A method and system for generating an elderly care evaluation scheme based on an AI large model

Abstract

The application discloses a kind of based on AI big model generation endowment evaluation scheme method and system, it is related to artificial intelligence and endowment service technical cross field, the specific steps of this method are as follows: by preset sensing device every 5 minutes collection old unstructured microfeature data and encryption upload;Call algorithm and carry out dimension transformation;Read cross-modal knowledge graph data quantization evaluation;Matching resource and calculating accessibility index screening target resource;AI big model integrates information and generates scheme containing multiple data and pushes to preset terminal.The application can capture the core physiological and behavioral data of the old people in real time without interference, and the full-link closed-loop protection ensures that privacy is not disclosed;Relying on cross-modal knowledge graph, multi-source data is efficiently linked, and the state of the old people is dynamically quantized and evaluated to adapt to changes;By calculating the accessibility index to screen practical resources, terminal hierarchical authorization and double authentication, it not only meets the information needs of each subject, but also ensures the safe and standardized circulation of the evaluation scheme.

Description

Technical Field

[0001] This invention relates to the intersection of artificial intelligence and elderly care service technology, specifically a method and system for generating elderly care assessment schemes based on a large AI model. Background Technology

[0002] With the accelerating aging of the global population and the continuous expansion of my country's elderly population, the demand for personalized and precise elderly care services is becoming increasingly prominent. Traditional elderly care assessments rely heavily on manual home visits and subjective judgments, which suffer from problems such as lagging data collection, limited assessment dimensions, and low efficiency. They are also unable to capture potential health risks and living needs in the daily behavior of the elderly in real time. At the same time, the uneven distribution of elderly care resources and the inaccurate matching of supply and demand further exacerbate the contradiction between supply and demand for elderly care services. There is an urgent need to leverage advanced technologies such as artificial intelligence and sensor technology to build an intelligent and automated elderly care assessment system to provide elderly care service solutions that meet their actual needs.

[0003] Existing technologies related to elderly care assessment often focus on structured health indicators in data collection, lacking a systematic capture of unstructured micro-features in the daily behavior of the elderly. This makes it difficult to comprehensively reflect key dimensions such as the elderly's muscle strength and environmental adaptability. Inadequate encryption mechanisms during data transmission and storage pose a risk of privacy breaches. The lack of precise correlation between features and assessment dimensions during dimensional transformation leads to distorted quantitative data. Resource matching is often based on simple geographical or service type matching, failing to consider key factors such as resource accessibility and intervention effectiveness, resulting in insufficient practicality of the matching results. Weak cross-modal data association capabilities prevent deep integration of individual elderly data, elderly care resources, and intervention cases. The lack of refined access control for terminals increases the risk of data misuse or information leakage. Furthermore, assessment scheme generation relies on manual integration, resulting in low efficiency and insufficient personalization, failing to meet the real-time and accuracy requirements of elderly care services.

[0004] In summary, existing elderly care assessment methods have significant shortcomings in terms of the comprehensiveness of data collection, the security of data, the accuracy of dimensional transformation, the scientific nature of resource matching, and the security of terminal access. This results in elderly care assessment solutions lacking specificity, reliability, and operability, failing to effectively address the core issue of accurately matching the supply and demand of elderly care services in the context of an aging population. Therefore, developing an intelligent elderly care assessment method and system that can accurately collect multi-dimensional micro-features, ensure data security, achieve efficient quantitative assessment and scientific resource matching, and ensure information security through hierarchical authorization has become an urgent need in the current elderly care service field. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for generating elderly care assessment plans based on an AI large-scale model. This involves collecting unstructured micro-feature data of the elderly through pre-set sensing devices, encrypting it at three levels, and uploading it to a data processing module. Multimodal algorithms are used to convert this data into structured data for quantitative assessment. This is combined with cross-modal knowledge graph matching of elderly care resources and calculation of resource accessibility. The AI ​​large-scale model integrates the assessment results, accessibility indicators, and target resources to generate an elderly care plan that includes risk warnings, which is then pushed to terminals with tiered authorization.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for generating elderly care assessment schemes based on AI large-scale models, the method comprising:

[0007] S100. Micro-feature Acquisition: Unstructured micro-feature data of the elderly's daily behavior is collected every 5 minutes using preset sensing devices. After collection, a summary corresponding to the unstructured micro-feature data is generated. The unstructured micro-feature data and the summary are encrypted using a three-level encryption mechanism of device end-transmission end-server to generate encrypted data. The encrypted data is transmitted to the data processing module via the SSL / TLS 1.3 protocol. The data processing module first processes the data using the decryption logic corresponding to the SSL / TLS 1.3 protocol, and then decrypts it using the device key to obtain the unstructured micro-feature data and the corresponding summary. The unstructured micro-feature data includes gait pause frequency, changes in the grip strength of tableware, and the reaction delay time for adjusting the night light when getting up at night.

[0008] S200, Dimensional Transformation: The multimodal semantic understanding algorithm is called to extract and map the unstructured micro-feature data obtained from the decryption in S100, so as to obtain the structured feature data of muscle strength decline warning degree and the structured feature data of environmental adaptability value corresponding to the quantitative assessment dimension.

[0009] S300, Quantitative Assessment: Receives the structured feature data of muscle strength decline warning degree and environmental adaptability value output by the dimension transformation, and reads the quantitative assessment dimension data of the past 24 hours in the cross-modal knowledge graph. Performs quantitative assessment through a weighted summation formula and outputs the quantitative assessment result.

[0010] S400, Resource Matching: Based on the output quantitative assessment results, the regional elderly care resource layer data is called through the cross-modal knowledge graph to match the output quantitative assessment results with the regional elderly care resources, calculate the accessibility index of the matched resources, and at the same time call the intervention effect layer data of the cross-modal knowledge graph to filter the target resources in the matched resources.

[0011] S500, Solution Generation: The AI ​​big model integrates the quantitative assessment results, calculated accessibility indicators, and selected target resource information to generate an elderly care assessment solution, which is then pushed to a preset terminal. The elderly care assessment solution includes risk warning data, priority resource data, implementation step data, and expected effect data.

[0012] Further, in step S100, the preset sensing device includes a gait sensing device, a pressure sensing device, and an infrared sensing device; the gait sensing device uses a flexible pressure sensor array, laid on the floor of the passageway from the elderly person's bedroom to the living room, and determines the gait pause frequency by the trigger time difference of adjacent sensors and the pressure peak distribution; the pressure sensing device is integrated inside the handle of the tableware, using a strain gauge pressure sensor, and obtains the change value of the tableware gripping force by accumulating the absolute value of the difference between five adjacent sampling points; the infrared sensing device starts timing when it detects a human infrared signal, and stops timing when the light brightness reaches a preset threshold, and the timing result is the nighttime night light adjustment response delay time.

[0013] Furthermore, in step S100, the unstructured micro-feature data and digest are encrypted using a three-level encryption mechanism involving the device end, the transmission end, and the server end:

[0014] Device-side encryption: After the preset sensing device collects unstructured micro-feature data, it generates the unstructured micro-feature data and a digest. Then, the device key is used to symmetrically encrypt the unstructured micro-feature data and the digest to generate encrypted data.

[0015] Encryption at the transmission end: Encrypted data is transmitted via SSL / TLS 1.3 protocol, and temporary session keys are dynamically updated during transmission. The update cycle is consistent with the data acquisition cycle in step S100.

[0016] Server-side decryption: The encryption and decryption module of the data processing module first processes the encrypted data transmitted by the transmission end through the decryption logic corresponding to the SSL / TLS 1.3 protocol, and then decrypts it with the device key to obtain unstructured micro-feature data and digest; the device key is generated by the key management center using the elderly identity identifier and device number, and each device corresponds to a unique device key and is automatically updated every 30 days.

[0017] Furthermore, in step S200, the calculation formula for the multimodal semantic understanding algorithm is as follows: ,in, Let be the correlation coefficient between the i-th type of unstructured micro-feature and the j-th quantitative evaluation dimension. Let k be the value of the i-th type of unstructured micro-feature acquired in the k-th sampling. It is the average value of n samplings of the i-th type of unstructured micro-feature. It is the mapping reference value of the j-th quantization dimension corresponding to the k-th collection. It is the average value of the nth mapping of the j-th quantization dimension, where n is the total number of collections within 24 hours.

[0018] Furthermore, in step S300, the process of calling the weighted summation formula and outputting the quantitative evaluation result is performed according to the following steps:

[0019] Targeted data extraction: Based on the identity of the target elderly, the quantitative assessment dimension data with timestamps within 24 hours before the current time is filtered out through the structured query interface of the cross-modal knowledge graph. The core includes the mapped structured feature data, forming an original data sequence with the structure of collection time-data type-value. This original data sequence is directly used as the input basis for algorithm startup and weight adaptation.

[0020] Algorithm Startup and Weight Adaptation: After starting the dynamic weight update algorithm, load the initial weight parameters: the initial weight for muscle weakness warning is 0.55, and the initial weight for environmental adaptability is 0.45. The dynamic weight update algorithm dynamically adjusts the weights of each dimension based on the changing trend of the preprocessed data sequence, using the following formula: , ,in, The real-time weight of the muscle weakness warning level at time t; The real-time weight of the environmental adaptability value at time t; The weight of the warning level of muscle weakness at time t-1; This is the weighting adjustment factor; The deviation between the preprocessed muscle strength data and the mean at time t; The deviation between the preprocessed data and the mean of the environmental adaptation dimension at time t; the preprocessed data is the data after standardization of the original data sequence.

[0021] Evaluation Result Generation and Output: Real-time weights and corresponding preprocessed data are used to calculate the quantitative evaluation result using a weighted summation formula, which is as follows: ,in, The quantitative evaluation result at time t. The real-time weight of the muscle weakness warning level at time t is [value]. The real-time weight of the environmental adaptability value at time t. , The data is preprocessed for the corresponding dimension; after the result is appended with an identity identifier and a timestamp, it passes the CRC-32 verification and is synchronously pushed to the resource matching module for S400 to call.

[0022] Furthermore, in step S400, the cross-modal knowledge graph adopts a three-level architecture of core node-related child node-related edge to call regional elderly care resource layer data, wherein the relationships between each layer and node are as follows:

[0023] Elderly Individual Data Layer: The elderly identity identifier is the core node, which is associated with three sub-nodes: unstructured micro-feature data, quantitative assessment dimension data, and historical assessment results. The association edge attributes between the sub-nodes and the core node are the attribution relationship and the corresponding timestamp.

[0024] Regional elderly care resource layer: with the unique resource code as the core node, it is associated with four sub-nodes: service capacity level, geographical coordinates, service item list and response time threshold, and the associated edge attributes are attribute descriptions;

[0025] Intervention effect layer: with the intervention case number as the core node, it is associated with four child nodes: unique resource code, service object identity identifier, intervention duration and intervention success rate, and the associated edge attribute is case association;

[0026] Knowledge Association Layer: Cross-layer association edges are constructed using a semantic similarity algorithm. The association edge attribute between the elderly person's identity identifier and the resource's unique code is the demand matching degree, and the association edge attribute between the elderly person's identity identifier and the intervention case number is the situation similarity degree. The weight value of the association edge ranges from 0 to 1, and associations with a weight ≥ 0.7 are prioritized.

[0027] Furthermore, in step S400, the quantitative assessment results are matched with regional elderly care resources by calling regional elderly care resource layer data through cross-modal knowledge graph, and the accessibility index of the matched resources is calculated according to the following steps:

[0028] Core parameters for matching resources are obtained through cross-modal knowledge graphs, including the actual travel time T from the resource to the service recipient and the average service response time of similar elderly care resources in the region. The resource's service coverage population P and its rated service capacity C are determined, along with time weights determined by expert weighting and big data training. and bearing weight ;

[0029] The formula for calculating the accessibility index of matching resources is: ,in, The available contribution value over time. The contribution value to accessibility in the service supply dimension;

[0030] The third step is to use the calculated accessibility index A as the basic parameter for resource screening, and simultaneously link it with the data of the cross-modal knowledge graph intervention effect layer to complete the accurate screening of target resources.

[0031] Furthermore, in step S500, the integration and output process of the large AI model includes the following four parts:

[0032] Data input: The AI ​​big model receives initial data from multiple sources, including quantitative assessment results output by S300, accessibility indicators calculated by S400 and selected target resources. It also covers the basic identity, health monitoring, living behavior, and previous assessment records of the elderly, as well as resource attributes, regional distribution, and supply and demand matching history in the elderly resource database.

[0033] Data preprocessing: The AI ​​big model cleans, deduplicatizes, and standardizes the initial data from multiple sources, removes invalid and abnormal data, and converts data of different formats into a standardized form that the model can recognize, providing high-quality data for subsequent calculations;

[0034] Module computation: The AI ​​big model performs collaborative computation through multiple functional sub-modules. The quantitative assessment sub-module outputs multi-dimensional risk quantification results for elderly care recipients; the accessibility index sub-module calculates accessibility-related indicators for elderly care resources; and the target resource screening sub-module combines the above results to screen out highly matched target resources.

[0035] Results Integration: The AI ​​big model cross-validates and corrects conflicts in the output results of each functional sub-module, integrates quantitative assessment results, accessibility indicators and target resource information into structured data, and generates an elderly care assessment plan based on this structured data, which includes risk warning data, priority resource data, implementation step data and expected effect data.

[0036] Furthermore, in step S500, the preset terminal employs a hierarchical authorization and access isolation mechanism, specifically designed to receive the elderly care assessment plan generated in S500.

[0037] Family member terminal: Authorization level is level 1. It can access risk warning data, priority resource data and implementation step data in the elderly care assessment plan. It has the right to receive real-time warning push, but no modification right.

[0038] Community terminal: The authorization level is level 2. In addition to level 1 permissions, it can access some data in the regional elderly care resource layer and has the authority to provide feedback on resource allocation progress and initiate application for plan adjustment.

[0039] Institutional terminal: The scope of permissions is limited to resource matching needs related to this institution, service confirmation instructions, and some assessment data of the corresponding elderly person;

[0040] The family terminal, community terminal, and institution terminal all have built-in security authentication modules. Before receiving a plan each time, dual authentication of the account password and SMS verification code is required. Only after successful authentication can the plan content be decrypted and viewed.

[0041] A system for generating elderly care assessment plans based on AI large-scale models, the system includes:

[0042] Micro-feature acquisition and encryption module: It is used to collect unstructured micro-feature data of the elderly’s daily behavior at a frequency of once every 5 minutes through gait sensing device, pressure sensing device and infrared sensing device, generate data summary, and complete data encryption, transmission and decryption through a three-level encryption mechanism of device end-transmission end-server.

[0043] The data processing and quantitative evaluation module is used to call the multimodal semantic understanding algorithm to transform the decrypted unstructured micro-feature data into structured feature data corresponding to the muscle weakness warning degree and environmental adaptability value, read relevant data from the cross-modal knowledge graph in the past 24 hours, and calculate and output the quantitative evaluation results through the dynamic weight update algorithm and weighted summation formula.

[0044] Cross-modal knowledge graph and resource matching module: It adopts a three-level architecture of core node-related child node-related edge to store data such as individual elderly data, regional elderly care resources, and intervention effects. It supports cross-level related queries, realizes the matching of quantitative assessment results with elderly care resources, calculates resource accessibility indicators and filters target resources.

[0045] AI Large Model Solution Generation Module: This module receives multi-source data such as quantitative assessment results, accessibility indicators, and target resources. After data cleaning, deduplication, standardization, and collaborative computation of multiple sub-modules, it integrates and generates an elderly care assessment solution that includes risk warnings, priority resources, implementation steps, and expected results.

[0046] Terminal interaction and permission management module: including family terminal, community terminal and institution terminal, adopts hierarchical authorization and permission isolation mechanism, built-in security authentication unit, used to receive and display elderly care assessment plan, support interactive operations within the permission scope, and terminal access requires dual authentication through account password and SMS verification code.

[0047] Compared with the prior art, the method of the present invention has the following beneficial effects:

[0048] I. This invention utilizes pre-set gait sensing, pressure sensing, and infrared sensing devices to continuously collect unstructured micro-feature data on the frequency of gait pauses, changes in the grip strength of tableware, and the delay in reaction time to adjusting lights at night, all within a 5-minute interval. This enables real-time, non-intrusive capture of key physiological and behavioral dimensions of the elderly's daily behavior, comprehensively covering core assessment elements such as muscle strength and environmental adaptability. After data collection, a three-level encryption mechanism is employed: device-transmission-server. The device symmetrically encrypts the data and digest using a dedicated key; the transmission end dynamically updates the temporary session key based on the SSL / TLS 1.3 protocol; and the server decrypts the data using a device management key generated by a key management center. Furthermore, the device key is automatically updated every 30 days, forming a closed-loop security protection system to ensure that the elderly's privacy data is not leaked throughout the entire process of collection, transmission, and storage. During the dimension transformation stage, a feature correlation coefficient formula is introduced, retaining only mapping relationships with a correlation coefficient ≥ 0.7. This accurately maps unstructured micro-features into quantitative data on muscle strength decline warning levels and environmental adaptability values, ensuring the accuracy and reliability of subsequent assessments from the data source.

[0049] Second, this invention constructs a deep association between an individual elderly data layer, a regional elderly care resource layer, an intervention effect layer, and a knowledge association layer through a three-level architecture of cross-modal knowledge graphs: layer-node-edge. It establishes cross-layer association edges using a semantic similarity algorithm, prioritizing associations with a weight ≥ 0.7 to achieve efficient and accurate linkage of multi-source data. During quantitative assessment, a dynamic weight update algorithm is activated, adjusting the real-time weights of muscle weakness warning level and environmental adaptability value based on the changing trends of the original data sequence within 24 hours. This is combined with a weighted summation formula to generate dynamic quantitative assessment results, ensuring the assessment results adapt to the real-time changes in the elderly's condition. In the resource matching stage, accessibility indicators including time and service supply dimensions are calculated, and target resources are screened using intervention success rate data from the intervention effect layer, ensuring that matched resources are both accessible and practical. Terminal access adopts a hierarchical authorization and permission isolation mechanism, with family members, community, and institutional terminals corresponding to different access permissions. Access to the plan requires dual authentication via account password and SMS verification code, ensuring both the information access needs of all relevant parties and the secure and standardized flow of the assessment plan.

[0050] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0052] Figure 1 A flowchart illustrating the steps of a method for generating elderly care assessment schemes based on a large AI model;

[0053] Figure 2 A module diagram of a system for generating elderly care assessment schemes based on a large AI model;

[0054] Figure 3 This is a flowchart illustrating the cross-modal knowledge graph invocation process for a method of generating elderly care assessment schemes based on a large AI model. Detailed Implementation

[0055] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0056] Example 1:

[0057] Implementation scenarios for methods that generate elderly care assessment plans based on AI large models.

[0058] To achieve accurate elderly care assessments and resource matching for seniors within its jurisdiction, a community adopted this AI-based big data model-driven elderly care assessment method. The process begins with micro-feature collection. Flexible pressure sensor arrays were installed on the floor along the passageway from the seniors' bedrooms to their living rooms as gait sensors. Strain gauge pressure sensors were integrated into the handles of commonly used tableware. Infrared sensors were also installed in the seniors' living spaces. These pre-set sensors continuously collected unstructured micro-feature data every 5 minutes, including the frequency of gait pauses, changes in tableware grip strength, and delays in adjusting nighttime lights. This comprehensive approach captured subtle aspects of the seniors' daily behavior, providing foundational data for subsequent assessments. After data collection, the device automatically generates a corresponding data digest, which is then encrypted using a three-level encryption mechanism: device-transmission-server. The device uses a dedicated device key to symmetrically encrypt the unstructured micro-feature data and the digest to generate encrypted data. The transmission end transmits the encrypted data via SSL / TLS 1.3 protocol, dynamically updating the temporary session key at a frequency consistent with the data collection cycle. The server first processes the transmitted encrypted data using the decryption logic corresponding to the SSL / TLS 1.3 protocol, then decrypts it using the device key to obtain the original unstructured micro-feature data and digest. The device key is generated by the key management center based on the elderly person's identity and device number. Each device has a unique key, which is automatically updated every 30 days, ensuring the security and privacy of the data throughout the collection, transmission, and storage process, preventing information leakage. Figure 1 As shown.

[0059] Next, a dimensional transformation is performed. The system calls a multimodal semantic understanding algorithm to extract and map features from the decrypted unstructured micro-feature data. The calculation formula for unstructured data such as gait, grip strength, and reaction delay to lighting adjustments is as follows: ,in, Let be the correlation coefficient between the i-th type of unstructured micro-feature and the j-th quantitative evaluation dimension. Let k be the value of the i-th type of unstructured micro-feature acquired in the k-th sampling. It is the average value of n samplings of the i-th type of unstructured micro-feature. It is the mapping reference value of the j-th quantization dimension corresponding to the k-th collection. It is the average value of n mappings of the j-th quantitative dimension, where n is the total number of collections within 24 hours. It is transformed into structured feature data of muscle strength decline warning degree and structured feature data of environmental adaptability value corresponding to the quantitative assessment dimension. This process achieves accurate mapping by calculating the correlation coefficient between unstructured micro-features and quantitative assessment dimensions, making the originally scattered and irregular micro-feature data standardized and quantifiable, thus laying a solid data foundation for subsequent quantitative assessment.

[0060] The system then proceeds to the quantitative assessment phase. It receives two types of structured feature data output from the dimensional transformation and, through a cross-modal knowledge graph structured query interface, filters the quantitative assessment dimension data from the previous 24 hours based on the target elderly person's identity. This forms a raw data sequence structured as collection time-data type-value, ensuring the timeliness and relevance of the assessment data. A dynamic weight update algorithm is then initiated, loading initial weight parameters of 0.55 for muscle weakness warning and 0.45 for environmental adaptability. Based on the changing trends of the preprocessed data after standardization of the raw data sequence, the dynamic weight update algorithm dynamically adjusts the weights of each dimension according to these trends. The formula is as follows: , ,in, The real-time weight of the muscle weakness warning level at time t; The real-time weight of the environmental adaptability value at time t; The weight of the warning level of muscle weakness at time t-1; This is the weighting adjustment factor; The deviation between the preprocessed muscle strength data and the mean at time t; To account for the deviation between the preprocessed data and the mean of the environmental adaptation dimension at time t, the weight allocation is made more consistent with the elderly person's current physical condition and behavioral performance. Then, using a weighted summation formula, the real-time weights and the corresponding preprocessed data for each dimension are calculated to obtain the quantitative assessment result at time t. The formula is as follows: ,in, The quantitative evaluation result at time t, The real-time weight of the muscle weakness warning level at time t is [value]. The real-time weight of the environmental adaptability value at time t. , The data is preprocessed for each corresponding dimension, and after the elderly person's identity and timestamp are added to the results, they are verified by CRC-32 and then pushed to the resource matching module to ensure the accuracy and completeness of the assessment results and provide a reliable basis for resource matching.

[0061] In the resource matching phase, the cross-modal knowledge graph adopts a three-tier architecture of core nodes, related child nodes, and related edges to access regional elderly care resource layer data. This knowledge graph includes an individual elderly data layer, a regional elderly care resource layer, an intervention effect layer, and a knowledge association layer. Cross-layer related edges are constructed using a semantic similarity algorithm, prioritizing the use of associations with a weight ≥ 0.7 to quickly locate elderly care resources highly aligned with the needs of the elderly. The system matches the quantitative assessment results with regional elderly care resources, first obtaining the actual travel time of the matched resources, the average service response time of similar elderly care resources in the region, the number of people served by the resource, the rated service capacity of the resource, and preset time and carrying capacity weights. Then, the accessibility index of the matched resources is calculated using the accessibility index calculation formula: ,in, The available contribution value over time. The accessibility contribution value of service supply directly reflects the convenience and guarantee capacity of resources in providing services to the elderly. Combined with data from the cross-modal knowledge graph intervention effect layer, target resources are selected to ensure that the matched resources not only meet the needs of the elderly but also have good service effectiveness and accessibility. Figure 3 As shown.

[0062] Finally, the solution generation process begins. The AI-powered big data model receives initial data from multiple sources, including quantitative assessment results, accessibility indicators, target resources, and basic information on the elderly person's identity, health monitoring, daily living behaviors, previous assessment records, and relevant data from the elderly care resource database. This comprehensive integration of various information related to elderly care is then processed through cleaning, deduplication, and standardization. Invalid and abnormal data are removed, and the data format is standardized to ensure high-quality input data and provide solid support for solution generation. Through collaborative computation of multiple functional sub-modules, such as the quantitative assessment sub-module, accessibility indicator sub-module, and target resource screening sub-module, the output results of each sub-module are cross-validated and conflict-corrected to ensure consistency and reliability. This data is then integrated into structured data to generate an elderly care assessment solution that includes risk warning data, priority resource data, implementation step data, and expected effect data. This provides clear, specific, and actionable elderly care guidance for the elderly, their families, and related service organizations, and is then pushed to preset terminals. The pre-set terminals include family terminal, community terminal, and institutional terminal, which adopt a hierarchical authorization and access isolation mechanism. Each terminal has a built-in security authentication module. Before receiving the plan, it is necessary to pass dual authentication of account password and SMS verification code. Only after successful authentication can the plan content be decrypted and viewed. This not only ensures the secure transmission of plan information, but also meets the information access needs of different entities, and facilitates the collaborative promotion of elderly care services by all parties.

[0063] In summary, this method for generating elderly care assessment plans based on an AI-powered large-scale model captures details of the elderly's daily behavior through micro-feature collection, ensures data security through a three-level encryption mechanism, and then uses a multimodal semantic understanding algorithm to transform data dimensions, turning unstructured data into quantifiable assessment indicators. It achieves accurate quantitative assessment using a dynamic weight update algorithm and a weighted summation formula, and combines cross-modal knowledge graphs to achieve efficient matching and accessibility calculation of elderly care resources. Finally, the AI-powered large-scale model integrates multi-source data to generate a comprehensive elderly care assessment plan and pushes it to corresponding terminals in a tiered manner. The entire process is interconnected, ensuring the authenticity, security, and accuracy of the data, while also achieving precise resource matching and personalized plan output, providing systematic and reliable methodological support for the scientific development of elderly care services.

[0064] Example 2:

[0065] System implementation scenarios for generating elderly care assessment schemes based on AI large-scale models.

[0066] The elderly care service platform has deployed a system based on an AI-powered large-scale model to generate elderly care assessment solutions. This system comprises a micro-feature acquisition and encryption module, a data processing and quantitative assessment module, a cross-modal knowledge graph and resource matching module, an AI-powered large-scale model solution generation module, and a terminal interaction and access control module. These modules work collaboratively to automate and precisely manage the entire elderly care assessment process, providing personalized, high-quality assessment services for seniors. Figure 2 As shown.

[0067] The micro-feature acquisition and encryption module is responsible for data acquisition and secure transmission. It is equipped with gait sensing, pressure sensing, and infrared sensing devices. The gait sensing device uses a flexible pressure sensor array, laid on the floor of the passageway from the bedroom to the living room of the platform's users, to detect the frequency of gait pauses and accurately capture the body state signals during the elderly's walking process. The pressure sensing device is integrated into the handle of the elderly's commonly used tableware. It is a strain gauge pressure sensor that can obtain the value of the change in the grip force of the tableware, reflecting the elderly's hand muscle strength. The infrared sensing device is used to detect the reaction delay time of the elderly when getting up at night to adjust the lights, reflecting the elderly's environmental adaptability. These devices collect unstructured micro-feature data every 5 minutes to ensure comprehensive and continuous recording of the elderly's daily behavioral details. After generating a data summary, it is processed through a three-level encryption mechanism: device-transmission-server. The device uses a unique device key to symmetrically encrypt the data and summary; during transmission, the SSL / TLS 1.3 protocol is used, and the temporary session key is dynamically updated according to the data collection cycle; after receiving the data, the server first processes it through the corresponding decryption logic, and then decrypts it with the device key to ensure the security of the data transmission process and avoid the leakage of sensitive information. The device key is generated by the key management center based on the elderly's identity and device number, and is automatically updated every 30 days to further enhance the level of data security.

[0068] The data processing and quantitative evaluation module receives unstructured micro-feature data decrypted by the micro-feature acquisition and encryption module, and calls a multimodal semantic understanding algorithm to extract and map features from the data, transforming the unstructured data into structured feature data for muscle weakness warning level and structured feature data for environmental adaptability value. The calculation formula of the multimodal semantic understanding algorithm is as follows: ,in, Let be the correlation coefficient between the i-th type of unstructured micro-feature and the j-th quantitative evaluation dimension. Let k be the value of the i-th type of unstructured micro-feature acquired in the k-th sampling. It is the average value of n samplings of the i-th type of unstructured micro-feature. It is the mapping reference value of the j-th quantization dimension corresponding to the k-th collection. This is the average value of n mappings for the j-th quantitative dimension, where n is the total number of data collections within 24 hours. This standardizes the data format and extracts its value, allowing the data to be directly used for subsequent assessment calculations. Next, this module extracts the quantitative assessment dimension data of the target elderly person within the past 24 hours through a structured query interface of a cross-modal knowledge graph, forming a raw data sequence and performing standardization processing to ensure the timeliness and consistency of the assessment data. The dynamic weight update algorithm is then activated, dynamically adjusting the weights of each dimension based on the changing trends of the preprocessed data sequence. The formula is: , ,in, The real-time weight of the muscle weakness warning level at time t; The real-time weight of the environmental adaptability value at time t; The weight of the warning level of muscle weakness at time t-1; This is the weighting adjustment factor; The deviation between the preprocessed muscle strength data and the mean at time t; To preprocess the environmental adaptation dimension data at time t to account for the deviation from the mean, the weighting allocation is made more consistent with the actual situation of the elderly. The quantitative assessment result is then calculated using a weighted summation formula, which is: ,in, The quantitative evaluation result at time t. The real-time weight of the muscle weakness warning level at time t is [value]. The real-time weight of the environmental adaptability value at time t. , The data is preprocessed for each corresponding dimension, and the results are verified by CRC-32 before being pushed to the cross-modal knowledge graph and resource matching module to ensure that the evaluation results are accurate and reliable, and to provide a scientific basis for resource matching.

[0069] The cross-modal knowledge graph and resource matching module adopts a three-tier architecture of core nodes, associated sub-nodes, and associated edges. It stores various types of data, including individual elderly data, regional elderly care resources, and intervention effects, constructing a comprehensive and tightly interconnected data network. Specifically, the individual elderly data layer uses the elderly's identity identifier as the core node, linking unstructured micro-feature data, quantitative assessment dimension data, and historical assessment results as sub-nodes, comprehensively preserving relevant information about the elderly. The regional elderly care resource layer uses the resource's unique code as the core node, linking service capacity level, geographical coordinates, service item list, and response time threshold as sub-nodes, clearly presenting the core attributes of elderly care resources. The intervention effect layer uses the intervention case number as the core node, linking resource unique codes, service recipient identity identifiers, intervention duration, and intervention success rate as sub-nodes, providing support for resource effect evaluation. The knowledge association layer constructs cross-layer associated edges through semantic similarity algorithms, clarifying the association attributes and weights between the elderly's identity identifier, resource unique code, and intervention case number. After receiving the quantitative assessment results, this module quickly matches the corresponding regional elderly care resources based on the relationships in the knowledge graph, obtains the core parameters of the matched resources, and calculates the accessibility index using the accessibility index calculation formula, which is: ,in, The available contribution value over time. The accessibility contribution value of the service supply dimension directly reflects the service convenience and supply capacity of resources. Combined with the data of the intervention effect layer, target resources are selected to achieve precise matching between elderly care resources and the needs of the elderly.

[0070] The AI-powered large-scale model solution generation module receives quantitative assessment results from the data processing and quantitative evaluation module, accessibility indicators and target resources from the cross-modal knowledge graph and resource matching module, and simultaneously collects multi-source data such as the basic identity of the elderly, health monitoring, daily living behaviors, previous assessment records, and resource attributes, regional distribution, and supply-demand matching history from the elderly care resource database. This comprehensive integration of relevant information provides ample data support for solution generation. After cleaning, deduplication, and standardization, this data is processed collaboratively by multiple functional sub-modules, including the quantitative evaluation sub-module, accessibility indicator sub-module, and target resource screening sub-module, to fully extract data value. The output results of each sub-module are cross-validated and conflict-corrected to ensure accuracy and consistency. This data is then integrated into structured data to generate an elderly care assessment solution that includes risk warning data, priority resource data, implementation step data, and expected effect data, providing comprehensive and personalized elderly care guidance.

[0071] The terminal interaction and access control module includes three preset terminals: family terminal, community terminal, and institutional terminal. It employs a hierarchical authorization and access isolation mechanism to meet the needs of different users. The family terminal has Level 1 authorization, allowing access to risk warning data, priority resource data, and implementation step data within the elderly care assessment plan, and receiving real-time alerts, but without modification permissions. This facilitates family members' timely understanding of the elderly's situation and cooperation in providing elderly care services. The community terminal has Level 2 authorization, possessing Level 1 permissions and access to some data at the regional elderly care resource level. It has the authority to provide feedback on resource allocation progress and initiate plan adjustment requests, facilitating community-level coordination of elderly care resources. The institutional terminal's permissions are limited to resource matching needs related to the institution, service confirmation instructions, and some assessment data for the corresponding elderly, ensuring the institution can efficiently provide targeted services. All terminals have a built-in security authentication module. Before receiving an elderly care assessment plan each time, users must enter their account password and verify an SMS verification code. Only after completing this dual authentication can the plan content be decrypted and viewed, effectively preventing information leakage risks, ensuring the security and privacy of plan information, and ensuring the orderly flow and use of elderly care service-related information under legal and compliant conditions.

[0072] In summary, this system for generating elderly care assessment solutions based on an AI-powered large-scale model constructs a complete elderly care assessment system from data collection to solution delivery through the collaborative operation of five functional modules. The micro-feature collection and encryption module ensures the comprehensiveness of data collection and the security of transmission; the data processing and quantitative assessment module achieves data transformation and accurate assessment; the cross-modal knowledge graph and resource matching module achieves efficient resource integration; the AI-powered large-scale model solution generation module outputs personalized assessment solutions; and the terminal interaction and permission management module ensures the secure transmission and accurate distribution of solutions. Each module performs its specific function while working closely together, achieving automation, intelligence, and precision throughout the entire elderly care assessment process. This effectively meets the diverse needs of the elderly, their families, communities, and elderly care institutions, providing strong system support for improving the quality of elderly care services.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for generating elderly care assessment schemes based on AI large-scale models, characterized in that, The specific steps of this method are as follows: S100, Micro-feature acquisition: Through preset sensing devices, unstructured micro-feature data of the elderly’s daily behavior are collected at a frequency of once every 5 minutes; After collection, a summary corresponding to the unstructured micro-feature data is generated, and the unstructured micro-feature data and the summary are encrypted and decrypted through a three-level encryption mechanism of device end-transmission end-server. S200, Dimensional Transformation: The multimodal semantic understanding algorithm is called to extract and map the unstructured micro-feature data in S100 to obtain the structured feature data of muscle strength decline warning degree and the structured feature data of environmental adaptability value corresponding to the quantitative assessment dimension. S300, Quantitative Assessment: Receives the structured feature data of muscle strength decline warning degree and environmental adaptability value output by the dimension transformation, and reads the quantitative assessment dimension data of the past 24 hours in the cross-modal knowledge graph. Performs quantitative assessment through a weighted summation formula and outputs the quantitative assessment result. S400, Resource Matching: Based on the output quantitative assessment results, the regional elderly care resource layer data is called through the cross-modal knowledge graph to match the output quantitative assessment results with the regional elderly care resources, calculate the accessibility index of the matched resources, and at the same time call the intervention effect layer data of the cross-modal knowledge graph to filter the target resources in the matched resources. S500, Solution Generation: The AI ​​big model integrates the quantitative assessment results, calculated accessibility indicators, and selected target resource information to generate an elderly care assessment solution, which is then pushed to a preset terminal. The elderly care assessment solution includes risk warning data, priority resource data, implementation step data, and expected effect data.

2. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S100, the preset sensing device includes a gait sensing device, a pressure sensing device, and an infrared sensing device. The gait sensing device uses a flexible pressure sensor array, which is laid on the floor of the passageway from the elderly person's bedroom to the living room. The frequency of gait pauses is determined by the trigger time difference and pressure peak distribution of adjacent sensors. The pressure sensing device is integrated inside the handle of the tableware and uses a strain gauge pressure sensor. The change in the gripping force of the tableware is obtained by accumulating the absolute values ​​of the differences between five adjacent sampling points. The infrared sensing device starts timing when it detects a human infrared signal and stops timing when the light brightness reaches a preset threshold. The timing result is the nighttime light adjustment response delay time.

3. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S100, the unstructured micro-feature data and digest are encrypted using a three-level encryption mechanism involving the device end, the transmission end, and the server end. Device-side encryption: After the preset sensing device collects unstructured micro-feature data, it generates the unstructured micro-feature data and a digest. Then, the device key is used to symmetrically encrypt the unstructured micro-feature data and the digest to generate encrypted data. Encryption at the transmission end: Encrypted data is transmitted via SSL / TLS 1.3 protocol, and temporary session keys are dynamically updated during transmission. The update cycle is consistent with the data acquisition cycle in step S100. Server-side decryption: The encryption and decryption module of the data processing module first processes the encrypted data transmitted by the transmission end through the decryption logic corresponding to the SSL / TLS 1.3 protocol, and then decrypts it with the device key to obtain unstructured micro-feature data and digest; the device key is generated by the key management center using the elderly identity identifier and device number, and each device corresponds to a unique device key and is automatically updated every 30 days.

4. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S200, the calculation formula for the multimodal semantic understanding algorithm is as follows: ,in, Let be the correlation coefficient between the i-th type of unstructured micro-feature and the j-th quantitative evaluation dimension. Let k be the value of the i-th type of unstructured micro-feature acquired in the k-th sampling. It is the average value of n samplings of the i-th type of unstructured micro-feature. It is the mapping reference value of the j-th quantization dimension corresponding to the k-th collection. It is the average value of the nth mapping of the j-th quantization dimension, where n is the total number of collections within 24 hours.

5. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S300, the process of calling the weighted summation formula and outputting the quantitative evaluation result is performed according to the following steps: Targeted data extraction: Based on the identity of the target elderly, the quantitative assessment dimension data with timestamps within 24 hours before the current time is filtered out through the structured query interface of the cross-modal knowledge graph. The core includes the mapped structured feature data, forming an original data sequence with the structure of collection time-data type-value. This original data sequence is directly used as the input basis for algorithm startup and weight adaptation. Algorithm Startup and Weight Adaptation: After starting the dynamic weight update algorithm, load the initial weight parameters: the initial weight for muscle weakness warning is 0.55, and the initial weight for environmental adaptability is 0.

45. The dynamic weight update algorithm dynamically adjusts the weights of each dimension based on the changing trend of the preprocessed data sequence, using the following formula: , ,in, The real-time weight of the muscle weakness warning level at time t; The real-time weight of the environmental adaptability value at time t; The weight of the warning level of muscle weakness at time t-1; This is the weighting adjustment factor; The deviation between the preprocessed muscle strength data and the mean at time t; The deviation between the preprocessed data and the mean of the environmental adaptation dimension at time t; the preprocessed data is the data after standardization of the original data sequence. Evaluation Result Generation and Output: Real-time weights and corresponding preprocessed data are used to calculate the quantitative evaluation result using a weighted summation formula, which is as follows: ,in, The quantitative evaluation result at time t. The real-time weight of the muscle weakness warning level at time t is [value]. The real-time weight of the environmental adaptability value at time t. , The data is preprocessed for the corresponding dimension; after the result is appended with an identity identifier and a timestamp, it passes the CRC-32 verification and is synchronously pushed to the resource matching module for S400 to call.

6. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S400, the cross-modal knowledge graph adopts a three-level architecture of core node-related child node-related edge to call regional elderly care resource layer data, wherein the relationships between each layer and node are as follows: Elderly Individual Data Layer: The elderly identity identifier is the core node, which is associated with three sub-nodes: unstructured micro-feature data, quantitative assessment dimension data, and historical assessment results. The association edge attributes between the sub-nodes and the core node are the attribution relationship and the corresponding timestamp. Regional elderly care resource layer: with the unique resource code as the core node, it is associated with four sub-nodes: service capacity level, geographical coordinates, service item list and response time threshold, and the associated edge attributes are attribute descriptions; Intervention effect layer: with the intervention case number as the core node, it is associated with four child nodes: unique resource code, service object identity identifier, intervention duration and intervention success rate, and the associated edge attribute is case association; Knowledge Association Layer: Cross-layer association edges are constructed using a semantic similarity algorithm. The association edge attribute between the elderly person's identity identifier and the resource's unique code is the demand matching degree, and the association edge attribute between the elderly person's identity identifier and the intervention case number is the situation similarity degree. The weight value of the association edge ranges from 0 to 1, and associations with a weight ≥ 0.7 are prioritized.

7. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S400, the method of calling regional elderly care resource layer data through cross-modal knowledge graph, matching the output quantitative assessment results with regional elderly care resources, and calculating the accessibility index of the matched resources according to the following steps: Core parameters for matching resources are obtained through cross-modal knowledge graphs, including the actual travel time T from the resource to the service recipient and the average service response time of similar elderly care resources in the region. The resource's service coverage population P and its rated service capacity C are determined, along with time weights determined by expert weighting and big data training. and bearing weight ; The formula for calculating the accessibility index of matching resources is: ,in, The available contribution value over time. The contribution value of accessibility in the service supply dimension; The third step is to use the calculated accessibility index A as the basic parameter for resource screening, and simultaneously link it with the data of the cross-modal knowledge graph intervention effect layer to complete the accurate screening of target resources.

8. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S500, the integration and output process of the large AI model includes the following four parts: Data input: The AI ​​big model receives initial data from multiple sources. The initial data from multiple sources includes the quantitative assessment results output by S300, the accessibility indicators calculated by S400 and the selected target resources. It also covers the basic identity of the elderly, health monitoring, living behavior, previous assessment records, as well as the resource attributes, regional distribution and supply and demand matching history in the elderly resource database. Data preprocessing: The AI ​​big model cleans, deduplicatizes, and standardizes the initial data from multiple sources, removes invalid and abnormal data, and converts data of different formats into a standardized form that the model can recognize, providing high-quality data for subsequent calculations; Module computation: The AI ​​big model performs collaborative computation through multiple functional sub-modules. The quantitative assessment sub-module outputs multi-dimensional risk quantification results for elderly care recipients; the accessibility index sub-module calculates accessibility-related indicators for elderly care resources; and the target resource screening sub-module combines the above results to screen out highly matched target resources. Results Integration: The AI ​​big model cross-validates and corrects conflicts in the output results of each functional sub-module, integrates quantitative assessment results, accessibility indicators and target resource information into structured data, and generates an elderly care assessment plan based on this structured data, which includes risk warning data, priority resource data, implementation step data and expected effect data.

9. The method for generating elderly care assessment schemes based on AI large-scale models according to claim 1, characterized in that, In step S500, the preset terminal employs a hierarchical authorization and access isolation mechanism, specifically designed to receive the elderly care assessment plan generated in S500. Family member terminal: Authorization level is level 1. It can access risk warning data, priority resource data and implementation step data in the elderly care assessment plan. It has the right to receive real-time warning push, but no modification right. Community terminal: The authorization level is level 2. In addition to level 1 permissions, it can access some data in the regional elderly care resource layer and has the authority to provide feedback on resource allocation progress and initiate application for plan adjustment. Institutional terminal: The scope of permissions is limited to resource matching needs related to this institution, service confirmation instructions, and some assessment data of the corresponding elderly person; The family terminal, community terminal, and institution terminal all have built-in security authentication modules. Before receiving a plan each time, dual authentication of the account password and SMS verification code is required. Only after successful authentication can the plan content be decrypted and viewed.

10. A system for generating elderly care assessment schemes based on AI large-scale models, the system being applicable to the method for generating elderly care assessment schemes based on AI large-scale models as described in any one of claims 1-9, characterized in that, The system includes: Micro-feature acquisition and encryption module: It is used to collect unstructured micro-feature data of the elderly’s daily behavior at a frequency of once every 5 minutes through gait sensing device, pressure sensing device and infrared sensing device, generate data summary, and complete data encryption, transmission and decryption through a three-level encryption mechanism of device end-transmission end-server. The data processing and quantitative evaluation module is used to call the multimodal semantic understanding algorithm to transform the decrypted unstructured micro-feature data into structured feature data corresponding to the muscle weakness warning degree and environmental adaptability value, read relevant data from the cross-modal knowledge graph in the past 24 hours, and calculate and output the quantitative evaluation results through the dynamic weight update algorithm and weighted summation formula. Cross-modal knowledge graph and resource matching module: It adopts a three-level architecture of core node-related child node-related edge to store individual data of the elderly, regional elderly care resources, and intervention effect data. It supports cross-level related queries, realizes the matching of quantitative assessment results with elderly care resources, calculates resource accessibility indicators and filters target resources; AI Large Model Solution Generation Module: This module receives quantitative assessment results, accessibility indicators, and target resource data. After data cleaning, deduplication, standardization, and collaborative computation among multiple sub-modules, it integrates and generates an elderly care assessment solution that includes risk warnings, priority resources, implementation steps, and expected results. Terminal interaction and permission management module: including family terminal, community terminal and institution terminal, adopts hierarchical authorization and permission isolation mechanism, built-in security authentication unit, used to receive and display elderly care assessment plan, supports interactive operations within the permission scope, and terminal access requires dual authentication through account password and SMS verification code.

Images