System and method for evaluating living habits of old people based on big data analysis

Abstract

The invention belongs to the technical field of big data assessment, and discloses an old people living habit assessment system and method based on big data analysis. The method comprises the following steps: collecting comprehensive life data of old people, and constructing to obtain a multi-dimensional big data set; performing comprehensive processing on the obtained multi-dimensional big data set to obtain a unified life data set; dynamically identifying the life mode of the old people from the unified life data set; a hybrid tree model is constructed, the life pattern of the old people is divided into a plurality of levels, and each level comprises a plurality of habit patterns; calculating different overlapping and crossing modes in the habit modes; evaluating the health degree of the old people according to the life mode of the old people; for different overlapping and crossing modes, obtaining association strength of the overlapping and crossing modes, and dynamically adjusting the evaluated health degree based on the association strength to obtain a health evaluation result; and outputting a health assessment result to an assessment terminal, and customizing a personalized health intervention scheme for the elderly.

Description

Technical Field

[0001] The present invention relates to the technical field of big data evaluation, and more specifically, to an evaluation system and method for the living habits of the elderly based on big data analysis. Background Art

[0002] With the continuous increase in the elderly population, how to provide high-quality health management and life care services for the elderly has become a major social problem to be solved; however, there are many deficiencies in the existing technology in the evaluation of the living habits of the elderly and health management, and it is difficult to meet the growing personalized needs of the elderly.

[0003] Firstly, the existing technology lacks the ability to comprehensively collect and integrate multi-dimensional living data of the elderly; the living conditions of the elderly involve multiple dimensions, but the existing technology often only focuses on data in a single dimension, such as physiological indicators, and it is difficult to comprehensively reflect the real living conditions of the elderly, resulting in one-sided evaluation results and low accuracy; secondly, the existing technology lacks the ability to dynamically identify and finely divide living patterns; the living patterns of the elderly are often hidden in a large amount of living data, with complex spatio-temporal characteristics and internal structures, but most of the existing technology relies on manual experience for pattern recognition, with low efficiency and difficulty in capturing detailed information of the patterns, restricting the in-depth evaluation of living habits; furthermore, the existing technology's health evaluation of the elderly is often too simplistic, and the accuracy and personalization degree of the evaluation results are not high; there are intricate mutual influences between different living patterns, but most of the existing evaluation methods are based on a single pattern for evaluation, ignoring the associations between patterns, resulting in evaluation results lacking pertinence and personalization degree; in addition, the existing evaluation methods are often static and one-time, and cannot dynamically adjust the evaluation results as the living habits of the elderly change, and it is difficult to provide continuous health management services for the elderly.

[0004] In view of this, the present invention proposes an evaluation system and method for the living habits of the elderly based on big data analysis to solve the above problems. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the existing technology and to achieve the above object, the present invention provides the following technical solutions: An evaluation system for the living habits of the elderly based on big data analysis, including: a data integration module, configured to receive comprehensive living data of the elderly collected by wearable devices, environmental sensors, and smart home devices, where the comprehensive living data includes physiological data, behavioral data, and environmental data; and construct a multi-dimensional big data set;

[0006] A unified processing module, configured to comprehensively process the obtained multi-dimensional big data set and form a unified data format to obtain a unified living data set;

[0007] A dynamic pattern recognition module, which is used to dynamically recognize the living patterns of the elderly from a unified living dataset by using an improved cellular automata algorithm;

[0008] A pattern division module, which is used to build a hybrid tree model, divide the living patterns of the elderly into several levels, and each level contains several habitual patterns; calculate different overlapping and cross patterns in the habitual patterns;

[0009] A comprehensive evaluation module, which is used to evaluate the health level of the elderly's living patterns; for different overlapping and cross patterns, obtain their association strengths, and dynamically adjust the evaluated health level based on the association strengths to obtain a health evaluation result; output the health evaluation result to an evaluation terminal; each module is connected by wired and / or wireless means to realize data transmission between modules.

[0010] Further, the method of performing comprehensive processing includes:

[0011] In a multi-dimensional big dataset, for each data dimension, calculate its interquartile range IQR = Q3 - Q1; where, Q3 is the upper quartile and Q1 is the lower quartile; determine the upper boundary value S1 and the lower boundary value S2 based on the interquartile range;

[0012] Mark the data points in the corresponding data dimension that exceed the upper boundary value or the lower boundary value as possible outliers;

[0013] Divide the multi-dimensional big dataset into several sub-datasets according to time or other identifiers;

[0014] For each data dimension in each sub-dataset, extract the values of all data points on the corresponding data dimension, and sort these values from small to large to obtain an ordered array; build a retrieval structure based on the ordered array;

[0015] Set a radius r. For each data dimension i, use the built retrieval structure to query the neighborhood range of the data point P on the corresponding data dimension; for each data dimension i, if the number of data points in the neighborhood of the data point P on the data dimension i is 0, set its own distance to infinity; otherwise, set its own distance where, m is the number of dimensions of the data point; δ is an influence parameter, and en(i) is the information entropy of the data dimension i; take the L2 norm of the self-distance and normalize it to obtain the outlier score of the data point P, and set an outlier score threshold; if the outlier score of the data point P is greater than the outlier score threshold, mark the data point P as a hidden outlier; if the data point P is marked as both a possible outlier and a hidden outlier, then regard the data point P as an outlier;

[0016] For each data dimension, calculate its Spearman-Pan coefficient rho; sort the data dimensions in descending order according to the value of rho, and select the top N data dimensions as reference dimensions; for the outliers in each subset of data, use the reference dimensions to replace the data points through K-nearest neighbor regression or decision tree regression; after re-integration, a unified living dataset is obtained.

[0017] Further, the method for constructing the retrieval structure includes:

[0018] Set the data structure of the retrieval node, where the data structure includes a pointer to the left subtree, a pointer to the right subtree, the value range corresponding to the retrieval node, and the number of data points within the value range;

[0019] For any retrieval node, determine the value range it needs to cover; if the upper and lower limits of the value range are equal, create a new node and set the number of data points within the corresponding value range to 1;

[0020] If the upper and lower limits of the value range are not equal, continue to split the value range. Specifically, calculate the middle value mid, create a new node ne, and set the value range of the new node ne to the original value range; recursively construct the left subtree of the new node ne, and define the value range it covers as the interval formed by the upper limit of the original value range and the middle value mid; recursively construct the right subtree of the new node ne, and define the value range it covers as the interval formed by mid + 1 and the lower limit of the original value range; through the recursive process, a complete retrieval structure is constructed.

[0021] Further, the method for identifying the living pattern includes:

[0022] Divide the unified living dataset into time windows of a fixed length according to the time series; construct a cell-like model of the elderly's living pattern, and define the cell state of the cells in the cell-like model; combine the data points within each time window into a cell, and the cell state is represented by a data vector composed of data points of different data types within the time window; define the coverage range of each cell, and use the data vector of the first time window as the initial cell state; define the state transition function, for each cell, according to its current cell state and the cell states of the cells within its coverage range, apply the state transition rule to calculate its own cell state at the next time step; at the same time, update the states of all cells; repeat until the preset number of iterations is reached; perform comprehensive clustering on the cell states after iteration to obtain state groups; analyze the characteristics of each state group and extract representative living patterns.

[0023] Further, the method for defining the state transition function includes:

[0024] Define a set of state transition rules R. Taking the cell state S(j, t) of cell j at time t, the cell states of all cells within its involved range, and the state transition rules R as inputs, calculate the new cell state S(j, t + 1) of cell j at time t + 1 as S(j, t + 1) = FP(S(j, t), {S(j′, t)|j'∈N(j)}, R); where S(j′, t) is the cell state of cell j' within the involved range of cell j; and FP() is the state transition function.

[0025] Furthermore, the way of performing comprehensive clustering includes:

[0026] Define the dimension of the clustering input as the dimension of the data vector corresponding to the cell state; set a mapping network and the grid size of the mapping network; initialize the weight vector of each neuron in the mapping network;

[0027] For the data vector x of each cell state, calculate the Euclidean distance between x and the weight vectors of all neurons; find the neuron c with the smallest Euclidean distance from x;

[0028] Define a neighborhood function where r_cy is the Euclidean distance between the position vector of neuron y and c on the grid; and σ(l) is the neighborhood radius at the current iteration number l;

[0029] For each neuron y, update its weight vector. The update formula for the weight vector is:

[0030] where w_y(l + 1) is the weight vector of neuron y at iteration number l + 1, w_y(l) is the weight vector of neuron y at the current iteration number l, is the learning rate at the current iteration number l;

[0031] Repeat the update until the preset number of iterations is reached; set a sliding time window, regularly update the cell-like model, cluster the neurons of the mapping network; and take the neurons corresponding to each cluster obtained by clustering and the data vectors of the cell states they map as a state group.

[0032] Furthermore, the way of dividing the living patterns of the elderly into several levels includes:

[0033] Construct a tree structure. According to the extracted living patterns, take all living patterns as the root nodes of the tree structure. For each living pattern, use the hierarchical clustering algorithm to divide it into sub-patterns, and take the sub-patterns as the sub-nodes of the original living pattern to construct an initial hierarchical pattern tree;

[0034] For each node in the hierarchical pattern tree, calculate its information gain, and select the top N1 features with the largest information gain as the key feature subset corresponding to the node; for each internal node in the hierarchical pattern tree, use the key feature subset to construct a hybrid tree model. Each leaf node of the hybrid tree model corresponds to a sub-pattern, and these sub-patterns are used as the child nodes of the current node and inserted into the hybrid tree model. For the leaf nodes of the hybrid tree model, use the key feature subset to construct a Gaussian mixture model, and each mixture component of the Gaussian mixture model corresponds to a habit pattern;

[0035] For each child node of each leaf node, calculate the Bayesian information criterion value of the Gaussian mixture model, and set a merging threshold; for all child nodes of the current leaf node, calculate their pairwise similarities;

[0036] Find two child nodes A and B with the largest probability distribution similarity, and merge them into a new child node C; calculate the Bayesian information criterion value of the Gaussian mixture model corresponding to the new child node C, denoted as the merging value; if the merging value is less than the Bayesian information criterion value of the child nodes before merging minus the merging threshold, accept the merger and use C to replace A and B as the new child nodes, otherwise do not merge.

[0037] Further, the calculation method of the overlapping cross patterns includes:

[0038] For each leaf node in the hybrid tree model, extract the mean vector and covariance matrix of each mixture component in its Gaussian mixture model; define an overlapping threshold, and for any two mixture components A1 and B1 in the Gaussian mixture model corresponding to each leaf node, calculate their Mahalanobis distance U(A1, B1); if the Mahalanobis distance is less than the overlapping threshold, it is considered that A1 and B1 overlap; repeat the pairwise calculation of the Mahalanobis distance for all sub-habit patterns under the same habit pattern to obtain all overlapping relationships within the habit pattern; for different habit patterns, calculate whether there is an overlapping relationship between them.

[0039] Based on all the obtained overlapping relationships, construct an overlapping graph, use each sub-habit pattern as a node, and if two sub-habit patterns overlap, connect an edge in the overlapping graph; find all the maximum cliques in the overlapping graph, and each maximum clique corresponds to an overlapping cross pattern.

[0040] Further, collect historical data, mine the association rules between life patterns and health conditions from the historical data, formalize the association rules, and construct a health assessment knowledge base for life patterns;

[0041] According to the health assessment knowledge base, initial health scores are set for different living patterns, which are the degrees of health; for each overlapping and intersecting pattern, calculate its association strength with other overlapping and intersecting patterns; construct a pattern association network, where the nodes in the pattern association network are overlapping and intersecting patterns, and the nodes are connected by edges and assigned weights; the weight of the edge is the association strength.

[0042] Traverse each node in the pattern association network and extract the association strength between the nodes connected to it.

[0043] Calculate the adjusted score PY = HG + ∑(GY × RT); where HG is the initial health score, GY is the degree of the node connected to the calculated node, and RT is the association strength between the corresponding node and the calculated node.

[0044] Map the adjusted score to the semantic health level, or perform mapping to obtain a curve; this is the health assessment result.

[0045] An assessment method for the living habits of the elderly based on big data analysis, which is implemented based on the above-mentioned assessment system for the living habits of the elderly based on big data analysis, includes: Step 1, receive the comprehensive living data of the elderly collected by wearable devices, environmental sensors, and smart home devices, and the comprehensive living data includes physiological data, behavioral data, and environmental data; construct a multi-dimensional big data set.

[0046] Step 2, perform comprehensive processing on the obtained multi-dimensional big data set and form a unified data format to obtain a unified living data set.

[0047] Step 3, use the improved cellular automata algorithm to dynamically identify the living patterns of the elderly from the unified living data set.

[0048] Step 4, construct a hybrid tree model, divide the living patterns of the elderly into several levels, and each level contains several habit patterns; calculate different overlapping and intersecting patterns in the habit patterns.

[0049] Step 5, for the living patterns of the elderly, evaluate their health degrees; for different overlapping and intersecting patterns, obtain their association strengths, and dynamically adjust the evaluated health degrees based on the association strengths to obtain a health assessment result; output the health assessment result to the assessment terminal.

[0050] The technical effects and advantages of the assessment system and method for the living habits of the elderly based on big data analysis of the present invention:

[0051] The present invention can comprehensively and accurately reflect the real living conditions of the elderly, breaking through the one-sided and subjective limitations of traditional assessment methods; through the comprehensive collection and intelligent processing of multi-source heterogeneous data, the system constructs a large dataset containing multi-dimensional information such as physiology, behavior, and environment, providing a sufficient data basis for assessment and ensuring the objectivity and comprehensiveness of the assessment results; secondly, it has the ability to automatically discover living patterns, can deeply mine the living rules hidden in the massive data, and provides refined pattern support for health assessment; the system uses algorithms to dynamically identify the living patterns of the elderly from multi-dimensional data, hierarchically divides the patterns, and refines them to the granularity of habit patterns, fully mining the internal structure and detailed information of the living patterns, providing rich pattern knowledge for assessment; furthermore, the health assessment method has a high degree of accuracy and personalization; a pattern association network is constructed, which can comprehensively consider the complex associations between living patterns, dynamically adjust the assessment results, and provide personalized suggestions for improving the living habits of the elderly, helping to improve the quality of life of the elderly. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 FIG. is a schematic diagram of the elderly living habit assessment system based on big data analysis of the present invention;

[0053] Figure 2 FIG. is a schematic diagram of the elderly living habit assessment method based on big data analysis of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] Embodiment 1

[0056] Please refer to Figure 1 As shown, the elderly living habit assessment system based on big data analysis in this embodiment includes:

[0057] A data integration module for receiving comprehensive living data of the elderly collected by wearable devices, environmental sensors, and smart home devices, where the comprehensive living data includes physiological data, behavioral data, and environmental data; and constructing a multi-dimensional large dataset;

[0058] A unified processing module for comprehensively processing the obtained multi-dimensional large dataset and forming a unified data format to obtain a unified living dataset;

[0059] A dynamic pattern recognition module, which is used to dynamically identify the living patterns of the elderly from a unified living dataset by using an improved cellular automata algorithm;

[0060] A pattern division module, which is used to construct a hybrid tree model, divide the living patterns of the elderly into several levels, and each level contains several habitual patterns; calculate different overlapping and cross patterns in the habitual patterns;

[0061] A comprehensive evaluation module, which is used to evaluate the health level of the elderly according to their living patterns; for different overlapping and cross patterns, obtain their association strength, and dynamically adjust the evaluated health level based on the association strength to obtain a health evaluation result; output the health evaluation result to an evaluation terminal to provide personalized suggestions for improving the living habits of the elderly; each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0062] Collect the physiological data of the elderly by using wearable devices, such as heart rate, body temperature, blood pressure, sleep quality, etc.; wearable devices include but are not limited to smart bracelets, smart watches, smart clothing, etc.; collect the data of the living environment of the elderly by using environmental sensors, such as temperature, humidity, noise, air quality, etc.; environmental sensors can be deployed in areas such as the rooms, living rooms, and bedrooms where the elderly live.

[0063] Collect the daily activity data of the elderly by using smart home devices, such as the usage of household appliances, the opening and closing status of doors and windows, movement trajectories, etc.; smart home devices include but are not limited to smart TVs, smart refrigerators, smart door locks, smart cameras, etc.; collect the living habit data of the elderly by using mobile terminal applications or network platforms, such as eating habits, exercise habits, social activities, etc.; these data can be obtained by the elderly filling in independently or by family members on their behalf; integrate and quantify the data from the above multiple sources, such as quantifying type data into numerical data points, etc.; construct a multi-dimensional big dataset containing the physiological data, environmental data, activity data, and living habit data of the elderly.

[0064] Unify the conversion of different data types (such as strings, numerical values, etc.) into numerical data recognizable by the model; since the data dimensions of different dimensions are different, normalization processing is required, and common methods include Min-Max normalization, Z-Score normalization, etc.; for categorical data, numerical encoding is required, such as One-Hot encoding, Label encoding, etc.

[0065] In the multi-dimensional big dataset, for each data dimension, calculate its interquartile range IQR = Q3 - Q1; where Q3 is the upper quartile and Q1 is the lower quartile; determine the upper boundary value S1 and the lower boundary value S2 based on the interquartile range.

[0066] The upper boundary value S1 = Q3 + k × IQR × (1 + α × (n - n0)); where k is a constant (usually taken as 1.5); α is a hyperparameter used to control the change rate of dimension adjustment, usually taking values from 0.1 to 0.5; n is the number of data dimensions; n0 is a preset dimension threshold. When the number of dimensions exceeds n0, the dimension adjustment starts to change. When the data dimension is high, the discrimination criterion for outliers is relaxed to avoid too strict judgment that may lead to a large number of data being judged as outliers. By adjusting α and n0, the sensitivity of the outlier discrimination criterion to the number of dimensions can be controlled;

[0067] The lower boundary value S2 = Q1 - k × IQR × (1 + β × |s| + γ × e), where β and γ are two control parameters used to control the sensitivity of the data; s is the skewness of the data corresponding to the data dimension, reflecting the asymmetry of the data distribution; e is the kurtosis of the data corresponding to the data dimension, reflecting the peakedness of the data distribution. Data points in the corresponding data dimension that exceed the upper boundary value or the lower boundary value are marked as potential outliers.

[0068] The large multi-dimensional dataset is segmented into several sub-datasets according to time or other identifiers. Each sub-dataset contains data records of multiple data dimensions under the same time period or identifier;

[0069] For each data dimension in each sub-dataset, the values of all data points on the corresponding data dimension are extracted, and these values are sorted from smallest to largest to obtain an ordered array. A retrieval structure is constructed based on the ordered array, the retrieval nodes of the retrieval structure are defined, and the data structure of the retrieval nodes is set. The data structure includes a pointer to the left subtree, a pointer to the right subtree, the value range corresponding to the retrieval node (represented by an interval), and the number of data points within the value range.

[0070] For any retrieval node, determine the value range it needs to cover. If the upper and lower limits of the value range are equal, it means that the retrieval node corresponds to a single value. Create a new node and set the number of data points within the corresponding value range to 1, indicating that there is one data point within the value range;

[0071] If the upper and lower limits of the value range are not equal, it indicates that the search node covers a value range, and then continue to split the value range. Specifically, calculate the middle value mid (average of the upper and lower limits), create a new node ne, and set the value range of the new node ne to the original value range; recursively construct the left subtree of the new node ne, and define the covered value range as the interval composed of the upper limit of the original value range and the middle value mid; recursively construct the right subtree of the new node ne, and define the covered value range as the interval composed of the middle value mid + 1 and the lower limit of the original value range; through the recursive process, construct a complete search structure, where each node corresponds to a value range and records the number of data points in this range.

[0072] Set a radius r to determine the neighborhood range of the data point P; the radius r can be a fixed value or dynamically adjusted according to the data distribution; for each data dimension i, use the constructed search structure to query the neighborhood range of the data point P in the corresponding data dimension; specifically, find the smallest search node on the search structure that contains the interval [P - r, P + r]; the number of data points of this search node is the number of data points in the neighborhood of the data point P in the data dimension i.

[0073] For each data dimension i, if the number of data points in the neighborhood of the data point P in the data dimension i is 0, it means that the data point P is an isolated point, and set its own distance to a large value (such as infinity); otherwise, set its own distance Among them, m is the number of dimensions of the data point; δ is the influence parameter used to control the influence degree of information entropy on the distance, en(i) is the information entropy of the data dimension i, and the information entropy can reflect the uniformity of the data distribution; for a more uniform dimension, en(i) is larger and the self - distance increases accordingly, which can improve the sensitivity to outliers in the uniformly distributed dimension; take the L2 norm of the self - distance and normalize it to obtain the outlier score of the data point P, and set an outlier score threshold; the outlier score threshold can be adjusted according to the data distribution or the expected proportion of outliers; if the outlier score of the data point P is greater than the outlier score threshold, then mark the data point P as a hidden outlier; if the data point P is marked as a potential outlier and a hidden outlier at the same time, then regard the data point P as an outlier.

[0074] For each data dimension, calculate its Spearman - Pan coefficient rho; sort the data dimensions in descending order according to the value of rho, and select the first N data dimensions as the reference dimensions; for the outliers in each subset of data, use the reference dimensions to replace the data point through K - nearest neighbor regression, decision tree regression or other regression models (calculate the new value to replace the data point corresponding to the original outlier); after re - integrating, a unified living data set is obtained.

[0075] The recognition methods of life patterns include:

[0076] Divide the unified life dataset into time windows of fixed length according to the time series, such as daily, weekly or monthly; construct a cell-like model of the elderly's life pattern, and define the cell state of the cell in the cell-like model; combine the data points within each time window into a cell, and the cell state is represented by a data vector composed of data points of different data types within the time window, that is, physiological indicators, activity types, environmental parameters, etc.

[0077] Define the scope of involvement of each cell. The scope of involvement can be several time windows before and after in time, or can include other relevant data within the same time period; take the data vector of the first time window as the initial cell state; define the state transition function. For each cell, according to its current cell state and the cell states of the cells within the scope of involvement, apply the state transition rule to calculate its own cell state at the next time step; at the same time, update the states of all cells; repeat until the preset number of iterations is reached.

[0078] The ways to define the state transition function include:

[0079] Based on the prior knowledge of the elderly's living habits and data characteristics, define a set of state transition rules R to describe how the cell state evolves over time; for example, there is a correlation between the cell states of adjacent time windows, environmental factors will affect the behavior state, and the physiological state is affected by the comprehensive influence of behavior and environment, etc.

[0080] Take the cell state S(j,t) of cell j at time t, the cell states of all cells within its scope of involvement, and the state transition rule R as inputs, and calculate the new cell state S(j,t + 1) of cell j at time t + 1;

[0081] S(j,t + 1) = FP(S(j,t),{S(j',t)|j'∈N(j)},R); where, S(j′,t) is the cell state of cell j' within the scope of involvement of cell j; FP() is the state transition function, which can be a machine learning model (such as RNN, LSTM, etc.) or a rule-based function;

[0082] Fusing data-driven machine learning methods and rule constraints based on domain knowledge helps to improve the accuracy and interpretability of pattern recognition.

[0083] Perform comprehensive clustering on the iterated cell states, group similar states into one category to obtain state groups; specifically, define the clustering input dimension as the dimension of the data vector corresponding to the cell state; set a mapping network, set the grid size (row × column) of the mapping network, and set it appropriately according to the data volume; initialize the weight vector of each neuron in the mapping network, usually using small random values.

[0084] For each data vector x of the cell state, calculate the Euclidean distance between x and the weight vectors of all neurons; find the neuron c with the minimum Euclidean distance from x; define a neighborhood function hc(l), which describes the topological neighborhood relationship between any neuron y and neuron c at the current iteration l.

[0085] Among them, r_cy is the Euclidean distance between the position vectors of neuron y and c on the grid; σ(l) is the neighborhood radius at the current iteration l, which is a decreasing function that controls the size of the neighborhood.

[0086] The decreasing function is usually: Among them, σ_0 is the initial neighborhood radius, and τ_σ is the decay rate control coefficient, which is used to control the decay rate of the neighborhood radius.

[0087] For each neuron y, update its weight vector. The update formula for the weight vector is:

[0088] Among them, w_y(l + 1) is the weight vector of neuron y at iteration l + 1, and w_y(l) is the weight vector of neuron y at the current iteration l. is the learning rate at the current iteration l (the same as σ(l)), which is a decreasing function used to control the convergence speed.

[0089] According to the neighborhood function and the learning rate, update the weight vectors of neuron c and the neurons in its neighborhood to move them in the direction of x; reduce the neighborhood radius and lower the learning rate to simulate the process of the network gradually converging; repeat the above process until the preset number of iterations is reached; during the above process, set a sliding time window to update the cell-like model regularly, introduce an adaptive mechanism, and dynamically adjust the state transition rules and neighborhood structure according to new data.

[0090] The neurons automatically form a topologically ordered mapping, and the data vectors corresponding to similar cell states are mapped to adjacent neurons. Cluster the neurons of the mapping network, such as K-means, hierarchical clustering, etc.; regard the neurons corresponding to each cluster obtained by clustering and the data vectors corresponding to the cell states they map as a state group; analyze the characteristics of each state group and extract representative living patterns; for example, regular work and rest patterns, frequent going out patterns, sedentary patterns, etc.

[0091] Specifically, extract features from each state group to extract the key features that describe the state group; the key features can include:

[0092] Physiological features: Statistical values (mean, variance, etc.) of physiological indicators such as heart rate, blood pressure, and sleep quality.

[0093] Behavioral features: statistical values of activity types (such as sedentary, walking, exercising, etc.) and their durations;

[0094] Environmental features: statistical values of environmental parameters such as temperature, humidity, noise, air quality, etc.;

[0095] Temporal features: time distribution of the occurrence of state groups (such as morning / evening, weekday / holiday, etc.);

[0096] Based on the extracted key features, perform pattern analysis on each state group:

[0097] Regularity analysis: Detect the periodic patterns of state groups over time, such as sleeping every night and going out every morning;

[0098] Duration analysis: Analyze the duration distribution of state groups, such as short activities and long sedentary periods;

[0099] Association analysis: Explore the association relationships between different dimensional features within state groups, such as improved sleep quality after exercise and reduced activities due to poor air quality.

[0100] Combined with domain knowledge, assign representative lifestyle pattern labels to each state group. For example, regular daily routine pattern: good physiological periodicity and regular activity schedule;

[0101] Sedentary pattern: mainly sedentary activities with less exercise;

[0102] Frequent going out pattern: mainly going out activities with a scattered activity time distribution;

[0103] Sleep problem pattern: poor sleep quality and frequent nocturnal activities;

[0104] Healthy pattern: good physiological indicators and moderate activities.

[0105] Through the above steps, representative lifestyle patterns can be extracted from the multi-dimensional life big data of the elderly, while complex spatio-temporal patterns can be captured, and it has good adaptability and interpretability, providing a basis for health assessment and personalized intervention.

[0106] The ways to divide the lifestyle patterns of the elderly into several levels include:

[0107] Construct a tree structure. According to the extracted lifestyle patterns, take all lifestyle patterns as the root nodes of the tree structure. For each lifestyle pattern, use hierarchical clustering algorithms (such as BIRCH, CURE, etc.) to divide it into sub-patterns, and take the sub-patterns as the child nodes of the original lifestyle pattern (parent pattern) to construct an initial hierarchical pattern tree.

[0108] For each node (pattern) in the hierarchical pattern tree, calculate the information gain of its feature set (physiological, behavioral, environmental, etc. features), and select the top N1 features with the largest information gain as the key feature subset corresponding to the node;

[0109] For each internal node (non-leaf node) in the hierarchical pattern tree, use the key feature subset to construct a hybrid tree model for subdividing the pattern. Each leaf node of the hybrid tree model corresponds to a sub-pattern, and these sub-patterns are used as the child nodes of the current node and inserted into the hybrid tree model. For the leaf nodes (sub-patterns) of the hybrid tree model, use the key feature subset to construct a Gaussian mixture model (GMM). Each mixture component of the Gaussian mixture model corresponds to a habitual pattern, and these habitual patterns are used as the child nodes of the current leaf node.

[0110] For the child nodes of each leaf node, calculate the Bayesian information criterion value (BIC) of the Gaussian mixture model, and set a merging threshold to determine whether the current pattern node needs to be merged. The merging threshold can be an empirical value or set adaptively according to the data distribution; for all the child nodes (mixture components of the GMM) of the current leaf node, calculate their pairwise similarity. The similarity can be the similarity of probability distributions, such as KL divergence, JS divergence, etc.; find the two child nodes (habitual patterns) A and B with the largest probability distribution similarity, merge them into a new child node C; calculate the Bayesian information criterion value (BIC) of the Gaussian mixture model corresponding to the new child node C, denoted as the merging value; if the merging value is less than the Bayesian information criterion value of the child nodes before merging minus the merging threshold, accept the merger and use C to replace A and B as the new child nodes (habitual patterns), otherwise do not merge.

[0111] The life patterns corresponding to the internal nodes are gradually subdivided layer by layer to form a hierarchical pattern tree. The top layer is the total pattern, the middle layer is the sub-pattern, and the bottom layer is the fine-grained habitual pattern; at the same time, taking advantage of the hybrid tree model, the pattern can be finely divided based on features, and the habitual pattern can be modeled using GMM; this hierarchical structure is conducive to analyzing the relationships between patterns of different granularities and can merge or refine the patterns according to needs, improving the interpretability and applicability of the model. At the same time, the hybrid tree model combines the advantages of decision trees and Gaussian mixture models and can perform both fine pattern division and probability modeling.

[0112] For each leaf node (corresponding to a habitual pattern) in the hybrid tree model, extract the mean vector and covariance matrix of each mixture component (sub-habitual pattern) in its Gaussian mixture model (GMM); define an overlap threshold to determine whether there is an overlap between two sub-habitual patterns; the overlap threshold can be an empirical value or set adaptively according to the data distribution.

[0113] For any two mixture components A1 and B1 in the Gaussian mixture model corresponding to each leaf node, calculate their Mahalanobis distance. where μA1 and μB1 are the mean vectors of A1 and B1 respectively, and are the covariance matrices of A1 and B1 respectively; if the Mahalanobis distance is less than the overlap threshold, it is considered that A1 and B1 overlap, and record their overlap relationship; repeat the pairwise calculation of the Mahalanobis distance for all sub-habit patterns under the same habit pattern to obtain all the overlap relationships within the habit pattern; for different habit patterns, calculate whether there is an overlap relationship between them.

[0114] Based on all the obtained overlap relationships, construct an overlap graph, with each sub-habit pattern as a node, and if two sub-habit patterns overlap, connect an edge in the overlap graph; find all the maximum cliques in the overlap graph, and each maximum clique corresponds to an overlapping cross pattern, that is, the intersection area of multiple sub-habit patterns; for each overlapping cross pattern, calculate the joint probability distribution of all the sub-habit patterns it contains as the probability model of the overlapping cross pattern.

[0115] The evaluation methods of the health level include:

[0116] Collect historical data (the comprehensive living data of the elderly in the past period that has been collected), mine the association rules between the living patterns and the health status from the historical data, formalize the association rules, and construct a health assessment knowledge base for the living patterns;

[0117] According to the health assessment knowledge base, set the initial health scores for different living patterns, that is, the health level; for example, the regular work and rest pattern gets a higher score, and the sedentary and less active pattern gets a lower score (quantified);

[0118] For each overlapping cross pattern, calculate its association strength with other overlapping cross patterns. The association strength can be represented by joint probability, conditional probability or other correlation measures; construct a pattern association network, where the nodes in the pattern association network are overlapping cross patterns, and the nodes are connected by edges and assigned weights; the weight of the edge is the association strength.

[0119] Traverse each node (pattern) in the pattern association network, and extract the association strength between the nodes (neighboring patterns) connected to it; it should be noted that there is more than one connected node, and there may be multiple; calculate the adjusted score PY = HG + ∑(GY × RT); where HG is the initial health score, GY is the degree of the node connected to the calculated node (the number of nodes connected to it); RT is the association strength between the corresponding node and the calculated node; through network propagation, the association strength of the neighboring patterns affects the score of the current living pattern, and the greater the association strength, the greater the impact on the score of the corresponding pattern.

[0120] Map the adjusted scores to semantic health levels, such as good, medium, poor, etc.; or perform mapping to obtain a curve; that is the health assessment result; generate a personalized health assessment report and improvement suggestions for the elderly; collect the feedback from the elderly to understand the effectiveness of the assessment result, and continuously optimize the health assessment knowledge base and association rules according to the feedback, and re-evaluate regularly to dynamically update the health assessment result.

[0121] Integrates domain knowledge, data-driven association analysis, and network propagation mechanism, can comprehensively consider the complex associations between lifestyle patterns, dynamically adjust health assessment, improve the accuracy and personalization of the assessment; and has good interpretability and sustainable optimization ability at the same time.

[0122] This embodiment can comprehensively and accurately reflect the real living conditions of the elderly, breaking through the one-sided and subjective limitations of traditional assessment methods; through the comprehensive collection and intelligent processing of multi-source heterogeneous data, the system constructs a large dataset containing multi-dimensional information such as physiology, behavior, and environment, providing a sufficient data basis for the assessment and ensuring the objectivity and comprehensiveness of the assessment result; secondly, it has the ability to automatically discover lifestyle patterns, can deeply explore the living rules hidden in the massive data, and provide refined pattern support for health assessment; the system uses algorithms to dynamically identify the lifestyle patterns of the elderly from multi-dimensional data, hierarchically divides the patterns, and refines them to the granularity of habit patterns, fully exploring the internal structure and detailed information of the lifestyle patterns, providing rich pattern knowledge for the assessment; furthermore, the health assessment method has a high degree of accuracy and personalization; a pattern association network is constructed, which can comprehensively consider the complex associations between lifestyle patterns, dynamically adjust the assessment result, and provide personalized improvement suggestions for the elderly's living habits, helping to improve the quality of life of the elderly.

[0123] Embodiment 2

[0124] Please refer to Figure 2 As shown, the parts not described in detail in this embodiment refer to the description content of Embodiment 1. Provide an elderly lifestyle assessment method based on big data analysis, including:

[0125] Step 1: Receive the comprehensive living data of the elderly collected by wearable devices, environmental sensors, and smart home devices, where the comprehensive living data includes physiological data, behavioral data, and environmental data; construct a multi-dimensional large dataset.

[0126] Step 2: Perform comprehensive processing on the obtained multi-dimensional large dataset and form a unified data format to obtain a unified living dataset.

[0127] Step 3: Dynamically identify the lifestyle patterns of the elderly from the unified living dataset using the improved cellular automata algorithm.

[0128] Step 4: Construct a hybrid tree model, divide the living patterns of the elderly into several levels, and each level contains several habitual patterns; calculate different overlapping and intersecting patterns in the habitual patterns;

[0129] Step 5: Evaluate the health level of the elderly's living patterns; for different overlapping and intersecting patterns, obtain their correlation strengths, and dynamically adjust the evaluated health level based on the correlation strengths to obtain a health assessment result; output the health assessment result to the evaluation terminal.

[0130] Embodiment 3

[0131] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-provided method for evaluating the living habits of the elderly based on big data analysis.

[0132] Since the electronic device introduced in this embodiment is the electronic device used to implement the method for evaluating the living habits of the elderly based on big data analysis in the embodiments of the present application, based on the method for evaluating the living habits of the elderly based on big data analysis introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various forms of change of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the method for evaluating the living habits of the elderly based on big data analysis in the embodiments of the present application, it falls within the scope of protection of the present application.

[0133] The above formulas are all dimensionless and take their numerical values for calculation. The formula is a formula obtained by collecting a large amount of data for software simulation to obtain the closest real situation. The preset parameters and threshold selection in the formula are set by those skilled in the art according to the actual situation.

[0134] The above is only the preferred implementation manner of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. An elderly living habit assessment system based on big data analysis, characterized in that, Including: A data integration module, which is used to receive the comprehensive life data of the elderly collected by wearable devices, environmental sensors, and smart home devices. The comprehensive life data includes physiological data, behavioral data, and environmental data; and construct a multi-dimensional big data set; A unified processing module, which is used to comprehensively process the obtained multi-dimensional big data set, and form a unified data format to obtain a unified life data set; A dynamic pattern recognition module, which is used to dynamically identify the life patterns of the elderly from the unified life data set by using an improved cellular automata algorithm; A pattern division module, which is used to construct a hybrid tree model, divide the life patterns of the elderly into several levels, and each level contains several habitual patterns; calculate different overlapping and intersecting patterns in the habitual patterns; A comprehensive evaluation module, which is used to evaluate the health status of the elderly's life patterns; for different overlapping and intersecting patterns, obtain their association strengths, and dynamically adjust the evaluated health status based on the association strengths to obtain a health evaluation result; output the health evaluation result to an evaluation terminal; each module is connected by wired and / or wireless means to realize data transmission between modules.

2. The elderly living habit assessment system based on big data analysis according to claim 1, wherein The method of the comprehensive processing includes: In the multi-dimensional big data set, for each data dimension, calculate its interquartile range IQR = Q3 - Q1; where Q3 is the upper quartile and Q1 is the lower quartile; determine the upper boundary value S1 and the lower boundary value S2 based on the interquartile range; Mark the data points in the corresponding data dimension that exceed the upper boundary value or the lower boundary value as possible outliers; Divide the multi-dimensional big data set into several sub-data sets according to time or other identifiers; For each data dimension in each sub-data set, extract the values of all data points in the corresponding data dimension, and sort these values from smallest to largest to obtain an ordered array; construct a retrieval structure based on the ordered array; Set a radius r. For each data dimension i, use the constructed retrieval structure to query the neighborhood range of the data point P in the corresponding data dimension. For each data dimension i, if the number of data points in the neighborhood of the data point P in data dimension i is 0, set its own distance to infinity; otherwise, set its own distance where m is the number of dimensions of the data points; δ is the influence parameter, en(i) is the information entropy of data dimension i; take the L2 norm of the self-distance and normalize it to obtain the anomaly score of the data point P, and set an anomaly score threshold. If the anomaly score of the data point P is greater than the anomaly score threshold, mark the data point P as a hidden outlier. If the data point P is marked as a potential outlier and a hidden outlier at the same time, then take the data point P as an outlier; For each data dimension, calculate its Spearman-Pan coefficient rho; sort the data dimensions in descending order according to the value of rho, and select the top N data dimensions as reference dimensions; for the outliers in each sub-data set, use the reference dimensions to replace the data points through K-nearest neighbor regression or decision tree regression; and then re-integrate to obtain a unified life data set.

3. The elderly living habit assessment system based on big data analysis according to claim 2, characterized in that The method of constructing the retrieval structure includes: Set the data structure of the retrieval node, and the data structure includes a pointer to the left subtree, a pointer to the right subtree, the value range corresponding to the retrieval node, and the number of data points within the value range; For any retrieval node, determine the value range it needs to cover; if the upper and lower limits of the value range are equal, create a new node and set the number of data points within the corresponding value range to 1; If the upper and lower limits of the value range are not equal, continue to split the value range. Specifically, calculate the middle value mid, create a new node ne, and set the value range of the new node ne to the original value range; recursively construct the left subtree of the new node ne, and define the covered value range as the interval formed by the upper limit of the original value range and the middle value mid; recursively construct the right subtree of the new node ne, and define the covered value range as the interval formed by mid + 1 and the lower limit of the original value range; through the recursive process, construct a complete retrieval structure.

4. The elderly living habit assessment system based on big data analysis according to claim 3, characterized in that The recognition method of the described living pattern includes: Dividing the unified living data set into time windows of a fixed length according to the time series; constructing a cell-like model of the elderly's living pattern, and defining the cell state of the cells in the cell-like model; combining the data points within each time window into a cell, and the cell state is represented by a data vector composed of data points of different data types within the time window; defining the involved range of each cell, and taking the data vector of the first time window as the initial cell state; defining a state transition function, for each cell, according to its current cell state and the cell states of the cells within the involved range, applying the state transition rule to calculate the cell state of itself at the next time step; simultaneously updating the states of all cells; repeating until the preset number of iterations is reached; comprehensively clustering the cell states after iteration to obtain state groups; analyzing the characteristics of each state group and extracting representative living patterns.

5. The elderly living habit assessment system based on big data analysis according to claim 4, characterized in that, The method of defining the state transition function includes: Defining a set of state transition rules R, taking the cell state S(j,t) of cell j at time t, all cell states within its involved range, and the state transition rules R as inputs, and calculating the new cell state S(j,t + 1) of cell j at time t + 1 = FP(S(j,t),{S(j′,t)|j'∈N(j)},R); where, S(j′,t) is the cell state of cell j' within the involved range of cell j; FP() is the state transition function.

6. The elderly living habit evaluation system based on big data analysis according to claim 5, characterized in that, The method of performing comprehensive clustering includes: Defining the clustering input dimension as the dimension of the data vector corresponding to the cell state; setting a mapping network and setting the grid size of the mapping network; initializing the weight vector of each neuron in the mapping network; For the data vector x of each cell state, calculating the Euclidean distance between x and the weight vectors of all neurons; finding the neuron c with the smallest Euclidean distance from x; Define a neighborhood function where \(r_{cy}\) is the Euclidean distance between the position vectors of neuron \(y\) and \(c\) on the grid; \(\sigma(l)\) is the neighborhood radius at the current iteration \(l\); For each neuron y, updating its weight vector, and the update formula of the weight vector is: where \(w_y(l + 1)\) is the weight vector of neuron \(y\) at iteration \(l + 1\), and \(w_y(l)\) is the weight vector of neuron \(y\) at the current iteration \(l\). is the learning rate at the current iteration \(l\); Repeating the update until the preset number of iterations is reached; setting a sliding time window, regularly updating the cell-like model, and clustering the neurons of the mapping network; taking the neurons corresponding to each cluster obtained by clustering and the data vectors of the cell states they map as a state group.

7. The elderly living habit assessment system based on big data analysis according to claim 6, characterized in that, The method of dividing the living pattern of the elderly into several levels includes: Construct a tree structure. According to the extracted life patterns, take all life patterns as the root nodes of the tree structure. For each life pattern, use the hierarchical clustering algorithm to divide it into sub-patterns, and take the sub-patterns as the child nodes of the original life pattern to construct an initial hierarchical pattern tree. For each node in the hierarchical pattern tree, calculate its information gain, and select the top N1 features with the largest information gain as the key feature subset corresponding to the node. For each internal node in the hierarchical pattern tree, use the key feature subset to construct a hybrid tree model. Each leaf node of the hybrid tree model corresponds to a sub-pattern, and insert these sub-patterns as the child nodes of the current node into the hybrid tree model. For the leaf nodes of the hybrid tree model, use the key feature subset to construct a Gaussian mixture model, and each mixture component of the Gaussian mixture model corresponds to a habit pattern. For each child node of each leaf node, calculate the Bayesian information criterion value of the Gaussian mixture model and set a merging threshold. For all child nodes of the current leaf node, calculate their pairwise similarities. Find two child nodes A and B with the largest probability distribution similarity, and merge them into a new child node C. Calculate the Bayesian information criterion value of the Gaussian mixture model corresponding to the new child node C, denoted as the merging value. If the merging value is less than the Bayesian information criterion value of the child nodes before merging minus the merging threshold, accept the merge and replace A and B with C as the new child nodes; otherwise, do not merge.

8. The elderly living habit assessment system based on big data analysis according to claim 7, characterized in that, The calculation method of the overlapping cross pattern includes: For each leaf node in the hybrid tree model, extract the mean vector and covariance matrix of each mixture component in its Gaussian mixture model. Define an overlapping threshold. For any two mixture components A1 and B1 in the Gaussian mixture model corresponding to each leaf node, calculate their Mahalanobis distance U(A1, B1). If the Mahalanobis distance is less than the overlapping threshold, it is considered that A1 and B1 overlap. Repeat calculating the Mahalanobis distance pairwise for all sub-habit patterns under the same habit pattern to obtain all overlapping relationships within the habit pattern. For different habit patterns, calculate whether there are overlapping relationships between them. Based on all the obtained overlapping relationships, construct an overlapping graph, take each sub-habit pattern as a node, and if two sub-habit patterns overlap, connect an edge in the overlapping graph. Find all the maximum cliques in the overlapping graph, and each maximum clique corresponds to an overlapping cross pattern.

9. The elderly living habit assessment system based on big data analysis according to claim 8, characterized in that, Collect historical data, mine the association rules between life patterns and health conditions from the historical data, formalize the association rules, and construct a health assessment knowledge base oriented to life patterns. According to the health assessment knowledge base, set an initial health score for different life patterns, that is, the degree of health. For each overlapping cross pattern, calculate its association strength with other overlapping cross patterns. Construct a pattern association network, where the nodes in the pattern association network are overlapping cross patterns, and the nodes are connected by edges and assigned weights. The weight of the edge is the association strength. Traverse each node in the pattern association network and extract the association strengths between the nodes connected to it. Calculate the adjusted score PY = HG + ∑(GY × RT); where HG is the initial health score, GY is the degree of the node connected to the calculated node; RT is the association strength between the corresponding node and the calculated node. Map the adjusted score to the semantic health level, or obtain a curve through mapping; this is the health assessment result.

10. The method for evaluating the living habits of the elderly based on big data analysis is implemented based on the system for evaluating the living habits of the elderly based on big data analysis according to any one of claims 1 to 9, and is characterized in that, It includes: Step 1: Receive the comprehensive life data of the elderly collected by wearable devices, environmental sensors, and smart home devices. The comprehensive life data includes physiological data, behavioral data, and environmental data; construct a multi-dimensional big data set. Step 2: Conduct comprehensive processing on the obtained multi-dimensional big data set and form a unified data format to obtain a unified life data set. Step 3: Dynamically identify the living patterns of the elderly from the unified life data set using the improved cellular automata algorithm. Step 4: Construct a hybrid tree model, divide the living patterns of the elderly into several levels, and each level contains several habitual patterns; calculate different overlapping and cross patterns in the habitual patterns. Step 5: Evaluate the health degree of the elderly's living patterns; for different overlapping and cross patterns, obtain their association strength, and dynamically adjust the evaluated health degree based on the association strength to obtain the health assessment result; output the health assessment result to the evaluation terminal.

Images