An old-age security management and control method and system based on multi-agent fusion hierarchical response

Abstract

The application discloses a pension safety management and control method and system based on multi-agent fusion hierarchical response, comprising the following steps: deploying multiple types of sensing devices to collect state sensing data of the elderly; constructing multiple functional agents to preliminarily analyze the state sensing data of the elderly and obtain each local risk judgment result with a confidence level; gathering each local risk judgment result into a fusion decision agent, and performing conflict detection and blackboard evidence debate negotiation to obtain the comprehensive risk level of the elderly; based on the comprehensive risk level of the elderly, a hierarchical response agent is used to match the comprehensive risk level and execute the corresponding hierarchical response strategy to obtain an execution result; the execution result and the change of the state sensing data of the elderly are fed back to the fusion decision agent, the individual historical baseline of the elderly is updated, the individual historical baseline of all the elderly is optimized, and the professional analysis model parameters of the multiple functional agents are optimized.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of smart elderly care and artificial intelligence, and more specifically, to an elderly care safety management and control method and system based on multi-agent fusion hierarchical response. Background Technology

[0002] As the aging population continues to grow, safety incidents such as falls, sudden cardiovascular and cerebrovascular diseases, gas poisoning, and fires are becoming increasingly frequent among those living alone or in institutional care settings. Furthermore, elderly individuals often lack self-rescue abilities after accidents, and guardians or healthcare professionals may delay detection, leading to serious consequences. How to implement 24 / 7, low-false-alarm, and closed-loop safety management for the elderly is a key technical challenge in the field of smart elderly care. Existing elderly care safety monitoring solutions mainly suffer from problems such as strong reliance on single-modal threshold alarms, lack of effective integration and conflict resolution of multi-source data, lack of tiered response, and inadequate closed-loop and confirmation mechanisms.

[0003] Therefore, existing technologies in the field of smart elderly care have shortcomings, and there is an urgent need for an elderly care safety management and control technology solution that can integrate multi-source heterogeneous sensing data, perform time reversal tracing and evidence debate and negotiation when modal conflicts occur, drive graded and precise response with risk levels with uncertainty, and have closed-loop optimization capabilities. Summary of the Invention

[0004] In view of the above problems, the purpose of this invention is to provide a method and system for elderly care safety management and control based on multi-agent fusion and hierarchical response, which can effectively solve the problems of strong dependence on single-modal threshold alarms, lack of effective fusion and conflict resolution of multi-source data, lack of hierarchical response, closed loop and confirmation in existing technologies in the field of smart elderly care.

[0005] The first aspect of this invention provides a multi-agent fusion hierarchical response method for elderly care safety management, comprising: Deploy various types of sensing devices to collect real-time status perception data of the elderly; Multiple functional intelligent agents are constructed to perform preliminary analysis on the state perception data of the elderly, and obtain multiple local risk assessment results with confidence levels. The results of the individual risk assessments are aggregated into a fusion decision-making intelligent agent, and conflict detection and blackboard evidence debate and negotiation are performed to obtain the comprehensive risk level of the elderly. Based on the comprehensive risk level of the elderly, the hierarchical response agent matches the comprehensive risk level and executes the corresponding hierarchical response strategy to obtain the execution result; The execution results and changes in the elderly person's state perception data are fed back to the fusion decision agent to update the individual historical baseline of the elderly person, optimize the individual historical baseline of all elderly persons, and optimize the professional analysis model parameters of the multiple functional agents.

[0006] In this solution, the results of the individual risk assessments are aggregated into a fusion decision-making intelligent agent, and conflict detection and blackboard evidence debate and negotiation are performed to obtain the comprehensive risk level of the elderly, including: Based on a unified spatiotemporal benchmark, the results of each local risk assessment are aligned to the same time axis and the same risk semantic space. Cross-modal conflicts in time and semantics between the results of each local risk assessment are detected, and the time period is divided into conflict-free time period and conflict time period according to the detection results. The results of each local risk assessment are mapped to a unified risk confidence level. The risk confidence levels for the conflict-free period are then fused using a preset method to obtain an initial consensus. For the conflict period, the original sensing signals within a preset number of consecutive time windows prior to the conflict period are traced back along the time sequence to reconstruct the event evolution trajectory, and the conflict period is determined based on the evolution trajectory to be caused by instantaneous sensor anomaly or gradual real risk. Using the initial consensus as the initial input, on the shared blackboard, the multiple functional agents question the local risk assessment results of other functional agents based on evidence from their own domains of responsibility and determine the intensity of the questioning. The fusion decision agent dynamically adjusts the weights of each local risk assessment result based on the intensity of the questioning and calculates the comprehensive risk level of the elderly person.

[0007] In this solution, the reconstructing of the event evolution trajectory and the determination based on the evolution trajectory whether the conflict period originates from a transient sensor anomaly or a gradual real risk include: Based on the time sequence of the elderly within a preset historical time period, a personalized daily event graph of the elderly is constructed. The personalized daily event graph is divided into nodes with daily events and edges with the transition relationships between events, and the normal time interval range of each transition relationship is marked. The original sensing signals within a preset number of consecutive time windows before the conflict period are identified as a preceding event sequence, and the original sensing signals within a preset number of consecutive time windows after the conflict period are identified as a subsequent event sequence. The preceding event sequence and the following event sequence are matched with the personalized daily event graph to obtain the preceding continuous matching path length and the following continuous matching path length. When the length of the preceding consecutive matching path is greater than or equal to a preset length threshold, and the length of the following consecutive matching path is greater than or equal to the preset length threshold, the conflict is determined to originate from the instantaneous sensor malfunction. The conflict is determined to originate from the asymptotic real risk when any of the following conditions are met: The preceding consecutive matching path length is less than the preset length threshold; the following consecutive matching path length is less than the preset length threshold; there is a transition in the preceding event sequence that is not present in the personalized daily event graph; there is a transition in the following event sequence that is not present in the personalized daily event graph; the time interval between adjacent events in the preceding event sequence or the following event sequence exceeds the corresponding normal time interval range.

[0008] The original sensory signals within the conflict period and a predetermined number of consecutive time windows are identified as event sequences, and the event sequences are matched with the personalized daily event map. When the event sequence has a continuous matching path in the personalized daily event graph, and the length of the continuous matching path is greater than or equal to a preset length threshold, and the preceding time window and the subsequent time window adjacent to the conflict period are both in the normal range, it is determined that the conflict originates from the instantaneous sensor malfunction. When the event sequence shows a non-existent transition in the personalized daily event graph, or the time interval between adjacent events exceeds the corresponding normal time interval range, or the length of the continuous matching path is less than the preset length threshold, the conflict is determined to originate from the progressive real risk.

[0009] This plan also includes: When tracing back the conflict period along the timeline, the video function agent records the elderly person's limb movements and movement trajectory in real time, and the vital sign detection function agent records abnormal vital sign data. When the time reversal tracing is performed again, the video function agent and the vital sign detection function agent will first search for and match the recorded limb movements, movement trajectories, and abnormal vital sign data of the elderly. If a match is found, the conflict period will be classified as a non-conflict period.

[0010] In this scheme, on a shared blackboard, multiple functional agents challenge the local risk assessment results of other functional agents based on evidence from their own domains of responsibility and determine the intensity of the challenge. The fusion decision agent dynamically adjusts the weights of each local risk assessment result based on the intensity of the challenge, including: The challenge is valid when the functional agent that raises the challenge can provide original perceptual evidence within its domain of responsibility that contradicts the challenged local risk assessment result, and the original perceptual evidence covers the time period on which the challenged local risk assessment result is based. The challenge is invalid if the functional agent raising the challenge fails to provide original perceptual evidence within its domain of responsibility that contradicts the challenged local risk assessment result, or if the original perceptual evidence does not cover the time period on which the challenged local risk assessment result is based. When the challenge is valid, the fusion decision-making agent will reduce the weight of the challenged local risk assessment result according to the intensity of the challenge, and the reduced weight will not be lower than the preset weight lower limit, and the challenge information will be recorded in the shared blackboard. When the challenge is found to be invalid, the fusion decision-making agent maintains the current weight of the challenged local risk assessment result unchanged.

[0011] In this scheme, based on the comprehensive risk level of the elderly person, the hierarchical response agent matches the comprehensive risk level and executes the corresponding hierarchical response strategy to obtain the execution result, including: The hierarchical response strategy is divided into three levels: Level 1 Response: Local notification, pushing comforting or reminder messages to the elderly's terminals; Level 2 Response: Send alarm notifications to the guardian's terminal and the caregiver's terminal, and initiate on-site verification; Level 3 Response: Based on the Level 2 response, coordinate with at least one of the following: medical emergency, fire, or community rescue, and continue to monitor until the risk is eliminated; The comprehensive risk level includes four levels: normal, attention, early warning, and emergency. The correspondence between these levels and the tiered response strategy is as follows: Normal and attention correspond to the Level 1 response, early warning corresponds to the Level 2 response, and emergency corresponds to the Level 3 response; Among them, "normal" means no risk, "attention" means continuous observation is required, "early warning" means that guardians or caregivers need to be notified, and "emergency" means that external rescue resources need to be coordinated. The tiered response strategy has escalation and degrade mechanisms: When a high-level response is not confirmed within a preset time or the risk continues to worsen, it will be automatically upgraded; when the risk level drops and is confirmed to be safe after review, it will be downgraded level by level and the record will be archived.

[0012] In this solution, the automatic escalation when a high-level response is not confirmed within a preset time period or the risk continues to worsen includes: The rate of risk deterioration is calculated based on the change in the overall risk level of the elderly over time. When the rate of risk deterioration exceeds the rate threshold and the estimated arrival time of the nearest schedulable disposal resource exceeds the duration threshold, while maintaining the execution of the current level response actions, the notification and resource reservation actions in the next level response actions are initiated in advance, while the on-site intervention and external rescue dispatch actions in the next level response actions are temporarily suspended. When the overall risk level is subsequently upgraded to the next level, the completed notifications and resource reservation results are directly inherited, and only the remaining response actions that have not yet been executed in that level are executed; When the overall risk level does not escalate and the risk decreases, the initiated notifications and resource reservations are revoked and archived, so that the response strength is continuously interpolated as the risk evolves rather than discretely jumping.

[0013] In this scheme, the execution results and changes in the elderly person's state perception data are fed back to the fusion decision agent to update the individual historical baseline of the elderly person and optimize the individual historical baselines of all elderly people. Simultaneously, the professional analysis model parameters of the multiple functional agents are optimized, including: The anonymized individual anomaly patterns are used as anomaly prototypes and uploaded to the global anomaly pattern library. After being filtered by the global abnormal pattern library, the abnormal prototype is migrated to the edge computing nodes of other elderly people in the same group as the target elderly person, so that the other edge computing nodes have the ability to identify the abnormal pattern during the cold start phase. The anomalous prototypes migrated to the target edge computing node are assigned confidence weights that decrease as the target individual's own sample accumulation increases, forming a dual-loop optimization of personalization and swarm intelligence to optimize the professional analysis model parameters of the multiple functional agents.

[0014] A second aspect of the present invention provides a multi-agent fusion hierarchical response elderly care safety management and control system, comprising: The multi-source sensing module is used to deploy various types of sensing devices to collect real-time status sensing data of the elderly. The multi-agent system assessment module is used to construct multiple functional agents to perform preliminary analysis on the elderly person's state perception data and obtain multiple local risk assessment results with confidence levels. The fusion decision module is used to aggregate the results of the local risk assessments into the fusion decision agent, and to perform conflict detection and blackboard evidence debate and negotiation to obtain the comprehensive risk level of the elderly. The graded response module is used to match the comprehensive risk level of the elderly person with the graded response agent and execute the corresponding graded response strategy to obtain the execution result. The execution feedback module is used to send the execution results and changes in the elderly person's state perception data back to the fusion decision agent, update the individual historical baseline of the elderly person, optimize the individual historical baseline of all elderly persons, and optimize the professional analysis model parameters of the multiple functional agents. Attached Figure Description

[0015] Figure 1 This invention provides a flowchart of a multi-agent fusion hierarchical response method for elderly care safety management. Figure 2 The flowchart of a multi-agent fusion hierarchical response elderly care safety management system provided by the present invention is shown. Detailed Implementation

[0016] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0018] Figure 1 The flowchart of a multi-agent fusion hierarchical response elderly care safety management method provided by the present invention is shown.

[0019] like Figure 1 As shown, this invention discloses a multi-agent fusion hierarchical response method for elderly care safety management, comprising: S101 deploys multiple types of sensing devices to collect real-time status perception data of the elderly; S102, Construct multiple functional intelligent agents to conduct preliminary analysis on the elderly's state perception data, and obtain multiple local risk assessment results with confidence levels; S103 integrates the results of risk assessments from various localities into a unified decision-making intelligent agent, and performs conflict detection and blackboard evidence debate and negotiation to obtain the comprehensive risk level of the elderly. S104, based on the comprehensive risk level of the elderly, the hierarchical response agent matches the comprehensive risk level and executes the corresponding hierarchical response strategy to obtain the execution result; S105 feeds back the execution results and changes in the elderly’s state perception data to the fusion decision agent, updates the individual historical baseline of the elderly, optimizes the individual historical baseline of all elderly, and optimizes the professional analysis model parameters of multiple functional agents.

[0020] According to an embodiment of the present invention, multiple sensing devices deployed in elderly care facilities are used to collect real-time state perception data of the elderly. The perception data includes at least two of the following: physiological signs data, behavioral activity data, and environmental status data. Multiple functional agents are constructed, including at least a physiological monitoring agent, a behavioral detection agent, and an environmental monitoring agent. Each functional agent performs preliminary analysis on data within its respective domain of responsibility based on its corresponding professional analysis model, and outputs a local risk assessment result with confidence. A fusion decision agent aggregates the local risk assessment results from each functional agent, aligns them to the same time axis and the same risk semantic space based on a unified spatiotemporal reference, detects cross-modal conflicts in time and semantics between the local risk assessment results, and divides the time period into conflict-free and conflict periods according to the detection results. Each local risk assessment result is mapped to a unified risk confidence representation, and the confidence of the conflict-free periods is fused using a preset method to obtain an initial consensus. For conflict periods, the original sense data within a preset number of consecutive time windows prior to the conflict period is traced back chronologically. The system recognizes signals, reconstructs the event evolution trajectory, and determines whether the conflict originates from instantaneous sensor anomalies or progressive real risks based on this trajectory. Using initial consensus as initial input, each functional agent challenges the local risk assessment results of other functional agents on a shared blackboard based on evidence from its own domain of responsibility. The fusion decision agent dynamically adjusts the weights of each local risk assessment result according to the intensity of the challenge, forming a consensus risk level. The hierarchical response agent matches and executes the corresponding hierarchical response strategy based on the comprehensive risk level. When the width of the uncertainty interval of the comprehensive risk level is higher than the preset difference threshold, an active confirmation sub-process is executed first to initiate a light interaction confirmation to the elderly person's terminal or caregiver's terminal. If no confirmation is received or no response is received within the time limit, the hierarchical response strategy is escalated according to the risk level. The hierarchical response strategy includes at least a first-level response, a second-level response, and a third-level response. The execution results of the hierarchical response strategy are collected, and the execution results and changes in the elderly person's state perception data are fed back to the fusion decision agent to update the individual historical baseline of the elderly person and optimize the individual historical baseline of all elderly persons, while also optimizing the professional analysis model parameters of each functional agent.

[0021] According to an embodiment of the present invention, the results of each local risk assessment are aggregated into a fusion decision-making agent, and conflict detection and blackboard evidence debate and negotiation are performed to obtain the comprehensive risk level of the elderly, including: Based on a unified spatiotemporal benchmark, the results of local risk assessments are aligned to the same time axis and the same risk semantic space. Cross-modal conflicts in time and semantics between the results of local risk assessments are detected, and the time period is divided into conflict-free time period and conflict time period according to the detection results. The results of risk assessments in each local area are mapped to a unified risk confidence level. For risk confidence levels in non-conflict periods, a preset method is used to merge them to obtain an initial consensus.

[0022] For the conflict period, the original sensing signals within a preset number of consecutive time windows before the conflict period are traced back along the time sequence to reconstruct the event evolution trajectory, and the conflict period is determined to be caused by instantaneous sensor anomalies or gradual real risks based on the evolution trajectory.

[0023] Using the initial consensus as the initial input, multiple functional agents on a shared blackboard challenge the local risk assessment results of other functional agents based on evidence from their own domains of responsibility and determine the intensity of the challenge. The fusion decision agent dynamically adjusts the weights of the local risk assessment results based on the intensity of the challenge and calculates the comprehensive risk level of the elderly.

[0024] It should be noted that a unified spatiotemporal benchmark refers to aligning the results of local risk assessments to the same time window according to timestamps and mapping their respective risk semantics to a unified risk level scale. During conflict detection, the risk level difference between each pair of different modalities is calculated. When the difference exceeds a preset difference threshold, it is determined that there is a cross-modal conflict. Time windows with conflicts are marked as conflict periods, otherwise they are non-conflict periods. The preset difference threshold is set by those skilled in the art according to actual needs.

[0025] Mapping to a unified risk confidence level means converting each local risk assessment result to the same risk confidence level scale based on the evidence type and its own confidence level of each local risk assessment result; the initial fusion is only performed on the confidence level of the conflict-free period, and the initial consensus obtained by fusion is used as the initial input for blackboard evidence debate and negotiation; when all periods are determined to be conflict-free periods, the initial consensus is used as the consensus risk level.

[0026] The preset method involves normalizing the confidence levels of each local risk assessment result within a conflict-free period to obtain the fusion weight corresponding to each local risk assessment result; mapping the risk levels in each local risk assessment result to risk values ​​according to a unified risk semantic space; performing a weighted average of the corresponding risk values ​​using each fusion weight to obtain an initial consensus; if the risk level is represented by a discrete level, the discrete level can be mapped to a continuous risk score first, and then a weighted average can be performed; if all confidence levels are zero, then each fusion weight is set to an equal weight.

[0027] The preset quantity is a pre-configured fixed parameter that represents the number of consecutive time windows to be traced back in time reversal tracing. This parameter is set during system initialization and is used to control the time span of the tracing.

[0028] The overall risk level is accompanied by an uncertainty range, the width of which reflects the degree of conflict between multiple sources of evidence.

[0029] According to embodiments of the present invention, reconstructing the event evolution trajectory and determining whether the conflict period originated from a transient sensor anomaly or a gradual real risk based on the evolution trajectory includes: Based on the time series of elderly people within a preset historical time period, a personalized daily event graph of the elderly is constructed. The personalized daily event graph is divided into daily events as nodes and the transition relationship between events as edges, and the normal time interval range of each transition relationship is marked. The original sensing signals within a preset number of consecutive time windows before the conflict period are identified as the preceding event sequence, and the original sensing signals within a preset number of consecutive time windows after the conflict period are identified as the following event sequence. The preceding event sequence and the following event sequence are matched with the personalized daily event graph to obtain the preceding continuous matching path length and the following continuous matching path length. If the length of the preceding consecutive matching path is greater than or equal to a preset length threshold, and the length of the subsequent consecutive matching path is greater than or equal to the preset length threshold, the conflict is determined to be caused by an anomaly in the instantaneous sensor. The conflict is determined to originate from the aforementioned asymptotic real risk when any of the following conditions are met: The preceding consecutive matching path length is less than a preset length threshold; the following consecutive matching path length is less than a preset length threshold; there is a transition in the preceding event sequence that is not present in the personalized daily event graph; there is a transition in the following event sequence that is not present in the personalized daily event graph; the time interval between adjacent events in the preceding or following event sequence exceeds the corresponding normal time interval range.

[0030] It should be noted that the personalized daily event map is obtained by segmenting and mining the state perception data of the elderly within a preset historical time period. Specifically, the original perception signal is divided into multiple event segments in chronological order, with each event segment corresponding to a daily activity, such as sleeping, waking up, eating, activity, and toileting. The transition relationship between events is established based on the temporal adjacency of the event segments. The normal time interval range of each transition relationship can be determined according to the statistical distribution of the time interval corresponding to that transition relationship in the historical record, for example, taking the mean plus or minus two standard deviations as the normal range.

[0031] The preset history can be set by those skilled in the art according to actual needs.

[0032] The preset number is a pre-configured parameter representing the number of consecutive time windows contained in both the preceding and following event sequences. A larger preset number results in more time windows participating in the matching and a more stable judgment result, but also requires more historical data; a smaller preset number results in more sensitive judgment, but is more susceptible to noise from a single time window.

[0033] The lengths of the preceding and following consecutive matching paths are used to measure whether the behavior of older adults is consistent with their individual historical norms before and after the conflict, respectively. The length of the preceding consecutive matching path reflects the normality of behavior before the conflict, while the length of the following consecutive matching path reflects whether behavior has returned to normal after the conflict.

[0034] The preset length threshold is a pre-defined parameter used to determine whether a continuous matching path is long enough.

[0035] When the lengths of two consecutive matching paths both reach or exceed the preset length threshold, it indicates that the behavior before and after the conflict is consistent with the elderly person's daily routine, and the conflict is isolated, thus it is judged as an instantaneous sensor anomaly. When the length of either path is insufficient, a transition not present in the personalized daily event map occurs, or the time interval between adjacent events exceeds the normal range, it indicates that the behavior has deviated from the norm, thus it is judged as a progressive real risk.

[0036] By employing the aforementioned personalized daily event graph matching method based on both preceding and subsequent events, combined with behavioral trajectory verification in both directions before and after a conflict, it is possible to effectively distinguish between instantaneous sensor anomalies and genuine progressive risks. This reduces misjudgments caused by noise in a single time window or occasional behavioral fluctuations, thereby improving the reliability of conflict nature determination. The determination results can be used to subsequently downweight the local risk assessment results corresponding to instantaneous sensor anomalies, or to upgrade the corresponding risk level of progressive genuine risks and trigger an escalation response.

[0037] According to an embodiment of the present invention, it further includes: When tracing back the conflict period along the timeline, the video function intelligent agent records the elderly person's limb movements and movement trajectory in real time, and the vital sign detection function intelligent agent records abnormal vital sign data. When the next execution time reversal and tracing occurs, the video function agent and the vital sign detection function agent will prioritize searching and matching the recorded limb movements, movement trajectories, and abnormal vital sign data of the elderly. If a match is found, the conflict period will be classified as a non-conflict period.

[0038] It should be noted that, during the time reversal and tracing process, the video function agent and the vital sign detection function agent respectively record the characteristic data of the historical conflict period for the behavioral modality and the physiological modality. Among them, limb movements and movement trajectories can be extracted through video analysis, posture estimation, target tracking, etc., and abnormal vital sign data can be collected through wearable devices or bedside monitoring devices; the above recorded data can serve as a sample library of the elderly person's individualized behavior and physiological characteristics. When a conflict period occurs again, the body movements, movement trajectories, and abnormal vital signs data of the current conflict period are first matched with the recorded historical feature samples. Matching can be achieved through feature similarity calculation, such as calculating the similarity of trajectory shape, consistency of movement type, and correlation of vital sign change trends. If the match is consistent, it indicates that the behavior and physiological manifestations of the current conflict period are highly similar to the historically recorded situation. In this case, the conflict period is more likely to be caused by the same known cause and can be classified as a non-conflict period. This avoids repeatedly entering into debate, negotiation, or escalation of response for repetitive patterns, thereby improving overall decision-making efficiency and reducing the false alarm rate.

[0039] According to an embodiment of the present invention, on a shared blackboard, multiple functional agents challenge the local risk assessment results of other functional agents based on evidence from their own domains of responsibility and determine the intensity of the challenge. A fusion decision agent dynamically adjusts the weights of each local risk assessment result based on the intensity of the challenge, including: The challenge is valid when the functional agent that raises the challenge can provide original perceptual evidence within its domain of responsibility that contradicts the challenged local risk assessment result, and the original perceptual evidence covers the time period on which the challenged local risk assessment result is based. The challenge is invalid if the functional agent raising the challenge fails to provide original perceptual evidence within its domain of responsibility that contradicts the challenged local risk assessment result, or if the original perceptual evidence does not cover the time period on which the challenged local risk assessment result is based. When a challenge is valid, the fusion decision-making agent will reduce the weight of the challenged local risk assessment result according to the intensity of the challenge, and the reduced weight will not be lower than the preset weight lower limit, and the challenge information will be recorded on the shared blackboard. When the challenge is found to be invalid, the fusion decision-making agent maintains the current weight of the challenged local risk assessment result.

[0040] It should be noted that the original perceptual evidence within the domain of responsibility refers to the original data collected by the sensors or data sources under the jurisdiction of the functional agent raising the question within the scope of its own responsibility and can be directly cited, rather than the results processed or paraphrased by other agents. For example, the behavior monitoring agent can only cite behavior sensor data such as vision, infrared, and pressure, and cannot directly cite the heart rate or blood oxygen data of the physiological monitoring agent. The time period on which the challenged local risk assessment results are based means that the time range of the original perception evidence collection should correspond to the time period of the original data on which the challenged local risk assessment results were based, rather than providing evidence outside the conflicting time period or only a part of the time period. The determination of the intensity of the challenge takes into account the confidence level of the contradictory evidence itself, the degree of deviation between the contradictory evidence and the challenged assessment in terms of risk confidence, and the proportion of the contradictory evidence covering the challenged period in the time window. The higher the confidence level, the greater the degree of deviation, and the higher the coverage ratio, the greater the intensity of the challenge, and the greater the reduction in the weight of the challenged assessment. The preset lower limit of weight is used to prevent the local risk assessment results of individual functional agents from being overly downgraded to the point of losing their eligibility to participate in the fusion due to being questioned, thereby maintaining the basic participation of multi-source evidence in the subsequent blackboard evidence debate and negotiation; The shared blackboard is a public data space where various functional agents share questioning information, evidence indexes, and weight adjustment records. After questioning information is recorded in the shared blackboard, other functional agents can check the basis and processing results of the questioning as a reference for their subsequent questioning, thus avoiding repeated questioning of the same content or ignoring valid questions. By setting the above conditions for valid challenges, invalid challenges that lack original evidence or have mismatched timeframes can be filtered out during multi-agent debate and negotiation. This ensures that dynamic weighting is only affected by real, traceable, time-related, and quantifiable evidence, thereby improving the convergence speed and reliability of consensus risk level calculation.

[0041] According to an embodiment of the present invention, based on the comprehensive risk level of the elderly, a hierarchical response agent performs comprehensive risk level matching and executes the corresponding hierarchical response strategy to obtain the execution result, including: The tiered response strategy is divided into three levels: Level 1 Response: Local notification, pushing comforting or reminder messages to the elderly's terminals; Level 2 Response: Send alarm notifications to the guardian's terminal and the caregiver's terminal, and initiate on-site verification; Level 3 Response: Based on Level 2 Response, it involves coordinating at least one of the following: medical emergency, fire, or community rescue, and continues to monitor until the risk is eliminated; The overall risk level includes four levels: normal, attention, early warning, and emergency. The correspondence between these levels and the tiered response strategies is as follows: Normal and attention correspond to Level 1 response, early warning corresponds to Level 2 response, and emergency corresponds to Level 3 response; Among them, "normal" means no risk, "attention" means continuous observation is required, "warning" means that guardians or caregivers need to be notified, and "emergency" means that external rescue resources need to be coordinated. The tiered response strategy has escalation and degrade mechanisms: When a high-level response is not confirmed within a preset time or the risk continues to worsen, it will be automatically upgraded; when the risk level drops and is confirmed to be safe after review, it will be downgraded level by level and the record will be archived.

[0042] It should be noted that the overall risk level is represented by a risk range using an upper risk bound and a lower risk bound. The lower risk bound is determined by the majority modal evidence that supports each other, while the upper risk bound is determined by the minority modal evidence that is conflicting. The difference between the upper risk bound and the lower risk bound is the width of the uncertainty range. The correspondence between the overall risk level and the tiered response strategy is determined in the following way: Pre-set a first risk threshold, a second risk threshold, and a third risk threshold; for example, these can be set based on historical risk event statistics, expert experience, or system operation strategies; divide the risk range into normal range, attention range, warning range, and emergency range; when the upper limit of the risk range is in the normal range or attention range, it corresponds to a level one response; when the upper limit of the risk range is in the warning range, it corresponds to a level two response; when the upper limit of the risk range is in the emergency range, it corresponds to a level three response.

[0043] According to an embodiment of the present invention, when a high-level response is not confirmed within a preset time period or the risk continues to worsen, automatic escalation is performed, including: The rate of risk deterioration is calculated based on the change in the comprehensive risk level of the elderly over time. When the rate of risk deterioration exceeds the rate threshold and the estimated arrival time of the nearest dispatchable resources exceeds the duration threshold, while maintaining the current level of response actions, the notification and resource reservation actions in the next level of response actions will be initiated in advance, while the on-site intervention and external rescue dispatch actions in the next level of response actions will not be executed for the time being. When the overall risk level is subsequently upgraded to the next level, the completed notifications and resource reservation results will be directly taken over, and only the remaining response actions that have not yet been executed in that level will be executed. When the overall risk level does not escalate and the risk decreases, cancel the initiated notifications and resource reservations and archive them, so that the response strength is continuously interpolated as the risk evolves rather than discretely jumping.

[0044] It should be noted that the preset duration is set by those skilled in the art based on actual needs.

[0045] The risk deterioration rate is calculated by obtaining the risk upper bound of the comprehensive risk level output in the current fusion cycle and the risk upper bound of the comprehensive risk level output in the previous fusion cycle; the risk upper bound of the current fusion cycle is subtracted from the risk upper bound of the previous fusion cycle, and then divided by the time interval between two adjacent fusion cycles to obtain the risk deterioration rate; the larger the risk deterioration rate, the faster the wind direction develops in an unfavorable direction. Dispatchable resources refer to the response forces that can be mobilized at any given time to respond to tiered response strategies, including at least one of nursing personnel, medical emergency forces, fire fighting forces, or community rescue forces.

[0046] The estimated arrival time of the nearest available dispatchable resource refers to the estimated time required for the nearest dispatchable resource to reach the target elderly person's location from the current moment. This time can be calculated based on the real-time location, task status, and dispatching strategy of the resource; for example, if the nearest caregiver is handling other events, that caregiver is not dispatchable, and the next available caregiver or resource should be selected for estimated arrival time calculation. Notification actions include sending early warning notifications to guardian terminals, caregiver terminals, or relevant institutions; resource reservation actions include initiating resource pre-notifications or reservations to medical emergency, fire, or community rescue institutions; on-site intervention actions include the arrival of caregivers to handle the situation; external rescue dispatch actions include the actual dispatch of ambulances or fire-fighting forces. Initiating notification actions and resource reservation actions in advance will not directly occupy on-site handling resources, but can significantly shorten the time required for the next level of response to take effect.

[0047] When the overall risk level is subsequently upgraded to the next level, since the notification and appointment have already been completed, the aforementioned results can be directly inherited, and only the remaining response actions that have not yet been executed at this level can be performed, thereby avoiding starting the response process from scratch; when the overall risk level is not upgraded and the risk decreases, the initiated notifications and resource appointments are revoked and archived to avoid unnecessary disturbance to relevant personnel and institutions, while releasing the reserved resources.

[0048] As a result, the response intensity no longer changes discretely according to the first-level response, second-level response, and third-level response, but smoothly transitions between adjacent levels as the risk evolves, realizing continuous interpolation-based response control, improving response timeliness, and reducing resource waste caused by false escalation.

[0049] According to an embodiment of the present invention, the execution results and changes in the elderly person's state perception data are fed back to the fusion decision agent to update the individual historical baseline of the elderly person and optimize the individual historical baseline of all elderly persons. Simultaneously, the professional analysis model parameters of multiple functional agents are optimized, including: The anonymized individual anomaly patterns are used as anomaly prototypes and uploaded to the global anomaly pattern library. After being filtered by the global abnormal pattern library, the abnormal prototypes are migrated to the edge computing nodes of other elderly people in the same group as the target elderly people, so that the other edge computing nodes have the ability to identify abnormal patterns during the cold start phase. The anomalous prototypes migrated to the target edge computing node are assigned confidence weights that decrease as the target individual's own sample accumulation increases, forming a dual-loop optimization of personalized and swarm intelligence to optimize the professional analysis model parameters of multiple functional agents.

[0050] It should be noted that changes in the state perception data of older adults include: Changes in physiological data: such as whether indicators like heart rate, blood oxygen, and respiratory rate return to normal range after the response is executed; Changes in behavioral activity data: such as whether activity level, posture, and position have returned to normal; The screening criteria include that the supporting sample size of the abnormal prototype is not less than the preset minimum sample size, the proportion of events confirmed as real risk events after review is not less than the preset confirmation rate threshold, and the similarity of the target elderly person's attributes with the applicable population labeled by the abnormal prototype is not less than the preset similarity threshold. By migrating selected anomalous prototypes to edge computing nodes of other elderly individuals within the same group, the individual historical baseline is optimized at the group level, enabling edge computing nodes in the cold start phase to possess prior recognition capabilities for similar anomalous patterns. The confidence weight of the migrated anomalous prototype in the target edge computing node gradually decreases as the node's own sample accumulation increases, gradually transitioning from prototype-driven to individual-driven data, forming a dual-loop optimization that combines personalized and collective intelligence.

[0051] Figure 2 The flowchart of a multi-agent fusion hierarchical response elderly care safety management system provided by the present invention is shown.

[0052] like Figure 2 As shown, this invention discloses a multi-agent fusion hierarchical response elderly care safety management and control system, comprising: The multi-source sensing module is used to deploy various types of sensing devices to collect real-time status sensing data of the elderly. The multi-agent system assessment module is used to construct multiple functional agents to perform preliminary analysis on the elderly's state perception data and obtain multiple local risk assessment results with confidence levels. The fusion decision module is used to aggregate the results of various local risk assessments into the fusion decision intelligent agent, and to perform conflict detection and blackboard evidence debate and negotiation to obtain the comprehensive risk level of the elderly. The graded response module is used to match the comprehensive risk level of the elderly with the corresponding graded response strategy by the graded response agent and obtain the execution result. The execution feedback module is used to feed back the execution results and changes in the elderly's state perception data to the fusion decision agent, update the individual historical baseline of the elderly, optimize the individual historical baseline of all elderly, and optimize the professional analysis model parameters of multiple functional agents.

[0053] According to an embodiment of the present invention, a multi-source sensing module collects real-time state perception data of the elderly through various sensing devices deployed in elderly care facilities. The perception data includes at least two of the following: physiological signs data, behavioral activity data, and environmental state data. A multi-agent system judgment module constructs multiple functional agents, including at least a physiological monitoring agent, a behavioral detection agent, and an environmental monitoring agent. Each functional agent performs preliminary analysis on the data belonging to its responsibility domain in the state perception data based on its corresponding professional analysis model, and outputs a local risk judgment result with confidence. A fusion decision module aggregates the local risk judgment results of each functional agent by a fusion decision agent. Based on a unified spatiotemporal benchmark, it aligns the local risk judgment results to the same time axis and the same risk semantic space, detects cross-modal conflicts between the local risk judgment results in terms of time and semantics, and divides the time period into conflict-free time periods and conflict periods according to the detection results. It maps each local risk judgment result to a unified risk confidence representation and fuses the confidence of the conflict-free time periods using a preset method to obtain an initial consensus. For the conflict period, it traces back a preset number of consecutive time windows before the conflict period along the time sequence. The system reconstructs the event evolution trajectory from the original perceived signals within the system and determines whether the conflict originates from an instantaneous sensor anomaly or a gradual real risk based on this trajectory. Using initial consensus as the initial input, each functional agent challenges the local risk assessment results of other functional agents on a shared blackboard based on evidence from its own domain of responsibility. The fusion decision agent dynamically adjusts the weights of each local risk assessment result based on the intensity of the challenge to form a consensus risk level. The tiered response module matches and executes the corresponding tiered response strategy based on the comprehensive risk level. When the width of the uncertainty interval of the comprehensive risk level is higher than the preset difference threshold, an active confirmation sub-process is executed first to initiate a light interaction confirmation to the elderly person's terminal or the caregiver's terminal. If no confirmation is received or no response is received within the time limit, the tiered response strategy is escalated according to the risk level. The tiered response strategy includes at least a first-level response, a second-level response, and a third-level response. The execution feedback module collects the execution results of the tiered response strategy and sends the execution results and changes in the elderly person's state perception data back to the fusion decision agent to update the individual historical baseline of the elderly person and optimize the individual historical baseline of all elderly persons, while also optimizing the professional analysis model parameters of multiple functional agents.

[0054] All information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals (including but not limited to signals transmitted between user terminals and other devices) involved in this application have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the "elderly person's state perception data" and "risk assessment results of various functional intelligent agents" involved in this disclosure were obtained with full authorization.

[0055] This invention discloses a multi-agent fusion hierarchical response method and system for elderly care safety management and control. It can overcome the shortcomings of existing elderly care safety monitoring, such as high false alarms in single-modality monitoring, difficulty in resolving multi-source conflicts, lack of hierarchical and confirmation response, and lack of closed-loop handling. It provides a multi-agent fusion hierarchical response method and system for elderly care safety management and control to improve the accuracy and robustness of risk identification in modal conflict scenarios, and accurately match handling resources according to the degree of risk, forming a sustainable and optimized safety management and control closed loop.

[0056] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0057] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0058] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0059] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A multi-agent fusion hierarchical response method for elderly care safety management, characterized in that, include: Deploy various types of sensing devices to collect real-time status perception data of the elderly; Multiple functional intelligent agents are constructed to perform preliminary analysis on the state perception data of the elderly, and obtain multiple local risk assessment results with confidence levels. The results of the individual risk assessments are aggregated into a fusion decision-making intelligent agent, and conflict detection and blackboard evidence debate and negotiation are performed to obtain the comprehensive risk level of the elderly. Based on the comprehensive risk level of the elderly, the hierarchical response agent matches the comprehensive risk level and executes the corresponding hierarchical response strategy to obtain the execution result; The execution results and changes in the elderly person's state perception data are fed back to the fusion decision agent to update the individual historical baseline of the elderly person, optimize the individual historical baseline of all elderly persons, and optimize the professional analysis model parameters of the multiple functional agents.

2. The multi-agent fusion hierarchical response method for elderly care safety management and control according to claim 1, characterized in that, The results of the individual risk assessments are aggregated into a fusion decision-making agent, and conflict detection and blackboard evidence debate and negotiation are performed to obtain the comprehensive risk level of the elderly, including: Based on a unified spatiotemporal benchmark, the results of each local risk assessment are aligned to the same time axis and the same risk semantic space. Cross-modal conflicts in time and semantics between the results of each local risk assessment are detected, and the time period is divided into conflict-free time period and conflict time period according to the detection results. The results of each local risk assessment are mapped to a unified risk confidence level. The risk confidence levels for the conflict-free period are then fused using a preset method to obtain an initial consensus. For the conflict period, the original sensing signals within a preset number of consecutive time windows prior to the conflict period are traced back along the time sequence to reconstruct the event evolution trajectory, and the conflict period is determined based on the evolution trajectory to be caused by instantaneous sensor anomalies or gradual real risks. Using the initial consensus as the initial input, on the shared blackboard, the multiple functional agents question the local risk assessment results of other functional agents based on evidence from their own domains of responsibility and determine the intensity of the questioning. The fusion decision agent dynamically adjusts the weights of each local risk assessment result based on the intensity of the questioning and calculates the comprehensive risk level of the elderly person.

3. The multi-agent fusion hierarchical response method for elderly care safety management and control according to claim 2, characterized in that, The reconstructed event evolution trajectory, and the determination based on the evolution trajectory whether the conflict period originated from a transient sensor anomaly or a gradual real risk, includes: Based on the time sequence of the elderly within a preset historical time period, a personalized daily event graph of the elderly is constructed. The personalized daily event graph is divided into nodes with daily events and edges with the transition relationships between events, and the normal time interval range of each transition relationship is marked. The original sensing signals within a preset number of consecutive time windows before the conflict period are identified as a preceding event sequence, and the original sensing signals within a preset number of consecutive time windows after the conflict period are identified as a subsequent event sequence. The preceding event sequence and the following event sequence are matched with the personalized daily event graph to obtain the preceding continuous matching path length and the following continuous matching path length. When the length of the preceding consecutive matching path is greater than or equal to a preset length threshold, and the length of the following consecutive matching path is greater than or equal to the preset length threshold, the conflict is determined to originate from the instantaneous sensor malfunction. The conflict is determined to originate from the asymptotic real risk when any of the following conditions are met: The preceding consecutive matching path length is less than the preset length threshold; the following consecutive matching path length is less than the preset length threshold; there is a transition in the preceding event sequence that is not present in the personalized daily event graph; there is a transition in the following event sequence that is not present in the personalized daily event graph; the time interval between adjacent events in the preceding event sequence or the following event sequence exceeds the corresponding normal time interval range.

4. The multi-agent fusion hierarchical response method for elderly care safety management and control according to claim 3, characterized in that, Also includes: When tracing back the conflict period along the timeline, the video function agent records the elderly person's limb movements and movement trajectory in real time, and the vital sign detection function agent records abnormal vital sign data. When the time reversal tracing is performed again, the video function agent and the vital sign detection function agent will first search for and match the recorded limb movements, movement trajectories, and abnormal vital sign data of the elderly. If a match is found, the conflict period will be classified as a non-conflict period.

5. The multi-agent fusion hierarchical response method for elderly care safety management and control according to claim 2, characterized in that, The process involves multiple functional agents on a shared blackboard challenging the local risk assessment results of other functional agents based on evidence from their own domains of responsibility and determining the intensity of the challenges. The fusion decision agent then dynamically adjusts the weights of each local risk assessment result based on the intensity of the challenges, including: The challenge is valid when the functional agent that raises the challenge can provide original perceptual evidence within its domain of responsibility that contradicts the challenged local risk assessment result, and the original perceptual evidence covers the time period on which the challenged local risk assessment result is based. The challenge is invalid if the functional agent raising the challenge fails to provide original perceptual evidence within its domain of responsibility that contradicts the challenged local risk assessment result, or if the original perceptual evidence does not cover the time period on which the challenged local risk assessment result is based. When the challenge is valid, the fusion decision-making agent will reduce the weight of the challenged local risk assessment result according to the intensity of the challenge, and the reduced weight will not be lower than the preset weight lower limit, and the challenge information will be recorded in the shared blackboard. When the challenge is found to be invalid, the fusion decision-making agent maintains the current weight of the challenged local risk assessment result unchanged.

6. The multi-agent fusion hierarchical response method for elderly care safety management and control according to claim 1, characterized in that, Based on the comprehensive risk level of the elderly person, the hierarchical response agent matches the comprehensive risk level and executes the corresponding hierarchical response strategy to obtain the execution result, including: The hierarchical response strategy is divided into three levels: Level 1 Response: Local notification, pushing comforting or reminder messages to the elderly's terminals; Level 2 Response: Send alarm notifications to the guardian's terminal and the caregiver's terminal, and initiate on-site verification; Level 3 Response: Based on the Level 2 response, coordinate with at least one of the following: medical emergency, fire, or community rescue, and continue to monitor until the risk is eliminated; The comprehensive risk level includes four levels: normal, attention, early warning, and emergency. The correspondence between these levels and the tiered response strategy is as follows: Normal and attention correspond to the Level 1 response, early warning corresponds to the Level 2 response, and emergency corresponds to the Level 3 response; Among them, "normal" means no risk, "attention" means continuous observation is required, "early warning" means that guardians or caregivers need to be notified, and "emergency" means that external rescue resources need to be coordinated. The tiered response strategy has escalation and degrade mechanisms: When a high-level response is not confirmed within a preset time or the risk continues to worsen, it will be automatically upgraded; when the risk level drops and is confirmed to be safe after review, it will be downgraded level by level and the record will be archived.

7. The multi-agent fusion hierarchical response method for elderly care safety management and control according to claim 6, characterized in that, The automatic escalation when a high-level response is not acknowledged within a preset time or the risk continues to worsen includes: The rate of risk deterioration is calculated based on the change in the overall risk level of the elderly over time. When the rate of risk deterioration exceeds the rate threshold and the estimated arrival time of the nearest schedulable disposal resource exceeds the duration threshold, while maintaining the execution of the current level response actions, the notification and resource reservation actions in the next level response actions are initiated in advance, while the on-site intervention and external rescue dispatch actions in the next level response actions are temporarily suspended. When the overall risk level is subsequently upgraded to the next level, the completed notifications and resource reservation results are directly inherited, and only the remaining response actions that have not yet been executed in that level are executed; When the overall risk level does not escalate and the risk decreases, the initiated notifications and resource reservations are revoked and archived, so that the response strength is continuously interpolated as the risk evolves rather than discretely jumping.

8. The multi-agent fusion hierarchical response method for elderly care safety management and control according to claim 1, characterized in that, The execution result and the changes in the elderly person's state perception data are fed back to the fusion decision agent to update the individual historical baseline of the elderly person and optimize the individual historical baseline of all elderly people. Simultaneously, the professional analysis model parameters of the multiple functional agents are optimized, including: The anonymized individual anomaly patterns are used as anomaly prototypes and uploaded to the global anomaly pattern library. After being filtered by the global abnormal pattern library, the abnormal prototype is migrated to the edge computing nodes of other elderly people in the same group as the target elderly person, so that the other edge computing nodes have the ability to identify the abnormal pattern during the cold start phase. The anomalous prototypes migrated to the target edge computing node are assigned confidence weights that decrease as the target individual's own sample accumulation increases, forming a dual-loop optimization of personalization and swarm intelligence to optimize the professional analysis model parameters of the multiple functional agents.

9. A multi-agent fusion hierarchical response elderly care safety management system, applied to the multi-agent fusion hierarchical response elderly care safety management method according to any one of claims 1-8, characterized in that, include: The multi-source sensing module is used to deploy various types of sensing devices to collect real-time status sensing data of the elderly. The multi-agent system assessment module is used to construct multiple functional agents to perform preliminary analysis on the elderly person's state perception data and obtain multiple local risk assessment results with confidence levels. The fusion decision module is used to aggregate the results of the local risk assessments into the fusion decision agent, and to perform conflict detection and blackboard evidence debate and negotiation to obtain the comprehensive risk level of the elderly. The graded response module is used to match the comprehensive risk level of the elderly person with the graded response agent and execute the corresponding graded response strategy to obtain the execution result. The execution feedback module is used to send the execution results and changes in the elderly person's state perception data back to the fusion decision agent, update the individual historical baseline of the elderly person, optimize the individual historical baseline of all elderly persons, and optimize the professional analysis model parameters of the multiple functional agents.

Images