Robot control system for linkage of multi-stage emergency response and old-age care community service

Abstract

The invention relates to the technical field of intelligent robot control, and discloses a multi-level emergency response and old-age care community service linked robot control system which comprises the steps that community resident real-time man-machine interaction data streams are acquired, and instantaneous service adaptability indexes are calculated; calculating a service adaptability-task complexity adaptation function value; executing a distributed auction algorithm to generate a task allocation scheme; optimizing resource allocation through a Nash equilibrium algorithm of the game theory; when an emergency event is detected, a multi-level linkage rapid redistribution mechanism is started; and converting the task allocation scheme into a robot control instruction sequence. According to the method, the problem of mismatching caused by mutual independence of user state evaluation and task scheduling in a traditional system is solved, and community multi-level linkage emergency response is realized.

Description

Technical Field

[0001] This invention relates to the field of integrated management technology for smart elderly care communities, and more specifically, to a robot control system that links multi-level emergency response with elderly care community services. Background Technology

[0002] In senior living communities, seniors with dementia require a variety of daily services, including medication reminders, rehabilitation training, and health checkups. However, the cognitive abilities of seniors with dementia exhibit significant individual differences and temporal fluctuations, with the same senior showing varying levels of cognitive load at different times of the day. Existing senior care service robot control systems process cognitive assessment and task scheduling as two separate modules, lacking a real-time cognitive-task adaptation mechanism. This can lead to the system assigning highly complex tasks during periods of cognitive decline in seniors, causing distress, resistance, or even service failure. Furthermore, when emergencies occur in the senior living community requiring the reallocation of service resources, existing systems cannot incorporate cognitive factors into resource scheduling decisions, resulting in low emergency response efficiency and decreased service quality. Summary of the Invention

[0003] This invention provides a robot control system that integrates multi-level emergency response with elderly care community services. It solves the mismatch problem caused by the independent cognitive assessment and task scheduling in traditional systems. It considers cognitive adaptability while taking into account task urgency and global resource utilization efficiency, and can maintain cognitive adaptability even in the event of emergencies.

[0004] This invention provides a robot control method that integrates multi-level emergency response with elderly care community services, comprising: The system acquires real-time human-computer interaction data streams from elderly people with dementia and calculates instantaneous cognitive load indicators by weighting and fusing touch screen response time, speech comprehension accuracy, and task completion rate. Obtain the attribute vector of the queue of service tasks to be executed, calculate the adaptation function value between cognitive load and task complexity, and the adaptation function value represents the degree of matching between the difference between cognitive load and task complexity through an exponential function; Based on the adaptation function value and the task urgency, a distributed auction algorithm is executed to generate an initial task allocation scheme. The resource allocation is optimized using the Nash equilibrium algorithm based on game theory, and the globally optimal task allocation scheme is output under the condition of satisfying the cognitive adaptability constraint. When an emergency event is detected, tasks are reallocated based on the cost of task interruption and the benefits of emergency response. The task allocation scheme is converted into a sequence of robot control instructions and the execution command is output.

[0005] Further calculations of instantaneous cognitive load indicators include: The touch screen response time is normalized and mapped to the [0,1] interval by upper and lower bound thresholds of the response time. Obtain data on the ratio of voice command comprehension accuracy to task completion rate; The three types of data are summed using weighted coefficients, where the weighted coefficients satisfy the constraint that the sum is 1 and non-negative. A sliding window mechanism was used to calculate cognitive load indicators within a time interval, with a window length of 5 minutes.

[0006] Further understanding the calculation of the fit function value between workload and task complexity includes: The adaptation function is defined as a negative exponential function of the absolute value of the difference between cognitive load and task complexity; Set the adaptation sensitivity parameter, and the parameter value range is [0.5, 10]; When the cognitive load perfectly matches the task complexity, the fit function value is 1; As the difference between cognitive load and task complexity increases, the fit function value approaches 0.

[0007] Further distributed auction algorithms include: Define a bidding function that comprehensively considers cognitive suitability, task urgency, and delay penalty; The competitive value of parallel computing by each elderly node for the target task; The task is to receive all bidding information and select the elderly person with the highest bid value; Detect and resolve conflicts when multiple tasks select the same elderly person; Iterate until all tasks are assigned or the maximum number of iterations is reached.

[0008] Further calculations of the delay penalty include: Calculate the ratio of the actual delay time of the task to the expected execution time; The urgency of the task is introduced as a penalty weighting coefficient; An exponential penalty is imposed when the delay exceeds the tolerance threshold; Ensure that the penalty function is monotonically increasing with respect to time and continuously differentiable.

[0009] Further game-theoretic Nash equilibrium algorithms include: Define the utility function for each older adult, including cognitive adaptation benefits and time costs; Set time capacity constraints, task quantity constraints, and minimum fit constraints; The optimal response strategy for each participant is calculated iteratively. The iteration terminates when the policy change is less than the convergence threshold. Verify and output a task allocation scheme that satisfies the Nash equilibrium condition.

[0010] Further task redistribution for emergency events includes: Define a task interruption cost function, which includes three components: schedule loss, cognitive mismatch cost, and restart cost; Define an emergency response benefit function, where the benefit value decays over time; Construct a redistribution optimization problem with the goal of minimizing interruption costs and maximizing emergency benefits; A stable matching algorithm is used to solve the problem and construct a bidirectional preference sequence between the elderly and the task. The reallocation decision was completed within the emergency response time window.

[0011] The generation of further robot control command sequences includes: Query the predefined action template library based on task type; The standard instruction sequence is adaptively adjusted based on the current cognitive load level; When the cognitive load is below the threshold, a simplified mode is activated to remove unnecessary steps and extend the execution time. The adjusted instruction sequence is encoded into a robot-executable format; Encapsulate the execution command containing robot identifier, user identifier, task identifier, and adaptive parameters.

[0012] Furthermore, it also includes: Construct an individual cognitive baseline model using an LSTM time-series prediction model; Predict the cognitive state curve for the next 6 hours based on historical cognitive performance data; The prediction curve is divided into high, medium, and low cognitive ability periods; The generation time period-task matching weight provides guidance for the auction algorithm.

[0013] The present invention also provides a control system for a service robot in an elderly care community, comprising: The data acquisition module is used to acquire real-time human-computer interaction data of elderly people with dementia; The cognitive assessment module is used to calculate instantaneous cognitive load indicators; The task scheduling module is used to perform distributed auctions and Nash equilibrium optimization. The emergency response module is used to detect emergency events and trigger task reassignment; The instruction conversion module is used to generate adaptive robot control instructions; The central scheduling server is used to coordinate the various modules and manage the task queue.

[0014] The beneficial effects of this invention are as follows: This invention's method, by employing a cognitive load-service complexity adaptation function, quantifies in real-time the matching degree between the cognitive abilities of elderly individuals with dementia and the complexity of service tasks, overcoming the mismatch problem caused by the independent cognitive assessment and task scheduling in traditional systems. By introducing an improved distributed auction algorithm and game-theoretic Nash equilibrium optimization, the system can consider both cognitive adaptability and task urgency and global resource utilization efficiency, addressing the issue of resistance from elderly individuals with dementia when assigned complex tasks during periods of cognitive decline. The emergency response mechanism based on matching theory ensures that cognitive adaptability is maintained even during emergencies, preventing service failures caused by neglecting cognitive factors during emergency handling. Attached Figure Description

[0015] Figure 1 This is a flowchart of a robot control method for linking multi-level emergency response with elderly care community services according to the present invention; Figure 2 This is a bar chart analyzing the composition of cognitive load indicators in this invention; Figure 3 This is a user-task adaptation heatmap of the present invention; Figure 4 This is a bar chart comparing the bidding function values ​​of the present invention; Figure 5 This is a line graph of the LSTM cognitive state prediction curve of the present invention; Figure 6 This is the Sankey diagram of the emergency redistribution process of the present invention. Detailed Implementation

[0016] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0017] At least one embodiment of the present invention discloses a robot control method that links multi-level emergency response with elderly care community services, such as... Figure 1 As shown, it includes the following steps: In the hardware environment of the elderly care community service robot system, including service robots equipped with touch screens and voice interaction modules, a central dispatch server, and a real-time data acquisition sensor network, the method of this implementation includes the following steps: Step 100: Obtain real-time human-computer interaction data streams of elderly people with dementia and calculate instantaneous cognitive load indicators.

[0018] Obtain touch screen response time series from the service robot's interaction module:

[0019] Voice command comprehension accuracy sequence and task completion sequence The real-time data stream, in which The number of sampling points. Based on the above three types of data, the instantaneous cognitive load index is calculated using a weighted fusion algorithm. :

[0020] in, , , The weighting coefficients for cognitive load calculation correspond to the weights of touch screen response time, speech understanding accuracy, and task completion, respectively, satisfying... and , and These are the upper and lower bounds of the response time, respectively. Indicates at time No. The cognitive load level of an individual user; the higher the value, the better the cognitive ability.

[0021] Furthermore, the time dimension in the cognitive load index formula is specifically reflected as follows: Time variable This represents the absolute time of system operation, in minutes, with a time range of [missing information]. ,in Minutes correspond to a 24-hour service cycle; , , Representing users respectively At any moment The touchscreen response time, speech understanding accuracy, and task completion rate all depend on specific points in time; the sampling frequency is set to once per minute to ensure timely capture of changes in cognitive state; in the time window sliding mechanism, the current moment... Cognitive load calculation based on time intervals The data within, of which The sliding window length of minutes ensures the sensitivity and continuity of cognitive load indicators to changes over time.

[0022] Furthermore, response time upper and lower bound thresholds and The determination method is as follows: based on the benchmark data collected during the system initialization phase, The average response time for a healthy adult to complete the same task is set to 2 seconds. The maximum tolerable time for a dementia-affected elderly person to complete the task at their lowest cognitive state is set at 30 seconds. The upper and lower bound thresholds for response time are based on statistical analysis of extensive experimental data to ensure the rationality and effectiveness of the response time normalization process. When the collected response time... At that time, Cut off as ;when At that time, Cut off as This is to ensure the stability of the normalized calculation.

[0023] The input to the aforementioned weighted fusion algorithm is the touch screen response time series. Accuracy sequence of voice command comprehension Task completion sequence and preset weighting coefficients , , The output is an instantaneous cognitive load index. .

[0024] Furthermore, weighting coefficients , , The specific determination method is as follows: First, collect interaction data from 100 elderly people with dementia over 30 days as training samples, including touch screen response time, speech comprehension accuracy, and task completion rate at different times each day; second, use a multi-objective optimization method to solve for the weight combination, with the objective function set as maximizing the Pearson correlation coefficient between the cognitive load index and the expert assessment of cognitive status; then, use the particle swarm optimization algorithm under constraints... and The search space is for the optimal weight combination. The number of particles was set to 30, and the maximum number of iterations was set to 200. Finally, the optimal weight combination was determined through cross-validation. , , The weighted combination resulted in a correlation coefficient of 0.82 between the cognitive load index and the expert assessment, which meets the practical requirements.

[0025] To ensure effective fusion of data with different dimensions, the weighted fusion algorithm performs the following preprocessing on the input data: touch screen response time (Unit: seconds) Min-max normalization was applied. Map time data to Range; Voice command comprehension accuracy and task completion rate Since it is proportional data, the range of values ​​is already... Within the interval, no additional normalization is required. This preprocessing ensures that the three types of interaction indicators with different properties are weighted and summed within the same numerical range, eliminating the impact of dimensional differences on cognitive load calculation.

[0026] It should be noted that touch screen response time This refers to the time interval from when the task prompt is displayed on the interface to when the elderly person completes the touch operation; the accuracy rate of voice command comprehension. This refers to the percentage of elderly people who correctly execute voice commands and the task completion rate. It refers to the proportion of task steps completed within a specified time.

[0027] In this embodiment of the application, in order to improve the robustness of the cognitive load index, before calculating the instantaneous cognitive load index, the original data is first subjected to sliding window smoothing with a window size of 5 sampling points, and median filtering is used to remove the interference of outliers.

[0028] Step 200: Obtain the attribute vector of the queue of service tasks to be executed, and calculate the cognitive load-service complexity adaptation function value.

[0029] Obtain the queue of tasks to be executed from the central scheduling server. Each of the tasks Includes task complexity urgency and estimated time Three attributes. Define the cognitive load-service complexity adaptation function. :

[0030] in, To adapt to the sensitivity parameter, the steepness of the fit function is controlled. A higher value indicates greater sensitivity to differences in cognitive load. , The closer the value is to 1, the better the cognitive load matches the task complexity.

[0031] Furthermore, the complete constraints of the cognitive load-service complexity adaptation function are as follows: Input constraints include Indicates time user Cognitive load level Indicates task The complexity level; parameter constraints include adaptation sensitivity parameters. The range of values ​​is limited to ,when The time function is insensitive to differences, resulting in insufficient discrimination. Overly strict time functions lead to system rigidity; output constraints ensure that cognitive load-service complexity matches the function value. ,when hour Indicates a complete match, when hour This indicates a complete mismatch; the time continuity constraint requires that the cognitive load-service complexity adaptation function be continuously differentiable in the time dimension, satisfying... Existence and boundedness ensure the smoothness of changes in cognitive states; matching threshold constraints are set. This is the minimum fit threshold for the task to be executable; matching schemes below this value will be automatically rejected by the system.

[0032] Furthermore, the theoretical basis and boundary condition handling for the matching threshold constraint are as follows: The selection of the threshold of 0.3 is based on statistical analysis of large-scale experimental data. The task failure rate exceeds 60%, and user satisfaction is below 40%, failing to meet the basic quality requirements of elderly care services; The system maintains a task success rate above 75% and user satisfaction above 70%, meeting practicality standards. When the fit of all pending tasks is below 0.3, the system activates a task delay mechanism, postponing the task until the user's cognitive state improves. The delay time is determined based on the LSTM prediction model. For urgent tasks (… When the fit is below 0.3, the system lowers the threshold to 0.2 and enables simplified execution mode, which improves the execution success rate by reducing task steps and adding auxiliary prompts. The threshold boundary processing adopts a lag mechanism, and the task allocation is only reactivated when the fit rises from below 0.3 to above 0.35 to avoid frequent task state switching.

[0033] Furthermore, adapt sensitivity parameters The method for determining the optimal parameter values ​​is as follows: The optimal parameter values ​​are searched on the training dataset using cross-validation, with the search range set to [value to be filled in]. The step size is 0.1. The evaluation metric is a weighted average of task success rate and user satisfaction, with a weighting ratio of 7:3. Experiments have verified that when... When the overall system performance reaches its optimal point, the cognitive load-service complexity adaptation function exhibits moderate sensitivity to the difference between cognitive load and task complexity. This effectively distinguishes the degree of matching while avoiding system rigidity caused by overly strict matching requirements. The adaptation sensitivity parameter value ensures that when cognitive load and task complexity are perfectly matched... When the difference is 0.4 It provides a reasonable matching metric.

[0034] Furthermore, adapt sensitivity parameters The constraint range and boundary conditions are handled as follows: When When the adaptive function is insensitive to differences in cognitive load and task complexity, the matching discrimination is insufficient, and the system cannot effectively select a suitable task allocation scheme; when At times, the fit function is too strict, only achieving a high fit when the cognitive load and task complexity are almost perfectly matched, leading to an overly rigid system and a sharp decrease in the number of executable tasks. In practical applications, when the calculated... Value exceeds When dealing with a range, boundary truncation is used, i.e. Set the time to 0.5. Set the time to 10; parameter The choice also needs to consider the characteristics of cognitive fluctuations in the elderly population; too small... The value ignores individual differences; an excessively large value... The value will amplify the effect of measurement error.

[0035] It should be noted that the task complexity The urgency level is determined through a comprehensive assessment of the number of task steps, the type of cognitive requirements, and the historical failure rate. Determined based on task type and the severity of consequences of delay; estimated time. In minutes.

[0036] Furthermore, task complexity The data preprocessing methods are as follows: First, collect the number of task operation steps. Cognitive requirement type coding:

[0037] and historical failure rate Three original indicators; Secondly, the number of operation steps is processed using min-max normalization. Map the number of steps to interval; Then, the cognitive requirement types are normalized after ordinal encoding, through:

[0038] Mapped to interval; Finally, the task complexity is calculated using a weighted fusion method:

[0039] The weighting coefficients are determined based on expert evaluation to ensure the effective integration of indicators of different natures.

[0040] Furthermore, the constraints and boundary conditions for the task complexity calculation formula are as follows: When When, truncate it to 1; when When, it is truncated to 20 to ensure the stability of the normalized calculation; when Not here When the range is within a certain range, the nearest mapping principle is used: values ​​less than 1 are mapped to 1, and values ​​greater than 5 are mapped to 5; when When set to 0, The time is set to 1; the final calculated task complexity is... Strictly ensure that Within the interval, when all input parameters are at their minimum values When all input parameters are at their maximum values .

[0041] Step 300: Convert the task scheduling scheme into a sequence of robot control instructions and output the execution command.

[0042] Based on the cognitive load-service complexity fit function value calculated in step 200, the task with the highest fit is selected and assigned to the corresponding elderly person. The selected task... Converted into robot control instruction sequence Each instruction It includes the service action type, interaction parameters, and execution duration. The output execution command format is:

[0043] in, Includes service content simplification strategies and interaction rhythm parameters adjusted based on the current cognitive load level.

[0044] The aforementioned task-instruction conversion decoding process includes the following steps: First, based on task type Query the predefined action template library to obtain standard instruction sequences. Secondly, based on the current level of cognitive load... Adaptive adjustment to standard instruction sequences, when Simplified mode is enabled at certain times to reduce interaction steps and extend single-step execution time. The standard mode is used; then, the adjusted instruction sequence is encoded into a robot-executable format, with each instruction... Includes action type coding Parameter array and execution time Finally, the encoded instruction sequence is encapsulated into a complete execution command. The command is sent to the corresponding service robot via the robot control API for execution.

[0045] Furthermore, the detailed implementation method of task-instruction conversion decoding is as follows: The first step, the template library construction phase, includes a predefined action template library containing 5 basic task templates, namely, medication reminder task templates. Rehabilitation training task template Health check task template Daily nursing task template Emergency Response Task Templates Each template contains a standard sequence of actions, parameter configuration, and time constraints; The second step, the template matching stage, is achieved through a hash table lookup mechanism. Template retrieval with time complexity The query key is the task type identifier. Returns the corresponding standard instruction sequence:

[0046] The third step, the adaptive adjustment phase, occurs when cognitive load... When, the instruction simplification rules include: removing unnecessary confirmation steps, breaking down complex actions into single actions, extending single-step execution time to 1.5 times the original time, and increasing the number of repetitions of voice prompts; when The standard instruction sequence remains unchanged. The fourth step is the instruction encoding stage, specifically the action type encoding. The parameter array is represented using 16-bit integers. Stored in JSON format, execution time Floating-point representation in milliseconds; The fifth step, the command encapsulation stage, encapsulates the encoded instruction sequence, robot identifier, user identifier, and adaptive parameters into an execution command in standard JSON format. ; Step 6, API call phase, via RESTful API interface Send execution commands, set the interface timeout to 5 seconds, and support asynchronous execution and status callback mechanisms.

[0047] In this embodiment of the application, in order to further optimize the task allocation effect, the following enhanced steps are introduced based on step 200: Step 201: Construct an individual cognitive baseline model using an LSTM time-series prediction model to generate a cognitive state prediction curve.

[0048] Based on historical cognitive performance data An individual cognitive baseline model for each elderly person was constructed using an LSTM temporal prediction model. The input layer of the LSTM temporal prediction model has a receptive dimension of [missing information]. eigenvectors , in For the first Cognitive load at any given time The rate of change of cognitive load, This represents the normalized time characteristics.

[0049] To ensure the stability and convergence of the input data for the LSTM model, the input feature vectors are preprocessed as follows: The value range is already in Within the range, no additional processing is required; for the rate of change of cognitive load Standardized processing is adopted, through Eliminating the influence of individual differences, among which and These represent the mean and standard deviation of the rate of change of cognitive load, respectively; time characteristics. pass Normalize and map the hours to Intervals ensure the effective representation of time-periodic characteristics.

[0050] The LSTM hidden layer contains 128 hidden units, and the state transitions are performed using the following update equation: Furthermore, the selection of LSTM model structure parameters is based on the following: the number of hidden units, 128, is determined through grid search. The optimal value determined within the range, the number of hidden units, can both fully capture the temporal dependence of cognitive states and avoid overfitting; input feature dimension It includes current cognitive load, rate of change, and time characteristics, covering the main factors affecting cognitive prediction; the sliding window length of 24 time steps corresponds to 24 hours of historical data, which can capture the daily cycle change pattern of the cognitive state of the elderly; the prediction window of 6 time steps corresponds to the prediction range of the next 6 hours, balancing prediction accuracy and the practicality of task scheduling.

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057] in, , , These are the forget gate, input gate, and output gate, respectively. In cellular state, In hidden state, and These are learnable parameters. The output layer is a fully connected layer that stores the hidden states of the LSTM. Mapping to predicted values: Furthermore, the specific implementation steps for LSTM model parameter initialization and training are as follows: First, the weight matrix , , , Using the Xavier initialization method, the weights are initially uniformly distributed. Random sampling in the middle, of which and These represent the input and output dimensions, respectively; the bias vector. , , , Initialized as a zero vector, but with a forget gate bias. The threshold is set to 1.0 to promote long-term memory retention; secondly, the training data is divided into training, validation, and test sets in an 8:1:1 ratio, with a batch size of 32; then, the backpropagation algorithm is used to update the parameters, and the gradient clipping threshold is set to 5.0 to prevent gradient explosion; finally, training is stopped early when the validation set loss does not decrease for 10 consecutive epochs, and the model weights with the minimum validation set loss are saved as the final parameters.

[0058] Furthermore, the necessity and specific implementation method of gradient clipping are as follows: During the backpropagation process of the LSTM network, the gradient propagates in a chain-like manner through the time steps, which easily leads to gradient explosion, especially when the sequence length is long (such as the 6-hour prediction window corresponding to 72 time steps in this application). The gradient value may grow exponentially, causing excessive parameter updates and destroying the learned features. Gradient clipping solves this problem by restricting the upper bound of the gradient norm. The specific calculation method is as follows: First, calculate the L2 norm of the gradients of all parameters. ,in For loss function, For the first There are several parameters; then it is determined whether the gradient norm exceeds the threshold. At that time, all gradients are scaled proportionally. The threshold of 5.0 is based on experimental verification. This value can effectively prevent gradient explosion while maintaining a sufficient learning rate. When the threshold is too small (such as 1.0), it will limit the model's learning ability. When the threshold is too large (such as 10.0), it cannot effectively prevent gradient explosion. The application of gradient clipping enables the LSTM model to converge stably in long sequence cognitive load prediction tasks, avoiding numerical instability problems during training.

[0059] Furthermore, the specific numerical calculation process of the Xavier initialization method is as follows: For the input dimension (Cognitive load, rate of change, time characteristics) and hidden layer dimensions The weight matrix is ​​initialized with boundary values ​​calculated as follows: Therefore, the weight values ​​are distributed uniformly. Random sampling is performed; for the weight matrix from hidden layer to hidden layer, Initialize the boundary value to The weight value is from Mid-sampling; output layer weight matrix , The boundary value is The weight value is from Mid-sampling; this initialization method ensures that the variance of the weight distribution remains stable during forward and backward propagation, avoiding gradient vanishing or gradient exploding problems.

[0060]

[0061] in Indicates prediction of the future The cognitive state at each time step.

[0062] The aforementioned LSTM time series prediction model employs a sliding window training mode with a window length of 24 time steps and a prediction window of 6 time steps. The optimization strategy uses the Adam optimizer, with a learning rate initialized to 0.001 and decaying by a factor of 0.9 every 50 epochs. The loss function combines mean squared error loss and trend consistency regularization. Furthermore, the selection of optimization parameters is based on the following: the Adam optimizer performs excellently in processing time-series data due to its adaptive learning rate characteristic; the initial learning rate of 0.001 is determined by... The grid search within the specified range is determined, and the initial learning rate ensures rapid model convergence and avoids oscillations. The learning rate decay coefficient of 0.9 and the decay interval of 50 epochs are set based on the trend of the validation set loss to ensure stable convergence in the later stages of training. Trend consistency weights are used. Through ablation experiments Within the selected range, the trend consistency weight can ensure the accuracy of trend prediction without overly restricting the precision of numerical prediction.

[0063] in To predict the number of time steps and Corresponding to the forecast window of the next 6 hours, For trend consistency weight and , For symbolic functions, The prediction time step is in minutes. Future time windows are generated based on the trained model parameters. Internal cognitive state prediction curve .

[0064] Furthermore, the complete parameter definition and time dimension of the LSTM loss function are as follows: number of prediction time steps. The range of values ​​is ,when Effective prediction is not possible when The prediction accuracy decreases significantly and the computational complexity becomes too high; the time step... Minutes represent a 1-hour time span for each prediction step, ensuring that the prediction granularity matches the task scheduling requirements; trend consistency weight. The constraints are ,when The weak trend constraint causes the forecast curve to oscillate. Excessive time-trend constraints negatively impact numerical prediction accuracy; the loss function value is constrained. The training convergence condition is set as follows: the change in validation set loss is less than 0.001 over 5 consecutive epochs; the time index constraint requires... Minutes are needed to ensure sufficient historical data for prediction; the prediction output time range is limited to [time range missing]. Minutes, which is the maximum cognitive state that can be predicted for the next 6 hours.

[0065] Furthermore, the rationality analysis of the LSTM prediction window constraint and the accuracy decay law are as follows: Prediction time steps The lower bound is set to 2 instead of 1 because Predicting only the cognitive state for the next hour has limited guiding significance for task scheduling and cannot support medium- to long-term task planning; prediction accuracy varies with the number of time steps. The decay follows an exponential decay law, specifically manifested as follows: ,when The prediction accuracy is approximately 0.73 when... The prediction accuracy is approximately 0.35; when The prediction accuracy is less than 0.3, which does not meet the reliability requirements of task scheduling. The upper limit of the prediction window of 6 hours is determined based on the diurnal cycle change characteristics of the cognitive state of the elderly. 6 hours covers half of the cognitive cycle and can capture the complete change process from cognitive peak to trough or from trough to peak. When a longer prediction window is required in practical applications, a sliding prediction strategy is adopted, and the prediction results are updated once an hour. The long-term prediction effect is achieved by combining multiple short-term predictions.

[0066] The aforementioned cognitive state prediction curve This needs to be decoded and converted into a basis for task scheduling decisions. The specific decoding process is as follows: First, the prediction curve is thresholded and segmented... Defined as a period of high cognitive ability, This is a period of moderate cognitive ability. The first step is to identify periods of low cognitive ability; the second step is to generate a task scheduling suggestion matrix based on the period segmentation results. ,in This indicates a recommendation during a specific time period. The arrangement complexity level is Finally, based on the complexity distribution of the current queue of tasks to be executed, personalized time-task matching weights are generated. This provides cognitive prediction guidance for subsequent auction algorithms.

[0067] Furthermore, the numerical decoding method for the cognitive state prediction curve is as follows: The first step is time discretization, which separates the continuous prediction curves. Samples are taken at 15-minute intervals to generate discrete time point sequences. ,in A point in time; The second step is to map cognitive ability levels for each time point. Predicted value The system is categorized into levels, with high cognitive ability periods assigned values. Assigning values ​​during periods of moderate cognitive ability Assigning values ​​during periods of low cognitive ability ; The third step is to establish a task complexity matching rule, which maps cognitive level to task complexity. correspond Highly complex tasks, correspond Medium-complexity tasks correspond Low-complexity tasks; The fourth step is to calculate the matching weights, based on the degree of matching between the cognitive level and the task complexity.

[0068] The weight is maximized when cognitive ability is matched with task complexity.

[0069] Furthermore, the boundary condition processing and continuity guarantee for cognitive state threshold segmentation are as follows: To avoid frequent switching of threshold boundaries, a lag mechanism is used to process boundary conditions. When a cognitive state transitions from a low level to a high level, it needs to remain above the new level threshold for three consecutive time points (45 minutes) before the level change is confirmed; when a cognitive state transitions from a high level to a low level, it also needs to remain below the new level threshold for three consecutive time points before the level change is confirmed; for values ​​near the threshold boundary ( or A buffer mechanism is introduced to keep the cognitive state within the buffer unchanged from the previous moment, thus avoiding frequent jumps in level. When the prediction curve exhibits abnormal fluctuations (the difference between adjacent time points exceeds 0.2), a three-point smoothing filter is applied. Eliminate the impact of abnormal fluctuations; the minimum duration of level changes is set to 30 minutes, meaning that any level change must last for at least two time points before it is accepted by the system, ensuring the stability of task scheduling decisions.

[0070] Step 202: Execute the improved distributed auction algorithm to generate a task-time matching matrix.

[0071] Based on the cognitive load-service complexity adaptation function value and mission urgency Define the bidding function :

[0072] in, This is an urgency weighting coefficient, used to adjust the influence of task urgency on the bidding function. For the task At any moment The delay penalty function, This is the penalty coefficient, used in the delayed penalty function to control the rate at which the penalty intensity increases. Indicates the elderly person's number. Indicates the task number.

[0073] Furthermore, the method for determining the parameters of the bidding function is as follows: urgency weight coefficient Through simulation experiments The search within the range is determined when It can balance the requirements of cognitive fit and task urgency, avoiding the over-prioritization of urgent tasks while neglecting cognitive fit; penalty coefficient The penalty coefficient is set to 0.5, determined based on statistical analysis of the impact of historical task delays on overall system efficiency. This ensures that the delay penalty plays an effective moderating role in the bidding function without dominating the decision-making process. Parameter combination. Extensive simulation verification has shown that it can effectively improve the response speed of emergency tasks while ensuring cognitive adaptation.

[0074] The aforementioned delay penalty function Defined as:

[0075] in, For the task At any moment The actual delay time and , Minutes for the task Expected execution time Minutes for the task The delay tolerance threshold, Depending on the urgency of the task, For the task The planned start time.

[0076] To ensure the consistency of the dimensions of the delay penalty function, the dimensional analysis of each term is as follows: The time delay ratio is dimensionless. The urgency weight is dimensionless. The exponent term in Since the time delay is dimensionless and standardized, the entire delay penalty function... It is a dimensionless value, consistent with other dimensionless terms in the bidding function. The delay penalty function uses a linear term. Reflects the basic delay ratio, urgency coefficient Increase the penalty weight for emergency tasks, exponential term Ensure that delays exceeding the tolerance threshold are punished exponentially.

[0077] Furthermore, the complete parameter constraints of the delay penalty function are as follows: Delay time constraint Ensure that penalties are only applied to actual delays, when The time indicates that the task will be executed in advance. Expected execution time constraint Minutes cover all task types, from simple reminders to complex care, when The task at the minute level is too simple to require precise scheduling. The task is too complex and needs to be broken down for execution; latency tolerance threshold constraint. minutes and satisfy This ensures that the tolerance threshold has a reasonable degree of leniency. Penalty function value constraint When the delay time approaches infinity, the penalty value also approaches infinity, reflecting the severe consequences of delay; the time continuity constraint requires... Regarding time It is monotonically increasing and continuously differentiable.

[0078] Furthermore, the specific numerical calculation example of the delay penalty function is as follows: For the medication reminder task ( minute, minute, When the delay time At the minute mark, the penalty value is calculated as follows:

[0079] When the delay time At minute 10:00, the penalty value is:

[0080] When the delay time At minute 10:00, the penalty value is:

[0081] This computational example demonstrates that the delay penalty function has a non-linear growth characteristic; the penalty is small for slight delays, but increases sharply once the tolerance threshold is exceeded, effectively guiding the system to prioritize tasks that are about to time out.

[0082] Generate through a distributed auction mechanism Task-Time Matching Matrix ,in Indicates the time period The task It is allocated to the corresponding elderly person.

[0083] Furthermore, the task-time matching matrix The process of decoding and converting into a specific execution plan is as follows: The first step, the matrix parsing stage, involves traversing the matching matrix. All Extract the user-task assignment pair from the elements. ; The second step, the time scheduling phase, involves determining a specific execution time window for each allocation pair based on task priority and user cognitive status. ,in The task start time. This is the task completion time; The third step is the resource allocation phase, which is based on task type. Allocate corresponding robot resources Service venues ; The fourth step is the parameter configuration phase, based on the user's current cognitive load level. Generate adaptive parameters This includes interaction mode, voice speed, and confirmation frequency; The fifth step, the instruction generation stage, converts the decoded allocation scheme into a standardized execution instruction format:

[0084] The sixth step, the conflict detection phase, checks whether there are time or resource conflicts in the generated execution plan. If a conflict is found, a reallocation mechanism is triggered. This decoding process ensures that the abstract output of the auction algorithm can be converted into a specific sequence of instructions that the robot system can directly execute.

[0085] The aforementioned improved distributed auction algorithm includes the following sub-steps: First, initialize the bidding matrix. Each elderly person's node is based on their own cognitive state. and the set of tasks to be assigned First, calculate the bid value; second, execute the parallel bidding process, with each node simultaneously broadcasting its bid value for the target task. Then, each task receives all bidding information and selects the elderly person with the highest bidding value as the initial allocation target; next, conflicts are detected and resolved, and when the same elderly person is selected by multiple tasks, the allocation is based on the task urgency. A second round of negotiation is performed; finally, the above process is iteratively executed until all tasks are assigned or the maximum number of iterations is reached, outputting the final task-time matching matrix. .

[0086] Furthermore, the detailed execution flow of the improved distributed auction algorithm is as follows: The first step, the system initialization phase, involves creating a node for each elderly person. Assign a unique identifier Configure communication ports and message queues, and establish a star topology network structure; The second step, the bidding calculation phase, involves each node calculating the bidding matrix in parallel. For the task The time complexity of the bidding calculation is The overall bidding computation complexity is ; The third step, the bidding broadcast phase, involves each node synchronously broadcasting bidding information via the TCP protocol. The message format is as follows: The broadcast timeout is set to 100 milliseconds; The fourth step, the bid collection phase, involves the central coordinator collecting and sorting all bid messages, and building a bid ranking list for each task. ; The fifth step, the conflict detection phase, checks for conflicts where a node is selected by multiple tasks. The conflict detection algorithm has a complexity of O(n log n). ; Step 6, Conflict Resolution Phase: For conflict nodes... Select urgency The highest-ranking task is assigned, and the other tasks become alternative bidders; Step 7, Convergence Judgment Phase: This occurs when all tasks have been assigned or the maximum number of iterations has been reached. The algorithm terminates when the time is right.

[0087] Step 203: Analyze the resource competition relationship using the Nash equilibrium algorithm of game theory and output the globally optimal allocation scheme.

[0088] The resource allocation among multiple elderly people and multiple tasks is modeled as a non-cooperative game, defining each elderly person... At any moment Utility function:

[0089] Constraints:

[0090]

[0091]

[0092]

[0093] in, For at any time Distributed to the elderly The task set, This is a time cost coefficient used to adjust the proportion of delay penalties in the total cost. Minutes for the elderly Time capacity limit, For the elderly The maximum number of tasks that can be handled simultaneously. The minimum fitness threshold is used. By iteratively solving for the Nash equilibrium point, the globally optimal task allocation scheme that maximizes the utility of all participants is obtained.

[0094] Furthermore, utility function The specific manifestations of the time dimension are as follows: Utility function time variables in This is reflected in four key aspects: First, the task set. It has a time stamp, indicating a specific moment. Distributed to the elderly The task combination supports dynamic task adjustment and real-time reallocation; secondly, cognitive load. The dynamic changes over time reflect the fluctuations in the cognitive state of older adults at different times, affecting the fitting function. The calculation results; third, the time cost item. Directly quantify the negative impact of task execution time on utility, among which For the task The estimated time consumption reflects the scarcity of time resources; fourth, the utility function value rate of change over time Reflecting the time sensitivity of utility in older adults, when Time indicates that utility increases over time. The time window constraint indicates that utility decreases over time; the time window constraint requires the utility function to remain non-negative within the task's validity period, i.e. ,in and These represent the start and end times of the task, respectively.

[0095] Furthermore, the complete constraints for the Nash equilibrium optimization problem are as follows: Time Capacity Constraints Ensure that the total time allocated to each elderly person does not exceed their capacity, among which Adjust dynamically based on individual health status and cognitive level; Task quantity constraints Limit the number of tasks assigned simultaneously to avoid cognitive overload. With cognitive load level There is a positive correlation; Fit constraints Ensure all task assignments meet the minimum matching requirements; assignments below the threshold will be rejected; resource balancing constraints. Ensure all tasks are assigned and not duplicated; time window constraints require tasks to be assigned. Must be by its deadline Execution begins before, i.e. Utility function value constraint Ensure that the utility of all participants is non-negative; when the utility becomes negative, the participant exits the game.

[0096] Furthermore, the time cost coefficient The determination method is as follows: the time cost coefficient reflects the weight of the impact of task execution time on overall utility, through... The optimal value for the parameter within the range was determined to be 0.3. When... In this way, the system can both prioritize cognitive adaptability and effectively control the time cost of task execution, avoiding task delays caused by excessive pursuit of perfect matching. The time cost coefficient is related to the trend consistency weight in step 201. They are completely different in function; the former controls the trade-off between time and cost, while the latter controls the constraints of predicting trends. They are independent of each other in terms of numerical values ​​and mechanisms of action.

[0097] The input to the aforementioned game-theoretic Nash equilibrium algorithm is the task-time matching matrix. Cognitive load values ​​of different elderly individuals Task complexity set and the set of expected time The output is the optimized task allocation scheme. Each of them Indicates allocation to the elderly The task set. The Nash equilibrium algorithm in game theory iteratively calculates the optimal response strategy for each participant until an equilibrium state is reached, meaning that no single participant can improve their utility by unilaterally changing their strategy.

[0098] Furthermore, the specific algorithm steps for iteratively solving the Nash equilibrium are as follows: Step 1, initialization phase, for each elderly person Set initial strategy This indicates the subset of tasks it accepts, and all strategies constitute the initial strategy combination. The second step, the optimal response calculation stage, is in the... In this iteration, for each participant Strategies to fix other participants By solving the optimization problem Calculate the optimal response strategy The third step, the strategy update phase, employs an asynchronous update mechanism, updating the strategy sequentially according to participant IDs. The policy update rules are as follows ,in For participants The feasible strategy space; the fourth step, the convergence judgment stage, calculates the policy change in two consecutive iterations. ,when and The convergence threshold is the convergence criterion for the Nash equilibrium iterative algorithm; convergence is considered achieved when the policy change is less than this value. The fifth step, the equilibrium verification phase, verifies whether the final policy combination satisfies the Nash equilibrium condition, i.e., for any participant... and any strategy They all Step 6, Result Output Stage: Combining Equilibrium Strategies Convert to task allocation scheme The maximum number of iterations is set to 100.

[0099] Furthermore, the specific calculation method for the convergence criterion of the Nash equilibrium algorithm is as follows: policy change. Using the Euclidean norm, for the policy vector... (in Indicates participants Accept the mission? The strategy change is calculated as follows: Convergence threshold The setting is based on experimental verification. When the policy change is less than this threshold, the utility improvement of subsequent iterations is less than 0.1%, which can be considered as achieving practical convergence. To prevent oscillations, a continuous convergence criterion mechanism is introduced, requiring that the following conditions be met in three consecutive iterations. Only then is convergence considered; the maximum number of iterations of 100 is based on complexity analysis. Participants and For tasks of this size, the algorithm typically converges within 30-50 iterations, with an upper limit of 100 iterations to ensure that the algorithm can still terminate in the worst case. When convergence is not achieved after reaching the maximum number of iterations, the strategy combination with the largest utility function value is selected as the approximate solution, and the convergence failure status is recorded for subsequent analysis.

[0100] In this embodiment of the application, in order to handle sudden emergency events, the method further includes the following step before step 300: Step 204: Reassign tasks for emergency events based on matching theory.

[0101] Upon detecting an emergency event signal, a stable matching algorithm is used to reallocate tasks based on the current task execution status and cognitive load level. The cost of task interruption is defined. and emergency response benefits Dynamic adjustment instructions are generated by solving the following time-dependent optimization problem:

[0102] Constraints: , , ,

[0103] in, For a moment The redistribution decision variables, when Time indicates task Distributed to the elderly , For emergency response at any time The profit value, Indicates the elderly person's number. Indicates the task number. This refers to the current moment.

[0104] Furthermore, the specific manifestation of optimizing the time dimension in the objective function is as follows: objective function time variables in Throughout the optimization process, this is reflected in three key aspects: First, the cost of interruption Over time Dynamic changes reflect the impact of task execution progress on interruption costs. As the task nears completion, the cost of interruption increases significantly. Second, decision variables It has a time stamp, indicating a specific moment. The allocation decision supports dynamic reallocation and real-time adjustment; Third, emergency response benefits The diminishing returns over time reflect the timeliness requirements of emergency response; delayed response will lead to diminishing returns; time window constraints. Limit the time range of the optimization solution, where For the moment an emergency occurs, The maximum response time is in minutes; Rate of change of the objective function value over time:

[0105] It reflects the time sensitivity of the system state and guides the optimization direction of real-time decision-making.

[0106] Furthermore, the complete constraints and time dimension of the emergency redistribution optimization problem are as follows: Task capacity constraints Limit each elderly person At any moment Maximum responsibility One task; task uniqueness constraint Ensure each task At any moment It will be allocated to only one elderly person; Decision variable constraints Binary properties representing allocation decisions; cognitive adaptability constraints Ensure the matching degree after reallocation meets the minimum requirements; time window constraints. Restricting reassignment operations in tasks Conducted within the effective time window; emergency response benefits Decays over time, satisfying ,in Basic return value, This is the emergency benefit attenuation coefficient, used to control the rate at which emergency response benefits decay over time. The timeframe for reassignment decisions must be based on the occurrence of the emergency event; optimizing time constraints requires that reassignment decisions must be made after the emergency event has occurred. Completed within minutes.

[0107] The aforementioned task interruption cost function Defined as:

[0108] in, For the task At any moment The current percentage of progress completed. For task complexity, For a moment The cognitive load-service complexity adaptation function value. For the task At any moment The cost of restarting , , The weighting coefficients of the interruption cost function correspond to the weights of time cost, cognitive switching cost, and restart cost, respectively, satisfying the following conditions: and .

[0109] Furthermore, the time dimension parameter of the interruption cost function is defined as follows: progress percentage. Monotonically increases over time, satisfying when This indicates the cumulative time characteristic of task completion progress; restart cost. It is directly proportional to the task execution time and is defined as follows: ,in For the task The start time, Minutes for the task The estimated total time; Adaptation function value Reflection Moment The real-time cognitive adaptation status varies with cognitive load. The fluctuations and dynamic changes occur; the interruption cost function value is constrained. The cost of interruption tends to reach its maximum when the schedule approaches 100%, and its minimum when the schedule is 0%; time continuity requirement. Regarding time Continuous differentiability ensures smooth cost changes and system stability.

[0110] Furthermore, the method for determining the weighting coefficients of the task interruption cost function is as follows: Weighting coefficients This reflects the dominant role of schedule loss in interruption costs, and analysis based on historical data shows that schedule loss is the cost factor that users are most concerned about. This demonstrates the significant impact of cognitive mismatch on interruption decisions and ensures that the reallocation process still considers cognitive suitability requirements. The corresponding restart cost receives a relatively small weight because restarting technical equipment requires fewer resources compared to human labor costs. Weighting scheme. Sensitivity analysis and expert evaluation have determined that the relative importance of various interruption costs can be reasonably quantified in emergency redistribution scenarios.

[0111] To ensure the consistency of the dimensions of the function, the restart cost is considered. Normalization is performed, through Map the restart costs of different tasks to The interval, in which This represents the maximum restart cost for all tasks in the system. This preprocessing ensures that the schedule loss term, the adaptability reciprocal term, and the restart cost term are all dimensionless values, eliminating the influence of different dimensions on cost calculation.

[0112] The task interruption cost function uses the schedule loss term. Quantify the loss of completed work, and the reciprocal of the adaptability term. Reflecting the additional costs arising from cognitive mismatch, restart the cost item. Consider the resource consumption of task reinitialization.

[0113] The inputs to the aforementioned stable matching algorithm include the current task execution state matrix and the real-time cognitive load level of each elderly person. The set of tasks to be reassigned and their attribute vectors, emergency event types and priorities, are used as the output to form a new task assignment scheme. The stable matching algorithm is based on the Gale-Shapley algorithm and constructs a bidirectional preference sequence between elderly people and tasks. The elderly people's preferences are ranked based on cognitive suitability, while the tasks' preferences are ranked based on execution efficiency and emergency response needs. The iterative matching process ensures the stability of the allocation results, meaning that there are no cases where elderly people and tasks mutually prefer each other but are not matched.

[0114] Furthermore, the detailed execution process of the Gale-Shapley stable matching algorithm is as follows: The first step, the preference sequence construction phase, involves creating a preference sequence for each elderly person. Construct a sequence of task preferences. According to the adaptive function value Sort in descending order; for each task Constructing the preference sequence of older adults According to execution efficiency and emergency response coefficient Arrange the products in descending order; The second step, the initialization phase, involves setting up a task receiving queue while all elderly individuals and tasks are in an idle state. and temporary matching relationships ; The third step, the proposal stage, involves each available senior citizen... The optimal task in its preference sequence that has not been proposed to. Send a marriage proposal request. The format of the marriage proposal message is: ; The fourth step is the task response phase, for each task... Collect all marriage proposals and select the elderly person whose preference ranking is highest. Accept for now, if There is already a matching object. and If the preference is higher, then reject. and accept ; The fifth step, the matching update phase, involves the rejected elderly individuals returning to an idle state and updating the temporary matching relationships. ; Step 6, Termination Phase: The algorithm terminates when all elderly individuals have a match or all feasible matches have been attempted, outputting a stable matching result; the algorithm's time complexity is O(n log n). ,in For the number of elderly people, The number of tasks.

[0115] On a certain morning in September 2024, from 9:00 AM to 12:00 PM, Chenguang Senior Living Community needed to assign five different types of service tasks to three elderly people with dementia. During the system operation, at 10:30 AM, an emergency occurred where Mr. Li experienced an abnormal heart rate, requiring task reassignment.

[0116] The participants' information is as follows: Grandma Zhang (user ID) Mild cognitive impairment; Grandpa Li (user ID) Moderate cognitive impairment; Aunt Wang (user ID) Mild cognitive impairment.

[0117] Step 100 Implementation Example: The system collected raw human-computer interaction data at 9:00 AM, based on weighting coefficients. , , The instantaneous cognitive load index was calculated.

[0118] Figure 2The study presents the cognitive load indicators for three elderly people with dementia, including three weighted components: touch screen response time, speech comprehension accuracy, and task completion rate.

[0119] Step 200 implementation example: The five currently pending task queue attributes are adapted to sensitivity parameters. Calculate the cognitive load-service complexity adaptation function value.

[0120] Figure 3 The result matrix of the cognitive load-service complexity adaptation function F_match is displayed, which intuitively shows the degree of adaptation of three users to five tasks with different complexities.

[0121] Step 202 Implementation Example: Based on bidding function Using parameters , The initial time delay penalty is 0, and the bidding results are calculated. After processing by the distributed auction algorithm, a task-time matching matrix is ​​generated. .

[0122] Figure 4 This demonstrates the rationality of the auction mechanism, where the highest bidder for each task will receive priority in allocation.

[0123] Figure 5 This demonstrates the predictive power of the LSTM model, providing forward-looking guidance for task scheduling.

[0124] Step 203 Implementation Example: Based on task-time matching matrix And utility function:

[0125] Using time cost coefficient The globally optimal allocation scheme is obtained by solving the Nash equilibrium algorithm. .

[0126] Step 204 Implementation Example: At 10:30, Grandpa Li ( If an abnormal heart rate occurs during rehabilitation training (60% completion), an immediate health check is required. The system detects the emergency and triggers the task reassignment mechanism.

[0127] Figure 6 This demonstrates the effectiveness of stable matching algorithms in emergency response.

[0128] Step 300 Implementation Example: Based on the redistribution scheme in step 204, the system converts the task into a sequence of robot control instructions. (Using Grandpa Li as an example...) Let's take the emergency health check task as an example for conversion.

[0129] This application example demonstrates that the method of the present invention can achieve a balance between cognitive adaptability and task efficiency in normal service scheduling, and can quickly reallocate resources when an emergency occurs to ensure the timely execution of critical tasks, while maintaining cognitive matching requirements to the greatest extent possible, thus verifying the effectiveness and practicality of the method.

[0130] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1.A method for controlling a robot for multi-stage emergency response and linkage with services in a senior community, characterized by, The method comprises the following steps: obtaining real-time human-computer interaction data stream of the old people with dementia, calculating the instantaneous cognitive load index by weighted fusion of touch screen response time, speech understanding accuracy and task completion degree; obtaining the attribute vector of the service task queue to be executed, and calculating the adaptive function value between the cognitive load and the task complexity, wherein the adaptive function value represents the matching degree of the difference between the cognitive load and the task complexity through an exponential function; performing a distributed bidding algorithm based on the adaptive function value and the task urgency to generate an initial task allocation scheme; optimizing the resource allocation through a game theory Nash equilibrium algorithm, and outputting a globally optimal task allocation scheme under the condition of meeting the cognitive adaptability constraint; when an emergency event is detected, redistributing the tasks based on the task interruption cost and the emergency response benefit; converting the task allocation scheme into a robot control instruction sequence and outputting an execution command. 2.The multi-stage emergency response and senior community service linkage robot control method of claim 1, wherein, The calculation of the instantaneous cognitive load index comprises: normalizing the touch screen response time, and mapping the response time to the interval [0, 1] through upper and lower threshold values; obtaining the proportion data of the speech instruction understanding accuracy and the task completion degree; performing weighted summation on the three types of data through weight coefficients, wherein the weight coefficients satisfy the constraint condition of sum being 1 and being non-negative; calculating the cognitive load index in the time interval through a sliding window mechanism, and the window length is 5 minutes. 3.The multi-stage emergency response and senior community service linkage robot control method of claim 1, wherein, The calculation of the adaptive function value between the cognitive load and the task complexity comprises: defining the adaptive function as a negative exponential function of the absolute value of the difference between the cognitive load and the task complexity; setting an adaptive sensitivity parameter, wherein the parameter value range is [0.5, 10]; when the cognitive load and the task complexity are completely matched, the adaptive function value is 1; when the difference between the cognitive load and the task complexity increases, the adaptive function value tends to 0. 4.The method of claim 1, wherein, The distributed bidding algorithm comprises: defining a bidding function, which comprehensively considers the cognitive adaptability, the task urgency and the delay penalty; each old person node calculates the bidding value for the target task in parallel; the task receives all the bidding information and selects the old person with the highest bidding value; detecting and solving the conflict of multiple tasks selecting the same old person; iteratively executing until all task allocation is completed or the maximum iteration number is reached. 5.The multi-stage emergency response and senior community service linkage robot control method of claim 4, wherein, The calculation of the delay penalty comprises: calculating the ratio of the actual delay time of the task to the expected execution time; introducing the task urgency as a penalty weight coefficient; applying an exponential penalty when the delay time exceeds the tolerance threshold; ensuring that the penalty function is monotonically increasing and continuously derivable with respect to time. 6.The method of claim 1, wherein, The game theory Nash equilibrium algorithm comprises: defining the utility function of each old person, which includes the cognitive adaptation benefit and the time cost; setting the time capacity constraint, the task quantity constraint and the minimum adaptability constraint; iteratively calculating the best response strategy of each participant; terminating the iteration when the strategy change is less than the convergence threshold; verifying and outputting the task allocation scheme that meets the Nash equilibrium condition. 7.The method of claim 1, wherein, The task redistribution of the emergency event comprises: defining a task interruption cost function, which includes three components of progress loss, cognitive mismatch cost and restart cost; defining an emergency response benefit function, wherein the benefit value decays over time; constructing a redistribution optimization problem, which aims to minimize the interruption cost and maximize the emergency benefit; The stable matching algorithm is used for solving, and a bidirectional preference sequence of the elderly and tasks is constructed. The re-allocation decision is completed within an emergency response time window. 8.The method of claim 1, wherein, The generation of the robot control instruction sequence includes: Querying a predefined action template library based on the task type; Adaptive adjustment of the standard instruction sequence according to the current cognitive load level; When the cognitive load is lower than a threshold, a simplified mode is enabled, unnecessary steps are deleted, and the execution time is extended; The adjusted instruction sequence is encoded into a robot executable format; An execution command containing a robot identifier, a user identifier, a task identifier and adaptive parameters is packaged. 9.The multi-stage emergency response and senior community service linkage robot control method of any one of claims 1 to 8, wherein, Further comprising: An individual cognitive baseline model is constructed by using an LSTM time series prediction model; A future 6-hour cognitive state curve is predicted based on historical cognitive performance data; The predicted curve is divided into high, medium and low cognitive ability periods; A period-task matching weight is generated to guide the bidding algorithm. 10.A senior community service robot control system, configured to perform the multi-stage emergency response and senior community service linkage robot control method of any one of claims 1-9. Comprise: A data acquisition module for acquiring real-time human-computer interaction data of the demented elderly; A cognitive evaluation module for calculating instantaneous cognitive load indicators; A task scheduling module for executing distributed bidding and Nash equilibrium optimization; An emergency response module for detecting emergency events and triggering task re-allocation; An instruction conversion module for generating adaptive robot control instructions; A central dispatch server for coordinating various modules and managing a task queue.

Images