for rescue path planning, which is modeled as a multi-constrained shortest path problem. Consider a set of geographical locations, including the homes of the elderly, medical institutions, and the locations of rescuers. Construct a weighted graph in which the edge weights contain multi-dimensional information such as distance, time, and traffic conditions. In emergency situations, it is necessary to find the optimal path from the rescue source to the destination, minimizing the weighted cost, including the time, distance, and other cost factors of the path. Taking time and distance as an example, a heuristic search algorithm is used to solve the shortest path problem, and the heuristic function is designed as: ,in represents the heuristic estimate, is a time-based heuristic estimate and is a distance-based heuristic estimate, and It is a balance parameter. Based on different emergency levels and resource types, multiple rescue path solutions are generated to form a resource allocation strategy set. Each strategy contains information such as resource allocation threshold, allocation order and rescue path.

[0101] Based on the resource allocation strategy set, dynamic intervention environment construction and rescue team configuration are carried out, and digital twin technology is used to construct a dynamic intervention environment. Information such as the elderly living environment, medical resource distribution and transportation network in the physical space is mapped into the virtual space to form a digital twin model, including the environmental topology, the elderly and related objects, resource entities and status monitoring points. The digital twin model is updated synchronously through real-time data to ensure that the virtual environment is consistent with the actual situation. Based on the resource allocation strategy set and the digital twin environment, a combinatorial optimization method is used to configure the rescue team. A set of rescuers is set, and each person has a capability vector, which represents the professional ability in different intervention tasks. The team configuration problem is modeled as a bipartite graph matching problem, and the optimization goal is: the matching degree between the maximum allocated personnel and the allocated characters and the ability and demand. The optimal personnel allocation plan is solved by the Hungarian algorithm, and combined with the resource allocation strategy, a complete dynamic intervention strategy is formed, including the intervention timing strategy, resource allocation strategy, personnel configuration strategy and intervention process strategy. The dynamic intervention strategy evaluates the current state in real time through the rule engine, adaptively adjusts the intervention parameters, ensures the accuracy and timeliness of the intervention measures, and provides comprehensive and intelligent home safety monitoring for the elderly.

[0102] The personalized safety risk assessment model is updated according to the dynamic intervention strategy to obtain a target safety monitoring model; Specifically, we conduct systematic correlation analysis and deep pattern mining on dynamic intervention strategies and historical intervention records, construct a sequence diagram reflecting the temporal effects of intervention measures, and provide a data basis for the evaluation of intervention measures. Combined with the data on changes in the health status of the elderly, we evaluate the effectiveness and adaptability of different intervention measures, establish a quantitative intervention effect measurement scale, and provide a scientific basis for the subsequent optimization of intervention strategies. Through the correlation analysis of the intervention effect measurement scale and the behavioral characteristics of the elderly, we establish a complex correlation network between risk factors, revealing the causal relationship and impact path between risk factors. Based on the correlation network, we perform health risk reasoning and probability calculation, construct a health risk probability tree, and realize the prediction and quantitative evaluation of potential health risks.

[0103] Based on the health risk probability tree, we optimize the allocation of monitoring resources, comprehensively consider the severity of risk, resource availability and intervention effect, and formulate preliminary monitoring strategy plans. We conduct multi-dimensional feasibility assessment and priority sorting on the preliminary plans to form a hierarchical monitoring strategy set with a hierarchical structure to adapt to different risk levels and resource conditions. Based on the hierarchical monitoring strategy set, we dynamically generate specific monitoring rules, and ensure the coordination between rules through consistency detection to form a complete target monitoring rule base. According to the rule base, we automatically arrange the daily monitoring process and build the decision logic, generate executable intelligent monitoring plans, and realize the intelligence and automation of the monitoring process.

[0104] Based on the intelligent monitoring solution, the risk assessment parameters are dynamically adjusted, and the monitoring conditions are updated according to the real-time monitoring data to form a temporary safety assessment strategy. The temporary strategy is applied to the global optimization process of the personalized safety risk assessment model. By balancing the weights of various factors and adjusting the parameter relationship, a high-precision and highly adaptable target safety monitoring model is finally obtained, providing comprehensive protection for the safety of the elderly at home.

[0105] In a specific embodiment, the updating of the personalized security risk assessment model according to the dynamic intervention strategy to obtain the target security monitoring model may specifically include the following steps: The dynamic intervention strategy and historical intervention records are associated and pattern mined to obtain an intervention effect sequence diagram, and the effectiveness and adaptability of the intervention measures are evaluated and tested based on the intervention effect sequence diagram and changes in the health status of the elderly to obtain an intervention effect measurement table; Perform risk factor association analysis based on the intervention effect measurement table and the behavioral characteristics of the elderly to obtain a risk factor association network, and perform potential health risk reasoning and probability estimation based on the risk factor association network to obtain a health risk probability tree; Based on the health risk probability tree, optimal allocation of monitoring resources and monitoring planning are performed to obtain a preliminary monitoring strategy plan, and feasibility assessment and priority sorting of monitoring measures are performed according to the preliminary monitoring strategy plan to obtain a hierarchical monitoring strategy set; Dynamically generate and check the consistency of monitoring rules based on the hierarchical monitoring strategy set to obtain a target monitoring rule base, and automatically arrange the daily monitoring process and build a decision tree according to the target monitoring rule base to obtain an intelligent monitoring solution; Based on the intelligent monitoring solution, risk assessment parameters are dynamically adjusted and monitoring conditions are updated to obtain a temporary safety assessment strategy. According to the temporary safety assessment strategy, the personalized safety risk assessment model is globally optimized and balanced to obtain a target safety monitoring model.

[0106] Specifically, the dynamic intervention strategy and historical intervention records are associated and pattern mined, and the time series data mining method is used to analyze the execution effect of the intervention strategy. First, the dynamic intervention strategy is represented as a sequence, where each intervention strategy contains information such as intervention type, execution time, and execution subject. At the same time, the historical intervention records are represented as a time series, where each record contains information such as intervention execution status, elderly people's reactions, and changes in health indicators. The time series similarity calculation method is used to analyze the matching degree between the intervention strategy and the execution record: ,in represents the temporal distance calculation function, Indicates intervention strategy, Represents the execution record, which can be calculated using the dynamic time warping algorithm. is a scale parameter that controls the similarity decay rate. A sequential pattern mining algorithm is used to identify frequently occurring intervention patterns from intervention records. An intervention pattern represents a sequence of intervention behaviors, and arrows represent temporal relationships. For each pattern, the frequency of the pattern appearing in the historical records is used as the support. Patterns with support greater than a preset threshold are selected as representative intervention patterns. Based on the mined intervention patterns, an intervention effect sequence diagram is constructed. In the diagram, nodes represent intervention behaviors and health states, edges represent state transition relationships, and edge weights represent transition probabilities, which are the number of times state A1 transitions to state B1 divided by the total number of times state A1 appears. The intervention effect sequence diagram clearly displays the sequence relationship between different intervention behaviors and the impact path on health states through a directed graph structure.

[0107] According to the intervention effect sequence diagram and the changes in the health status of the elderly, the effectiveness of the intervention measures and the adaptability test are evaluated using a multi-index evaluation method. Define the health status change vector ,in and Represent the health status vectors before and after the intervention. The health status vector contains multidimensional features such as physiological indicators, activity ability, and cognitive status. For intervention measures, the effectiveness score of the intervention measures is calculated: the weight vector is multiplied by the health status change vector. The adaptability test uses the cross-validation method to divide the elderly into different types of groups based on age, health status, and living habits. For each group and intervention measure, the adaptability score is calculated: the mean of the effectiveness score in the group is divided by the standard deviation. The higher the adaptability score, the better the adaptability of the intervention measure in the group. The results of effectiveness and adaptability evaluation are combined to construct an intervention effect measurement table. The measurement table is in matrix form, with rows representing intervention measures, columns representing evaluation indicators, and matrix elements representing the scores of the intervention on the indicators. Through normalization, the scores of each indicator are mapped to the [0,1] interval, and finally a standardized intervention effect measurement table is obtained, which provides a quantitative basis for subsequent analysis.

[0108] Based on the intervention effect measurement table and the behavioral characteristics of the elderly, risk factor association analysis was performed. Association analysis identifies the association between risk factors by calculating the mutual information value and conditional dependency probability between risk factors. and The strength of the correlation between , calculated as: ;in, represents the mutual information function, Represents the information entropy function. This formula calculates the standardized mutual information between two risk factors, reflecting the degree of association between them. The higher the mutual information value, the stronger the association between the two risk factors. According to the results of the association analysis, a risk factor association network is constructed. The association network is a weighted undirected graph. The nodes represent risk factors, the edges represent the association between risk factors, and the weights of the edges represent the strength of the association. Based on the risk factor association network, potential health risk reasoning is performed. Health risk reasoning uses the Bayesian network model to calculate the health risk probability under different risk factor combinations through conditional probability. Suppose the health risk event is , the risk factor set is , then the health risk probability is calculated as: ; This formula represents the probability of a health risk event occurring under the condition that the states of each risk factor are known. The conditional probability is obtained through historical data learning and expert knowledge fusion, reflecting the degree of influence of risk factors on health risks. By enumerating different state combinations of risk factors, calculating the corresponding health risk probabilities, and constructing a health risk probability tree. The health risk probability tree is a tree structure, where the root node represents the initial state, the leaf node represents the final health risk state, the intermediate node represents the risk factor state combination, the branch represents the state transition path, and the cumulative product of the probabilities on each path represents the probability of occurrence of the path.

[0109] Based on the health risk probability tree, the monitoring resources are optimally allocated. Monitoring resources include sensor equipment, caregivers, medical resources, etc. Resource allocation balances risk control effects and resource costs by solving multi-objective optimization problems. Suppose the monitoring resource allocation scheme is , and its objective function is: in, is the balance parameter, Representation scheme The risk control effect function is Representation scheme The goal is to minimize , that is, maximizing the risk control effect while controlling costs. Based on the results of resource optimization allocation, formulate a monitoring plan, clarify the deployment location, monitoring frequency and data collection strategy of various monitoring equipment, and form a preliminary monitoring strategy plan. Conduct a feasibility assessment on the preliminary monitoring strategy plan, considering the acceptance of the elderly, the difficulty of technical implementation and environmental adaptability. Suppose the weight formula is used as the evaluation function to calculate the feasibility assessment results. According to the feasibility assessment results, prioritize the monitoring measures to form a hierarchical monitoring strategy set. Priority sorting takes into account the risk control effect, urgency and resource requirements of the measures to ensure that key monitoring measures are implemented first.

[0110] Based on the hierarchical monitoring strategy set, monitoring rules are dynamically generated. Monitoring rules include trigger conditions, execution operations, and feedback mechanisms. Template filling and rule induction methods are used to generate rules to convert strategies into executable rule statements. The generated rules are checked for consistency, conflicts and overlaps between rules are identified, and optimization adjustments are made. Consistency detection verifies the behavioral consistency of the rule set in various situations through rule comparison and logical reasoning to ensure that no contradictory decisions are made. Based on the optimized rule set, a target monitoring rule base is formed. Based on the target monitoring rule base, daily monitoring processes are automatically orchestrated. Process orchestration builds a complete monitoring workflow by defining the execution order, conditional branches, and information flow of monitoring activities. The goal of process orchestration is to optimize the efficiency of activity execution and the timeliness of risk response. Through process modeling and optimization, a decision tree is constructed to clarify the decision path and execution strategy in various situations and form an intelligent monitoring plan. Based on the intelligent monitoring plan, risk assessment parameters are dynamically adjusted. Parameter adjustment updates the key parameters in the risk assessment model by analyzing monitoring data and risk event feedback. Suppose the parameter set of the risk assessment model is , the parameter update function is: ;in, is the learning rate, Representation parameters The loss gradient of the model. This formula implements the gradient descent update of the parameters, making the model better adaptable to new data and situations. At the same time, the monitoring conditions are updated, the threshold settings and trigger rules are adjusted, and a temporary safety assessment strategy is formed. According to the temporary safety assessment strategy, the personalized safety risk assessment model is globally optimized. Global optimization improves the accuracy and adaptability of the model by adjusting the model structure and retraining the parameters. The optimization process uses a cross-validation method to ensure the stability of the model in different elderly groups and different scenarios. The optimized model is balanced to balance sensitivity and specificity, ensuring that the model can identify risks in a timely manner without generating too many false alarms, and finally forming a target safety monitoring model.

[0111] This embodiment achieves precise monitoring by constructing a risk assessment model, avoiding the one-size-fits-all problem of traditional monitoring systems. Through the environmental perception network constructed by multiple types of sensors, intelligent identification of abnormal behaviors and risk classification warnings are realized, greatly improving the accuracy and timeliness of warnings. According to the analysis of behavioral trajectories, safe areas are automatically divided and intelligently configured to form virtual safety protection areas to achieve key monitoring of high-risk areas. Self-optimization is carried out according to the intervention effect and changes in health status, and the risk assessment model is continuously updated to improve the long-term monitoring effect. Dynamic intervention strategies are formulated in combination with external medical resource information to achieve efficient allocation of medical resources in emergency situations. A hierarchical response authority strategy is adopted to protect the privacy and dignity of the elderly to the greatest extent while ensuring safety.

[0112] Embodiment 2; See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, and a method for home safety monitoring of retired elderly people based on the Internet of Things is provided, including: S1. Deploy multiple types of sensors in the elderly’s home environment to collect and analyze data around the clock, and obtain behavioral data with activity feature labels; S2. Performing correlation analysis based on the behavioral data and the health record information of the elderly to obtain a personalized safety risk assessment model; S3. According to the personalized safety risk assessment model, real-time monitoring and pattern recognition are performed on the daily activities of the elderly to obtain a behavior trajectory analysis report; S4. Based on the behavior trajectory analysis report, the living area of ​​the elderly is divided into security levels and intelligently configured to obtain a virtual security guard area and adaptive reminder rules; S5. According to the activity feature label, the virtual security guard area and the adaptive reminder rule, the behavior data is graded and processed and a warning signal is generated to obtain graded warning information and a response authority strategy; S6. Based on the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information, an integrated analysis is performed to obtain a dynamic intervention strategy, and the personalized safety risk assessment model is updated according to the dynamic intervention strategy to obtain a target safety monitoring model.

[0113] The present embodiment also provides a home safety monitoring device for retired elderly people based on the Internet of Things. The home safety monitoring device for retired elderly people based on the Internet of Things includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the home safety monitoring method for retired elderly people based on the Internet of Things in the above-mentioned embodiments.

[0114] This embodiment also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for home safety monitoring of retired elderly people based on the Internet of Things.

[0115] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0116] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A home safety monitoring system for retired elderly people based on the Internet of Things, characterized in that: include: Collection and processing module: deploy multiple types of sensors in the elderly’s home environment to collect and analyze data around the clock, and obtain behavioral data with activity feature labels; Model building module: performing correlation analysis based on the behavioral data and the health record information of the elderly to obtain a personalized safety risk assessment model; Analysis module: Based on the personalized safety risk assessment model, the daily activities of the elderly are monitored in real time and pattern recognized to obtain a behavior trajectory analysis report; Division module: Based on the behavior trajectory analysis report, the living area of ​​the elderly is divided into security levels and intelligently configured to obtain virtual security guard areas and adaptive reminder rules; Classification module: classifies the behavior data and generates warning signals according to the activity feature labels, the virtual security guard area and the adaptive reminder rules, and obtains graded warning information and response authority strategies; Optimization module: Based on the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information, an integrated analysis is performed to obtain a dynamic intervention strategy, and the personalized safety risk assessment model is updated according to the dynamic intervention strategy to obtain a target safety monitoring model.

2. The home safety monitoring system for retired elderly people based on the Internet of Things according to claim 1 is characterized in that: The above method deploys multiple types of sensors in the elderly's home environment to collect and analyze data around the clock, and obtains behavioral data with activity feature labels, including: Conduct spatial planning and sensor demand analysis on the elderly’s home environment, obtain a sensor distribution plan, and deploy multiple types of sensor devices based on the sensor distribution plan to build an environmental perception network; Continuously collecting environmental parameters and activity signals through the environmental perception network to obtain an original perception data stream, and performing signal filtering and timing alignment processing on the original perception data stream to obtain cleaned structured data; Based on the structured data, spatiotemporal feature calculation and behavior sequence segmentation are performed to obtain a basic behavior unit set, and a pattern recognition algorithm and context association analysis are applied to the basic behavior unit set to obtain behavior semantic features; Activity classification and attribute labeling are performed according to the behavior semantic features to obtain preliminary behavior labels, and the preliminary behavior labels are verified and enhanced in combination with the historical behavior pattern library to obtain behavior data with activity feature labels.

3. The home safety monitoring system for retired elderly people based on the Internet of Things according to claim 1 is characterized in that: The personalized safety risk assessment model is obtained by performing correlation analysis based on the behavior data and the health record information of the elderly, including: Performing structured analysis and feature extraction on the health record information of the elderly to obtain a health status feature vector, and identifying health risk factors based on the health status feature vector to obtain an initial health risk rating; Based on the initial health risk rating and the behavior data, a correlation analysis is performed on the relationship between the health status and the behavior pattern to obtain a risk association rule set and a health behavior characteristic model, and according to the risk association rule set and the health behavior characteristic model, the health status and the behavior pattern are cross-validated and weighted to obtain a health behavior risk association map; Calculating the importance and prioritizing the risk factors based on the health behavior risk association map to obtain a risk factor weight matrix, and quantifying the levels of potential risk scenarios based on the risk factor weight matrix and the activity feature labels to obtain an initial risk assessment matrix; Based on the initial risk assessment matrix, a time series pattern analysis is performed on the risk prediction to obtain a set of risk evolution rules, and according to the set of risk evolution rules, a Markov chain modeling and risk transfer probability calculation are performed on the initial risk assessment matrix to obtain a dynamic risk probability distribution; Based on the dynamic risk probability distribution, risk prediction accuracy assessment and threshold optimization are performed to obtain a risk model adjustment parameter set, and based on the risk model adjustment parameter set, risk assessment multi-feature fusion and model training verification are performed to obtain a personalized safety risk assessment model.

4. The home safety monitoring system for retired elderly people based on the Internet of Things according to claim 1 is characterized in that: According to the personalized safety risk assessment model, real-time monitoring and pattern recognition are performed on the daily activities of the elderly to obtain a behavior trajectory analysis report, including: Real-time collection and signal processing of elderly activity data are performed to obtain an original behavior data stream, and normal behavior definition and abnormality detection are performed on the elderly activities based on the original behavior data stream and the personalized safety risk assessment model to obtain an activity status marking result; Based on the activity state marking results, time and space feature extraction and activity type recognition are performed on the behavior events to obtain a multi-dimensional behavior feature set, and based on the multi-dimensional behavior feature set, the activity patterns of the elderly are analyzed and conventional behavior modeling is performed to obtain a behavior pattern library; Based on the behavior pattern library and historical activity data, a behavior time series model is constructed and trend analysis is performed to obtain a behavior change trajectory diagram, and behavior deviation calculation and risk scoring are performed based on the behavior change trajectory diagram and preset health and safety indicators to obtain a behavior risk measurement scale; Based on the behavior risk measurement table and the personalized safety risk assessment model, risk level classification and urgency ranking are performed to obtain a risk event priority queue, and monitoring tasks are intelligently allocated and monitored in parallel through the risk event priority queue to obtain a distributed monitoring task network; Based on the distributed monitoring task network, behavior data statistics aggregation and pattern extraction are performed to obtain structured behavior records, and activity association analysis and safety trend prediction are performed based on the structured behavior records to obtain a behavior trajectory analysis report.

5. The home safety monitoring system for retired elderly people based on the Internet of Things according to claim 1 is characterized in that: Based on the behavior trajectory analysis report, the living area of ​​the elderly is divided into security levels and intelligently configured to obtain a virtual security guard area and adaptive reminder rules, including: Performing spatiotemporal distribution statistics and frequency analysis on the behavior trajectory analysis report to obtain a behavior pattern feature matrix, and performing living area correlation calculation and classification based on the behavior pattern feature matrix and environmental factor data to obtain an initial area function set; Based on the initial regional function set, heat map modeling and regional clustering analysis are performed on the correlation between the activity trajectories of the elderly to obtain a behavior density distribution map, and safety risk assessment and level division are performed based on the behavior density distribution map to obtain a preliminary safety area division result; Based on the preliminary safety area division results, high-risk area identification and abnormal behavior analysis are performed to obtain a risk area association matrix, and daily activity pattern modeling and scenario deduction are performed based on the risk area association matrix to obtain a safety monitoring process map; Based on the safety monitoring process map and the health status indicators of the elderly, regional safety warning thresholds are set and monitoring points are deployed to obtain a regional monitoring strategy set, and based on the regional monitoring strategy set, a virtual safety guarding environment is constructed and sensors are configured to obtain a virtual safety guarding area; Based on the virtual safety guard area and the caregiver authority setting, the safety warning behavior is extracted and parameterized to obtain adaptive reminder rules.

6. The home safety monitoring system for retired elderly people based on the Internet of Things according to claim 1 is characterized in that: The step of performing graded classification processing and early warning signal generation on the behavior data according to the activity feature label, the virtual security guard area and the adaptive reminder rule to obtain graded early warning information and response authority strategy includes: Performing scenario analysis and feature extraction on the behavior data to obtain a behavior feature vector set, and performing abnormal behavior severity assessment and priority sorting according to the behavior feature vector set and the activity feature labels to obtain a graded abnormal behavior task list; Automatically selecting and configuring the early warning algorithm and parameters based on the hierarchical abnormal behavior task list and the virtual security guard area to obtain a differentiated early warning scheme, and performing parallel analysis and processing of abnormal behaviors and accuracy verification according to the differentiated early warning scheme to obtain an initial early warning data set; Based on the initial warning data set and the adaptive reminder rules, the warning information is hierarchically organized and notified to control the marking, so as to obtain a warning information package with multi-level protection, and the response authority is finely divided and the strategy is generated according to the multi-level protection warning information package to obtain an initial response authority rule set; Based on the initial response authority rule set and monitoring scenario requirements, response process state machine modeling and conversion rule definition are performed to obtain a dynamic response state diagram, and response operation risk assessment and threshold setting are performed according to the dynamic response state diagram to obtain response risk control parameters; Based on the response risk control parameters, early warning signals are generated and distributedly pushed to obtain a distributed early warning instruction set, and protocol design and performance verification of the response process are performed according to the distributed early warning instruction set to obtain hierarchical early warning information and response authority strategy.

7. The home safety monitoring system for retired elderly people based on the Internet of Things according to claim 1 is characterized in that: The dynamic intervention strategy is obtained by integrating and analyzing the behavior trajectory analysis report, the response authority strategy, the processing record of the graded warning information and the external medical resource information, including: Performing time series correlation and pattern extraction on the behavior trajectory analysis report and the processing record of the graded warning information to obtain a health risk feature matrix, and performing abnormal state correlation calculation and classification according to the health risk feature matrix and the response authority strategy to obtain an initial risk scenario set; Based on the initial risk scenario set, trend chart modeling and type cluster analysis are performed on the health status change trend of the elderly to obtain a health status change chart, and medical needs assessment and urgency classification are performed based on the health status change chart to obtain a preliminary intervention needs classification result; Based on the preliminary intervention demand classification results, high-risk state identification and intervention effect analysis are performed to obtain an intervention demand association matrix, and emergency response mode modeling and plan deduction are performed according to the intervention demand association matrix to obtain an intervention process map; Based on the intervention process map and external medical resource information, medical resource allocation thresholds are set and rescue path planning is performed to obtain a resource allocation strategy set, and based on the resource allocation strategy set, a dynamic intervention environment is constructed and a rescue team is configured to obtain a dynamic intervention strategy.

8. The home safety monitoring system for retired elderly people based on the Internet of Things according to claim 1 is characterized in that: The step of updating the personalized safety risk assessment model according to the dynamic intervention strategy to obtain a target safety monitoring model includes: The dynamic intervention strategy and historical intervention records are associated and pattern mined to obtain an intervention effect sequence diagram, and the effectiveness and adaptability of the intervention measures are evaluated and tested based on the intervention effect sequence diagram and changes in the health status of the elderly to obtain an intervention effect measurement table; Perform risk factor association analysis based on the intervention effect measurement table and the behavioral characteristics of the elderly to obtain a risk factor association network, and perform potential health risk reasoning and probability estimation based on the risk factor association network to obtain a health risk probability tree; Based on the health risk probability tree, optimal allocation of monitoring resources and monitoring planning are performed to obtain a preliminary monitoring strategy plan, and feasibility assessment and priority sorting of monitoring measures are performed according to the preliminary monitoring strategy plan to obtain a hierarchical monitoring strategy set; Dynamically generate and check the consistency of monitoring rules based on the hierarchical monitoring strategy set to obtain a target monitoring rule base, and automatically arrange the daily monitoring process and build a decision tree according to the target monitoring rule base to obtain an intelligent monitoring solution; Based on the intelligent monitoring solution, risk assessment parameters are dynamically adjusted and monitoring conditions are updated to obtain a temporary safety assessment strategy. According to the temporary safety assessment strategy, the personalized safety risk assessment model is globally optimized and balanced to obtain a target safety monitoring model.

9. Home safety monitoring equipment for retired elderly people based on the Internet of Things, characterized by: The home safety monitoring device for retired elderly people based on the Internet of Things comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the IoT-based home safety monitoring device for retired elderly people implements the IoT-based home safety monitoring system for retired elderly people described in any one of claims 1-8.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, a method for home safety monitoring of retired elderly people based on the Internet of Things is implemented.

Images