Smart Elderly Care Service Management Method Based on Deep Information Integration

Abstract

The present invention relates to the technical field of intelligent elderly care, and further relates to an intelligent elderly care service management method based on in-depth information integration. The method includes: Step 1: Collect the health data of each elderly person through various sensors; take the health data of the same elderly person collected at the same moment as a sample, and for each sample, extract the features of the sample; form all the samples into a sample data set; form the features of all the samples into a feature data set; Step 2: Perform data fusion on the sample data set to obtain a fused data set; Step 3: Perform matrix feature analysis on each element in the time series to screen out the elderly people with health problems and send information to the elderly people to prompt them that there are health problems. The present invention performs health data fusion through a multi-layer Boltzmann machine, provides continuous monitoring, personalized services, and timely problem warnings, so as to improve the scientific nature and effectiveness of intelligent elderly care services.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent elderly care, and particularly relates to an intelligent elderly care service management method based on deep information integration. Background Art

[0002] With the continuous aggravation of social aging, the health concerns and elderly care service needs of the elderly population have become increasingly prominent. In order to better meet the health management and living needs of the elderly, various intelligent elderly care service management methods and systems have gradually emerged. These methods and systems aim to provide more comprehensive, personalized and effective health monitoring and elderly care services for the elderly through technical means such as information technology, sensors, data analysis and artificial intelligence.

[0003] Before the rise of intelligent elderly care service methods, traditional elderly health monitoring methods mainly relied on regular medical examinations and self-reports of the elderly. These methods have the following limitations: discontinuity and untimely: traditional examinations are usually carried out regularly and cannot provide continuous health data monitoring. Once the elderly have health problems, they may need to wait for the next appointment to be processed, which may cause serious consequences. Subjectivity: The self-reporting method is easily affected by the subjective consciousness and memory of the elderly, and there may be problems of inaccurate or incomplete information. Resource waste: Regular medical examinations require a large amount of medical resources, including time and manpower, and problems may not be found every time.

[0004] In recent years, the progress of sensor technology has provided new opportunities for the health monitoring of the elderly. For example, the widespread application of wearable devices (such as smart watches) and health sensors (such as sphygmomanometers, heart rate monitors) can achieve continuous collection of physiological data. However, these technologies also have some problems: rich data but difficult to manage: the data generated by sensors is huge, and effective data management and analysis methods are needed to extract useful information from it. Isolated data sources: Multiple sensors and devices may work independently, making it difficult to integrate and comprehensively analyze information. Summary of the Invention

[0005] The main purpose of the present invention is to provide an intelligent elderly care service management method based on deep information integration, which performs health data fusion through a multi-layer Boltzmann machine, provides continuous monitoring, personalized services and timely problem warnings, so as to improve the scientificity and effectiveness of intelligent elderly care services.

[0006] In order to solve the above problems, the technical solution of the present invention is realized as follows:

[0007] An intelligent elderly care service management method based on deep information integration, the method includes:

[0008] Step 1: Collect the health data of each elderly person through various sensors; Take the health data of the same elderly person collected at the same moment as a sample. For each sample, extract the features of the sample; Compose all the samples into a sample data set; Compose the features of all the samples into a feature data set;

[0009] Step 2: Perform data fusion on the sample data set to obtain a fused data set, specifically including: Initialize a multi-layer Boltzmann machine, initialize the connection weights and biases; Input the sample data set as input data into the multi-layer Boltzmann machine; Define the state of the multi-layer Boltzmann machine as the state of the entire network, including the states of the visible layer and multiple hidden layers; Use the Metropolis-Hastings sampling method to perform Gibbs sampling to obtain the sampling results to capture the state of the multi-layer Boltzmann machine; Based on the sampling results, calculate the gradient of the energy function with respect to all parameters; Adopt an adaptive learning rate strategy to update the parameters; Jointly model the hidden state representation of the multi-layer Boltzmann machine and the sample data set to obtain the fused data set;

[0010] Step 3: Take the fused data sets corresponding to the sample data sets at all moments within a time period and compose them into a time series in chronological order. Each element in the time series is a fused data set. Perform sequence feature analysis on the time series to determine whether there are any problems with the health of the elderly during this time period; If so, then perform matrix feature analysis on each element in the time series to screen out the elderly with health problems and send information to the elderly to prompt them that their health has problems.

[0011] Further, the health data at least includes: blood pressure, heart rate, blood sugar, body temperature, respiratory rate, blood oxygen saturation, and intraocular pressure.

[0012] Further, after collecting the health data of each elderly person through various sensors in Step 1, data preprocessing will also be performed on the collected health data, specifically including: performing missing value processing, outlier processing, duplicate value processing, and normalization processing on the collected health data in sequence.

[0013] Further, the energy function of the multi-layer Boltzmann machine in Step 2 is expressed using the following formula:

[0014] ;

[0015] where, is the sample data set, is the state vector of the visible layer in the multi-layer Boltzmann machine, containing M visible units, representing the features of the sample data set, and each visible unit represents the value of a feature; is the state vector of the first hidden layer, containing N hidden units; each Represents the state of a neuron in the first hidden layer; are the state vectors of the additional hidden layers, each represents the state of the neurons in the kth hidden layer; To connect the visible layer visible units and the first hidden layer The connection weights between neurons; To connect hidden layers and The connection weights between hidden layers; is the bias of the visible layer, which is used to adjust the visible layer The activation threshold of visible units; is the bias of the first hidden layer, which is used to adjust the first hidden layer The activation threshold of each neuron; is the energy function.

[0016] Furthermore, in step 2, the state of the multilayer Boltzmann machine is defined as the state of the entire network, thereby obtaining a multilayer Boltzmann state machine; the multilayer Boltzmann state machine is determined by the following formula:

[0017] ;

[0018] ;

[0019] in, is the state activation probability from the visible layer to the first hidden layer;

[0020] here, represents the first hidden layer The state of the neurons, when the visible layer state is given The probability that it is activated when The first hidden layer The state of a neuron is 0 or 1, which represents the activation state or inactivation state of the neuron respectively; is the first hidden layer The bias of each neuron; For the hidden layers to The state activation probability of hidden layers; among them, ; Indicates the The hidden layer The state of the neurons, given the state of the previous hidden layer The probability that it is activated when Indicates the connection The hidden layer the connection weight between the th neuron of the th neuron and the th neuron of the th hidden layer; the bias of the

[0021] Furthermore, when initializing the connection weights, initialize the connection weights as random values sampled from a Gaussian distribution; when initializing the bias of the hidden layer, initialize the bias to zero.

[0022] Furthermore, the process of using the Metropolis-Hastings sampling method to perform Gibbs sampling in step 2 includes:

[0023] Suppose sampling from the conditional distribution is achieved by using the random walk method and randomly changing one or more elements of ; next, calculate the acceptance probability for determining whether to accept the candidate state, and the formula is as follows:

[0024] ;

[0025] where is the conditional distribution of the current state, is the conditional distribution of the candidate state, is the transition probability from the current state to the candidate state, is the transition probability from the candidate state to the current state; according to the acceptance probability , accept the candidate state with a certain probability. If , then accept the candidate state; otherwise, accept the candidate state with a probability of If the candidate state is not accepted, keep the current state unchanged; repeat the above steps until the set number of times is reached; the final sampling result is:

[0026] .

[0027] Furthermore, the method for calculating the gradient of the energy function with respect to all parameters based on the sampling result includes: for the gradient calculation of the connection weight , use the following formula for calculation:

[0028] ;

[0029] where represents the expectation when the sample data set is used as the input of the Boltzmann machine, represents the expectation when the feature data set is used as the input of the Boltzmann machine; for the connection weight Gradient calculation of , which is calculated using the following formula:

[0030] ;

[0031] For the visible layer bias Gradient calculation is performed using the following formula:

[0032] ;

[0033] For the hidden layer bias Gradient calculation of , which is calculated using the following formula:

[0034] .

[0035] Furthermore, the method of using an adaptive learning rate strategy to update parameters in step 3 includes: Let the parameter to be updated be , where can be the connection weight , visible layer bias or hidden layer bias ; Initialize the learning rate to a positive value less than 0.1, the first moment estimate , the second moment estimate , the time step , the first exponential decay rate to the first set value, the second exponential decay rate to the second set value; In each training iteration, calculate the gradient of the parameter , increase the time step ; Use the following formula to update the first moment estimate and the second moment estimate:

[0036] ;

[0037] ;

[0038] Since in the initial iteration, and estimates tend to be zero, so bias correction is needed to reduce the bias of the estimates. The corrected first moment estimate and second moment estimate are:

[0039] ;

[0040] ;

[0041] Then use the following formula to update the parameter :

[0042] ;

[0043] where, is a positive constant used to avoid the denominator being zero.

[0044] Furthermore, the method for performing sequence feature analysis on the time series in step 3 to determine whether there are health problems of the elderly during this time period includes: calculating the value of the determinant of the covariance matrix of the time series and comparing it with a set determination threshold. If it exceeds the set determination threshold, it is determined that there are health problems of the elderly; the method for performing matrix feature analysis on each element in the time series to screen out the elderly with health problems includes: calculating the Z-score of each element in the matrix corresponding to each element in the time series, and taking the element with the Z-score exceeding the set score value as an abnormal element, and the elderly corresponding to this abnormal element is the elderly with health problems.

[0045] The intelligent elderly care service management method based on information deep integration of the present invention has the following beneficial effects: The multi-layer Boltzmann machine of the present invention performs deep learning and model training on various collected health data, and fuses these data together to obtain a more comprehensive and complex data representation. This data fusion process makes full use of the advantages of multi-layer neural networks, can discover the potential correlations and features between data, thereby improving the scientificity and accuracy of data. The data fusion of the multi-layer Boltzmann machine also promotes data integration and comprehensive analysis. Modern health management needs to process a large amount of health data, including data from different sensors and devices. The multi-layer Boltzmann machine can seamlessly integrate these multi-source data together for comprehensive analysis. This helps medical professionals better understand the health status of the elderly and make more informed medical decisions. The data fusion of the multi-layer Boltzmann machine also helps in timely problem identification and warning. By deeply analyzing the fused data, the system can detect the health problems and risk trends of the elderly. Once potential problems are identified, the system can immediately issue an alarm or notify the elderly, their families or medical professionals so as to take necessary measures. This timely problem identification and warning helps reduce the occurrence of emergencies and improve the quality of life of the elderly. The present invention collects a variety of health data through various sensors, including indicators such as blood pressure, heart rate, blood sugar, body temperature, respiratory rate, blood oxygen saturation and intraocular pressure, and these data reflect the physiological state of the elderly. Different from traditional regular examinations, the present invention can monitor the health status of the elderly in real time, capture the immediate information of health changes, and discover potential problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic flowchart of the method for the intelligent elderly care service management method based on information deep integration provided by the embodiment of the present invention. Detailed implementation manners

[0047] To enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with 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 of 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 scope of protection of the present invention.

[0048] Embodiment 1: Refer to Figure 1 , a smart elderly care service management method based on in-depth information integration, the method comprising:

[0049] Step 1: Collect the health data of each elderly person through various sensors; take the health data of the same elderly person collected at the same moment as a sample, and for each sample, extract the features of the sample; form a sample data set with all the samples; form a feature data set with the features of all the samples;

[0050] Various sensors are used to monitor the health data of the elderly in real time, and such data may include but are not limited to heart rate, body temperature, blood pressure, physical activity, blood sugar, etc. The data collected by the sensors is digital and is sampled according to a set time interval or event trigger. Extracting features from each sample is to transform the original health data into a form that can be used for subsequent deep learning and data analysis. Feature extraction can adopt various methods, including statistical features (such as mean, standard deviation), frequency domain analysis (such as Fourier transform), time domain analysis (such as autocorrelation function), etc. These features describe the key aspects of the data, such as the distribution, periodicity, trend, etc. of the data. By integrating the data collected by different sensors into a sample data set, multiple health indicators can be considered comprehensively to form a comprehensive health data set. The advantage of doing so is that the health status of the elderly can be understood more comprehensively, rather than just focusing on a single indicator.

[0051] Step 2: Perform data fusion on the sample data set to obtain a fused data set, specifically including: initializing a multi-layer Boltzmann machine, initializing connection weights and biases; inputting the sample data set as input data into the multi-layer Boltzmann machine; defining the state of the multi-layer Boltzmann machine as the state of the entire network, including the states of the visible layer and multiple hidden layers; using the Metropolis-Hastings sampling method to perform Gibbs sampling to obtain sampling results to capture the state of the multi-layer Boltzmann machine; calculating the gradients of the energy function with respect to all parameters based on the sampling results; adopting an adaptive learning rate strategy to update the parameters; jointly modeling the hidden state representation of the multi-layer Boltzmann machine and the sample data set to obtain a fused data set;

[0052] The multi-layer Boltzmann machine (Multi-Layer Boltzmann Machine) is a deep learning model composed of multiple layers, with each layer containing multiple neurons (nodes). These neurons are connected together through connection weights and bias terms. The multi-layer Boltzmann machine is based on the principle of Boltzmann distribution and is used to learn complex non-linear relationships between data. Metropolis-Hastings sampling and Gibbs sampling are Markov chain Monte Carlo methods used to sample from complex distributions. They are used to simulate the state of the multi-layer Boltzmann machine, including the visible layer and multiple hidden layers. To update the parameters of the model, it is necessary to calculate the gradients of the energy function with respect to all parameters. The adaptive learning rate strategy is a method for dynamically adjusting the size of the learning rate to better converge to the optimal solution during training.

[0053] The multi - layer Boltzmann machine is used for data fusion, integrating the sample data set from Step 1 with its feature representation. This fusion process helps to establish a higher - level abstract representation, capturing the complex relationships and patterns in the data. This contributes to enhancing the data's representational ability, thus better understanding the health status of the elderly. The multi - layer Boltzmann machine represents the associations between data by learning connection weights and bias terms. This helps the model to automatically learn and extract features without the need for manually designed feature engineering. This automatic feature learning can better adapt to the non - linear relationships in the data, improving the model's performance. Metropolis - Hastings sampling and Gibbs sampling methods are used to train the multi - layer Boltzmann machine. They enable the model to capture the high - order relationships in the data distribution, thereby enhancing the model's representational ability. The adaptive learning rate strategy helps to effectively update the model parameters, ensuring that the model gradually converges to the optimal solution during training. The multi - layer Boltzmann machine allows the integration of multi - source data from different sensors. This is one of its most crucial functions. The health data of the elderly usually comes from multiple sensors, such as heart rate monitoring, body temperature detection, and movement activity monitoring. Through the multi - layer Boltzmann machine, these heterogeneous data can be organically combined to form a comprehensive data representation, more comprehensively reflecting the overall health status of the elderly. This integration of multi - source data is crucial for accurately monitoring and diagnosing the health problems of the elderly and is one of the key innovations of this method.

[0054] The multi - layer Boltzmann machine learns the high - order relationships of data through multiple hidden layers. This means that it can capture more complex patterns and inter - relationships in the data, rather than just simple linear relationships. This ability is very important for identifying early signs or potential patterns of health problems, as many health problems have complex multi - factor influences. The multi - layer Boltzmann machine automatically learns and extracts features of the data, which are very valuable for subsequent analysis and diagnosis. Through deep learning, it can form multi - level feature representations, from raw data to more high - level abstract features. These abstract features can better describe the health status of the elderly, providing richer information for further data analysis.

[0055] Step 3: Take the fusion data sets corresponding to the sample data sets at all moments within a period of time and form a time series in chronological order. Each element in the time series is a fusion data set. Conduct sequence feature analysis on the time series to determine whether there are health problems among the elderly within this period. If so, then conduct matrix feature analysis on each element in the time series to screen out the elderly with health problems and send information to the elderly to prompt them that their health is problematic.

[0056] The main function of this step is to monitor the health problems of the elderly. Through the analysis of time series, it is possible to detect whether there are abnormalities or trend changes in the health status of the elderly. This makes it possible to detect health problems at an early stage, so as to take appropriate intervention measures. If the time series analysis reveals health problems of the elderly, an alarm or notification can be generated to inform medical staff or the elderly themselves. This helps to achieve personalized alarms and reminders, pay timely attention to potential problems, and provide timely medical intervention or advice. If potential problems are detected, further matrix feature analysis can help determine which elderly people may be facing health problems. This can improve the accuracy of diagnosis and help distinguish which elderly people need more urgent medical attention. Through time series analysis, data from multiple time points can be comprehensively considered to form a comprehensive assessment of the health status of the elderly. This helps elderly care service providers better understand the overall health status of the elderly in order to develop personalized care plans and health management strategies.

[0057] In this step, a time series is formed by combining the fusion data sets corresponding to the sample data sets at all moments within a period of time. A time series is a collection of data arranged in chronological order, and each element is a fusion data set. The principles of time series analysis include using statistical and data mining techniques to study the changing trends, periodicity, correlation, etc. of data over time. Each element in the time series is a fusion data set, and feature analysis can be performed on the entire time series to understand whether there are problems with the health status of the elderly during this period. This includes identifying abnormal patterns, trend changes or emergencies. Sequence feature analysis can use techniques such as statistical methods and machine learning algorithms to find abnormalities. If potential health problems are found in the time series, further matrix feature analysis is performed on each time series element (fusion data set). This analysis process can include using linear algebra and statistical methods to explore the patterns, relationships and abnormalities within the data set. This helps to more accurately determine which elderly people may be facing health problems.

[0058] Embodiment 2: The health data at least includes: blood pressure, heart rate, blood sugar, body temperature, respiratory rate, blood oxygen saturation and intraocular pressure.

[0059] Specifically, blood pressure is the pressure exerted by the heart to pump blood into the arteries, usually expressed in millimeters of mercury (mmHg). Monitoring blood pressure can help identify cardiovascular problems such as hypertension and hypotension. By recording blood pressure values, timely intervention can be carried out to maintain a healthy blood pressure level. Heart rate is the number of times the heart beats per minute, usually expressed as the number of pulses per minute. Heart rate monitoring is crucial for assessing heart health and detecting irregular heartbeats or arrhythmias. The normal heart rate range varies depending on age and health status. Blood glucose is the concentration of glucose in the blood, usually expressed in millimoles per liter (mmol / L) or milligrams per deciliter (mg / dL). Monitoring blood glucose levels is very important for diabetes patients' management and can help adjust insulin treatment and diet. Body temperature represents the temperature of the body, usually expressed in degrees Celsius (°C) or degrees Fahrenheit (°F). Abnormal body temperature may be an indication of fever or hypothermia and may be related to infections or other health problems. Respiratory rate represents the number of breaths per minute. The normal respiratory rate of an adult is usually between 12 and 20 breaths per minute. Monitoring the respiratory rate helps evaluate the function of the respiratory system. Blood oxygen saturation is the saturation of oxygen in the blood, usually expressed as a percentage (%). Monitoring blood oxygen saturation can help detect respiratory problems and hypoxemia. Intraocular pressure is the pressure inside the eyeball, usually expressed in millimeters of mercury (mmHg). Intraocular pressure monitoring is very important for the diagnosis and management of eye diseases such as glaucoma. The monitoring and analysis of these comprehensive health data can provide a comprehensive assessment of the health status, helping medical staff and elderly care service providers better understand the physiological conditions of the elderly. At the same time, time series analysis and deep learning of these data can better monitor and diagnose the health problems of the elderly, provide personalized elderly care services, and contribute to improving the quality of life and health level of the elderly.

[0060] Example 3: After collecting the health data of each elderly person through various sensors in Step 1, data preprocessing will also be performed on the collected health data, specifically including: performing missing value processing, outlier processing, duplicate value processing, and normalization processing on the collected health data in sequence.

[0061] Specifically, missing values refer to data points or information that were not obtained during the collection process. The health data of the elderly may have missing values due to various reasons, such as sensor failures, non-cooperation of the elderly, equipment damage, etc. In dealing with missing values, appropriate methods need to be adopted to fill or process the missing values. Common methods include mean filling, interpolation, deleting missing values, or filling according to the time series pattern of the data. Outliers refer to data points that are significantly different from other data points or deviate from the normal range. The health data of the elderly may contain outliers, which may be caused by sensor errors, measurement errors, or health problems. Outlier processing aims to detect and correct these outliers to avoid adverse effects on subsequent analysis and models. The processing methods include truncation, smoothing, or replacing outliers. Duplicate values refer to exactly the same data points that appear during the collection process. These duplicate values may interfere with data analysis because they do not provide additional information. In dealing with duplicate values, it is necessary to detect and delete or merge duplicate values to ensure the uniqueness and consistency of the data. Health data usually involves different measurement units and ranges. To perform effective data analysis, it is necessary to standardize this data to a unified scale. Standardization usually includes mean normalization and standard deviation normalization of the data to ensure that different features have similar scales, thus preventing certain features from dominating in the analysis.

[0062] Embodiment 4: The energy function of the multi-layer Boltzmann machine in step 2 is represented by the following formula:

[0063] ;

[0064] where, is the sample data set, is the state vector of the visible layer in the multi-layer Boltzmann machine, containing M visible units, representing the features of the sample data set, and each visible unit represents the value of a feature; is the state vector of the first hidden layer, containing N hidden units; each represents the state of a neuron in the first hidden layer; are all state vectors of additional hidden layers, and each represents the state of a neuron in the k-th hidden layer; is the connection weight between the th visible unit of the visible layer and the th neuron of the first hidden layer; is the connection weight between the th hidden layer and the th hidden layer; is the bias of the visible layer, used to adjust the activation threshold of the th visible unit of the visible layer; is the bias of the first hidden layer, used to adjust the activation threshold of the th neuron in the first hidden layer; is the energy function.

[0065] Specifically, the connection term between the visible layer and the first hidden layer ( ): This part represents the connection weights between each visible unit in the visible layer and each neuron in the first hidden layer and their interactions. It reflects the information transfer and mutual influence between the visible layer and the first hidden layer. These connection weights will be adjusted during model training to learn the correlation relationships between data. The connection terms between each hidden layer ( ): This part represents the connection weights and their interactions between every two adjacent hidden layers. It reflects the information transfer and mutual influence between different hidden layers. These connection weights will also be adjusted during model training to capture the high-order correlation relationships of data. The bias term of the visible layer ( ): This part contains the bias terms of each visible unit in the visible layer . The bias term is used to adjust the activation threshold of visible units and affect whether they will be activated. The existence of the bias term enables the model to adapt to different data distributions and features. The bias term of the first hidden layer ( ): This part contains the bias terms of each neuron in the first hidden layer . The bias term is used to adjust the activation threshold of neurons in the first hidden layer and affect whether they will be activated. The existence of the bias term enables the model to adapt to different data distributions and features.

[0066] The energy function of the multi-layer Boltzmann machine is responsible for capturing high-order relationships and features in the data. In the management of intelligent elderly care services, the health data of the elderly comes from multiple sensors, such as blood pressure, heart rate, blood sugar, etc. These data may be very complex in their original form, but the energy function of the multi-layer Boltzmann machine can automatically extract key features in the data by learning connection weights and bias terms, achieving data dimensionality reduction and abstract representation. This helps to integrate different types of health data into a consistent representation, providing a better data foundation for subsequent intelligent elderly care services. Through the optimization process of the energy function, the multi-layer Boltzmann machine can learn the patterns and correlation relationships in the data. This is crucial for monitoring and diagnosing the health problems of the elderly. In the management of intelligent elderly care services, being able to detect potential health problems early is a key task. The energy function of the multi-layer Boltzmann machine helps the model identify abnormal patterns and trend changes, so that it can issue alarms more timely and provide personalized health advice. Combining the analysis results of the multi-layer Boltzmann machine, intelligent elderly care services can better understand the health status and needs of the elderly. The learning process of the energy function can generate an in-depth understanding of the health status of the elderly, including potential risk factors and trends. Based on this information, elderly care services can provide more personalized care plans, medication management, and life advice to meet the specific needs of each elderly person. The energy function of the multi-layer Boltzmann machine allows the integration of multi-source health data to form a comprehensive health assessment. This helps elderly care service providers to better understand the overall health status of the elderly, rather than just a single indicator. This comprehensive health management is the core of the intelligent elderly care service method. By integrating data at multiple levels into a model, it is possible to pay more comprehensive attention to the health and quality of life of the elderly.

[0067] Example 5: In step 2, the state of the multi-layer Boltzmann machine is defined as the state of the entire network, obtaining a multi-layer Boltzmann state machine; the multi-layer Boltzmann state machine is determined by the following formula:

[0068] ;

[0069] ;

[0070] where is the state activation probability from the visible layer to the first hidden layer;

[0071] here, represents the state of the rd neuron in the first hidden layer, and the probability that it is activated when the visible layer state is given; the state of the rd neuron in the first hidden layer, taking values of 0 or 1, representing the activated state or non-activated state of the neuron respectively; is the bias of the -th neuron in the first hidden layer; is the state activation probability from the -th hidden layer to the -th hidden layer; where, ; represents the state of the -th neuron in the -th hidden layer, and it is the probability of being activated when given the state of the previous hidden layer; represents the connection weight between the -th neuron in the -th hidden layer and the -th neuron in the -th hidden layer; is the bias of the -th neuron in the -th hidden layer.

[0072] Specifically, and These probabilities represent the activation probabilities of neurons at different levels during the deep information integration process. In the intelligent elderly care service management method, these probabilities can be used to represent the extraction of different-level features of the elderly's health data. For example, the first hidden layer can learn basic physiological features, while subsequent hidden layers can learn more advanced features, such as potential health problem patterns. This hierarchical feature extraction helps to understand and interpret the elderly's health data. is the sigmoid function, which is commonly used in neural networks to introduce non-linearity. It maps the linearly combined input to the probability value range of (0, 1). In the intelligent elderly care service management method, this non-linearity helps the model capture the non-linear relationships between complex health data. This means that the model can better adapt to different patterns and features in the elderly's health data, including sudden events and potential health risks. and The connection weights are used to measure the connection strength and direction between neurons. In the intelligent elderly care service management method, these weight values are learned through the training process to capture various correlation relationships in the elderly's health data. This enables the model to automatically discover useful features and patterns in the data, providing support for the diagnosis and prediction of health problems. and The bias terms are used to adjust the activation thresholds of neurons. In the intelligent elderly care service management method, the existence of bias terms allows the model to be adjusted according to different distributions and features of the data. This is very important for personalized elderly care services and health management because the health conditions and needs of each elderly person may be different.

[0073] The formula of the multi-layer Boltzmann state machine describes the connection weights, bias terms, and activation probabilities between neurons, which helps the model learn key features and high-order relationships in the data. In the intelligent elderly care service management method, the health data of the elderly comes from multiple sensors, such as blood pressure, heart rate, blood sugar, etc. This data may be very complex in its original form, but the formula of the multi-layer Boltzmann state machine helps the model automatically extract and represent the key features of this data, thus achieving data dimensionality reduction and abstraction, and helping to integrate different types of health data into a consistent representation. The activation probability calculation of the multi-layer Boltzmann state machine helps the model understand the health status and needs of the elderly. By calculating the activation probabilities of neurons, the method can understand the feature extraction process of different samples and provide personalized care plans, medication management, and life advice based on this information. This helps to meet the specific needs of each elderly person and improve the quality of elderly care services. The activation probability calculation of the multi-layer Boltzmann state machine helps to monitor and diagnose the health problems of the elderly. By analyzing the activation patterns of neurons, the method can detect potential health problems early and take measures in advance. This is one of the important goals of intelligent elderly care service management to ensure that the health status of the elderly is timely concerned and treated.

[0074] Example 6: When initializing the connection weights, initialize the connection weights as random values sampled from a Gaussian distribution; when initializing the bias of the hidden layer, initialize the bias to zero.

[0075] Specifically, the connection weights are initialized to random values sampled from a Gaussian distribution. This means that the initial value of each connection weight is random and follows the statistical properties of the Gaussian distribution. The Gaussian distribution is a common probability distribution that has two parameters: the mean and the standard deviation. In this initialization method, the weights are randomly initialized to random values from the Gaussian distribution. The reason for doing this is to introduce randomness and avoid having overly similar weights at the beginning of training. If all connection weights are initialized to the same value, it may limit the network's ability to learn from the data. Initializing the connection weights to random values helps break the symmetry of the weights, enabling the model to better learn different features and patterns in the data. This randomness helps accelerate the training of the model and improve its generalization ability, allowing it to better adapt to new data. The biases of the hidden layers are initialized to zero. Initializing the biases of the hidden layers to zero is a simple and effective method. The bias term is used to adjust the activation threshold of the neuron, and initializing them to zero means that no additional offset is introduced initially. This means that at the initial stage of training, the activation state of the neuron is not affected by the bias, and the model will start learning the data from a relatively balanced state. Initializing the biases of the hidden layers to zero helps ensure that the initial state of the model is relatively neutral and does not bias towards any specific direction. This helps the model better explore the data distribution without being affected by the initial bias, thus accelerating the training process.

[0076] Example 7: The process of using the Metropolis-Hastings sampling method to perform Gibbs sampling in step 2 includes:

[0077] Suppose sampling from the conditional distribution is achieved by using the random walk method and randomly changing one or more elements of ; Next, calculate the acceptance probability for determining whether to accept the candidate state, and the formula is as follows:

[0078] ;

[0079] where, is the conditional distribution of the current state, is the conditional distribution of the candidate state, is the transition probability from the current state to the candidate state, is the transition probability from the candidate state to the current state; According to the acceptance probability , accept the candidate state with a certain probability. If , accept the candidate state; otherwise, accept the candidate state with probability . If the candidate state is not accepted, keep the current state unchanged; repeat the above steps until the set number of times is reached; The final sampling result is:

[0080] 。

[0081] Specifically, and these two terms respectively represent the conditional probability distributions of the current state and the candidate state under the condition of a given visible layer state . They are used to measure the likelihood of the first hidden layer state under the current data observation. These probability distributions are affected by the model parameters and the data, reflecting the degree of fit of the model to the data. and these two terms represent the transition probabilities from the current state to the candidate state and from the candidate state to the current state . They describe the change probabilities between states, and this change is usually controlled by the learning algorithm of the model. These transition probabilities enable the model to effectively explore the state space. The acceptance probability is a crucial calculation, which is used to decide whether to accept the candidate state . If , the candidate state is accepted, otherwise the candidate state is accepted with probability . This process is based on the Metropolis - Hastings algorithm, which ensures the rationality and stability of state changes. If the acceptance probability is high, it is more likely to accept the candidate state, otherwise it may be rejected. The role of the whole process is to use the Metropolis - Hastings sampling method to sample the state of the first hidden layer from the conditional distribution , while ensuring that the state changes are reasonable. This sampling process is part of the Gibbs sampling, which is used for the training of the multi - layer Boltzmann machine. It has the following key roles: Through random walks and transition probabilities, this sampling process allows the model to explore the state space, thus better learning the distribution of the data. The Metropolis - Hastings algorithm can, to a certain extent, avoid the model falling into local optimal solutions through the calculation of the acceptance probability . Through multiple samplings, it can help the model gradually converge to the true distribution of the data, improving the stability of training.

[0082] Example 8: The method for calculating the gradient of the energy function with respect to all parameters based on the sampling results includes: For the calculation of the gradient of the connection weight , the following formula is used for calculation:

[0083] ;

[0084] wherein, represents the expectation when the sample data set is used as the input of the Boltzmann machine, represents the expectation when the feature data set is used as the input of the Boltzmann machine; for the gradient calculation of the connection weight where it is calculated using the following formula:

[0085] ;

[0086] For the gradient calculation of the visible layer bias it is calculated using the following formula:

[0087] ;

[0088] For the gradient calculation of the hidden layer bias where it is calculated using the following formula:

[0089] .

[0090] Specifically, for the gradient calculation of the connection weight : It measures the model's fitness by comparing the expectations on the sample data set and the feature data set. This gradient calculation helps optimize the model, thereby better understanding the health data of the elderly. By updating the connection weight, the model can more accurately capture the key features and patterns in the data, and thus improve the detection and prediction of the health problems of the elderly. For the gradient calculation of the connection weight where : This formula is used to calculate the gradient of the connection weight between different hidden layers. It considers the connection from the th hidden layer to the th hidden layer. The update of the connection weight between different hidden layers of the multi-layer Boltzmann machine helps the model to understand the health data of the elderly more deeply. This deep information integration can improve the data's abstract representation ability, making the model more suitable for analyzing the health trends and problems of the elderly. For the gradient calculation of the visible layer bias : This formula is used to calculate the gradient of the bias of the visible layer. It considers the activation of the visible layer units. The gradient calculation of the visible layer bias helps the model better adapt to the health data of different elderly. By updating the bias of the visible layer, the model can more precisely adjust the activation threshold of the data to capture the individual differences and health changes of the elderly. For the gradient calculation of the hidden layer bias where : This formula is used to calculate the biases The gradient. It takes into account the activation of neurons in the hidden layer. Calculating the bias gradient of the hidden layer helps the model better capture the patterns and features of healthy data. By updating the bias of the hidden layer, the model can adjust the activation threshold of neurons in the hidden layer to better match the health status of the elderly.

[0091] Example 9: The method of using an adaptive learning rate strategy to update parameters in step 3 includes: Let the parameter to be updated , where can be the connection weight , the visible layer bias or the hidden layer bias ; Initialize the learning rate as a positive value less than 0.1, the first moment estimate , the second moment estimate , the time step , the first exponential decay rate as the first set value, the second exponential decay rate as the second set value; In each training iteration, calculate the gradient of the parameter , increase the time step ; Use the following formula to update the first moment estimate and the second moment estimate:

[0092] ;

[0093] ;

[0094] Since in the initial iteration, and 's estimates tend to be zero, so bias correction is needed to reduce the bias of the estimates. The corrected first moment estimate and second moment estimate are:

[0095] ;

[0096] ;

[0097] Then use the following formula to update the parameter :

[0098] ;

[0099] where is a positive constant used to avoid the case of a zero denominator.

[0100] Specifically, in deep learning, the goal of the model is to minimize the loss function, and the update direction of the parameters is determined by calculating the parameter gradients. The learning rate determines the step size of parameter updates. An overly large learning rate may lead to unstable training, while an overly small learning rate will result in slow convergence. The subject of this invention is a smart elderly care service management method based on the integration of depth information, which requires the model to quickly and accurately adapt to changes in the health data of the elderly. Therefore, the adaptive adjustment of the learning rate is very important in this context to ensure that the model can effectively provide elderly care services under different circumstances. The learning rate in this strategy is adaptive and will be dynamically adjusted according to the performance of the model at different time steps. This is achieved through the first moment estimation and the second moment estimation to implement. This dynamically adjusted learning rate helps to solve the problem of difficult learning rate setting during the training process. In the elderly care service management method, the health data of the elderly may show different fluctuations and changes. Therefore, the model needs to be able to automatically adjust the learning rate according to the actual situation to better adapt to these changes. Due to the nature of the exponentially weighted moving average, the estimated values in the initial iterations tend to be biased towards zero. Therefore, bias correction is necessary to reduce the bias of the estimation. In the smart elderly care service management method, the data may have different characteristics, and the model needs to be able to adapt to these characteristics without being affected by the bias in the initial stage. Therefore, bias correction helps to ensure that the model can converge stably throughout the training process.

[0101] Embodiment 10: The method for performing sequence feature analysis on the time series in step 3 to determine whether there are problems with the health of the elderly during this period includes: calculating the value of the determinant of the covariance matrix of the time series and comparing it with a set determination threshold. If it exceeds the set determination threshold, it is determined that there are problems with the health of the elderly; the method for performing matrix feature analysis on each element in the time series to screen out the elderly with health problems includes: calculating the Z-score of each element in the matrix corresponding to each element in the time series, and taking the element with the Z-score exceeding the set score value as an abnormal element, and the elderly corresponding to this abnormal element is the elderly with health problems.

[0102] Specifically, the covariance matrix is a mathematical tool used to describe the correlation between different features. In health monitoring, different health data (such as blood pressure, heart rate, blood sugar, etc.) may be correlated with each other, and these correlations may change when the health status of the elderly changes.

[0103] The determinant value can be regarded as a "sensitivity indicator" of the covariance matrix. When the determinant value of the covariance matrix is large, it indicates that the correlation between data is strong, and the healthy data may show large fluctuations or anomalies. This may have an impact on the health of the elderly. The Z-score is a standardized measure used to quantify the deviation of a data point from the mean. In this context, the Z-score is used to measure whether the data at each time point is abnormal compared to other time points. When the absolute value of the Z-score is large, it means that the corresponding data point deviates from the mean, which may reflect abnormal changes in the health status of the elderly. The decision threshold is a pre-set threshold used to determine when a health problem is considered to have occurred. A determinant value exceeding the decision threshold implies that the change in the covariance matrix may be related to a health problem and requires attention. If the determinant value does not exceed the decision threshold, the system further analyzes the Z-score to identify abnormal data points. By analyzing the Z-score, the system can identify the data points that are abnormal in the time series. These abnormal data points may have an impact on the health of the elderly. Combining the Z-score and the decision threshold, the system can screen out which elderly people may have health problems during this period, so as to provide them with more attention and medical services. By integrating this principle with the intelligent elderly care service management method, this method can be used to monitor the health status of the elderly in real time and timely detect potential health problems. This helps to provide personalized elderly care services to meet the different health needs of the elderly, while reducing unnecessary interventions and medical costs. Through in-depth information integration, the system can comprehensively understand the health status of the elderly, provide more accurate elderly care advice and services for them, thereby improving the quality and efficiency of elderly care services.

[0104] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A smart elderly care service management method based on deep information integration, characterized by: The method comprises: Step 1: Collect the health data of each elderly person through various sensors; take the health data of the same elderly person collected at the same time as a sample, extract the features of each sample; form all samples into a sample data set; form the features of all samples into a feature data set; Step 2: Perform data fusion on the sample data set to obtain a fused data set, specifically including: initializing a multi-layer Boltzmann machine, initializing the connection weights and biases; inputting the sample data set as input data into the multi-layer Boltzmann machine; defining the state of the multi-layer Boltzmann machine as the state of the entire network, including the states of the visible layer and multiple hidden layers; using the Metropolis-Hastings sampling method to perform Gibbs sampling to obtain sampling results to capture the state of the multi-layer Boltzmann machine; based on the sampling results, calculating the gradient of the energy function with respect to all parameters; using an adaptive learning rate strategy to update the parameters; and jointly modeling the hidden state representation of the multi-layer Boltzmann machine with the sample data set to obtain a fused data set. Step 3: The fused datasets corresponding to the sample datasets at all moments within a time period are organized into a time series in chronological order. Each element in the time series is a fused dataset. Sequence feature analysis is performed on the time series to determine whether any elderly person has health problems within the time period. If so, matrix feature analysis is performed on each element in the time series to screen out elderly people with health problems and send a message to the elderly person to remind them of the health problems. The energy function of the multi-layer Boltzmann machine in step 2 is expressed as follows: Among them, v is the sample data set, is the state vector of the visible layer in the multi-layer Boltzmann machine, contains M visible units, representing the characteristics of the sample data set, and each visible unit v i Indicates the value of a feature; h (1) is the state vector of the first hidden layer, which contains N hidden units; each represents the state of a neuron in the first hidden layer; h (2) ,h (3) ,...,h (n) are the state vectors of the additional hidden layer, each h (k) represents the state of the neurons in the kth hidden layer; W ij is the connection weight between the i-th visible unit of the visible layer and the j-th neuron of the first hidden layer; W (k) is the connection weight between the kth hidden layer and the k+1th hidden layer; b i is the bias of the visible layer, which is used to adjust the activation threshold of the i-th visible unit in the visible layer; c j is the bias of the first hidden layer, which is used to adjust the activation threshold of the jth neuron in the first hidden layer; E(v,h (1) ,h (2) ,…,h (n) ) is the energy function; In step 2, the state of the multilayer Boltzmann machine is defined as the state of the entire network, thereby obtaining a multilayer Boltzmann state machine; the multilayer Boltzmann state machine is determined by the following formula: in, is the state activation probability from the visible layer to the first hidden layer; here, represents the state of the jth neuron in the first hidden layer, and the probability of it being activated when given the visible layer state v; The state of the jth neuron in the first hidden layer takes the value 0 or 1, indicating the activation or inactivation of the neuron, respectively; is the bias of the jth neuron in the first hidden layer; is the state activation probability from the kth hidden layer to the k+1th hidden layer; Represents the state of the jth neuron in the kth hidden layer. Given the state h of the previous hidden layer (k-1) The probability that it is activated when represents the connection weight between the j'th neuron in the k-1th hidden layer and the j"th neuron in the kth hidden layer; is the bias of the jth neuron in the kth hidden layer; Based on the sampling results, the method of calculating the gradient of the energy function with respect to all parameters includes: for the connection weight W ij The gradient calculation is calculated using the following formula: Among them, data represents the expectation when the sample data set is used as the input of the Boltzmann machine, <·> model represents the expectation when the feature dataset is used as the input of the Boltzmann machine; h sample is the sampling result; for the connection weight W (k) The gradient calculation of , where k>1, is calculated using the following formula: For the visible layer bias b i The gradient calculation is calculated using the following formula: For the hidden layer bias The gradient calculation of , where k>1, is calculated using the following formula:

2. The smart elderly care service management method based on deep information integration according to claim 1 is characterized in that: The health data includes at least: blood pressure, heart rate, blood sugar, body temperature, respiratory rate, blood oxygen saturation and intraocular pressure.

3. The smart elderly care service management method based on deep information integration as claimed in claim 2 is characterized in that: After collecting the health data of each elderly person through various sensors in step 1, the collected health data will also be preprocessed, specifically including: missing value processing, outlier processing, duplicate value processing and standardization processing of the collected health data in turn.

4. The smart elderly care service management method based on deep information integration according to claim 1 is characterized in that: When initializing the connection weights, the connection weights are initialized to random values sampled from a Gaussian distribution; when initializing the bias of the hidden layer, the bias is initialized to zero.

5. The smart elderly care service management method based on deep information integration as claimed in claim 4 is characterized in that: The process of performing Gibbs sampling using the Metropolis-Hastings sampling method in step 2 includes: Suppose that from the conditional distribution P(h (1) |v) sampling, using the random walk method, by randomly changing h (1) Next, the acceptance probability α is calculated to determine whether to accept the candidate state. The formula is as follows: Among them, P(h (1) |v) is the conditional distribution of the current state, is the conditional distribution of candidate states, is the transition probability from the current state to the candidate state, is the transition probability from the candidate state to the current state; according to the acceptance probability α, the candidate state is accepted with a certain probability. If α ≥ 1, the candidate state is accepted. Otherwise, the candidate state is accepted with probability α. If the candidate state is not accepted, the current state remains unchanged. Repeat the above steps until the set number of times is reached. The final sampling result is h sample =[h (1) ,h (2) ,…,h (n) ]; Candidate status.

6. The smart elderly care service management method based on deep information integration as claimed in claim 1 is characterized in that: The method of using the adaptive learning rate strategy to update the parameters in step 3 includes: setting the update parameter θ, where θ is connected to the weight W ij 、W (k) , visible layer bias b i or hidden layer bias Initialize the learning rate z to a positive value less than 0.1, the first-order moment estimate m = 0, the second-order moment estimate v = 0, the time step t = 0, the first exponential decay rate β1 to the first set value, and the second exponential decay rate β2 to the second set value; in each training iteration, calculate the gradient of the parameters Increase the time step t = t + 1; use the following formula to update the first-order moment estimate and the second-order moment estimate: Since the estimates of m and v are biased towards zero in the initial iteration, bias correction is required to reduce the bias of the estimate. The corrected first-order moment estimate and second-order moment estimate are: Then use the following formula to update the parameter θ: Here, ∈ is a positive constant used to avoid the denominator being zero.

7. The smart elderly care service management method based on deep information integration according to claim 6 is characterized in that: In step 3, the method for performing sequence feature analysis on the time series to determine whether any elderly people have health problems within the time period includes: calculating the value of the determinant of the covariance matrix of the time series, comparing it with a set judgment threshold, and if it exceeds the set judgment threshold, it is determined that there are elderly people with health problems; the method for performing matrix feature analysis on each element in the time series to screen out elderly people with health problems includes: calculating the Z-score of each element in the matrix corresponding to each element in the time series, and treating the element with a Z-score exceeding the set score value as an abnormal element, and the elderly person corresponding to the abnormal element is an elderly person with health problems.

Images