An old-age home operation optimization and risk management method based on digital twinning

Abstract

The application discloses a kind of based on digital twinning old-age home operation optimization and risk management method, it is related to wisdom old-age care technical field.The method includes: obtaining the individual vulnerability response relationship of each in-hospital old person, the relationship is the mapping from environmental disturbance parameter to behavior instability degree;According to the preset period, generate the combination of antagonistic environmental disturbance parameter;In digital twin system, the expected instability degree of all in-hospital old people under the action of the disturbance combination is calculated in parallel, obtain the group vulnerability distribution, and determine the in-hospital old person with the highest expected instability degree as the key care object from it;Generate camouflage care task, and with minimizing the total care service delay cost as the goal, assign the task to the optimal caregiver and insert it into its task sequence;After task execution, the vulnerability response relationship is incrementally updated according to real behavior data.The present application is helpful for the transformation of care resources from after-response to pre-protection.

Description

Technical Field

[0001] This invention relates to the field of smart elderly care technology, and more specifically, to a method for optimizing the operation and managing risks of elderly care institutions based on digital twins. Background Technology

[0002] As society ages, elderly care institutions face increasing pressure in managing operational safety. Safety incidents such as falls, sudden illnesses, and unusual wandering are often unexpected, while the allocation of caregivers is usually relatively fixed. How to achieve timely control of safety risks under limited resources has become a real problem in the operation of elderly care institutions.

[0003] In recent years, digital twin technology has been gradually introduced into the management of elderly care institutions. Chinese patent CN119152640A discloses a safety risk management system for elderly care institutions based on digital twin technology. This system integrates video images, meteorological data, and behavioral characteristics of the elderly to identify environmental hazards such as water stains on the ground and glass doors, and performs correlation analysis with the posture and gait of the elderly. When fall characteristics or accidental event characteristics are detected, a warning command is generated and an alarm is triggered.

[0004] The aforementioned existing technologies have the following shortcomings: First, the scheduling of care resources occurs after the risk is detected. There is a response delay from the system's identification of the anomaly to the arrival of caregivers on site. For events such as falls that can cause serious consequences in a short period of time, the delay may lead to aggravation of the injury. Second, in existing solutions, digital twins mainly serve the functions of visualizing the physical environment and monitoring the status. They do not play a connecting role between the pre-simulation of risk evolution and the decision-making on the scheduling of care resources, and cannot complete the pre-deployment of resources while the elderly are still in a normal state. Therefore, a digital twin-based method for optimizing the operation and managing risks of elderly care institutions is proposed to address the above issues. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for optimizing the operation and managing risks of elderly care institutions based on digital twins. This method aims to address the problems in existing risk management methods for elderly care institutions where the scheduling of care resources lags behind the occurrence of risks, and where digital twins fail to establish a connection between risk pre-simulation and resource scheduling.

[0006] To achieve the above objectives, this invention provides a method for optimizing the operation and managing risks of elderly care institutions based on digital twins, comprising the following steps: S1: Obtain the individual vulnerability response relationship for each elderly resident in the facility. This vulnerability response relationship is a mapping from environmental disturbance parameters to the degree of behavioral instability of the elderly resident. Establishing this mapping relationship allows for the calculation of the corresponding degree of behavioral instability for any given environmental disturbance parameter input, thereby reflecting the behavioral stability characteristics of different elderly individuals under various environmental perturbations.

[0007] Furthermore, the environmental disturbance parameters are composed of the rate of change of environmental parameters collected by non-contact sensors. These environmental parameters include at least two of the following: spatial illuminance, regional pedestrian density, background sound pressure level, and temperature and humidity. Describing these environmental parameters in the form of rates of change can capture the dynamic fluctuation characteristics of the environmental state and more accurately characterize the environmental micro-disturbances that may cause behavioral instability in the elderly.

[0008] Furthermore, the process of obtaining the individual vulnerability response relationship is as follows: Obtain the sequence of environmental disturbance parameters and the corresponding behavioral instability labels within a preset time window prior to the occurrence of historical safety incidents involving the elderly person in the facility; Using the static vulnerability characteristics of the elderly person in the hospital as a condition, the mapping from the environmental disturbance parameter sequence to the behavioral instability degree label is learned to obtain the individual vulnerability response relationship.

[0009] Through the above learning process, the establishment of vulnerability response relationship fully considers the elderly person's own static vulnerability characteristics such as age, fall history, and cognitive impairment level, thereby realizing the representation of different individual-differentiated stress response patterns.

[0010] Furthermore, the behavioral instability label is calculated from at least two of the following indicators for the elderly person in the hospital within the preset time window: gait symmetry deviation, movement trajectory entropy, and abnormal dwell time in the designated safe area.

[0011] The above indicators reflect the degree to which the behavior of the elderly in the hospital deviates from their daily baseline from three different dimensions: postural stability, movement regularity, and spatial behavior pattern. The comprehensive use of multiple indicators helps to quantify the behavioral instability of the elderly more comprehensively.

[0012] S2: According to a preset cycle, under the condition of satisfying the physical constraints of the internal space of the elderly care institution, generate a combination of adversarial environmental disturbance parameters that increases the highest degree of behavioral instability among the elderly population in the institution.

[0013] This combination of adversarial environmental perturbation parameters is not a description of the real environmental state, but rather a set of virtual environmental parameter change rate combinations actively constructed by the system in the digital twin space. Its purpose is to apply a standardized virtual stress test to all elderly residents in the hospital in order to identify the most vulnerable individuals in the group.

[0014] Furthermore, the process of generating the adversarial environmental disturbance parameter combination is as follows: Obtain the sensitivity coefficient of each elderly resident to different environmental parameters; Within the feasible perturbation domain that satisfies the aforementioned spatial physical constraints, generate a value that satisfies the target value. The combination of environmental parameter change rates that is maximized is taken as the combination of adversarial environmental disturbance parameters, wherein... For the first Sensitivity coefficient of elderly residents in the facility For the first The expected instability of elderly residents in the facility.

[0015] This target value reflects the sum of the instability of all elderly residents under sensitivity weighting. By maximizing it, the generated combination of adversarial environmental disturbance parameters can simultaneously stimulate the responses of multiple elderly residents to their respective sensitive environmental factors, thereby maximizing the difference in vulnerability within the group.

[0016] Furthermore, the sensitivity coefficient is extracted from the individual vulnerability response relationship of the elderly person in the hospital, characterizing the intensity of their behavioral instability response to changes in environmental parameters of different dimensions. The higher the sensitivity coefficient, the more sensitive the elderly person in the hospital is to changes in the corresponding environmental parameters, and the more likely they are to exhibit behavioral instability under similar environmental perturbations.

[0017] S3: In the digital twin system, the adversarial environmental disturbance parameters are combined and input into the vulnerability response relationship of each elderly person in the hospital. The expected instability of each elderly person in the hospital under the disturbance scenario is calculated in parallel to obtain the group vulnerability distribution. The elderly person in the hospital with the highest expected instability is determined from the group vulnerability distribution as the key care target for the current cycle.

[0018] This step, through a virtual survey of all elderly residents in the facility within a digital twin space, identifies the most vulnerable individuals within the group, providing a clear target for subsequent resource pre-deployment.

[0019] Furthermore, the process of identifying the key care recipients is as follows: Determine whether the expected instability of the elderly patient with the highest expected instability exceeds a preset instability threshold; If the threshold is exceeded, the elderly person will be identified as a key care recipient, indicating that the vulnerability shown by the elderly person in the current virtual stress test exceeds the normal fluctuation range before the occurrence of historical security events. If the time limit is not exceeded, the subsequent steps of the current cycle will be terminated, indicating that all elderly residents in the hospital currently have good anti-interference ability and there is no need to start resource pre-deployment.

[0020] S4: Based on the personal care file of the key care recipient, generate a disguised care task that involves close proximity to the area where the key care recipient is located.

[0021] This disguised care task is presented on the caregiver's terminal in the form of routine care services, hiding the purpose of security protection within the routine care process, so that caregivers can conduct close-range surveillance of high-risk elderly people while performing their normal care work.

[0022] Furthermore, the process of generating the disguised care task is as follows: Extract daily care preferences and habitual actions from the personal care records of the key care recipients; From the preset care task rule library, select one care event that matches the daily care preferences and habitual actions and includes the caregiver staying and observing in the designated area, and use it as the display content of the disguised care task.

[0023] Through the above matching process, the disguised care tasks are highly consistent with the daily routines of the elderly in terms of content, and no alarm signs or emergency markers are introduced on the caregiver's end, thus completing the pre-deployment of safety resources without creating an additional tense atmosphere.

[0024] S5: Based on the real-time location of the on-duty caregiver and the current care task sequence, with the goal of minimizing the sum of the delay costs of care services for all elderly people in the hospital, determine the target caregiver who will perform the spoofed care task and the insertion position of the task in the target caregiver's care task sequence, and send the spoofed care task to the target caregiver's terminal.

[0025] This step, through global optimization and scheduling, seamlessly integrates safety prevention tasks into daily care processes in a way that minimizes the impact on other care services throughout the hospital, balancing the timeliness of safety management with the continuity of care services.

[0026] Furthermore, the process of determining the target caregiver and insertion location is as follows: A care scheduling optimization model is constructed, which aims to minimize the weighted sum of the delay times of all care tasks, and sets a higher safety weight for the spoofed care tasks than for regular care tasks. Based on the real-time location of each on-duty caregiver and the current care task sequence, the care scheduling optimization model is solved to obtain the target caregiver assignment and task sequence insertion position of the masquerade care task.

[0027] By assigning higher safety weights to surreptitious care tasks, the scheduling model prioritizes the timely execution of these tasks during the solution process, even if other routine care tasks may experience slight delays, thereby ensuring the timeliness of safety protection at the global optimization level.

[0028] S6: Within a preset time window after the execution of the disguised care task, obtain the real behavioral data of the key care recipient.

[0029] If the behavioral instability of the key care recipient does not exceed the preset instability threshold, it indicates that the current investigation is a false alarm. The disturbance amplitude of the subsequent generated adversarial environmental disturbance parameter combination is reduced to suppress excessive investigation and reduce unnecessary scheduling overhead.

[0030] If the behavioral instability of the key care recipient exceeds the preset instability threshold and subsides after intervention via the surreptitious care task, it indicates that the investigation was an effective early warning. The elderly person did indeed experience a brief period of behavioral instability, but it was resolved in time because the caregiver was already nearby. At this point, using the combination of environmental disturbance parameters and the actual behavioral instability data, the individual vulnerability response relationship of the key care recipient is incrementally updated, making the prediction of similar disturbance patterns more accurate.

[0031] The technical effects and advantages of this invention are as follows: First, addressing the issue of existing technologies where care resource allocation lags behind risk occurrence, this invention constructs individual vulnerability response relationships and periodically generates adversarial environmental disturbance parameter combinations to proactively conduct parallel stress tests on the digital twins of all elderly residents in a digital twin space. This method does not rely on passively predicting the future behavior of the elderly; instead, it identifies the most vulnerable care recipients in the system by observing which elderly person first exhibits behavioral instability under the same virtual disturbance scenario. This approach helps shift the focus of safety management from "who will be at risk" to "who is least vulnerable to risk," moving the timing of care resource deployment from reactive response to proactive prevention.

[0032] Secondly, addressing the issue in existing technologies where digital twins fail to bridge the gap between risk pre-simulation and resource allocation, this invention generates disguised care tasks, packaging safety precautions as routine care services such as providing drinking water and environmental inspections. These tasks do not include alarm indicators on the caregiver's terminal, but their execution location and time are precisely designed, allowing caregivers to conduct close-range surveillance of high-risk elderly individuals while performing their regular care duties. This approach transforms digital twins from a visual monitoring tool into a proactive element in scheduling decisions, facilitating precise pre-positioning of care resources without disrupting the normal operation of elderly care institutions.

[0033] Third, this invention constructs a feedback loop to continuously observe the elderly person's actual behavior after the sham care task is performed. In cases of false alarms, the system suppresses and adjusts the generation of subsequent adversarial environmental disturbance parameter combinations; in cases of valid warnings, the system uses the current disturbance parameter combination and actual behavioral instability data as incremental samples to incrementally update the individual vulnerability response relationship. Through the co-evolution of the adversarial disturbance generation strategy and the individual vulnerability model, the system's accuracy in capturing early signs of behavioral instability in the elderly person is gradually improved. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention; Figure 2 A flowchart for constructing the individual vulnerability response relationship of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Example 1 As attached Figures 1 to 2 The method for optimizing the operation and managing risks of elderly care institutions based on digital twins is shown and runs on the digital twin platform already built by the elderly care institution.

[0037] The platform has been connected to the following systems: environmental sensor network, in-hospital elderly positioning and behavior perception system, caregiver terminal, and care task management system.

[0038] An environmental sensor network is used to collect environmental parameter data for each functional area. An in-hospital elderly positioning and behavior perception system is used to acquire real-time location and behavioral characteristic data for each elderly resident. Caregiver terminals are used to receive care task instructions and provide feedback on execution status. A care task management system is used to maintain the current care task sequence and execution progress for each on-duty caregiver.

[0039] The specific implementations of the above systems belong to the existing technology in this field.

[0040] S1: Obtain the individual vulnerability response relationship for each elderly resident in the facility. A vulnerability response relationship is established for each elderly resident in the facility, which defines a mapping from environmental disturbance parameters to the degree of behavioral instability.

[0041] The environmental disturbance parameter is a multi-dimensional vector, composed of the rate of change of environmental parameters collected by non-contact sensors. The environmental parameters include at least two of the following: spatial illuminance, area pedestrian density, background sound pressure level, and temperature and humidity.

[0042] Take spatial illuminance as an example. Light sensors deployed in corridors, activity rooms, and living rooms collect illuminance values ​​at a fixed frequency. The difference between two consecutive values ​​divided by the time interval is the rate of change of spatial illuminance.

[0043] The rate of change of population density in the area is obtained by dividing the difference in the number of people in the area at adjacent moments, as statistically analyzed by millimeter-wave radar, by the time interval. The rate of change of background sound pressure level is calculated in the same way by an environmental microphone. The rate of change of temperature and humidity is calculated in the same way by an environmental temperature and humidity sensor.

[0044] The degree of behavioral instability is a quantitative measure of how much an elderly person deviates from their daily behavioral baseline. A value of 0 indicates complete consistency with the daily baseline. A value of 1 indicates that the degree of deviation has reached the maximum value within the historical observation period.

[0045] To establish the mapping relationship, the system first collects historical data for each elderly person.

[0046] The data comes from past safety incidents involving the elderly person, including falls, sudden physical discomfort, or prolonged abnormal wandering. For each incident, the system extracts the environmental disturbance parameter sequence and corresponding behavioral instability labels from 15 minutes prior to the time of the incident. The sampling interval is 1 second, with a total of 900 sampling points.

[0047] The degree of behavioral instability is calculated from at least two of the following three indicators: gait symmetry deviation, movement trajectory entropy, and abnormal dwell time in a specified safe area.

[0048] Gait symmetry deviation measures changes in walking posture. The system collects gait point cloud data through pressure-sensing flooring or LiDAR and extracts symmetry indices for the left and right gait cycles. The average symmetry index for the first 30 days after admission is used as the daily baseline. The system calculates the absolute value of the difference between the current value and the baseline value. Then, it uses maximum-minimum normalization, with the historical maximum and minimum values ​​within the 30-day observation period as the reference range.

[0049] The trajectory entropy measures the degree of disorder in a movement path. The system collects a sequence of movement trajectory points using positioning sensors and calculates the Shannon entropy. The average trajectory entropy over all 15-minute time windows within the previous 30 days is used as the baseline value. The system calculates the absolute value of the difference between the current value and the baseline value, and then performs normalization on the difference.

[0050] The system specifies abnormal dwell time in designated safe zones, targeting high-risk areas such as restrooms. It calculates the duration of a single, continuous stay by an elderly person in such areas, then subtracts the average dwell time for the same time period. The average dwell time is the average of dwell times in the same area during the previous 30 days. Positive differences are normalized.

[0051] The three indicators are normalized and then weighted and summed to serve as a label for the degree of behavioral instability.

[0052] The weights of each indicator were determined as follows: Pearson correlation analysis was performed on the values ​​of each indicator and the event labels in all historical safety incidents involving the elderly person. The absolute values ​​of the correlation coefficients of each indicator were taken. These absolute values ​​were then normalized so that the sum of the weights was 1.

[0053] After collecting information on at least 50 historical security incidents involving the elderly person, the system began to establish vulnerability response relationships.

[0054] As one implementation method, the system employs a conditional regression model. The model uses static vulnerability features as conditional variables as inputs. Static vulnerability features include at least one of the following: age, history of falls, level of cognitive impairment, and balance ability score.

[0055] During training, the sequence of environmental perturbation parameters within a 15-minute time window is used as the independent variable. The corresponding behavioral instability level label is used as the supervision signal. The model outputs the predicted value, and the mean squared error between the predicted and label values ​​is calculated. Then, the parameters are updated along the gradient direction that reduces the error.

[0056] After multiple iterations, training is complete when the mean squared error converges or reaches the preset number of training rounds. The model is stored as a file. Inputting environmental perturbation parameters will output the corresponding expected instability.

[0057] As another implementation method, the system uses a multidimensional lookup table.

[0058] First, the rate of change of each environmental parameter is discretized according to a preset interval. For example, the historical range of the rate of change of illuminance is -100 lux per second to 100 lux per second. This can be evenly divided into 5 intervals, corresponding to five levels: "sudden darkening", "significant darkening", "remaining basically unchanged", "significant brightening", and "sudden brightening". Other environmental parameters are processed similarly.

[0059] The Cartesian product of each dimension interval constitutes a finite number of discretized combinations. For each combination, the system retrieves all corresponding sampling points from historical data. The average value of the behavioral instability labels of these sampling points is taken as the output value of that combination.

[0060] Then, using the static fragility feature as the index key, the correspondence between the discretized combination and the average instability output value is organized into a lookup table.

[0061] During a query, the system maps continuous input values ​​to corresponding discrete intervals. If a value is outside the range, it is mapped to the nearest boundary interval. The corresponding output value is then read based on the static vulnerability feature index.

[0062] S2: Generate a combination of adversarial environmental disturbance parameters according to a preset period. The periodic probing phase runs every 10 minutes. The goal is to synthesize a set of adversarial environmental perturbation parameters. This can be understood as the system proactively constructing a harsh but plausible environmental scenario within a digital twin space to test which elderly person is most likely to become unstable.

[0063] S2 is executed in two phases.

[0064] The first stage involves extracting the sensitivity coefficient for each elderly person.

[0065] The system measures sensitivity dimension by dimension from the vulnerability response relation. Taking spatial illuminance as an example: the rate of change of other dimensions is set to zero, and only a small unit increment perturbation is applied to illuminance. This perturbation vector is input into the vulnerability response relation to obtain the output value. The difference between the output values ​​with and without perturbation is calculated. The ratio of this difference to the unit increment is the sensitivity coefficient of the elderly person's illuminance.

[0066] The higher the coefficient, the more sensitive the elderly person is to changes in light. Using the same method, the sensitivity coefficients for each dimension are measured one by one, forming a vector. The larger the component value of each dimension in the vector, the more sensitive the person is to that dimension. The magnitude of the vector reflects the overall sensitivity of the elderly person.

[0067] The second stage involves generating a combination of adversarial environmental perturbation parameters.

[0068] The generation process is constrained by the feasible perturbation domain. The feasible perturbation domain is a bounded closed region of the rate of change of each dimension. The upper and lower bounds are set based on the maximum range of actual change of each environmental parameter in the most recent 90 days.

[0069] The combination must also meet spatial physical constraints. That is, when the rate of change of illuminance in a certain area exceeds the threshold for sudden change in illuminance, the rate of change of pedestrian density in that area at the same moment must not exceed the threshold for sudden change in pedestrian flow. The two thresholds are each taken as the 95th percentile of the absolute value of their respective historical rate of change.

[0070] As one implementation method, the problem is transformed into an optimization solution.

[0071] The objective is to find a combination within the feasible perturbation domain that maximizes the sum of the products of the sensitivity coefficients and the expected instability for all elderly individuals. For each elderly individual, the sensitivity coefficient is taken as the vector magnitude. The expected instability is obtained from the vulnerability response relationship input to the candidate combination. The two are multiplied and summed. A larger sum indicates that the combination is more efficient at distinguishing vulnerable individuals.

[0072] The solution employs an iterative search, using the rate of change vector of the current real-world environment as the initial reference point.

[0073] In each round of the search, one candidate combination is generated in both the positive and negative directions for each dimension. The generation method involves increasing or decreasing the step size by one unit in the current dimension component, while keeping other dimensions unchanged. The initial step size is one-tenth of the feasible region range for that dimension.

[0074] Calculate the sum of the products corresponding to each candidate combination. Select the maximum value. If this maximum value is better than the current benchmark value, use it as the new benchmark and continue the search while maintaining the step size. If the values ​​of all candidate combinations are not better than the current benchmark value, reduce the step size by half.

[0075] The search terminates when the step size is reduced to less than one percent of the feasible region, or when the number of search rounds reaches 50. The current optimal benchmark point is output as the combination of adversarial environmental perturbation parameters.

[0076] As another implementation method, the system generates perturbation combinations through clustering.

[0077] Calculate the cosine similarity of the sensitivity coefficient vectors of every two elderly individuals. Elderly individuals with a similarity greater than 0.8 are grouped into the same sensitivity group.

[0078] For each sensitive group, the average sensitivity vector is calculated as the group's sensitivity direction. The amplitude is taken as the average of the perturbation amplitudes of each member of the group before the historical event. A candidate combination is generated along this direction.

[0079] The candidate combinations for each group are weighted and summed according to their proportion of the population. If all components of the synthesized combination are within the feasible perturbation domain, it is adopted directly. If any component exceeds the boundary, it is pruned to the nearest boundary value.

[0080] S3: Perform population vulnerability assessment and identify key care recipients within a digital twin system. S3 uses the perturbation combination generated by S2 to conduct parallel investigations on all elderly people to find the most vulnerable link.

[0081] The digital twin system maintains a digital twin instance for each senior citizen. Each instance carries the senior citizen's vulnerability response relationships.

[0082] The system encapsulates the parameters of the adversarial environmental disturbance and distributes them to all instances simultaneously. Each instance independently calculates the expected instability under the disturbance and returns the result.

[0083] This process is equivalent to asking all digital twins the same question: How unstable would you be in the face of this set of environmental mutations? The set of all answers constitutes the population vulnerability distribution.

[0084] The system identifies key individuals requiring care from the distributions. Specifically, the system iterates through the distribution list, comparing each distribution to find the one with the highest expected instability.

[0085] Then determine whether the highest value exceeds the preset instability threshold T.

[0086] The method for determining T is as follows: Collect all historical values ​​of behavioral instability prior to the occurrence of security events. Sort these values ​​from smallest to largest. Take the 90th percentile as T. This means that the elderly person's vulnerability must exceed the level of the state before 90% of historical events to be considered abnormal.

[0087] If the highest expected instability exceeds T, the elderly person is identified as a key care recipient for the current cycle, and subsequent steps continue. If it does not exceed T, indicating that the current group has good resilience, the current cycle is terminated, and the next cycle begins again from S2.

[0088] As another implementation method, when there are a large number of elderly people, they can be divided into layers first.

[0089] The stratification was based on static vulnerability characteristics. Elderly individuals with a positive history of falls or whose cognitive impairment level exceeded a preset threshold were classified into the high-vulnerability stratum. The rest were classified into the normal stratum.

[0090] The investigation is conducted in two phases. First, the highly vulnerable layer is investigated. If the highest expected instability of this layer exceeds T, it is directly identified as a high-risk area, and the investigation terminates. Only when the highest value of the highly vulnerable layer does not exceed T does the second phase, investigating the regular layers, commence. Finally, the results from the two layers are combined, and the global highest value is compared with T.

[0091] S4: Generate disguised caregiving tasks S4 generates a disguised care task for the pre-deployment of care resources.

[0092] The core objective of this mission is to provide close, preventative care for high-risk elderly individuals. However, it is presented to the public as a routine care service, without any warning signs. This allows for resource deployment without creating a tense atmosphere or disrupting normal operations.

[0093] As one implementation method, the system generates tasks by matching a care task rule base.

[0094] First, extract daily preferences and habits from the personal care records of key care recipients. This includes preferred times for drinking water, health check-ups, and frequently used walking routes and resting locations.

[0095] Then, the most suitable care reason is selected from the preset task rule base. The rule base is compiled by the care supervisor. Each rule includes the task name, applicable conditions, description of the operation content, standard duration, and recommended execution area.

[0096] During the matching process, the system calculates the match degree between the elderly person's current time period and location and the various rule conditions. The current time period is divided into six time periods based on the hour: morning, forenoon, afternoon, late afternoon, evening, and night. The current location is obtained by the positioning system. The match degree is obtained by weighted average of the time period overlap and the regional spatial distance score.

[0097] The system filters out rules with a matching degree exceeding a preset threshold and whose operation content includes a pause and observation action. The rule with the highest matching degree is then selected.

[0098] The operational description, standard duration, and recommended areas of this rule constitute the content of a disguised care task.

[0099] For example, an elderly person who habitually drinks water at 10 a.m. is in their room. The system matches them with a "drinking water care" task. The caregiver receives the instruction: "Please go to room 302 to provide Grandma Wang with drinking water and observe her drinking and complexion."

[0100] As another implementation method, the system directly calls the elderly person's daily care plan.

[0101] The task with the closest execution time to the current moment and an inactive status is identified from the plan. If the task content already contains keywords such as "stay" or "observe," it is used directly. If not, a stay / observation step is automatically appended to the end of the task, for example, "After completing the above operations, please stay and observe in this area for five minutes." The task execution time is also increased accordingly.

[0102] S5: Disguised Pre-deployment Scheduling of Care Resources The goal of S5 is to assign surreptitious care tasks to the most suitable caregivers in a way that minimizes the impact on overall hospital services.

[0103] The system constructs a care scheduling optimization model for this purpose. The model finds the optimal caregiver assignment and sequence insertion position for masquerading tasks. The goal is to minimize the total delay cost of tasks across the entire hospital.

[0104] In the model, each task has a delay duration and a delay cost weight.

[0105] The delay duration is the difference between the actual start time of the task and the planned start time. If no timeout occurs, the delay duration is zero.

[0106] The delay cost weight for regular tasks is 1. The safety weight for masquerading tasks is set to 10. This means that when calculating the total cost, the cost of a masquerading task for every unit of delay is 10 times that of a regular task. The solution algorithm will prioritize ensuring its timely execution.

[0107] The total latency cost is the sum of the latency of each task multiplied by its weight.

[0108] As one implementation method, the solution is obtained by enumeration.

[0109] The system obtains the current task sequence of each caregiver. It identifies all possible insertion positions in the sequence, including the beginning, end, between tasks, and after the completion time of the currently executing task.

[0110] For each candidate location, the system simulates the insertion of a camouflage task. After insertion, all tasks at that location and those following it in the original sequence are shifted sequentially to the next position. The delay time for the shifted task is equal to the execution time of the camouflage task, plus the estimated time for the caregiver to move from the previous task point to the target point. The movement time is calculated by dividing the straight-line distance between the two points by the average walking speed of 1.2 meters per second.

[0111] The system calculates the total hospital-wide delay cost for each option. The option with the lowest cost is selected. The corresponding caregiver and insertion position for this option are considered the optimal solution.

[0112] Once the plan is finalized, the system integrates the task into the target caregiver's schedule. It then distributes the task to their mobile device. The device only displays the task name, operation description, and suggested execution time.

[0113] As another implementation method, when the number of caregivers and tasks is large, an iterative optimization method of neighborhood search can be used.

[0114] The system first constructs an initial solution. For example, tasks are assigned to the caregiver who is closest and has the fewest tasks, and placed at the end of their sequence.

[0115] The neighborhood of a solution is generated by two operations: reassigning it to another caregiver, or moving it forward or backward one position in the current caregiver's sequence.

[0116] Starting from the initial solution, repeatedly search the neighborhood. If a neighboring solution with a lower total delay cost is found, move to it and continue the search. If no neighboring solution is better than the current solution, stop searching. The current solution is used as the final assignment scheme.

[0117] S6: Game-theoretic self-evolutionary feedback loop S6 is activated after the task is completed. It is a continuously running feedback learning mechanism that enables the system's detection accuracy and scheduling capabilities to continuously evolve.

[0118] The system continuously acquires real-time behavioral data of the elderly at 1-second intervals for 15 minutes after task execution. Using the same method as S1, the degree of behavioral instability at each sampling time is calculated. The maximum value within these 15 minutes is taken as the actual degree of behavioral instability for this investigation.

[0119] As one implementation method, different strategies are executed based on the comparison result between the actual value and the threshold T.

[0120] Scenario 1: The actual value does not exceed T. The system determines this detection as a false alarm. To reduce the probability of subsequent false alarms, the boundary values ​​of each dimension of the feasible perturbation domain are uniformly multiplied by a contraction coefficient, such as 0.9. This will reduce the overall amplitude of the perturbation combination generated in the next cycle.

[0121] Scenario 2: The actual value exceeds T, but falls back below T within the same time window. The system determines this as a valid warning. This indicates that the elderly person did experience a brief period of vulnerability, but the risk was mitigated because the caregiver was already in place.

[0122] At this point, the system pairs the combination of perturbation parameters used in this instance with the actual instability value, creating a high-value incremental sample. This sample is then used to update the vulnerability response relationship model using small-step gradient descent. The learning rate is set to 0.1 times the initial training learning rate to avoid drastic model fluctuations caused by a single sample. After the update, the model becomes more sensitive to predicting this type of perturbation.

[0123] As an alternative implementation, the system does not use threshold triggering, but instead performs continuous smooth fine-tuning.

[0124] Regardless of whether the actual value exceeds T, the current data pair is treated as an incremental sample for small-step updates to the model. Each sample is assigned an exponentially decaying weight. This weight is determined by the time interval between the sample collection time and the current update time, as well as a 7-day time constant. Samples from more recent times have a smaller impact on the model.

[0125] The system also continuously adjusts the generation intensity of subsequent disturbance combinations based on the difference between the actual value and T.

[0126] When the difference is greater than zero, a contraction factor is multiplied by the boundary value of the feasible perturbation domain. The contraction factor is determined by the difference and a preset positive coefficient of the adjustment rate. The larger the difference, the smaller the contraction factor, and the more severely the perturbation amplitude is compressed in the next cycle. When the difference is less than or equal to zero, the detection intensity remains unchanged.

[0127] Through this continuous game between adversarial generation and vulnerability models, the accuracy of system detection improves over time.

[0128] Through the complete operation of S1 to S6 described above, this invention establishes a complete intelligent closed loop on the information layer of the existing digital twin platform, encompassing offline modeling, online detection, pre-deployment of camouflage, and continuous self-evolution. This closed loop operates entirely at the software level, requiring no additional hardware. It enables a fundamental shift in the security management of elderly care institutions from passive response to proactive defense.

[0129] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing the operation and managing risks of elderly care institutions based on digital twins, characterized in that, Includes the following steps: S1: Obtain the individual vulnerability response relationship for each elderly person in the hospital, wherein the vulnerability response relationship is a mapping of environmental disturbance parameters to the degree of behavioral instability of the elderly person in the hospital; S2: According to a preset cycle, under the condition of satisfying the physical constraints of the internal space of the elderly care institution, generate a combination of adversarial environmental disturbance parameters that increases the highest degree of behavioral instability among the elderly population in the institution. S3: In the digital twin system, the adversarial environmental disturbance parameters are combined and input into the vulnerability response relationship of each elderly person in the hospital. The expected instability of each elderly person in the hospital is calculated in parallel to obtain the group vulnerability distribution. The elderly person in the hospital with the highest expected instability is identified from the group vulnerability distribution as the key care target for the current cycle. S4: Based on the personal care file of the key care recipient, generate a disguised care task that involves close contact with the key care recipient in the area where the key care recipient is located; S5: Based on the real-time location of the on-duty caregiver and the current care task sequence, with the goal of minimizing the sum of the delay costs of care services for all elderly people in the hospital, determine the target caregiver who will perform the spoofed care task and the insertion position of the task in the target caregiver's care task sequence, and send the spoofed care task to the target caregiver's terminal.

2. The method for optimizing the operation and managing risks of elderly care institutions based on digital twins according to claim 1, characterized in that, The environmental disturbance parameters mentioned in step S1 are composed of the rate of change of environmental parameters collected by non-contact sensors. The environmental parameters include at least two of the following: spatial illuminance, regional pedestrian density, background sound pressure level, and temperature and humidity.

3. The method for optimizing the operation and managing risks of elderly care institutions based on digital twins according to claim 1, characterized in that, Step S1 involves obtaining the individual vulnerability response relationship, including: Obtain the sequence of environmental disturbance parameters and the corresponding behavioral instability labels within a preset time window prior to the occurrence of historical safety incidents involving the elderly person in the facility; Using the static vulnerability characteristics of the elderly person in the hospital as a condition, the mapping from the environmental disturbance parameter sequence to the behavioral instability degree label is learned to obtain the individual vulnerability response relationship.

4. The method for optimizing the operation and managing risks of elderly care institutions based on digital twins according to claim 3, characterized in that, The behavioral instability label is calculated from at least two of the following indicators for the elderly person in the hospital within the preset time window: gait symmetry deviation, movement trajectory entropy, and abnormal dwell time in a designated safe area.

5. The method for optimizing the operation and managing risks of elderly care institutions based on digital twins according to claim 1, characterized in that, Step S2 generates the adversarial environmental disturbance parameter combination, including: Obtain the sensitivity coefficient of each elderly resident to different environmental parameters; Within the feasible perturbation domain that satisfies the aforementioned spatial physical constraints, generate a value that satisfies the target value. The combination of environmental parameter change rates that is maximized is taken as the combination of adversarial environmental disturbance parameters, wherein... For the first Sensitivity coefficient of elderly residents in the facility For the first The expected instability of elderly residents in the facility.

6. The method for optimizing the operation and managing risks of elderly care institutions based on digital twins according to claim 5, characterized in that, The sensitivity coefficient is extracted from the individual vulnerability response relationship of the elderly person in the hospital and characterizes the intensity of their behavioral instability response to changes in environmental parameters of different dimensions.

7. The method for optimizing the operation and managing risks of elderly care institutions based on digital twins according to claim 1, characterized in that, Step S3 involves identifying the key care recipients, including: Determine whether the expected instability of the elderly patient with the highest expected instability exceeds a preset instability threshold; If the number exceeds the limit, the individual is identified as a key care recipient; if the number does not exceed the limit, the subsequent steps of the current cycle are terminated.

8. The method for optimizing the operation and managing risks of elderly care institutions based on digital twins according to claim 1, characterized in that, Step S4 generates the disguised care task, including: Extract daily care preferences and habitual actions from the personal care records of the key care recipients; From the preset care task rule library, select one care event that matches the daily care preferences and habitual actions and includes the caregiver staying and observing in the designated area, and use it as the display content of the disguised care task.

9. The method for operational optimization and risk management of elderly care institutions based on digital twins according to claim 1, characterized in that, Step S5, which determines the target caregiver and insertion location, includes: A care scheduling optimization model is constructed, which aims to minimize the weighted sum of the delay times of all care tasks, and sets a higher safety weight for the spoofed care tasks than for regular care tasks. Based on the real-time location of each on-duty caregiver and the current care task sequence, the care scheduling optimization model is solved to obtain the target caregiver assignment and task sequence insertion position of the masquerade care task.

10. The method for operational optimization and risk management of elderly care institutions based on digital twins according to any one of claims 1 to 9, characterized in that, It also includes step S6: Within a preset time window after the execution of the disguised care task, the real behavioral data of the key care recipient is obtained; If the actual behavioral data indicates that the degree of behavioral instability of the key care recipient does not exceed the preset instability threshold, then the perturbation amplitude of the subsequently generated adversarial environmental disturbance parameter combination will be reduced. If the real behavioral data indicates that the degree of behavioral instability of the key care recipient exceeds the preset instability threshold and falls back after intervention by the disguised care task, then the individual vulnerability response relationship of the key care recipient is incrementally updated using the current combination of environmental disturbance parameters and the real behavioral instability data.

Images