A remote monitoring system and method for elderly care facilities

Abstract

The present invention relates to the field of health monitoring technology, specifically to a remote supervision system and method for elderly care facilities, the system comprising a behavior pattern analysis module, a health monitoring optimization module, a risk assessment dynamic module, a causal intervention assessment module, a personalized threshold setting module, a network behavior monitoring module, and a genetic strategy optimization module. In the present invention, by adopting advanced technologies such as graph neural network algorithm, support vector machine and random forest algorithm, dynamic Bayesian network algorithm, counterfactual theory and Douglas causal model, machine learning algorithm, network theory and genetic algorithm, beneficial effects are brought, and behavior patterns and their changes are identified through deep learning, health risks are effectively predicted and evaluated, and time series data processing and dynamic evaluation technology are used to achieve continuous monitoring and timely intervention of the health status of the elderly, improve the quality of life and safety of the elderly, and the application of genetic algorithm optimizes health management strategies to make resource allocation more efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of health monitoring, and in particular to a system and method for remotely monitoring elderly care facilities. Background Art

[0002] The field of health monitoring technology is a field that involves the use of various technologies to monitor and maintain the health and well-being of the elderly. This field focuses specifically on how to effectively use technology to improve the quality and efficiency of services in elderly care facilities while ensuring the safety and comfort of the elderly. Health monitoring technology includes remote medical monitoring, activity and behavior tracking, environmental monitoring, etc., aiming to assist medical professionals and elderly care facility managers to better care for the elderly by providing real-time data and insights.

[0003] Among them, the main purpose of the remote monitoring system for elderly care facilities is to improve the quality and efficiency of elderly care services through technical means. This system can monitor the health status, activity patterns and environment of the elderly in real time, so as to promptly detect potential health problems or safety hazards. The goal of the system is to prevent accidents and health problems through early intervention, while reducing the burden on the staff of the elderly care facilities and improving the efficiency of care. In addition, these systems can also provide valuable data on the daily life patterns of the elderly, helping medical and care staff to better understand and meet their needs.

[0004] In terms of behavior pattern and health status monitoring, traditional systems often rely on simple statistical methods and regular monitoring, lacking in-depth data analysis and prediction capabilities. This leads to incomplete and inaccurate analysis of the behavior and health status of the elderly, making it difficult to detect potential health risks in a timely manner, and increasing the difficulty of monitoring the health of the elderly. In addition, traditional systems also lack flexibility and personalization in resource allocation and health intervention strategies, resulting in the inability to use resources most effectively, affecting the overall health management of the elderly. These deficiencies lead to untimely monitoring of the health status of the elderly and intervention measures in actual operations, thus affecting the quality of life and safety of the elderly. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a remote monitoring system and method for elderly care facilities.

[0006] In order to achieve the above-mentioned object, the present invention adopts the following technical scheme: a remote supervision system for elderly care facilities includes a behavior pattern analysis module, a health monitoring optimization module, a risk assessment dynamic module, a causal intervention assessment module, a personalized threshold setting module, a network behavior monitoring module, and a genetic strategy optimization module;

[0007] The behavior pattern analysis module uses a graph neural network algorithm to build a dynamic behavior graph model based on the behavior data and social interaction data of the elderly, and identifies behavior patterns and changes through deep learning to generate a comprehensive overview of behavior patterns;

[0008] The health monitoring optimization module uses support vector machine and random forest algorithms based on a comprehensive overview of behavioral patterns to analyze long-term health records, monitor real-time health status, identify health trends and risks, and generate health pattern insights;

[0009] The dynamic risk assessment module is based on health pattern insights and uses a dynamic Bayesian network algorithm to analyze continuous health monitoring data, extract key risk indicators using time series data processing methods, and conduct real-time risk assessment and trend prediction through model construction and dynamic assessment technology to generate dynamic risk prediction indicators;

[0010] The causal intervention assessment module is based on dynamic risk prediction indicators, uses counterfactual theory and Douglas causal model for causal inference, analyzes the impact of multiple health intervention measures, evaluates intervention effects, and generates intervention impact analysis;

[0011] The personalized threshold setting module uses machine learning algorithms to analyze historical and real-time health data based on intervention impact analysis, performs personalized threshold adjustment and anomaly detection, and generates customized health threshold indicators;

[0012] The network behavior monitoring module is based on customized health threshold indicators and adopts network theory to conduct structured analysis of daily activities and behavior patterns, map social and behavioral relationships through network construction technology, identify social clusters of behavior patterns using community detection algorithms, and analyze behavior dynamics through behavioral analysis methods to generate network behavior insight maps;

[0013] The genetic strategy optimization module is based on the network behavior insight map and uses genetic algorithms to encode and optimize health management strategies. At the same time, it integrates health monitoring parameters with resource allocation plans, and through cyclic selection and refinement of evolutionary mechanisms, it mines the optimal management strategy and generates a strategy optimization plan.

[0014] As a further solution of the present invention, the comprehensive overview of behavioral patterns includes behavioral frequency distribution, key social circle analysis, and key behavioral event identification; the health pattern insights include physiological indicator trends, health abnormality indicators, and individual health risk levels; the dynamic risk prediction indicators include health status probability scores, short-term risk predictions, and long-term health risk predictions; the intervention measure impact analysis includes intervention strategy effect scores, health improvement contribution, and intervention measure optimization plans; the customized health threshold indicators include individualized physiological indicator thresholds, health risk warning indicators, and instant health assessment standards; the network behavior insight map includes social interaction pattern maps, key behavioral network nodes, and social activity impact assessments; the strategy optimization plan includes an optimized monitoring parameter set, a resource allocation plan, and an intervention strategy adjustment plan.

[0015] As a further solution of the present invention, the behavior pattern analysis module includes a graph structure construction submodule, a feature extraction submodule, and a behavior recognition submodule;

[0016] The graph structure construction submodule is based on the behavioral data and social interaction data of the elderly, adopts graph theory algorithms, uses the NetworkX library to build a dynamic behavioral graph model, adds nodes and edges through the add_node and add_edge methods, sets node weights to reflect the frequency of behavior, uses degree_centrality and betweenness_centrality functions to calculate node centrality, and generates a behavioral social network graph;

[0017] The feature extraction submodule is based on the behavioral social network graph, adopts feature engineering methods, uses the Pandas library to preprocess the data, including cleaning and normalization, uses the PCA function of the Scikit-learn library to perform principal component analysis, extracts node degrees and clustering coefficients, and generates a behavioral feature data set;

[0018] The behavior recognition submodule is based on the behavior feature dataset, adopts the graph neural network algorithm, uses the PyTorchGeometric library to build a GNN model, defines the model layer including GraphConv and ReLU activation functions, sets training parameters, and applies the model to learn and identify behavior patterns to generate a comprehensive overview of the behavior patterns.

[0019] As a further solution of the present invention, the health monitoring optimization module includes a health pattern recognition submodule, a real-time monitoring submodule, and a health pattern analysis submodule;

[0020] The health pattern recognition submodule is based on a comprehensive overview of behavioral patterns, adopts a support vector machine algorithm, uses the SVC function in the Scikit-learn library, sets the kernel function to a radial basis function, adjusts the penalty parameter C to 1.0, and the kernel parameter gamma to auto, classifies and recognizes patterns in long-term health records, analyzes health trends and potential risks, and generates health pattern classification data;

[0021] The real-time monitoring submodule uses the random forest algorithm based on the health pattern classification data, through the RandomForestClassifier function in the Scikit-learn library, sets the number of trees to 100, analyzes the real-time health monitoring data, identifies the current health status and changes, and generates real-time health monitoring results;

[0022] The health pattern analysis submodule builds a prediction model based on real-time health monitoring results using the Python programming language and the Scikit-learn library, uses RandomForestClassifier as the classifier to set parameters, where n_estimators is set to 100, max_depth is set to 10, and the train_test_split method is used to divide the training set and the test set. The fit method is applied to train the model, and the predict method is used for prediction to generate health pattern insights.

[0023] As a further solution of the present invention, the risk assessment dynamic module includes a DBN model building submodule, a time series data processing submodule, and a risk dynamic assessment submodule;

[0024] The DBN model building submodule is based on health pattern insights and uses Python's pgmpy library to execute the dynamic Bayesian network algorithm. The model is built by defining health status nodes and transition probabilities. The nodes represent multiple health indicators. The transition probabilities are calculated based on historical data to represent the changes in health status over time. The network is built using the add_node and add_edge methods, and the conditional probability table is set to reflect the probability relationship between nodes to generate the DBN model configuration.

[0025] The time series data processing submodule is configured based on the DBN model, uses Python's Pandas library to perform time series analysis, processes continuous health monitoring data, sets data time index, applies the rolling method to calculate the moving average, executes the diff method to perform data differentiation, and generates time series feature data;

[0026] The risk dynamic assessment submodule performs risk assessment based on time series feature data using the DBN model, calculates the probability distribution of health indicators through the VariableElimination of the pgmpy library, analyzes the dynamic changes of health risks, predicts short-term and long-term health risks using probability query and inference methods, and generates dynamic risk prediction indicators.

[0027] As a further solution of the present invention, the causal intervention assessment module includes a causal model construction submodule, a counterfactual analysis submodule, and an impact assessment submodule;

[0028] The causal model construction submodule is based on dynamic risk prediction indicators, adopts counterfactual theory, constructs a causal model through Python's CausalInference library, sets causal variables and outcome variables, uses the statistical_methods method to perform covariate balance and matching, and constructs a causal relationship model;

[0029] The counterfactual analysis submodule is based on the causal relationship model, adopts the Douglas causal model, uses Python's DoWhy library to perform counterfactual analysis, creates counterfactual scenarios, including the situation where the target health intervention is not performed, compares the difference between the actual intervention results and the counterfactual results, uses the identify and estimate methods to analyze the intervention effect, and generates counterfactual analysis results;

[0030] The impact assessment submodule comprehensively evaluates the impact of various health interventions based on the results of counterfactual analysis, uses Python's statistical analysis library to perform regression analysis and difference comparison, analyzes the effect of each intervention, and generates an intervention impact analysis.

[0031] As a further solution of the present invention, the personalized threshold setting module includes a health data analysis submodule, a dynamic threshold adjustment submodule, and an anomaly detection submodule;

[0032] The health data analysis submodule analyzes historical health data based on intervention impact analysis, using support vector machines and random forest algorithms, using SVC and RandomForestClassifier functions in Python's Scikit-learn library, setting the kernel function of SVC to radial basis function, the penalty parameter C to 1.0, the number of trees of RandomForestClassifier to 100, and the maximum depth to 10, to identify and analyze health patterns in historical data and generate health trend analysis results;

[0033] The dynamic threshold adjustment submodule uses a decision tree algorithm to dynamically adjust the threshold of real-time health data based on the health trend analysis results, uses Python's DecisionTreeClassifier function, sets the maximum depth of the tree to 5, and the splitting criterion to information gain, analyzes real-time health data, dynamically adjusts the health threshold to match the individual's health status, and generates a dynamically adjusted health threshold;

[0034] The anomaly detection submodule applies an anomaly detection algorithm to monitor the health status based on the dynamically adjusted health threshold. It uses Python's IsolationForest function, sets the number of trees to 100 and the anomaly ratio to 0.01, monitors real-time health data, identifies data points that deviate from the normal range, and generates customized health threshold indicators.

[0035] As a further solution of the present invention, the network behavior monitoring module includes a network construction submodule, a community detection submodule, and a behavior analysis submodule;

[0036] The network construction submodule is based on customized health threshold indicators and adopts network theory. Through Python's NetworkX library, nodes representing the behaviors of the elderly and edges representing the interactions between behaviors are created. The add_node and add_edge methods are used to build the network. The weight of the edge is determined by the frequency and intensity of the elderly's behavior. The network structure representing daily activities and social interactions is constructed to generate network structure data.

[0037] The community detection submodule uses a community detection algorithm based on network structure data, creates a Louvain method through Python's Community library, identifies social clusters and associated behavioral pattern groups in the network by optimizing the modularity in the social structure, and analyzes the social structure characteristics to generate community detection results;

[0038] The behavior analysis submodule uses the behavior analysis method based on the community detection results to perform statistical analysis on the behavior data through Python's Pandas library, calculates the behavior frequency, duration and social interaction pattern, analyzes the changes and dynamics of the behavior pattern, and generates a network behavior insight map.

[0039] As a further solution of the present invention, the genetic strategy optimization module includes a parameter encoding submodule, a fitness function design submodule, and a strategy selection submodule;

[0040] The parameter encoding submodule is based on the network behavior insight map, uses a genetic algorithm, and is executed through Python's DEAP library to encode health monitoring parameters and resource allocation schemes into genetic algorithm gene sequences. The create function is used to set the encoding rules and length of the gene sequence, including the value range of the monitoring parameters and the quantitative standard of resource allocation, to generate the initial genetic code;

[0041] The fitness function design submodule designs a fitness function to evaluate the effect of the health management strategy based on the initial genetic coding, uses the Fitness function in the DEAP library of Python, sets the coverage, cost-effectiveness and user satisfaction of the strategy, calculates the fitness score of each strategy, evaluates the utility and feasibility of the strategy according to the score, and generates a fitness evaluation result;

[0042] The strategy selection submodule selects and optimizes the health management strategy based on the fitness evaluation results by applying the evolutionary mechanism of the genetic algorithm. It executes the select, mate and mutate functions through the DEAP library of Python to select the population, cross and mutate the genes, and iterates to capture the optimal health management strategy and generate a strategy optimization solution.

[0043] A remote supervision method for elderly care facilities is provided, and the remote supervision method for elderly care facilities is performed based on the remote supervision system for elderly care facilities, and comprises the following steps:

[0044] S1: Based on the behavioral data and social interaction data of the elderly, a graph neural network algorithm is used to build a dynamic behavioral graph model and analyze behavioral patterns in combination with deep learning technology. The Python NetworkX library is used to create nodes and edges, and the centrality of the nodes is calculated to generate behavioral social network analysis.

[0045] S2: Based on the behavioral social network analysis, support vector machines and random forest algorithms are used to classify and identify long-term health records, and real-time health monitoring data is used to analyze health status. Python's Scikit-learn library is used for data processing and model training to generate health trend and risk analysis;

[0046] S3: Based on the health trend and risk analysis, the dynamic Bayesian network algorithm is used to analyze the continuous health monitoring data, and the dynamic Bayesian network is constructed using the Python pgmpy library to process time series data and extract key risk indicators to generate dynamic risk prediction indicators;

[0047] S4: Based on the dynamic risk prediction indicators, counterfactual theory and Douglas causal model are used to perform causal inference and evaluation on the impact of differentiated health intervention measures. Causal relationship modeling and counterfactual analysis are performed using Python's CausalInference library and DoWhy library to compare the actual intervention results with those without intervention, thereby generating an impact assessment of the intervention measures.

[0048] S5: Based on the impact assessment of the intervention measures, a machine learning algorithm is used to conduct a comprehensive analysis of historical and real-time health data, and a decision tree algorithm is applied through the Python Scikit-learn library to perform personalized threshold adjustment and anomaly detection to determine the health status of individuals and generate customized health thresholds;

[0049] S6: Based on the customized health threshold, a genetic algorithm is used to optimize the health management strategy. The DEAP library of Python is used to perform parameter encoding, fitness function design and strategy selection. The health management strategy is optimized and refined through an evolutionary mechanism to generate a strategy optimization solution.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are:

[0051] In the present invention, by adopting advanced technologies such as graph neural network algorithm, support vector machine and random forest algorithm, dynamic Bayesian network algorithm, counterfactual theory and Douglas causal model, machine learning algorithm, network theory and genetic algorithm, significant beneficial effects are brought. By deep learning to identify behavioral patterns and their changes, the system can more effectively predict and evaluate health risks. By using time series data processing and dynamic evaluation technology, the system can realize continuous monitoring and timely intervention of the health status of the elderly, improve the quality of life and safety of the elderly. The application of genetic algorithm optimizes health management strategies, makes resource allocation more efficient, and enhances the overall performance of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a system flow chart of the present invention;

[0053] Figure 2 It is a schematic diagram of the system framework of the present invention;

[0054] Figure 3 It is a flow chart of the behavior pattern analysis module of the present invention;

[0055] Figure 4 This is a flow chart of the health monitoring optimization module of the present invention;

[0056] Figure 5 It is a flow chart of the risk assessment dynamic module of the present invention;

[0057] Figure 6 is a flow chart of the causal intervention assessment module of the present invention;

[0058] Figure 7 This is a flow chart of a personalized threshold setting module of the present invention;

[0059] Figure 8 This is a flow chart of the network behavior monitoring module of the present invention;

[0060] Fig. 9 It is a flow chart of the genetic strategy optimization module of the present invention;

[0061] Fig.10 It is a schematic diagram of the method steps of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0064] Embodiment 1

[0065] See also Figure 1 to Figure 2 ,The present invention provides a technical solution: a remote supervision system for elderly care facilities includes a behavior pattern analysis module, a health monitoring optimization module, a risk assessment dynamic module, a causal intervention assessment module, a personalized threshold setting module, a network behavior monitoring module, and a genetic strategy optimization module;

[0066] The behavior pattern analysis module uses graph neural network algorithms to build a dynamic behavior graph model based on the behavior data and social interaction data of the elderly, and identifies behavior patterns and changes through deep learning to generate a comprehensive overview of behavior patterns;

[0067] The health monitoring optimization module uses support vector machine and random forest algorithms to analyze long-term health records, monitor real-time health status, identify health trends and risks, and generate health pattern insights based on a comprehensive overview of behavioral patterns;

[0068] The dynamic risk assessment module is based on health pattern insights and uses a dynamic Bayesian network algorithm to analyze continuous health monitoring data, extract key risk indicators using time series data processing methods, and conduct real-time risk assessment and trend prediction through model construction and dynamic assessment technology to generate dynamic risk prediction indicators;

[0069] The causal intervention assessment module uses counterfactual theory and Douglas causal model to make causal inferences based on dynamic risk prediction indicators, analyze the impact of various health interventions, evaluate the intervention effects, and generate intervention impact analysis;

[0070] The personalized threshold setting module uses machine learning algorithms to analyze historical and real-time health data based on intervention impact analysis, conduct personalized threshold adjustment and anomaly detection, and generate customized health threshold indicators;

[0071] The network behavior monitoring module is based on customized health threshold indicators and adopts network theory to conduct structured analysis of daily activities and behavior patterns. It maps social and behavioral relationships through network construction technology, identifies social clusters of behavioral patterns using community detection algorithms, and analyzes behavioral dynamics through behavioral analysis methods to generate network behavior insight maps.

[0072] The genetic strategy optimization module is based on the network behavior insight map and uses genetic algorithms to encode and optimize health management strategies. At the same time, it integrates health monitoring parameters with resource allocation plans, and through cyclic selection and refinement through evolutionary mechanisms, it discovers the optimal management strategy and then generates a strategy optimization plan.

[0073] The comprehensive overview of behavioral patterns includes behavioral frequency distribution, key social circle analysis, and key behavioral event identification. Health pattern insights include physiological indicator trends, health abnormality indicators, and individual health risk levels. Dynamic risk prediction indicators include health status probability scores, short-term risk predictions, and long-term health risk predictions. Intervention impact analysis includes intervention strategy effectiveness scores, health improvement contribution, and intervention optimization plans. Customized health threshold indicators include individualized physiological indicator thresholds, health risk warning indicators, and instant health assessment standards. Network behavior insight maps include social interaction pattern maps, key behavioral network nodes, and social activity impact assessments. Strategy optimization plans include optimized monitoring parameter sets, resource allocation plans, and intervention strategy adjustment plans.

[0074] In the behavior pattern analysis module, the daily behavior data and social interaction data of the elderly are collected as input. The data format is time series data, which contains information such as timestamp, activity type, activity duration, and social contacts participating in the activity. Through the graph neural network algorithm, specifically, advanced technologies such as graph convolutional network (GCN) or graph attention network (GAT) are used to convert these data into a dynamic behavior graph model. In this process, nodes represent individual behavioral events or social interactions, and edges represent the relationship strength between events or individuals. The algorithm iteratively learns node features and topological structures and optimizes weight parameters to capture complex behavior patterns and social dynamics. Deep learning frameworks such as TensorFlow or PyTorch are used to implement model training and optimization, and the network weights are adjusted using backpropagation and gradient descent algorithms. The final output of this module is a series of identified behavior patterns and their changing trends. These patterns are encoded into a comprehensive overview file, which provides basic data for subsequent health monitoring, effectively reveals the behavioral habits and social patterns of the elderly, and provides a basis for personalized health management.

[0075] The health monitoring optimization module is based on a comprehensive overview of behavioral patterns, combined with the long-term health records of the elderly, which exist in the form of structured data tables and contain information such as physiological indicators, medical examination results, and daily health status assessments. The module uses support vector machine (SVM) and random forest (RF) algorithms for data analysis. SVM distinguishes health states by building a decision plane that maximizes the boundary in the feature space, while random forest improves the accuracy and stability of predictions by building multiple decision trees and performing ensemble learning. The detailed operation of the algorithm includes steps such as feature selection, model training, cross-validation, and hyperparameter tuning. Cross-validation is used to evaluate model performance and ensure that the model has good generalization ability. Through these methods, the module can monitor real-time health status, identify health trends and risks, and generate health pattern insight reports. The report not only summarizes the health status and changing trends of the elderly, but also provides a scientific basis for early risk warning and health intervention.

[0076] The dynamic risk assessment module uses the dynamic Bayesian network (DBN) algorithm to analyze continuous health monitoring data, which is recorded in the form of time series and covers various physiological indicators and health events of the elderly. DBN can handle the uncertainty and variability of time series data, and describe the evolution of health status by building state transition models and observation models. The module uses advanced time series data processing methods, such as sliding window technology, to extract key time points and events, so as to accurately identify key risk indicators. Dynamic assessment techniques, including filtering, smoothing and prediction, are used to update the estimate of health status and future risk trends in real time. Through this series of refined operations, the dynamic risk prediction indicators generated by the module provide personalized health risk assessment for the elderly, making medical intervention more timely and targeted, and greatly improving the efficiency and effectiveness of health management.

[0077] The causal intervention assessment module integrates dynamic risk prediction indicators and uses counterfactual theory and Douglas causal model for in-depth analysis. This process involves complex mathematical modeling and statistical analysis techniques. The module first defines potential outcome variables and intervention variables, and then uses causal inference algorithms to estimate potential outcomes under different health interventions. In this process, the module addresses the problem of bias in observational data by adjusting confounding variables and using propensity score matching. Through these detailed mathematical operations, the module can accurately evaluate the effects of different health interventions and generate an intervention impact analysis report. This report not only provides scientific intervention recommendations for medical professionals, but also provides customized strategies for the health management of the elderly, thereby maximizing the effect of health interventions.

[0078] The personalized threshold setting module uses machine learning algorithms to analyze historical and real-time health data based on the unique health status and risk level of each elderly person. These data are in various formats, including continuous physiological indicator measurements, health event records, etc. The algorithms in the module, such as decision trees, neural networks, or ensemble learning methods, are trained to learn the mapping relationship from health data to health status. By analyzing the correlation between these data and health events, the module can identify the factors that have the greatest impact on the health status of the elderly, and adjust the threshold accordingly to achieve anomaly detection. Ultimately, the customized health threshold indicators generated by the module can provide more accurate health monitoring and early warning for each elderly person, ensuring the timeliness and effectiveness of health management measures.

[0079] The network behavior monitoring module uses network theory to conduct a structured analysis of the social activities and behavior patterns of the elderly by analyzing customized health threshold indicators and daily behavior patterns. The network constructed by the module not only contains nodes (representing individuals or behavioral events) and edges (representing social or behavioral relationships), but also uses community detection algorithms to identify social clusters in behavioral patterns. This analysis reveals the core structure and key members of the elderly's social activities, while at the same time, through behavioral dynamics analysis, deeply understands the changing trends and social influences of individual behaviors. The generated network behavior insight map not only provides an intuitive view of social and behavioral relationships for the elderly, but also provides important information for community service providers and family caregivers to optimize social activities and health interventions to promote social participation and healthy living for the elderly.

[0080] The genetic strategy optimization module uses genetic algorithms to encode and optimize health management strategies based on the network behavior insight map. By simulating natural selection and genetic mechanisms such as selection, crossover and mutation, the module continuously iterates in multiple generations of strategies to screen out the management strategy with the highest fitness. In this process, the genetic algorithm regards health monitoring parameters and resource allocation schemes as chromosomes, and each chromosome represents a potential management strategy. The effect of each strategy is evaluated by defining a fitness function, which takes into account multiple dimensions such as health improvement, resource utilization efficiency and user satisfaction. In each generation of strategies, selective pressure forces excellent strategies to be retained, while new strategy variants are introduced through crossover and mutation to increase diversity and prevent the algorithm from converging to the local optimal solution too early. As the iteration proceeds, the module is able to mine more refined and personalized health management strategies, which not only take into account the health status and personal preferences of the elderly, but also optimize the allocation and use of resources, ensuring the sustainability and effectiveness of health management measures. Finally, the strategy optimization scheme generated by the genetic strategy optimization module is presented in the form of a report, which provides a scientific and systematic solution for the health management of the elderly, achieves the optimization of health management, and helps to improve the quality of life and health level of the elderly.

[0081] See also Figure 3 ,The behavior pattern analysis module includes a graph structure building submodule, a feature extraction submodule, and a behavior recognition submodule;

[0082] The graph structure construction submodule is based on the behavioral data and social interaction data of the elderly, adopts graph theory algorithms, uses the NetworkX library to build a dynamic behavioral graph model, adds nodes and edges through the add_node and add_edge methods, sets node weights to reflect the frequency of behavior, and uses degree_centrality and betweenness_centrality functions to calculate node centrality to generate a behavioral social network graph;

[0083] The feature extraction submodule is based on the behavioral social network graph and adopts feature engineering methods. It uses the Pandas library to preprocess the data, including cleaning and normalization, and uses the PCA function of the Scikit-learn library to perform principal component analysis, extract node degrees and clustering coefficients, and generate a behavioral feature data set.

[0084] The behavior recognition submodule is based on the behavior feature dataset and adopts the graph neural network algorithm. It uses the PyTorchGeometric library to build a GNN model, define the model layer including GraphConv and ReLU activation functions, set training parameters, and apply the model to learn and identify behavior patterns to generate a comprehensive overview of the behavior patterns.

[0085] In the graph structure construction submodule, the behavioral data and social interaction data of the elderly are collected as input. These data include individual activity records, interaction frequency, timestamps, etc., and the format is time series or event logs. The NetworkX library is used, which is a powerful Python library dedicated to creating, manipulating complex graph structures and analyzing networks. Each node added by the add_node method represents a specific behavior or social interaction event. The addition of nodes is based on unique events or interactions in the data. The add_edge method is used to connect these nodes. The addition of edges is based on the relationship between events or interactions, such as sequential occurrence, individuals participating in the same activity, etc. When building a graph model, the setting of node weights reflects the frequency or importance of behaviors or interactions, which can be determined by analyzing repeated patterns or frequent social interactions in behavioral data. The centrality of nodes can be calculated through the degree_centrality and betweenness_centrality functions in NetworkX. These indicators reveal the most active or important nodes in the graph, that is, which behaviors or individuals occupy a core position in the social network. After completing these steps, the generated behavioral social network graph provides an intuitive and structured view for subsequent analysis, revealing the patterns of elderly behavior and social interactions and their connections with each other. This graphical representation not only helps to understand the social and behavioral patterns of individuals, but also provides a basis for further feature extraction and behavior recognition analysis.

[0086] In the feature extraction submodule, based on the constructed behavioral social network graph, feature engineering methods are used to preprocess and analyze the data in the network graph to ensure the quality and applicability of the data. The Pandas library, a powerful data analysis and processing library, is used to clean the data in the network graph, including removing outliers, filling missing data, and normalizing the data format to ensure the consistency and accuracy of the data. Next, the Scikit-learn library, a widely used machine learning library, provides a PCA (Principal Component Analysis) function, which is a dimensionality reduction technique used to extract key features from complex data and reduce the dimensions of the data set while retaining the most important information. In this submodule, PCA is used to extract key features of nodes in the graph, such as node degree and clustering coefficient, which reflect the connectivity of nodes and the closeness of social clusters. Through this series of preprocessing and feature extraction steps, the generated behavioral feature dataset contains the key features of the elderly's behavior and social interaction, providing accurate and meaningful input data for further behavior identification and analysis. These feature datasets can reveal the essential characteristics of the elderly's behavior patterns and provide a solid foundation for identifying and predicting behavior patterns.

[0087] In the behavior recognition submodule, based on the behavior feature dataset, the graph neural network (GNN) algorithm is used for in-depth analysis and recognition. The PyTorch Geometric library, a PyTorch extension library specifically for graph neural network research, provides various tools and methods required to build GNN models. When defining the model, the GraphConv layer is included, which is a graph convolution operation that can capture the dependencies between nodes in the graph; the ReLU activation function is used to increase the nonlinear ability of the model, enabling the model to capture more complex patterns. During the training process, appropriate training parameters are set, such as learning rate, batch size, number of iterations, etc. These parameters have an important impact on the training effect and convergence speed of the model. By applying the trained GNN model to learn and recognize the behavior pattern, the model can identify specific behavior patterns based on the input behavior feature dataset and predict future behavior trends. The generated comprehensive overview of the behavior pattern provides an in-depth analysis and understanding of the behavior and social interaction of the elderly, reveals the potential behavior patterns and their changes, and provides valuable insights into the health management and social participation of the elderly. This comprehensive overview not only helps healthcare providers and family caregivers better understand and support the elderly, but also provides researchers with an important resource for studying the behavior of the elderly.

[0088] For example, monitoring the daily activities and social interactions of the elderly to predict health risks and improve the quality of life. Data items include the frequency of walking, eating, social activities, and the number of interactions with family, friends, and nursing facility staff. The simulated values ​​are 10 walking times per day, three meals per day, 5 social activities per week, 2 interactions with family, 1 interaction with friends, and 3 interactions with staff. The process of applying deep learning algorithms and convolutional neural networks involves taking these activity and interaction data as input, extracting features through multiple Conv2D and MaxPooling2D layers, classifying with Dense layers, and training the model using the Adam optimizer and categorical_crossentropy loss function. The fit method is used to train the model, and the evaluate method is used to verify the accuracy of the model. The resulting deep analysis model can predict the health status and social needs of the elderly based on their behavior and social interaction patterns, and provide personalized health management and social activity recommendations, thereby improving the quality of life and health level of the elderly in nursing facilities.

[0089] See also Figure 4 ,The health monitoring optimization module includes a health pattern recognition submodule, a real-time monitoring submodule, and a health pattern analysis submodule;

[0090] The health pattern recognition submodule is based on a comprehensive overview of behavioral patterns and adopts the support vector machine algorithm. Through the SVC function in the Scikit-learn library, the kernel function is set to the radial basis function, the penalty parameter C is adjusted to 1.0, and the kernel parameter gamma is set to auto. It classifies and recognizes patterns in long-term health records, analyzes health trends and potential risks, and generates health pattern classification data;

[0091] The real-time monitoring submodule uses the random forest algorithm based on the health pattern classification data, through the RandomForestClassifier function in the Scikit-learn library, setting the number of trees to 100, to analyze the real-time health monitoring data, identify the current health status and changes, and generate real-time health monitoring results;

[0092] The health pattern analysis submodule builds a prediction model based on real-time health monitoring results using the Python programming language and the Scikit-learn library. It uses RandomForestClassifier as the classifier to set parameters, with n_estimators set to 100 and max_depth set to 10. It also uses the train_test_split method to divide the training set and the test set. It uses the fit method to train the model and uses the predict method to make predictions to generate health pattern insights.

[0093] In the health pattern recognition submodule, long-term health records of the elderly are first collected and sorted. These records include data such as physiological parameters (such as heart rate, blood pressure, etc.), activity level, eating habits, sleep quality, etc. The format is a structured data table, where each row represents a record at a time point and each column represents a health indicator. The support vector machine (SVM) algorithm, especially the SVC function in the Scikit-learn library, is used to classify and identify these health data. SVM is a supervised learning algorithm that is suitable for binary and multi-classification problems. Here, SVM is used to distinguish different modes of health status, such as normal, warning, high risk, etc. The kernel function selects the radial basis function (RBF) because it can handle classification problems in nonlinear feature space. The penalty parameter C is set to 1.0 to control the degree of penalty for misclassification, and the kernel parameter gamma is set to auto to let the algorithm automatically select the most appropriate value. In this way, the module is able to learn and identify potential health patterns from long-term health records and analyze health trends and risks. Finally, this submodule generates health pattern classification data, which reveals the patterns and trends of individual health status and provides a basis for subsequent health monitoring and intervention.

[0094] In the real-time monitoring submodule, real-time health monitoring data is received. These data come from wearable devices or health monitoring systems, including but not limited to real-time health indicators such as heart rate, number of steps, and sleep monitoring. The data format is a time series, and the values ​​of multiple health indicators are recorded at each time point. The random forest algorithm is implemented through the RandomForestClassifier function in the Scikit-learn library. Random forest is an ensemble learning method that improves the accuracy and robustness of the overall prediction by building multiple decision trees and aggregating their prediction results. In this submodule, the number of trees is set to 100 to ensure the generalization ability and prediction performance of the model. By analyzing real-time health data, this submodule can identify the current health status and key changes, and provide real-time feedback for immediate intervention. The generated real-time health monitoring results help to detect health problems in a timely manner and prevent potential health risks from developing into more serious conditions.

[0095] In the health pattern analysis submodule, real-time health monitoring results are comprehensively used to build a prediction model to analyze health patterns and provide future health trend predictions. Using Python and RandomForestClassifier in the Scikit-learn library, the n_estimators parameter is set to 100 to build a sufficient number of decision trees, and the max_depth parameter is set to 10 to limit the depth of the tree to avoid overfitting. The data is divided into training and test sets by the train_test_split method to ensure that the model can make effective predictions on unseen data. The fit method is used to train the model to learn patterns and associations in health data. The predict method is applied to new or real-time health data to predict health status. The health pattern insights generated by this submodule reveal the potential trends and risk factors of the health status of the elderly, providing a scientific basis for the formulation of preventive measures and intervention strategies, and helping to improve the quality of life and health of the elderly.

[0096] In the health monitoring and management system of the elderly care facility, a comprehensive set of submodules are implemented to improve the health level and quality of life of residents. First, the health pattern recognition submodule analyzes the long-term health records of residents, including heart rate, blood pressure, activity level and other data, through the support vector machine algorithm to identify health risk patterns, such as potential risk of cardiovascular disease or trend of decreased activity. Then, the real-time monitoring submodule uses the random forest algorithm to analyze the health data collected from wearable devices in real time, such as abnormal heart rate increase, to immediately identify changes in health status. Finally, the health pattern analysis submodule uses the random forest classifier to compare historical and real-time data, predict future health trends, and recommend corresponding intervention measures. For example, for residents who are detected to be at risk of cardiovascular disease, the system recommends increasing targeted exercise and dietary adjustments; for residents with reduced activity, it is recommended to participate in more social activities to increase activity. Through the collaborative work of these submodules, the system can not only provide personalized health monitoring and management, but also detect and prevent potential health problems in a timely manner, greatly improving the overall health status and quality of life of the elderly in the elderly care facility.

[0097] See also Figure 5 ,The risk assessment dynamic module includes the DBN model building submodule, the ,time series data processing submodule, and the risk dynamic assessment submodule;

[0098] The DBN model building submodule is based on health pattern insights and uses Python's pgmpy library to execute the dynamic Bayesian network algorithm. The model is built by defining health status nodes and transition probabilities. The nodes represent multiple health indicators. The transition probabilities are calculated based on historical data to represent the changes in health status over time. The network is built using the add_node and add_edge methods, and the conditional probability table is set to reflect the probability relationship between nodes to generate the DBN model configuration.

[0099] The time series data processing submodule is based on the DBN model configuration, uses Python's Pandas library to perform time series analysis, processes continuous health monitoring data, sets the data time index, applies the rolling method to calculate the moving average, executes the diff method to perform data differentiation, and generates time series feature data;

[0100] The risk dynamic assessment submodule uses the DBN model to perform risk assessment based on time series feature data, calculates the probability distribution of health indicators through the VariableElimination of the pgmpy library, analyzes the dynamic changes of health risks, uses probabilistic query and inference methods to predict short-term and long-term health risks, and generates dynamic risk prediction indicators.

[0101] In the DBN model building submodule, Python's pgmpy library is used to execute the dynamic Bayesian network algorithm. The process begins with defining health state nodes, each of which represents a health indicator such as heart rate, blood pressure, or activity level. These nodes are connected by transition probabilities calculated based on historical health data, representing the changes in health status over time. Through the add_node and add_edge methods, nodes and edges are added to the network to build a complete DBN model. The conditional probability table (CPT) of each node is set to reflect the probabilistic relationship between nodes, estimated based on historical data. Finally, this submodule generates a DBN model configuration for representing and predicting the evolution of health status over time, providing a basis for subsequent risk assessment.

[0102] In the time series data processing submodule, Python's Pandas library is used to process continuous health monitoring data. At this stage, the data is set with a time index to facilitate time series analysis. The rolling method is applied to calculate the moving average, which is used to smooth the time series data and reduce the impact of short-term fluctuations. The diff method is used to calculate the difference between consecutive measurements to identify changes in health indicators over time. Through these operations, time series feature data are generated, which highlight the main trends and patterns of health indicators over time, providing rich information for risk assessment.

[0103] In the risk dynamic assessment submodule, DBN model and time series feature data are used to dynamically assess health risks. The probability distribution of health indicators is calculated through the VariableElimination algorithm of the pgmpy library, and the dynamic changes of these indicators over time are analyzed. Probabilistic query and inference methods are used to predict short-term and long-term health risks based on current and past health data. Finally, this submodule generates dynamic risk prediction indicators, which provide elderly care facilities with a time-dynamic view of residents' health risks, so that health interventions can be more timely and targeted, thereby improving the effectiveness of preventive measures.

[0104] In the remote monitoring system of the elderly care facility, the DBN model building submodule, the time series data processing submodule and the risk dynamic assessment submodule work together to provide residents with a comprehensive health monitoring and risk assessment solution. By collecting health indicators such as heart rate, blood pressure, number of steps and sleep data of residents, the system uses Python's pgmpy library to build a dynamic Bayesian network (DBN) model, which contains health state nodes and transition probabilities calculated based on historical data to reflect the changes in health status over time. Subsequently, the system uses the Pandas library to process these continuous health monitoring data, sets the time index and applies the moving average and data difference methods to generate time series feature data, which reveals the main trends and pattern changes of health indicators. In the risk dynamic assessment stage, the system again uses the DBN model to analyze the time series feature data through the VariableElimination algorithm of the pgmpy library, calculate the probability distribution of health indicators and predict short-term and long-term health risks. For example, for a resident with a persistently high heart rate and blood pressure and a significantly reduced number of steps, the system predicts an increase in the risk of cardiovascular disease. Based on these dynamic risk assessment results, the system provides the medical team of the nursing facility with detailed health reports and personalized intervention recommendations, such as adjusting the residents' diet and exercise plans, or conducting more in-depth medical examinations. This comprehensive health monitoring and assessment method not only improves the nursing facility's understanding of the residents' health status, but also makes targeted intervention measures more timely and effective, thereby significantly improving the residents' health management and quality of life.

[0105] See also Figure 6 ,The causal intervention assessment module includes a causal model building submodule, a counterfactual analysis submodule, and an impact assessment submodule;

[0106] The causal model construction submodule is based on dynamic risk prediction indicators, adopts counterfactual theory, builds a causal model through Python's CausalInference library, sets causal variables and outcome variables, uses the statistical_methods method to perform covariate balance and matching, and builds a causal relationship model;

[0107] The counterfactual analysis submodule is based on the causal relationship model and adopts the Douglas causal model. It uses Python's DoWhy library to perform counterfactual analysis, create counterfactual scenarios, including the situation where the target health intervention is not implemented, compare the difference between the actual intervention results and the counterfactual results, use the identify and estimate methods to analyze the intervention effect, and generate counterfactual analysis results;

[0108] The impact assessment submodule comprehensively evaluates the impact of various health interventions based on the results of counterfactual analysis, uses Python's statistical analysis library to perform regression analysis and difference comparison, analyzes the effect of each intervention, and generates an impact analysis of the interventions.

[0109] In the causal model construction submodule, counterfactual theory and Python's CausalInference library are used to construct a causal model based on the dynamic risk prediction indicators of residents in nursing facilities. This process involves determining causal variables (such as specific health interventions) and outcome variables (such as improved health indicators). By using the statistical_methods method, covariates are balanced and matched to ensure that the estimation of causal effects is not affected by confounding variables. The constructed causal relationship model can reveal the direct connection between health interventions and expected health outcomes, providing a basis for further counterfactual analysis.

[0110] In the counterfactual analysis submodule, the constructed causal model is used to perform counterfactual analysis using the Douglas causal model and Python's DoWhy library. Scenarios were created that included the failure to implement the target health intervention, and the differences between the actual intervention results and these counterfactual scenarios were compared. The effects of specific interventions were analyzed through the identify and estimate methods, revealing the extent to which the interventions affected the expected health outcomes. The generated counterfactual analysis results provide an empirical basis for evaluating the effectiveness of health interventions.

[0111] In the impact assessment submodule, based on the results of counterfactual analysis, Python's statistical analysis library was used to comprehensively evaluate the impact of various health interventions. By performing regression analysis and difference comparison, the effect of each intervention was systematically analyzed, revealing the relative contribution of different interventions to improving the health status of residents. The generated intervention impact analysis report provides data-driven decision support for the management and medical team of the nursing facility, enabling them to select and implement the most effective health intervention strategies based on evidence.

[0112] In the remote monitoring system of the elderly care facility, the causal model construction submodule, the counterfactual analysis submodule, and the impact assessment submodule jointly provide a comprehensive analytical framework for the evaluation of health interventions. In response to the problem of hypertension, the system first collects and analyzes the baseline health data of residents, including age, gender, lifestyle habits, and blood pressure readings before the intervention, through the causal model construction submodule. Using the CausalInference library, this submodule builds a causal model that clarifies the causal relationship between interventions (such as regular physical activity, healthy diet, and health education) and blood pressure control, and excludes potential biases through the balance and matching of covariates. Subsequently, the counterfactual analysis submodule uses the DoWhy library to create a series of hypothetical scenarios that simulate what the blood pressure of residents would be like if these health interventions were not implemented. By comparing the actual post-intervention results with these counterfactual scenarios, this submodule reveals the actual effects of the interventions and provides direct evidence for evaluating the effectiveness of the interventions. Finally, the impact assessment submodule synthesizes the results of the counterfactual analysis and deeply evaluates the specific impact of each intervention on blood pressure control by performing regression analysis and difference comparison. This phase of analysis helps nursing facility management and medical teams understand which interventions are most effective and how health management strategies should be adjusted to maximize residents’ health and well-being.

[0113] See also Figure 7 ,The personalized threshold setting module includes a health data analysis submodule, a dynamic threshold adjustment submodule, and anomaly detection submodule;

[0114] The health data analysis submodule uses support vector machines and random forest algorithms to analyze historical health data based on intervention impact analysis. It uses the SVC and RandomForestClassifier functions in Python's Scikit-learn library, sets the SVC kernel function to the radial basis function, the penalty parameter C to 1.0, the number of trees of the RandomForestClassifier to 100, and the maximum depth to 10 to identify and analyze health patterns in historical data and generate health trend analysis results.

[0115] The dynamic threshold adjustment submodule uses a decision tree algorithm to dynamically adjust the threshold of real-time health data based on the health trend analysis results. It uses Python's DecisionTreeClassifier function, sets the maximum depth of the tree to 5, and the splitting criterion to information gain. It analyzes real-time health data, dynamically adjusts the health threshold to match the individual's health status, and generates a dynamically adjusted health threshold.

[0116] The anomaly detection submodule applies an anomaly detection algorithm to monitor the health status based on the dynamically adjusted health threshold. It uses Python's IsolationForest function, sets the number of trees to 100 and the anomaly ratio to 0.01, monitors real-time health data, identifies data points that deviate from the normal range, and generates customized health threshold indicators.

[0117] In the health data analysis submodule, support vector machines and random forest algorithms are used to conduct in-depth analysis of historical health data based on the impact analysis of intervention measures. Through Python's Scikit-learn library, the SVC and RandomForestClassifier functions are used specifically. For the historical health data set, the module performs a series of detailed operations. The support vector machine algorithm is implemented through the SVC function, with the radial basis function set as the kernel function and the penalty parameter C set to 1.0. This configuration aims to distinguish different health states by building a decision boundary in a high-dimensional space. During the specific implementation of the radial basis function kernel, each data point is mapped to a higher-dimensional space, in which hyperplanes can be used to separate data points belonging to different categories. The penalty parameter C controls the tightness of this separation boundary, thereby maintaining the generalization ability of the model while ensuring classification accuracy. The random forest algorithm is implemented through the RandomForestClassifier function, where the number of trees is set to 100 and the maximum depth is 10. This means that the model will build 100 decision trees, each of which is trained on a random subset of the data set. The maximum depth limit helps prevent the model from overfitting. During the execution process, each tree independently analyzes the data and gives a prediction result, and finally the final classification result is determined by a voting mechanism. This method improves the accuracy and robustness of health pattern recognition by integrating the predictions of multiple simple models. The combination of these two algorithms enables the health data analysis submodule to effectively identify and analyze health patterns in historical data and generate health trend analysis results. These results can not only reveal the changing trend of the health status of individuals or groups over time, but also provide a scientific basis for subsequent intervention measures.

[0118] In the dynamic threshold adjustment submodule, the decision tree algorithm is used and implemented through Python's DecisionTreeClassifier function. The maximum depth of the tree is set to 5, and the splitting criterion uses information gain. In this process, the decision tree algorithm dynamically adjusts the threshold of real-time health data by analyzing the results of health trend analysis. During the construction of the decision tree, the splitting of each node is based on the principle of maximizing information gain, and the features that can best classify the data are selected. The maximum depth limit helps to avoid excessive model complexity and thus reduce the risk of overfitting. The dynamic threshold is adjusted by continuously comparing the real-time data with the decision tree model and adjusting the threshold according to the judgment criteria on the decision tree path to better adapt to the individual's current health status. This method enables the health monitoring system to respond more sensitively to subtle changes in health status and provide support for personalized health management.

[0119] In the anomaly detection submodule, the IsolationForest function is used, and the number of trees is set to 100 and the anomaly ratio is set to 0.01. IsolationForest is an outlier detection method based on random forests, and its core idea is to isolate outliers. In practice, the algorithm randomly selects a feature and then randomly selects a cut value of the feature to build a decision tree. Outliers are isolated earlier in these trees and therefore have a shorter path length. By setting the anomaly ratio, the algorithm can adaptively determine the threshold based on the characteristics of the data set to identify outliers. In the context of health monitoring, this means that the system can effectively identify data points that deviate from the normal health state, providing the possibility for timely intervention.

[0120] Taking the remote monitoring system of elderly care facilities as the background, imagine a scenario in which the system collects health data including heart rate, blood pressure, sleep quality and activity. These data items specifically include heart rate (beats / minute), blood pressure (mmHg), sleep quality (sleep score, 0-100), and activity (number of steps). The simulated data example is: heart rate 75 beats / minute, blood pressure 120 / 80mmHg, sleep quality score 85, and daily activity 8000 steps.

[0121] In the health data analysis submodule, the system first uses these historical health data as input and analyzes them using support vector machines and random forest algorithms. The support vector machine algorithm configures the radial basis function and penalty parameter C=1.0 through the SVC function, processes heart rate and blood pressure data, and determines the boundaries of health status. The random forest algorithm uses the RandomForestClassifier function, configures 100 trees and a maximum depth of 10, analyzes sleep quality and activity data, and identifies health trends and abnormal patterns. This analysis process comprehensively evaluates the individual's physiological parameters and living habits, and generates comprehensive reports containing health trends and potential risks. These reports provide scientific basis for nursing staff in nursing facilities to develop or adjust personalized health management plans.

[0122] In the dynamic threshold adjustment submodule, the system uses a decision tree algorithm to process real-time health data based on the results of the health trend analysis. With a maximum depth of 5 for the decision tree and information gain as the splitting criteria, the system analyzes real-time data, such as heart rate and blood pressure for the day, and compares them with the set health thresholds. In this way, the system is able to dynamically adjust the thresholds. For example, if a resident's heart rate is continuously higher than the individual average, the system will lower the heart rate threshold to detect potential health problems earlier.

[0123] The anomaly detection submodule uses the IsolationForest algorithm, sets the number of trees to 100 and the anomaly ratio to 0.01 to monitor real-time health data. When data points of heart rate, blood pressure, sleep quality, or activity deviate from the set dynamic threshold, the system will mark these data points as abnormal. For example, if a resident's activity suddenly drops from an average of 8,000 steps per day to less than 2,000 steps, the system will identify this change as abnormal behavior, trigger an alarm, and notify caregivers for further evaluation.

[0124] See also Figure 8 ,The network behavior monitoring module includes a network construction submodule, a community detection submodule, and a behavior analysis submodule;

[0125] The network construction submodule is based on customized health threshold indicators and adopts network theory. Through Python's NetworkX library, it creates nodes representing the behaviors of the elderly and edges representing the interactions between behaviors. The add_node and add_edge methods are used to build the network. The weight of the edge is determined by the frequency and intensity of the elderly's behavior. The network structure representing daily activities and social interactions is constructed to generate network structure data.

[0126] The community detection submodule uses the community detection algorithm based on network structure data and creates the Louvain method through Python's Community library. It optimizes the modularity in the social structure to identify social clusters and associated behavioral pattern groups in the network, analyzes the social structure characteristics, and generates community detection results.

[0127] The behavior analysis submodule uses the behavior analysis method based on the community detection results to perform statistical analysis on the behavior data through Python's Pandas library. It calculates the behavior frequency, duration, and social interaction patterns, analyzes the changes and dynamics of the behavior patterns, and generates a network behavior insight map.

[0128] In the network construction submodule, a complex network structure representing the daily activities and social interactions of the elderly is constructed through network theory and Python's NetworkX library. This process uses the elderly's behaviors as nodes, the interactions between behaviors as edges, and the weights of the edges are determined based on the frequency and intensity of the behaviors. Specifically, first, a node is added for each behavior through the add_node method, such as "walking", "reading", "gardening", etc. Each node represents a specific activity participated in by the elderly. Subsequently, the add_edge method is used to add edges based on the frequency and intensity of interactions between behaviors to connect related nodes. For example, if an elderly person often engages in gardening activities after walking, the weight of the edge between the two activities will be higher. The weight of the edge is calculated by comprehensively analyzing the frequency of behaviors and the intensity of associations between behaviors, which reflects the intensity of interactions between different behaviors. Through this series of operations, a network structure that can reflect the daily life patterns and social interaction patterns of the elderly in detail is constructed, and the generated network structure data provides a basis for subsequent community detection and behavior analysis.

[0129] In the community detection submodule, based on the constructed network structure data, a community detection algorithm is used to identify social clusters and associated behavioral pattern groups in the network. This process is implemented through Python's Community library, and the Louvain method is used to optimize the modularity in the social structure. The Louvain method is a community detection algorithm based on modularity optimization. It effectively identifies the community structure in the network by iteratively assigning nodes to communities that maximize modularity. In the specific implementation, the algorithm first calculates the modularity contribution of each node in the network and assigns the nodes to the community based on this. Then, the algorithm evaluates whether the total modularity of the network can be further improved by merging communities. If so, the merger is performed. This process is repeated until the modularity cannot be improved by merging communities. In this way, the algorithm is able to reveal the inherent social structure of the behavioral patterns of the elderly, identify groups of individuals with similar behavioral patterns, and generate community detection results. These results not only reveal the patterns of social interactions among the elderly, but also provide a basis for understanding the characteristics of social structure.

[0130] In the behavior analysis submodule, based on the community detection results, the behavior analysis method is used to conduct in-depth statistical analysis of the behavior data of the elderly. Through Python's Pandas library, this submodule calculates and analyzes the behavior frequency, duration, and social interaction pattern. Specifically, the behavior data is first imported into Pandas's DataFrame, and then various statistical functions such as mean() and sum() are used to calculate indicators such as the average frequency and total duration of each behavior. Next, based on the community detection results, the behavior patterns in different social clusters are compared and analyzed to identify the commonalities and differences of the behavior patterns. In addition, the trend of behavior patterns over time and the impact of social interaction patterns on behavior changes are analyzed. Through these analyses, a network behavior insight map is generated, which not only provides a detailed description of the behavior patterns of the elderly, but also reveals the dynamic changes behind the behavior patterns and the impact of social interactions. These insights are of great value in designing targeted health interventions and social activities, and help improve the quality of life and social satisfaction of the elderly.

[0131] Assume that the system collects data including the frequency and duration of daily walking, reading, and gardening activities, as well as the number of social interactions with other residents. For example, the average frequency of walking activities is 2 times a day, and the average duration is 30 minutes each time; the average frequency of reading activities is 3 times a day, and the average duration is 1 hour each time; the average frequency of gardening activities is 3 times a week, and the average duration is 1.5 hours each time; the average number of social interactions with other residents is 5 times a day. Through the network construction submodule, the system creates a network structure containing these activities and interactions, where nodes represent activities, edges represent the relationship between activities, and the weight of the edge reflects the strength of the association of activities. The community detection submodule reveals social clusters formed based on common behavior patterns. For example, residents who often walk and garden together form a community. The behavior analysis submodule provides in-depth insights into behavior patterns, such as the high correlation between gardening activities and the number of social interactions, indicating that gardening is an effective activity to enhance social interaction. These analysis results provide a scientific basis for elderly care facilities to design activities and social programs that are more suitable for residents' needs, thereby optimizing the daily life and social experience of the elderly.

[0132] See also Fig. 9 ,The genetic strategy optimization module includes parameter encoding submodule, fitness function design submodule, and strategy selection submodule;

[0133] The parameter encoding submodule is based on the network behavior insight graph and uses a genetic algorithm. It is executed through Python's DEAP library to encode health monitoring parameters and resource allocation schemes into genetic algorithm gene sequences. The create function is used to set the encoding rules and length of the gene sequence, including the value range of the monitoring parameters and the quantitative standard of resource allocation, to generate the initial genetic code.

[0134] The fitness function design submodule designs a fitness function to evaluate the effect of the health management strategy based on the initial genetic coding. It uses the Fitness function in the DEAP library of Python to set the coverage, cost-effectiveness and user satisfaction of the strategy, calculates the fitness score of each strategy, evaluates the utility and feasibility of the strategy based on the score, and generates the fitness evaluation result.

[0135] The strategy selection submodule selects and optimizes the health management strategy based on the fitness evaluation results by applying the evolutionary mechanism of the genetic algorithm. It executes the select, mate, and mutate functions through the Python DEAP library to select the population, cross and mutate genes, and iterates to capture the optimal health management strategy and generate a strategy optimization plan.

[0136] In the parameter encoding submodule, a series of refined operations were performed using a genetic algorithm based on the network behavior insight graph to encode health monitoring parameters and resource allocation schemes into gene sequences in the genetic algorithm. Through the create function of Python's DEAP library, this submodule sets the encoding rules and length of the gene sequence, including the value range of monitoring parameters such as heart rate, normal range of blood pressure, number of daily activity steps, etc., as well as quantitative standards for resource allocation, such as the number of health monitoring devices that can be allocated to each elderly person and the degree of attention of medical staff. Each part of the gene sequence represents a health monitoring parameter or resource allocation decision, and its value is encoded in a form that can be operated by the genetic algorithm. In this way, the initial genetic encoding is generated, which provides a basis for the subsequent fitness function design and the selection and optimization of health management strategies.

[0137] In the fitness function design submodule, based on the initial genetic coding, a fitness function is carefully designed to evaluate the effectiveness of the health management strategy. Using the Fitness function in Python's DEAP library, this submodule sets multiple evaluation dimensions for each health management strategy, including the coverage, cost-effectiveness, and user satisfaction of the strategy. Coverage evaluates the proportion of the elderly population that the strategy can cover, cost-effectiveness analyzes the ratio of the economic input to the health benefits of implementing the strategy, and user satisfaction reflects the acceptance and satisfaction of the elderly with health management measures. By combining these dimensions, the fitness function calculates the fitness score of each strategy. A high score means that the strategy performs better in terms of coverage, cost-effectiveness, and user satisfaction. This process generates fitness evaluation results, which provides an important basis for screening and optimizing health management strategies.

[0138] In the strategy selection submodule, based on the fitness evaluation results, the evolutionary mechanism of the genetic algorithm is applied to select and optimize the health management strategy. Through the select, mate and mutate functions implemented by the DEAP library, this submodule performs a series of operations such as population selection, gene crossover and mutation. Population selection is based on the fitness evaluation results, and the strategy with good performance is preferentially selected as the father of the next generation. The gene crossover (mate) operation simulates the mating process in biological genetics, partially swaps the gene sequences of two strategies, and produces offspring with new characteristics. The mutation (mutate) operation introduces new mutations and increases the diversity of strategies by randomly changing certain genes in the strategy gene sequence. Through repeated iterations of this evolutionary process, the submodule can capture the most optimized health management strategy and generate a strategy optimization solution. While ensuring the comprehensiveness and accuracy of health monitoring for the elderly, these solutions also take into account the effective use of resources and the actual acceptance of the elderly, providing scientific and efficient health management solutions for elderly care facilities.

[0139] The system collects health monitoring parameters including heart rate, blood pressure, activity level, as well as resource allocation information such as the number of available health monitoring devices and the ratio of medical personnel. For example, the normal range of heart rate is set at 60-100 beats per minute, the normal range of blood pressure is set at systolic blood pressure 90-140mmHg and diastolic blood pressure 60-90mmHg, and the target number of daily activity steps is set at 5000-10000 steps. In terms of resource allocation, the number of health monitoring devices that can be allocated to each elderly person is set at 1-2, and the number of professional medical personnel equipped for every 10 elderly people is set at 1-3. Through the parameter encoding submodule, this information is encoded into the gene sequence of the genetic algorithm. The fitness function design submodule comprehensively evaluates the performance of different health management strategies in terms of coverage, cost-effectiveness and user satisfaction, and selects the strategy with the best overall performance. The strategy selection submodule continuously optimizes the health management strategy through the evolutionary mechanism of the genetic algorithm, and finally generates a set of health management plans that take into account the comprehensiveness and accuracy of health monitoring while taking into account the effective use of resources and the acceptance of the elderly, effectively improving the scientificity and practicality of health management in elderly care facilities.

[0140] See also Fig.10 A remote supervision method for elderly care facilities is provided. The remote supervision method for elderly care facilities is performed based on the remote supervision system for elderly care facilities, and comprises the following steps:

[0141] S1: Based on the behavioral data and social interaction data of the elderly, a graph neural network algorithm is used to build a dynamic behavioral graph model and analyze behavioral patterns in combination with deep learning technology. The Python NetworkX library is used to create nodes and edges, and the centrality of the nodes is calculated to generate behavioral social network analysis.

[0142] S2: Based on behavioral social network analysis, support vector machines and random forest algorithms are used to classify and identify long-term health records, and real-time health monitoring data is used to analyze health status. Python's Scikit-learn library is used for data processing and model training to generate health trend and risk analysis;

[0143] S3: Based on health trend and risk analysis, the dynamic Bayesian network algorithm is used to analyze continuous health monitoring data. The dynamic Bayesian network is constructed using Python's pgmpy library to process time series data and extract key risk indicators to generate dynamic risk prediction indicators.

[0144] S4: Based on dynamic risk prediction indicators, counterfactual theory and Douglas causal model are used to make causal inference and evaluation of the impact of differentiated health interventions. Causal modeling and counterfactual analysis are performed using Python's CausalInference library and DoWhy library to compare the actual intervention results with those without intervention, and then generate an impact assessment of the interventions.

[0145] S5: Based on the impact assessment of intervention measures, a machine learning algorithm is used to conduct a comprehensive analysis of historical and real-time health data, and a decision tree algorithm is applied through the Python Scikit-learn library to perform personalized threshold adjustment and anomaly detection, determine the health status of individuals, and generate customized health thresholds;

[0146] S6: Based on customized health thresholds, a genetic algorithm is used to optimize the health management strategy. The Python DEAP library is used to perform parameter encoding, fitness function design, and strategy selection. The health management strategy is optimized and refined through the evolutionary mechanism to generate a strategy optimization solution.

[0147] The health management system for the elderly, which was formed by the comprehensive application of advanced technologies such as graph neural networks, machine learning algorithms, dynamic Bayesian networks, and causal inference models, has brought significant beneficial effects. In-depth understanding of the behavioral patterns and social interactions of the elderly provides accurate data support, which helps to identify key health influencing factors and thus design more effective health intervention measures. The system can accurately identify long-term trends and potential risks in health records, providing a scientific basis for personalized health monitoring and timely intervention. In addition, the dynamic risk assessment mechanism enables health management to respond sensitively to changes in health status, adjust intervention strategies in a timely manner, and effectively prevent health risks. The application of causal inference ensures that the effect evaluation of intervention measures is more accurate, providing a solid foundation for the formulation of effective health management plans. The health management strategy optimized by machine learning and genetic algorithms not only improves the accuracy and adaptability of personalized health management, but also significantly improves the quality of life of the elderly and reduces the overall cost of health management.

[0148] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A remote monitoring system for elderly care facilities, characterized in that: The system includes a behavior pattern analysis module, a health monitoring optimization module, a risk assessment dynamic module, a causal intervention assessment module, a personalized threshold setting module, a network behavior monitoring module, and a genetic strategy optimization module; The behavior pattern analysis module uses a graph neural network algorithm to build a dynamic behavior graph model based on the behavior data and social interaction data of the elderly, and identifies behavior patterns and changes through deep learning to generate a comprehensive overview of behavior patterns; The behavior pattern analysis module includes a graph structure construction submodule, a feature extraction submodule, and a behavior recognition submodule; The graph structure construction submodule is based on the behavioral data and social interaction data of the elderly, adopts graph theory algorithms, uses the NetworkX library to build a dynamic behavioral graph model, adds nodes and edges through the add_node and add_edge methods, sets node weights to reflect the frequency of behavior, uses degree_centrality and betweenness_centrality functions to calculate node centrality, and generates a behavioral social network graph; The feature extraction submodule is based on the behavioral social network graph, adopts feature engineering methods, uses the Pandas library to preprocess the data, including cleaning and normalization, uses the PCA function of the Scikit-learn library to perform principal component analysis, extracts node degrees and clustering coefficients, and generates a behavioral feature data set; The behavior recognition submodule is based on the behavior feature dataset, adopts the graph neural network algorithm, uses the PyTorchGeometric library to build a GNN model, defines the model layer including GraphConv and ReLU activation functions, sets training parameters, applies the model to learn and recognize behavior patterns, and generates a comprehensive overview of behavior patterns; The health monitoring optimization module uses support vector machine and random forest algorithms based on a comprehensive overview of behavioral patterns to analyze long-term health records, monitor real-time health status, identify health trends and risks, and generate health pattern insights; The health monitoring optimization module includes a health pattern recognition submodule, a real-time monitoring submodule, and a health pattern analysis submodule; The health pattern recognition submodule is based on a comprehensive overview of behavioral patterns, adopts a support vector machine algorithm, uses the SVC function in the Scikit-learn library, sets the kernel function to a radial basis function, adjusts the penalty parameter C to 1.0, and the kernel parameter gamma to auto, classifies and recognizes patterns in long-term health records, analyzes health trends and potential risks, and generates health pattern classification data; The real-time monitoring submodule uses the random forest algorithm based on the health pattern classification data, through the RandomForestClassifier function in the Scikit-learn library, sets the number of trees to 100, analyzes the real-time health monitoring data, identifies the current health status and changes, and generates real-time health monitoring results; The health pattern analysis submodule uses Python programming language and Scikit-learn library to build a prediction model based on real-time health monitoring results, uses RandomForestClassifier as the classifier to set parameters, where n_estimators is set to 100, max_depth is set to 10, and uses the train_test_split method to divide the training set and the test set, applies the fit method to train the model, and uses the predict method to predict and generate health pattern insights; The dynamic risk assessment module is based on health pattern insights and uses a dynamic Bayesian network algorithm to analyze continuous health monitoring data, extract key risk indicators using time series data processing methods, and conduct real-time risk assessment and trend prediction through model construction and dynamic assessment technology to generate dynamic risk prediction indicators; The risk assessment dynamic module includes a DBN model construction submodule, a time series data processing submodule, and a risk dynamic assessment submodule; The DBN model building submodule is based on health pattern insights and uses Python's pgmpy library to execute the dynamic Bayesian network algorithm. The model is built by defining health status nodes and transition probabilities. The nodes represent multiple health indicators. The transition probabilities are calculated based on historical data to represent the changes in health status over time. The network is built using the add_node and add_edge methods, and the conditional probability table is set to reflect the probability relationship between nodes to generate the DBN model configuration. The time series data processing submodule is configured based on the DBN model, uses Python's Pandas library to perform time series analysis, processes continuous health monitoring data, sets data time index, applies the rolling method to calculate the moving average, executes the diff method to perform data differentiation, and generates time series feature data; The risk dynamic assessment submodule uses the DBN model to perform risk assessment based on time series feature data, calculates the probability distribution of health indicators through the VariableElimination of the pgmpy library, analyzes the dynamic changes of health risks, predicts short-term and long-term health risks using probability query and inference methods, and generates dynamic risk prediction indicators; The causal intervention assessment module is based on dynamic risk prediction indicators, uses counterfactual theory and Douglas causal model for causal inference, analyzes the impact of multiple health intervention measures, evaluates intervention effects, and generates intervention impact analysis; The causal intervention assessment module includes a causal model construction submodule, a counterfactual analysis submodule, and an impact assessment submodule; The causal model construction submodule is based on dynamic risk prediction indicators, adopts counterfactual theory, constructs a causal model through Python's CausalInference library, sets causal variables and outcome variables, uses the statistical_methods method to perform covariate balance and matching, and constructs a causal relationship model; The counterfactual analysis submodule is based on the causal relationship model, adopts the Douglas causal model, uses Python's DoWhy library to perform counterfactual analysis, creates counterfactual scenarios, including the situation where the target health intervention is not performed, compares the difference between the actual intervention results and the counterfactual results, uses the identify and estimate methods to analyze the intervention effect, and generates counterfactual analysis results; The impact assessment submodule comprehensively evaluates the impact of various health interventions based on the results of counterfactual analysis, uses Python's statistical analysis library to perform regression analysis and difference comparison, analyzes the effect of each intervention, and generates an intervention impact analysis; The personalized threshold setting module uses machine learning algorithms to analyze historical and real-time health data based on intervention impact analysis, performs personalized threshold adjustment and anomaly detection, and generates customized health threshold indicators; The personalized threshold setting module includes a health data analysis submodule, a dynamic threshold adjustment submodule, and an anomaly detection submodule; The health data analysis submodule analyzes historical health data based on intervention impact analysis, using support vector machines and random forest algorithms, using SVC and RandomForestClassifier functions in Python's Scikit-learn library, setting the SVC kernel function to a radial basis function, the penalty parameter C to 1.0, the number of trees of the RandomForestClassifier to 100, and the maximum depth to 10, to identify and analyze health patterns in historical data and generate health trend analysis results; The dynamic threshold adjustment submodule uses a decision tree algorithm to dynamically adjust the threshold of real-time health data based on the health trend analysis results, uses Python's DecisionTreeClassifier function, sets the maximum depth of the tree to 5, and the splitting criterion to information gain, analyzes real-time health data, dynamically adjusts the health threshold to match the individual's health status, and generates a dynamically adjusted health threshold; The anomaly detection submodule applies an anomaly detection algorithm to monitor the health status based on the dynamically adjusted health threshold, uses Python's IsolationForest function, sets the number of trees to 100 and the anomaly ratio to 0.01, monitors real-time health data, identifies data points that deviate from the normal range, and generates customized health threshold indicators; The network behavior monitoring module is based on customized health threshold indicators and adopts network theory to conduct structured analysis of daily activities and behavior patterns, map social and behavioral relationships through network construction technology, identify social clusters of behavior patterns using community detection algorithms, and analyze behavior dynamics through behavioral analysis methods to generate network behavior insight maps; The network behavior monitoring module includes a network construction submodule, a community detection submodule, and a behavior analysis submodule; The network construction submodule is based on customized health threshold indicators and adopts network theory. Through Python's NetworkX library, nodes representing the behaviors of the elderly and edges representing the interactions between behaviors are created. The add_node and add_edge methods are used to build the network. The weight of the edge is determined by the frequency and intensity of the elderly's behavior. The network structure representing daily activities and social interactions is constructed to generate network structure data. The community detection submodule uses a community detection algorithm based on network structure data, creates a Louvain method through Python's Community library, identifies social clusters and associated behavioral pattern groups in the network by optimizing the modularity in the social structure, and analyzes the social structure characteristics to generate community detection results; The behavior analysis submodule uses the behavior analysis method based on the community detection results to perform statistical analysis on the behavior data through Python's Pandas library, calculates the behavior frequency, duration and social interaction pattern, analyzes the behavior pattern changes and dynamics, and generates a network behavior insight map; The genetic strategy optimization module uses genetic algorithms to encode and optimize health management strategies based on network behavior insight graphs, integrates health monitoring parameters with resource allocation plans, and uses evolutionary mechanisms to cyclically select and refine, thereby mining the optimal management strategy and generating a strategy optimization plan. The genetic strategy optimization module includes a parameter encoding submodule, a fitness function design submodule, and a strategy selection submodule; The parameter encoding submodule is based on the network behavior insight map, uses a genetic algorithm, and is executed through Python's DEAP library to encode health monitoring parameters and resource allocation schemes into genetic algorithm gene sequences. The create function is used to set the encoding rules and length of the gene sequence, including the value range of the monitoring parameters and the quantitative standard of resource allocation, to generate the initial genetic code; The fitness function design submodule designs a fitness function to evaluate the effect of the health management strategy based on the initial genetic coding, uses the Fitness function in the DEAP library of Python, sets the coverage, cost-effectiveness and user satisfaction of the strategy, calculates the fitness score of each strategy, evaluates the utility and feasibility of the strategy according to the score, and generates a fitness evaluation result; The strategy selection submodule selects and optimizes the health management strategy based on the fitness evaluation results by applying the evolutionary mechanism of the genetic algorithm. It executes the select, mate and mutate functions through the DEAP library of Python to select the population, cross and mutate the genes, and iterates to capture the optimal health management strategy and generate a strategy optimization solution.

2. The remote monitoring system for elderly care facilities according to claim 1, characterized in that: The comprehensive overview of behavioral patterns includes behavioral frequency distribution, key social circle analysis, and key behavioral event identification; the health pattern insights include physiological indicator trends, health abnormality indicators, and individual health risk levels; the dynamic risk prediction indicators include health status probability scores, short-term risk predictions, and long-term health risk predictions; the intervention measure impact analysis includes intervention strategy effectiveness scores, health improvement contribution, and intervention measure optimization plans; the customized health threshold indicators include individualized physiological indicator thresholds, health risk warning indicators, and instant health assessment standards; the network behavior insight map includes social interaction pattern maps, key behavioral network nodes, and social activity impact assessments; the strategy optimization plan includes an optimized monitoring parameter set, a resource allocation plan, and an intervention strategy adjustment plan.

3. A remote supervision method for elderly care facilities, characterized in that: The remote monitoring system for elderly care facilities is implemented according to any one of claims 1-2.

Images