A remote nursing analysis system for rehabilitation state evaluation

Abstract

The application discloses a remote nursing analysis system for rehabilitation state evaluation, relates to the technical field of remote nursing data analysis, and is used for solving the problem of poor continuous evaluation of rehabilitation state. The application performs time alignment on physiological data, training data, posture data and nursing record execution time, generates a confidence marker, identifies missing, out-of-order and noise segments, improves the credibility of rehabilitation evaluation data, organizes training tasks, rest tasks, follow-up tasks and reevaluation tasks into a nursing scheme graph based on a sleep state unit and individual baseline parameters, compiles a nursing state machine, outputs a candidate nursing scheme set and a trajectory prediction record, and under the constraints of safety, resources and dependence, screens feasible schemes and selects a target nursing scheme. Then, the target scheme is compiled into a task queue, a task receipt is recycled to generate an updated sample entry, individual baseline parameters and model parameters are updated, and nursing strategy is continuously and adaptively optimized.

Description

Technical Field

[0001] This invention relates to the field of remote nursing data analysis technology, and more specifically, to a remote nursing analysis system for assessing rehabilitation status. Background Technology

[0002] In the fields of remote nursing and home-based rehabilitation, nursing management methods based on wearable device monitoring and online follow-up have been widely adopted. These systems typically access physiological data, training data, posture data, and nursing records via computer programs. They then generate a summary display of rehabilitation progress at the nursing end, and based on this, schedule training reminders, follow-up assessments, and nursing interventions.

[0003] Existing solutions focus on improving data collection coverage and alarm response speed, generating assessment conclusions through rule thresholds or model inferences; and enabling remote guidance in conjunction with task assignment. The technology involves the collaborative management of digital data processing and nursing procedures.

[0004] However, existing technologies have limitations in the closed-loop assessment and intervention process. For example, there is an inherent contradiction between the discreteness of multi-source data collection and the continuity of patients' actual behavior and physiological changes; device interruptions, network jitter, and sensor noise can lead to missing data, out-of-order data, and fluctuations. This makes it difficult to consistently align data from different sources under a unified time reference, and this temporal inconsistency makes it difficult for the system to form a continuous, stable, and traceable rehabilitation trajectory. Consequently, nursing recommendations are easily affected by abnormal fragments and require frequent adjustments, making it difficult to ensure consistency between training, rest, follow-up, and reassessment, thus weakening the reliability of remote nursing assessments and the feasibility of interventions in complex home environments. Summary of the Invention

[0005] To overcome the aforementioned deficiencies in the prior art, the following solution is proposed to address the problem of poor continuous assessment of rehabilitation status in the aforementioned background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A remote nursing analysis system for rehabilitation status assessment includes a time-series access module, a sleep modeling module, a plan generation module, a verification and deduction module, and a closed-loop update module.

[0008] The timing access module is used to access physiological data, training data, posture data, and nursing records, perform normalization and time alignment, and output a sequence of rehabilitation data frames with confidence markers.

[0009] The sleep modeling module is used to extract a time series set of sleep parameters from the rehabilitation data frame sequence, slice it to generate sleep state units and establish individual baseline parameters;

[0010] The plan generation module is used to construct a nursing plan atlas and compile a nursing state machine. Based on the plan generation model trained by machine learning, it takes sleep state units, individual baseline parameters and rehabilitation data frame sequences as inputs and outputs a set of candidate nursing plans and trajectory prediction records.

[0011] The verification and simulation module is used to verify the set of candidate nursing solutions by applying safety constraints, resource constraints, and dependency mutual exclusion constraints based on confidence markers, and to select the target nursing solution by combining trajectory prediction records.

[0012] The closed-loop update module is used to compile the target care plan into a task queue and output it, receive task receipts to generate updated sample entries, update individual baseline parameters based on the updated sample entries, and update the plan generation model.

[0013] Furthermore, the timing access module performs normalization and time alignment and generates a sequence of rehabilitation data frames, including:

[0014] The timestamps in physiological data, training data, posture data, and nursing records are converted to a unified format and mapped to a unified time base.

[0015] Resample each data source according to a uniform sampling period and generate a frame index; generate missing test markers for segments where the time interval between adjacent frames exceeds a preset range;

[0016] Generate out-of-order tags for segments with reversed timestamps or frame indexes; generate noise tags for amplitude abrupt changes, saturation intervals, or continuous zero-value intervals;

[0017] Confidence tags are generated based on missing test tags, out-of-order tags, and noise tags and written into the rehabilitation data frame sequence.

[0018] Furthermore, the sleep modeling module extracts a temporal set of sleep parameters from the rehabilitation data frame sequence, slices them to generate sleep state units, and establishes individual baseline parameters, including:

[0019] Heart rate, respiratory, blood oxygen, body movement, and skin temperature parameters are extracted from the rehabilitation data frame sequence to form a time series set of sleep parameters;

[0020] Perform time-consistency on the time series set of sleep parameters and generate change indicators;

[0021] The slice boundaries are determined based on the change indicators and sleep state units are generated. The sleep state units include slice identifier, start and end time, sleep parameter vector, change indicators, and confidence flags.

[0022] Construct a stable set of slices from slices whose confidence markers do not contain missing test markers and noise markers;

[0023] Individual baseline parameters are generated based on statistics from a stable set of slices.

[0024] Furthermore, the solution generation module includes:

[0025] Define training tasks, rest tasks, follow-up tasks, and reassessment tasks as task nodes and write the task type, execution window, duration, and intensity parameters;

[0026] Generate dependency edges, mutual exclusion edges, trigger edges, and fallback edges between task nodes to form a nursing plan graph;

[0027] A nursing state machine is generated based on the nursing plan graph. The nursing state machine includes a set of state identifiers and a set of transition conditions. The transition conditions are generated by mapping trigger edges and backoff edges.

[0028] The model generates a set of candidate nursing plans by inputting sleep state units, individual baseline parameters, and rehabilitation data frame sequences into the scheme.

[0029] A trajectory prediction record is generated for the candidate nursing plan set. The trajectory prediction record includes a state sequence, a state start and end time sequence, a trigger event sequence, and a resource usage sequence.

[0030] Furthermore, the verification and deduction module performs safety constraints, resource constraints, and dependency mutual exclusion constraints verification on the candidate nursing plan set based on confidence markers, including:

[0031] Dependency constraints are generated based on the dependency edges of the nursing plan graph, and mutual exclusion constraints are generated based on the mutual exclusion edges of the nursing plan graph.

[0032] Resource constraints are generated based on the execution window and duration of task nodes;

[0033] Safety constraints are generated based on sleep state units and individual baseline parameters;

[0034] The task nodes in the candidate nursing plan set are matched with the start and end times of the sleep state unit to establish the task slice association relationship.

[0035] When the confidence flag of the sleep state unit corresponding to the task slice association contains a missing test flag or a noise flag, the nursing state machine transition corresponding to that task node is disabled.

[0036] Candidate care solutions that simultaneously satisfy safety constraints, resource constraints, dependency constraints, and mutual exclusion constraints, and are not prohibited from enabling transfer, are added to the set of feasible care solutions.

[0037] Furthermore, the verification and simulation module combines trajectory prediction records to select the target nursing plan, including:

[0038] For each feasible care plan in the set of feasible care plans, read the trajectory prediction record and parse it to obtain the state sequence, trigger event sequence and resource consumption sequence;

[0039] The number of backoff edge triggers is counted based on the trigger event sequence, and a backoff count is generated.

[0040] The number of resource conflict segments is counted based on the resource occupancy sequence, and a conflict count is generated.

[0041] Lexicographical comparison is performed within the set of feasible nursing solutions, first comparing backoff counts and then comparing conflict counts. If the comparison results are the same, the number of consecutive segments between the training state and the rest state in the state sequence is compared, and the feasible nursing solution with the best comparison result is output as the target nursing solution.

[0042] Furthermore, the closed-loop update module compiles the target care plan into a task queue and outputs it, receives task acknowledgments to generate updated sample entries, updates individual baseline parameters based on the updated sample entries, and updates the plan generation model, including:

[0043] The target nursing plan is expanded into a task queue, which includes a task identifier, a target status identifier, a distribution time tag, an execution window, and a receipt judgment condition.

[0044] Receive task receipts from the task queue and generate updated sample entries. The updated sample entries include the target care plan identifier, task receipt set, sleep state unit summary, rehabilitation data frame sequence summary, and actual execution status sequence.

[0045] The stable slice set is reconstructed based on the updated sample entries, and the individual baseline parameters are updated; the scheme generation model is incrementally trained based on the updated sample entries to update the model parameters.

[0046] Furthermore, the trajectory prediction records generated by the scheme generation module include:

[0047] The set of candidate nursing solutions is expanded according to the nursing state machine according to the control cycle, and the start and end time series of the state is generated based on the execution window and duration of the task node.

[0048] Map the transition conditions of the triggering edge and the backtracking edge to a sequence of triggering events and write them into the trajectory prediction record;

[0049] Based on resource constraints, the resource usage segments of task nodes in each control cycle are analyzed and a resource usage sequence is generated;

[0050] Write the state identifiers of the nursing state machine into the state sequence in chronological order and align them with the start and end time sequences of the states to output trajectory prediction records.

[0051] Furthermore, the closed-loop update module's process of receiving task feedback also includes:

[0052] Create a receipt index for each task identifier in the task queue. The task receipt includes a receipt timestamp, completion flag, timeout flag, and exception flag.

[0053] Sort the task receipts by their issuance time label and establish a task status mapping with the target status identifier of the nursing status machine.

[0054] The effective interval of the task status is determined based on the execution window and the receipt timestamp, and the actual execution status sequence is generated.

[0055] Write the actual execution state sequence into the updated sample entry.

[0056] Furthermore, the closed-loop update module generates a model based on the update scheme for the updated sample entries, including:

[0057] The model input entries are constructed by extracting sleep state unit summaries, individual baseline parameters, and rehabilitation data frame sequence summaries from the updated sample entries.

[0058] The monitoring items are constructed by extracting target care plan identifiers, task receipt sets, and actual execution status sequences from the updated sample items.

[0059] A sorting label is generated for the candidate nursing plan set based on the task feedback set. The sorting label is determined by the completion mark ratio, the number of timeout marks, the number of abnormal marks, and the number of backtracking edge triggers.

[0060] Incremental training is performed on the scheme generation model based on the model input entries and sorting labels to update the model parameters.

[0061] The technical effects and advantages of the remote nursing analysis system for rehabilitation status assessment of the present invention are as follows:

[0062] This invention achieves unified time alignment and trusted labeling management of multi-source data in home rehabilitation scenarios by constructing a remote nursing closed loop that includes time-series access, sleep modeling, scheme generation, verification and deduction, and closed-loop update. It can identify and gate missing, out-of-order, and noisy segments, reducing the interference of abnormal data on assessment and decision-making.

[0063] Based on this, a computable rehabilitation state input is formed based on sleep state units and individual baseline parameters. Training tasks, rest tasks, follow-up tasks, and reassessment tasks are organized into a nursing plan atlas and compiled into a nursing state machine. The plan generation model is used to output a set of candidate nursing plans and trajectory prediction records, so that the nursing strategy has the ability to be structured and deducible. The verification and deduction module selects feasible plans and selects the target nursing plan under safety constraints, resource constraints, and mutual exclusion constraints, reducing frequent regressions and state oscillations.

[0064] Finally, the closed-loop update module compiles the target care plan into a task queue and collects task receipts, generating updated sample entries to continuously update individual baseline parameters and plan generation model parameters. This allows the care plan to continuously and adaptively adjust with sleep fluctuations, changes in compliance, and training execution feedback, thereby improving the reliability of remote rehabilitation assessment, the feasibility of nursing interventions, and the stability of long-term operation. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the structure of a remote nursing analysis system for assessing rehabilitation status according to the present invention. Detailed Implementation

[0066] 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.

[0067] In order to achieve the above objectives, Figure 1 A schematic diagram of the structure of a remote nursing analysis system for rehabilitation status assessment according to the present invention is provided, which specifically includes a time-series access module, a sleep modeling module, a scheme generation module, a verification and deduction module, and a closed-loop update module;

[0068] The timing access module is used to access physiological data, training data, posture data, and nursing records, perform normalization and time alignment, and output a sequence of rehabilitation data frames with confidence markers.

[0069] The sleep modeling module is used to extract a time series set of sleep parameters from the rehabilitation data frame sequence, slice it to generate sleep state units and establish individual baseline parameters;

[0070] The plan generation module is used to construct a nursing plan atlas and compile a nursing state machine. Based on the plan generation model trained by machine learning, it takes sleep state units, individual baseline parameters and rehabilitation data frame sequences as inputs and outputs a set of candidate nursing plans and trajectory prediction records.

[0071] The verification and simulation module is used to verify the set of candidate nursing solutions by applying safety constraints, resource constraints, and dependency mutual exclusion constraints based on confidence markers, and to select the target nursing solution by combining trajectory prediction records.

[0072] The closed-loop update module is used to compile the target care plan into a task queue and output it, receive task receipts to generate updated sample entries, update individual baseline parameters based on the updated sample entries, and update the plan generation model.

[0073] The time-series access module is used to access physiological data, training data, posture data, and nursing records, perform normalization and time alignment, and output a sequence of rehabilitation data frames with confidence markers. Specifically, the implementation is as follows:

[0074] In this embodiment, the timing access module is deployed on the edge computing node or cloud computing node of the remote nursing analysis system. It accesses physiological data, training data, posture data, and nursing records through the data communication interface, and performs normalization and time alignment processing on the four types of input records.

[0075] The timing access module parses the timestamp field of each input record, completes the unified format conversion, and obtains a unified timestamp. Then, the unified timestamp is mapped to a unified time base. The unified time base is configured by the system to determine the base start time and maintain global consistency.

[0076] It should be noted that before unified timestamp mapping, the timing access module maintains clock calibration parameters for each data source. The clock calibration parameters include clock offset and drift correction. The clock offset is obtained by measuring the round-trip time of the handshake message and aligning it with the local clock when the data source is first accessed. During operation, it is re-estimated at a fixed calibration cycle to correct long-term drift. When a calibration failure is detected or the drift correction exceeds the allowable time jitter boundary, the timing access module sets the confidence flag of the corresponding frame of the data source to an out-of-order state and retains the original timestamp for traceability.

[0077] The timing access module is configured with a unified sampling period. Based on the unified sampling period, a frame index sequence is generated on a unified time base. The frame index consists of a frame identifier and a frame time. Each data source is resampled according to the unified sampling period. During resampling, the input record with the smallest time difference from the frame time is selected as the intra-frame sample value of the frame at each frame time. The intra-frame sample value is then bound to the corresponding frame index, thereby forming the frame structure of the rehabilitation data frame sequence. The frame structure retains the original timestamp, unified timestamp, frame index, and data payload field. The data payload field carries the sample values ​​or record content corresponding to physiological data, training data, posture data, and nursing records, respectively.

[0078] The window length of the sampling window is configured by the system parameter table. Intra-frame sampled values ​​are generated only when the unified timestamp of the input record falls into the sampling window. When there is no input record in the sampling window, the data source payload for that frame is set to empty and a missing measurement mark is generated. When there are multiple input records with the same time difference in the sampling window, the one with the higher priority is selected as the intra-frame sampled value according to the data source priority table. The data source priority table is configured by the system parameter table and remains compatible during version updates.

[0079] During the resampling and frame index generation process, missing test markers, out-of-order markers, and noise markers are generated according to rules, specifically including:

[0080] The generation of missing test markers is based on the time interval between adjacent frames. When the intra-frame sample value of the corresponding data source is not obtained at a certain frame time, a missing test marker is generated in that frame.

[0081] When the effective sampling interval between adjacent frames exceeds the preset range, a missing test mark is generated for the frame interval covered by the interval, and the starting frame index and ending frame index of the missing test interval are recorded. The preset range is determined by the unified sampling period and the allowable time jitter boundary and written into the system parameter table.

[0082] The system parameter table is stored on non-volatile storage media, written by the configuration process during deployment, and loaded at runtime. The system parameter table includes at least the unified sampling period, allowable time jitter boundary, mutation threshold, saturation value interval boundary, continuous zero value judgment length, change threshold, minimum slice duration, maximum slice duration, control period, resource capacity parameters, safety boundary parameters, and the applicable scope of the data source for each parameter. Among them, the unified sampling period, control period, and resource capacity parameters are configured by the nursing terminal or preset according to the scenario. The mutation threshold and saturation value interval boundary are jointly determined by the sensor protocol range information and deployment parameters. The change threshold and safety boundary parameters are jointly determined by the individual baseline parameters and the scenario preset boundary and written into the system parameter table.

[0083] The generation of out-of-order markers is based on the determination of timestamp order and frame index order:

[0084] When the timestamps of input records from the same data source are reversed, an out-of-order marker is generated for the reversed records;

[0085] When the frame indexes written from the same data source are reversed or the same frame index is written repeatedly, an out-of-order flag is generated for the conflicting frames and the conflicting frame index is recorded.

[0086] The generation of noise markers is based on the determination of amplitude abrupt changes, saturation intervals, and continuous zero-value intervals: amplitude abrupt changes are achieved by comparing the change amplitude of the same parameter sample values ​​in adjacent frames.

[0087] When the magnitude of the change exceeds the mutation threshold, noise markers are generated and the mutation direction and magnitude are recorded.

[0088] The saturation range is determined based on the upper and lower limits of the sensor's range. When the sampled value falls into the saturation range, a noise marker is generated. The continuous zero value range is achieved by detecting the continuous range where the same parameter has a sampled value of zero or an invalid constant in consecutive frames. When the length of the continuous range reaches the continuous zero value judgment length, a noise marker is generated for each frame in the continuous range and the start and end frame indices of the range are recorded. The mutation threshold, saturation range boundary, and continuous zero value judgment length are all configured by the system parameter table and remain consistent during the processing.

[0089] After the missing test marker, out-of-order marker, and noise marker are generated, the timing access module generates confidence markers based on these markers and writes them into the rehabilitation data frame sequence, forming a rehabilitation data frame sequence output with confidence markers. The confidence markers are expressed using a combined encoding method, with the encoded fields corresponding to the three states of missing test markers, out-of-order markers, and noise markers, and are written according to deterministic rules:

[0090] When a missing test flag exists, the missing test status field of the confidence flag is set to invalid;

[0091] When an out-of-order tag exists, the out-of-order status field of the confidence tag is set to invalid and associated with the conflict frame index record; when a noise tag exists, the noise status field of the confidence tag is set to invalid.

[0092] When none of the three types of markers exist, the confidence marker's three status fields are all set to valid.

[0093] The final output rehabilitation data frame sequence is organized with the frame index as the primary key. Each frame contains the frame index, frame time, intra-frame sampled values ​​of physiological data, intra-frame sampled values ​​of training data, intra-frame sampled values ​​of posture data, intra-frame sampled values ​​of nursing records, missing test markers, out-of-order markers, noise markers, and confidence markers. This rehabilitation data frame sequence is used as input to the sleep modeling module, enabling the sleep modeling module to extract the time series set of sleep parameters under a unified time reference and further generate sleep state units.

[0094] The sleep modeling module is used to extract a time series set of sleep parameters from the rehabilitation data frame sequence, slice it to generate sleep state units, and establish individual baseline parameters. Specifically, the implementation is as follows:

[0095] The sleep modeling module takes the rehabilitation data frame sequence output by the time-series access module as input. The rehabilitation data frame sequence is organized by frame index and carries frame time and confidence marker. In each frame, the sleep modeling module extracts heart rate parameters, respiratory parameters, blood oxygen parameters, body movement parameters, and skin temperature parameters from physiological data and posture data, and forms a time-series set of sleep parameters with frame time as time label. Among them, the heart rate parameter is taken from the heart rate sample value of each frame, the respiratory parameter is taken from the respiratory sample value of each frame, the blood oxygen parameter is taken from the blood oxygen sample value of each frame, the body movement parameter is taken from the body movement intensity sample value or posture change amplitude sample value of each frame, and the skin temperature parameter is taken from the skin temperature sample value of each frame.

[0096] The sleep modeling module performs timeline consistency checks and adjustments on the time series set of sleep parameters to achieve time consistency, including:

[0097] When duplicate samples occur at the same frame time, one sample is retained according to the rule of smaller time difference or higher source priority. When a frame confidence flag contains a missing test flag, the missing test status is written into the sleep parameter time series set entry corresponding to that frame and the frame time is retained. The original missing test is not replaced by interpolation. When a frame confidence flag contains a noise flag, the noise status is written into the sleep parameter time series set entry corresponding to that frame and the original sample value is retained for tracing. At the same time, it is used for confidence flag aggregation in subsequent slices.

[0098] The above five types of parameters serve as commonly used physiological and behavioral inputs for home sleep monitoring and wearable sleep technology, and can form a continuous time-series data foundation in the home setting.

[0099] After obtaining the time-consistent set of sleep parameters, the sleep modeling module generates change indicators and determines slice boundaries accordingly. The generation of change indicators adopts a combination of frame-by-frame temporal differencing and stability determination.

[0100] The sampled values ​​of heart rate, respiratory, blood oxygen and skin temperature parameters in adjacent frames are calculated to determine the amplitude of change in body motion intensity in adjacent frames or the amplitude of fluctuation in short frame segments. Then, the amplitude of change of each parameter is compared with the corresponding change threshold in the system parameter table. If any parameter satisfies the change amplitude exceeding the change threshold in several consecutive frames, the starting frame time of the consecutive frame segment is recorded as a candidate point of slice boundary.

[0101] If all parameters do not exceed the change threshold within several consecutive frames, then the frame segment is recorded as a stable interval and used to extend the current slice.

[0102] It should be noted that the sleep modeling module applies both minimum and maximum slice duration constraints to the slice length. Both are preset by the system parameter table and fixed with the system version. The number of frames for several consecutive frames is configured by the system parameter table to suppress slice fragmentation caused by occasional fluctuations in a single frame. The final confirmation of candidate points for slice boundaries requires that the interval between adjacent candidate points be no less than the minimum slice duration and no greater than the maximum slice duration. If the candidate points are too dense, the candidate point with the larger peak value of the change indicator is retained as the slice boundary.

[0103] After the slice boundaries are determined, the sleep modeling module generates sleep state units on the time series set of sleep parameters according to the slice boundaries. Each sleep state unit includes a slice identifier, start and end times, sleep parameter vectors, change indicators, and confidence markers, including:

[0104] The slice identifier consists of the slice number and the date identifier. The start and end times are taken from the first and last frame times covered by the slice. The sleep parameter vector consists of a statistical summary of each parameter within the slice. The statistical summary includes at least the representative values ​​of heart rate, respiratory, blood oxygen, body movement, and skin temperature parameters. The representative values ​​are obtained by sampling values ​​from within the slice according to the statistical rules specified in the system parameter table.

[0105] In this embodiment, the statistical rule adopts the median rule: the sampled values ​​of the same parameter within the slice are collected in chronological order and sorted. The sampled value at the middle position after sorting is taken as the representative value. When the number of sampled values ​​within the slice is even, the sampled value that is closer to the midpoint of the slice among the two middle sampled values ​​is taken as the representative value. At the same time, the discrete statistics of the parameter are output as an additional quality field. The discrete statistics adopt the quantile interval width, which is obtained by taking the difference between the sampled values ​​at the upper quantile position and the lower quantile position after sorting.

[0106] The change indicator is written into the peak value and duration of the change of each parameter in the slice, which is used as the trajectory prediction record input for the subsequent scheme generation model;

[0107] The confidence marker is generated by aggregating the missing test markers, out-of-order markers, and noise markers of each frame within the slice according to the union rule. If any frame within the slice contains a missing test marker, then the slice's confidence marker contains a missing test marker. If any frame within the slice contains a noise marker, then the slice's confidence marker contains a noise marker.

[0108] After generating sleep state units, the sleep modeling module selects a set of stable slices based on confidence markers and establishes individual baseline parameters, specifically including:

[0109] The stable slice set consists of sleep state units whose confidence markers do not contain missing markers or noise markers. During screening, the continuity of the start and end times of the slices is checked, excluding slices that cross obvious time discontinuities to ensure consistency in the baseline estimation source. Individual baseline parameters are generated based on statistics from the stable slice set. These statistics are calculated separately for each parameter dimension and written into the individual baseline parameter structure.

[0110] For heart rate, respiratory, blood oxygen, and skin temperature parameters, calculate the baseline central statistic and baseline discrete statistic respectively. The baseline central statistic uses the median or mean, and the baseline discrete statistic uses the quantile interval width or standard deviation.

[0111] Baseline activity level statistics and baseline fluctuation statistics are calculated for body movement parameters. The former uses the median or mean of the representative values ​​of the slice, while the latter uses the median or upper quantile statistics of the fluctuation amplitude within the slice. The sleep modeling module outputs individual baseline parameters and sleep state units together as part of the input of the protocol generation module, enabling the protocol generation model to jointly model sleep state units and rehabilitation data frame sequences and generate a set of candidate care protocols under the constraints of individual baseline parameters.

[0112] The protocol generation module is used to construct a nursing protocol atlas and compile a nursing state machine. Based on the protocol generation model trained by machine learning, it takes sleep state units, individual baseline parameters, and rehabilitation data frame sequences as input, and outputs a set of candidate nursing protocols and trajectory prediction records. Specifically, the implementation is as follows:

[0113] The solution generation module takes the sleep state unit output by the sleep modeling module, individual baseline parameters, and rehabilitation data frame sequence output by the time sequence access module as input. It first establishes a nursing task dictionary and constructs a nursing solution map.

[0114] The nursing task dictionary is used to limit the fields and value range of task nodes. The task node fields include task type, execution window, duration, and intensity parameter. The task type is limited to training task, rest task, follow-up task, and reassessment task.

[0115] The execution window consists of a start time and an end time, which are used to limit the time range during which the task is allowed to start; the duration is used to limit the time span from the start to the end of the task.

[0116] Intensity parameters are used to characterize the load level of a training task or the recovery level of a rest task.

[0117] The solution generation module instantiates each task into task nodes based on the nursing task dictionary, and generates dependency edges, mutual exclusion edges, trigger edges, and fallback edges between task nodes to form a nursing solution graph, including:

[0118] Dependency edges are generated by prerequisite rules, which limit a task node to only become available after its preceding task node is completed. Mutual exclusion edges are generated by resource conflict rules, which determine whether two task nodes are mutually exclusive when they overlap in time and occupy the same resource type. Resource types include available time resources for patients, available time resources for nursing staff, and resources occupied by training equipment.

[0119] The available time resources for patients are generated by the patient's schedule configuration, and the available time resources for nursing staff are generated by the nursing staff's shift configuration. The resources occupied by training equipment are generated by the training equipment registry, which contains the available time periods and the maximum concurrent usage. The resource capacity parameters are determined by the above-mentioned available time period set and the maximum concurrent usage and are written into the system parameter table for consistent use by resource conflict rules, resource constraint resolution, and conflict count generation.

[0120] The trigger edge is generated by the trigger condition rule, which takes the deviation between the change indicator in the sleep state unit and the individual baseline parameter as input and outputs the activation condition of the task node.

[0121] The fallback edge is generated by the fallback condition rule. The fallback condition rule takes the confidence marker in the rehabilitation data frame sequence and the abnormal receipt in the training data as input, and outputs the fallback condition for switching from the current task node to the alternative task node. This makes the nursing plan map have a computable dependency structure, mutual exclusion structure, condition triggering structure and fallback structure, which meets the structured input requirements of subsequent compilation and trajectory inference.

[0122] After the nursing plan atlas is constructed, the plan generation module compiles the nursing state machine based on the nursing plan atlas and calls the plan generation model to output a set of candidate nursing plans;

[0123] The nursing state machine includes a set of state identifiers and a set of transition conditions. The set of state identifiers is generated by mapping task nodes in the nursing plan graph, with each task node corresponding to a state identifier. The set of transition conditions is generated by mapping trigger edges and backoff edges, with trigger edges mapping to normal transition conditions and backoff edges mapping to backoff transition conditions.

[0124] Normal transition conditions are composed of the satisfaction of dependent edges, the arrival of the execution window, and the judgment results of the triggering condition rules. Fallback transition conditions are composed of the judgment results of the fallback condition rules. The scheme generation module performs unified feature processing on sleep state units, individual baseline parameters, and rehabilitation data frame sequences. Sleep state units provide sleep parameter vectors and change indicators. Individual baseline parameters provide baseline central statistics and baseline discrete statistics for each sleep parameter. Rehabilitation data frame sequences provide task completion sequences of training data, motion quality segments of posture data, and confidence marker segments. Then, the above features are input into the scheme generation model to output a set of candidate nursing schemes.

[0125] Each candidate nursing plan in the candidate nursing plan set contains at least the task node parameter assignment results and edge set activation results of the nursing plan graph, and is bound to its corresponding nursing state machine one by one, so that trajectory prediction records can be generated one by one for the candidate nursing plan set.

[0126] The solution generation module generates a trajectory prediction record while outputting a set of candidate nursing solutions. The trajectory prediction record includes a state sequence, a state start and end time sequence, a trigger event sequence, and a resource usage sequence.

[0127] Specifically, the solution generation module expands the nursing state machine corresponding to each candidate nursing solution according to the control cycle. The control cycle is configured by the system parameter table and kept fixed. Based on the execution window and duration of the task node, a state start and end time sequence is generated. Each item in the state start and end time sequence corresponds to the expected start time and expected end time of the state identifier on the time axis. The transition conditions of the trigger edge and the backoff edge are mapped to the trigger event sequence. Each item in the trigger event sequence includes the event type identifier, the trigger time, and the associated state identifier.

[0128] The scheme generation module analyzes the resource occupancy segments of task nodes in each control cycle based on resource constraints and generates a resource occupancy sequence. Resource constraints are composed of the available time period of resource type, maximum concurrent occupancy, and mutual exclusion rules. Resource occupancy segments are used to describe the occupancy range of patient available time resources, nursing staff available time resources, and training equipment occupancy resources of each task node within a certain control cycle.

[0129] Finally, the state identifiers of the nursing state machine are sorted according to the state start and end time series and written into the state sequence. The state sequence is aligned with the state start and end time series to output the trajectory prediction record. This allows each candidate nursing plan to have a verifiable and deducible structured trajectory description, and the feasibility screening and target plan selection can be performed by confidence markers and constraints.

[0130] The verification and simulation module is used to verify the set of candidate nursing solutions based on confidence markers, applying safety constraints, resource constraints, and dependency mutual exclusion constraints. It then combines trajectory prediction records to select the target nursing solution. Specifically, the implementation is as follows:

[0131] The verification and simulation module is deployed on the computing node of the remote nursing analysis system. It takes the set of candidate nursing plans and trajectory prediction records output by the plan generation module as input, and reads the nursing plan map, nursing state machine, sleep state unit, individual baseline parameters and confidence markers in the rehabilitation data frame sequence.

[0132] The verification and deduction module first constructs dependency constraints and mutual exclusion constraints, including:

[0133] Dependency constraints are generated directly from the dependency edges of the nursing plan graph. Only after the starting task node of the dependency edge is completed can the ending task node enter the enableable state.

[0134] Mutual exclusion constraints are directly generated from the mutual exclusion edges of the nursing plan graph. When the execution windows of two task nodes connected by the mutual exclusion edge overlap, they are prohibited from being enabled at the same time.

[0135] Then, resource constraints and safety constraints are constructed: resource constraints are derived from the execution window and duration of each task node. The verification and simulation module expands each task node into an occupied interval, checks whether the occupied interval completely falls within the execution window, and checks whether the number of occupied intervals in the same time period exceeds the resource capacity parameter. The resource capacity parameter is preset by the system parameter table and can include the patient available time resource capacity and the nursing service resource capacity.

[0136] Safety constraints are generated from sleep state units and individual baseline parameters. The verification and deduction module checks the deviation of each sleep parameter vector of each sleep state unit from the baseline central statistic and baseline discrete statistic of the individual baseline parameters. When the deviation exceeds the safety boundary specified in the system parameter table, a safety limit associated with the time interval of that sleep state unit is generated. The safety limit is then mapped to a constraint on the upper limit of the training task intensity parameter, the upper limit of the training task duration, or the training task activation condition. This ensures that the candidate nursing plan has a deterministic and verifiable safety boundary during the verification phase. This structured expression of the nursing process, which describes the nursing process using task dependencies, mutual exclusion relationships, and time relationships, is consistent with the workflow modeling methods commonly used in clinical pathways and nursing plans. For example, the plan can define resources to express conditional actions, dependencies, and time relationships, thereby supporting verification and deduction.

[0137] After the constraint construction is completed, the verification and deduction module performs gating verification based on the confidence flag and generates a set of feasible nursing solutions. Specifically, the verification and deduction module performs time overlap matching between each task node in the candidate nursing solution set and the start and end time of the sleep state unit. The time overlap matching is based on whether the execution window of the task node overlaps with the start and end time of the sleep state unit, and forms a task slice association relationship.

[0138] When a task node overlaps with multiple sleep state units, the sleep state unit with the longest overlap duration is selected as the associated slice of the task node. If the overlap durations are the same, the associated slice is selected based on the principle that the start time is closer to the start time of the task node's execution window.

[0139] The verification and deduction module reads the confidence markers of the associated slices. When the confidence markers contain missing or noise markers, the nursing state machine transition corresponding to that task node is prohibited. This prohibition is achieved by setting the entry transition condition of the corresponding state of the task node to be unsatisfactory, and recording the set of prohibited transition markers within the candidate nursing plan to ensure traceability in subsequent trajectory analysis. For each candidate nursing plan, the verification and deduction module checks whether the safety constraints, resource constraints, dependency constraints, and mutual exclusion constraints are simultaneously satisfied based on the gating results. If satisfied, the candidate nursing plan is added to the feasible nursing plan set; otherwise, a rejection reason code is written. The rejection reason code can distinguish at least five types of situations: unsatisfactory safety constraints, unsatisfactory resource constraints, unsatisfactory dependency constraints, unsatisfactory mutual exclusion constraints, and unreachable due to prohibited transitions. This ensures that the generation of the feasible nursing plan set has a deterministic basis and is verifiable.

[0140] It should be noted that, in order to ensure the verifiability of the nursing state machine deduction, this embodiment converts the nursing process into a finite state structure and performs constraint verification on its transition conditions. This is consistent with the implementation path of using finite state machines for simulation and property verification in medical workflows.

[0141] After obtaining the set of feasible nursing care options, the verification and simulation module selects the target nursing care option based on the trajectory prediction records. For each feasible nursing care option in the set, the module reads the trajectory prediction records, parses them to obtain the state sequence, trigger event sequence, and resource occupancy sequence, and generates backtracking and conflict counts accordingly, including:

[0142] The rollback count is obtained by traversing the trigger event sequence and counting the number of event entries corresponding to the rollback edge. Each event entry in the trigger event sequence contains an event type identifier and an associated status identifier. When the rollback edge corresponding to the event type identifier is triggered, it is counted in the rollback count.

[0143] The conflict count is obtained by traversing the resource occupancy sequence and checking whether the resource occupancy segment within the same control period exceeds the resource capacity parameter. When it exceeds the resource capacity parameter, the control period is counted as a resource conflict segment and accumulated to form the conflict count.

[0144] Then, a lexicographical comparison is performed. The comparison order is fixed as follows: first compare the backoff count, then compare the conflict count. The one with the smaller backoff count takes precedence. If the backoff counts are the same, the one with the smaller conflict count takes precedence.

[0145] When the backoff count and the conflict count are the same, the number of consecutive segments between the training state and the resting state in the state sequence is further compared. The number of consecutive segments is obtained by scanning the state sequence in chronological order, merging adjacent identical state identifiers into one segment and counting them. In this embodiment, a smaller number of consecutive segments is given priority to reduce frequent state switching.

[0146] The final output is the optimal feasible nursing plan as the target nursing plan. The target nursing plan identifier and its corresponding elimination control record are output to the closed-loop update module to support the closed-loop process of subsequent task queue compilation and receipt-driven update.

[0147] The closed-loop update module is used to compile the target care plan into a task queue and output it, receive task acknowledgments to generate update sample entries, update individual baseline parameters based on the update sample entries, and update the plan generation model. Specifically, the implementation is as follows:

[0148] The closed-loop update module receives the target nursing plan and its corresponding nursing state machine from the previous processing link, expands the nursing state machine into a task queue, and outputs it to the patient-side execution device and the nursing-side management device. The expansion process adopts deterministic compilation rules.

[0149] Using the state identifier of the nursing state machine as the source of the target state identifier, a task record is generated for each state identifier, and a unique task identifier is assigned to the task record. The task record is written with a distribution time tag, execution window, and acknowledgment conditions. The distribution time tag is taken from the planned start time of the target nursing plan or the distribution time determined by the system scheduling strategy. The execution window is determined by the planned start and end time corresponding to the state and expressed by the start time and end time. The acknowledgment conditions are configured as a machine-readable set of conditions according to the task type. The training task corresponds to the joint satisfaction of the completion acknowledgment in the training data and the effective continuous segment of the posture data. The rest task corresponds to the decrease of the body movement parameter to the allowable range of the individual baseline parameter and the maintenance until the end of the execution window. The follow-up task corresponds to the addition of a follow-up event entry in the nursing record. The re-evaluation task corresponds to the addition of a re-evaluation conclusion entry in the nursing record.

[0150] The task queue is organized with the task identifier as the primary key to ensure consistency in receipt retrieval and traceability. The way this class tracks the progress of medical and nursing workflow execution by task records is consistent with the general practice of expressing the task execution status in a consistent manner in medical workflows.

[0151] Furthermore, after the task queue is issued, the closed-loop update module receives the task receipts and forms a receipt index. The task receipt includes a receipt timestamp, completion flag, timeout flag, and exception flag. The receipt timestamp is generated by the execution device or nursing terminal management device and written after being corrected by a unified time base. The receipt index uses the task identifier as the index key, records the receipt set corresponding to the same task identifier, and retains the receipt arrival order and receipt source identifier.

[0152] The closed-loop update module sorts task receipts by their issuance timestamps and establishes a task status mapping with the target status identifier of the nursing state machine. The task status mapping uses the target status identifier in the task record as the mapping value, so that each receipt can be traced back to the corresponding status identifier. Based on the execution window and the receipt timestamp, the closed-loop update module determines the effective interval of the task status and generates the actual execution status sequence.

[0153] When there is a valid completion mark in the receipt set, the timestamp of that receipt is used as the actual end time of the state, and the issuance timestamp or execution window start time of the task is used as the actual start time of the state.

[0154] If a completion receipt does not exist when the execution window ends, the timeout flag will be set to valid and the execution window termination time will be used as the actual termination time.

[0155] When an exception is marked as valid, an exception state fragment is written into the actual execution state sequence and the time of the exception is recorded for updating the sample entry for tracing. The actual execution state sequence is organized in chronological order into a sequence structure of state identifiers and their corresponding effective intervals, and updated sample entries are written to form a verifiable execution trajectory.

[0156] When generating updated sample entries, the closed-loop update module writes the target care plan identifier, task receipt set, actual execution status sequence and sleep state unit summary, and rehabilitation data frame sequence summary together. The sleep state unit summary includes at least the slice identifier, start and end time, sleep parameter vector, change indicator, and confidence marker summary fields. The rehabilitation data frame sequence summary includes at least the frame time coverage, training data execution segment summary, posture data quality segment summary, and confidence marker distribution summary.

[0157] The closed-loop update module then reconstructs the stable slice set based on the updated sample entries and updates the individual baseline parameters, including:

[0158] From the sleep state units associated with the updated sample entries, sleep state units whose confidence markers do not contain missing markers and noise markers are selected as a stable slice set. On the candidate set, the baseline central statistic and baseline discrete statistic of individual baseline parameters are updated according to preset statistical rules. Historical statistics are retained during the update to achieve traceable rolling updates.

[0159] The closed-loop update module converts the updated sample entries into incremental training input and supervision entries for the protocol generation model. The input entries consist of a summary of the sleep state unit, individual baseline parameters, and a summary of the rehabilitation data frame sequence. The supervision entries consist of a set of task feedback and a sequence of actual execution states. This triggers incremental training of the protocol generation model to update the model parameters, thereby achieving a closed-loop adaptive update of the target nursing protocol execution results to the generation of subsequent candidate nursing protocols. This type of incremental training method, which updates the model step by step based on continuously arriving data, is a common implementation path for online learning or incremental learning.

[0160] For example, if a patient undergoing home rehabilitation after discharge experiences a nighttime interruption in wearing the device, resulting in a loss of blood oxygen parameters, the timing access module generates a missing measurement marker and writes it into a confidence marker in the corresponding frame. Simultaneously, if the device becomes loose overnight, causing a sudden change in heart rate amplitude, a noise marker is generated and written into the confidence marker. The sleep modeling module then generates sleep state units and selects a stable slice set to establish individual baseline parameters. After the plan generation module outputs a set of candidate nursing plans and trajectory prediction records, the verification and deduction module performs time overlap matching between the next day's training task nodes and the start and end times of the sleep state units. For task nodes whose associated sleep state unit confidence markers contain missing measurement markers or noise markers, the nursing state machine transition is disabled, thereby eliminating the corresponding candidate nursing plans and selecting the target nursing plan. The closed-loop update module distributes tasks to the task queue and receives task receipts to generate updated sample entries, which are used to update individual baseline parameters and update the plan generation model.

[0161] For example, a short-term offline event at the training terminal for home rehabilitation subjects caused delayed receipts and a uniform timestamp reversal. The timing access module generated out-of-order markers and wrote them into confidence markers to maintain traceability. In the sleep state unit generated by the sleep modeling module, some slice change indicators continuously deviated from the individual baseline parameters and converged into noise markers. The solution generation module still output multiple sets of candidate care solutions. The verification and deduction module performed lexicographical comparisons based on backtracking and conflict counts and compared the number of consecutive segments between the training state and the rest state when necessary to reduce frequent backtracking and switching. The closed-loop update module wrote the anomaly markers and timeout markers into the update sample entries and formed the actual execution state sequence, driving the reconstruction of the stable slice set and the update of the solution generation model.

[0162] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0163] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0164] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0165] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0166] In conclusion, the above description is only 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 remote nursing analysis system for assessing rehabilitation status, characterized in that: It includes a timing access module, a sleep modeling module, a solution generation module, a verification and simulation module, and a closed-loop update module; The timing access module is used to access physiological data, training data, posture data, and nursing records, perform normalization and time alignment, and output a sequence of rehabilitation data frames with confidence markers. The sleep modeling module is used to extract a time series set of sleep parameters from the rehabilitation data frame sequence, slice it to generate sleep state units and establish individual baseline parameters; The plan generation module is used to construct a nursing plan atlas and compile a nursing state machine. Based on the plan generation model trained by machine learning, it takes sleep state units, individual baseline parameters and rehabilitation data frame sequences as inputs and outputs a set of candidate nursing plans and trajectory prediction records. The verification and simulation module is used to verify the set of candidate nursing solutions by applying safety constraints, resource constraints, and dependency mutual exclusion constraints based on confidence markers, and to select the target nursing solution by combining trajectory prediction records. The closed-loop update module is used to compile the target care plan into a task queue and output it, receive task receipts to generate updated sample entries, update individual baseline parameters based on the updated sample entries, and update the plan generation model.

2. The remote nursing analysis system for rehabilitation status assessment according to claim 1, characterized in that: The timing access module performs normalization and time alignment and generates a sequence of rehabilitation data frames, including: The timestamps in physiological data, training data, posture data, and nursing records are converted to a unified format and mapped to a unified time base. Resample each data source according to a uniform sampling period and generate a frame index; generate missing test markers for segments where the time interval between adjacent frames exceeds a preset range; Generate out-of-order tags for segments with reversed timestamps or frame indexes; generate noise tags for amplitude abrupt changes, saturation intervals, or continuous zero-value intervals; Confidence tags are generated based on missing test tags, out-of-order tags, and noise tags and written into the rehabilitation data frame sequence.

3. The remote nursing analysis system for rehabilitation status assessment according to claim 2, characterized in that: The sleep modeling module extracts a temporal set of sleep parameters from the rehabilitation data frame sequence, slices them to generate sleep state units, and establishes individual baseline parameters, including: Heart rate, respiratory, blood oxygen, body movement, and skin temperature parameters are extracted from the rehabilitation data frame sequence to form a time series set of sleep parameters; Perform time-consistency on the time series set of sleep parameters and generate change indicators; The slice boundaries are determined based on the change indicators and sleep state units are generated. The sleep state units include slice identifier, start and end time, sleep parameter vector, change indicators, and confidence flags. Construct a stable set of slices from slices whose confidence markers do not contain missing test markers and noise markers; Individual baseline parameters are generated based on statistics from a stable set of slices.

4. The remote nursing analysis system for rehabilitation status assessment according to claim 3, characterized in that: The solution generation module includes: Define training tasks, rest tasks, follow-up tasks, and reassessment tasks as task nodes and write the task type, execution window, duration, and intensity parameters; Generate dependency edges, mutual exclusion edges, trigger edges, and fallback edges between task nodes to form a nursing plan graph; A nursing state machine is generated based on the nursing plan graph. The nursing state machine includes a set of state identifiers and a set of transition conditions. The transition conditions are generated by mapping trigger edges and backoff edges. The model generates a set of candidate nursing plans by inputting sleep state units, individual baseline parameters, and rehabilitation data frame sequences into the scheme. A trajectory prediction record is generated for the candidate nursing plan set. The trajectory prediction record includes a state sequence, a state start and end time sequence, a trigger event sequence, and a resource usage sequence.

5. A remote nursing analysis system for rehabilitation status assessment according to claim 4, characterized in that: The verification and deduction module performs safety constraints, resource constraints, and dependency mutual exclusion constraints verification on the candidate nursing plan set based on confidence markers, including: Dependency constraints are generated based on the dependency edges of the nursing plan graph, and mutual exclusion constraints are generated based on the mutual exclusion edges of the nursing plan graph. Resource constraints are generated based on the execution window and duration of task nodes; Safety constraints are generated based on sleep state units and individual baseline parameters; The task nodes in the candidate nursing plan set are matched with the start and end times of the sleep state unit to establish the task slice association relationship. When the confidence flag of the sleep state unit corresponding to the task slice association contains a missing test flag or a noise flag, the nursing state machine transition corresponding to that task node is disabled. Candidate care solutions that simultaneously satisfy safety constraints, resource constraints, dependency constraints, and mutual exclusion constraints, and are not prohibited from enabling transfer, are added to the set of feasible care solutions.

6. A remote nursing analysis system for rehabilitation status assessment according to claim 5, characterized in that: The verification and simulation module, combined with trajectory prediction records, selects the target nursing plan, including: For each feasible care plan in the set of feasible care plans, read the trajectory prediction record and parse it to obtain the state sequence, trigger event sequence and resource consumption sequence; The number of backoff edge triggers is counted based on the trigger event sequence, and a backoff count is generated. The number of resource conflict segments is counted based on the resource occupancy sequence, and a conflict count is generated. Lexicographical comparison is performed within the set of feasible nursing solutions, first comparing backoff counts and then comparing conflict counts. If the comparison results are the same, the number of consecutive segments between the training state and the rest state in the state sequence is compared, and the feasible nursing solution with the best comparison result is output as the target nursing solution.

7. A remote nursing analysis system for rehabilitation status assessment according to claim 6, characterized in that: The closed-loop update module compiles the target care plan into a task queue and outputs it. It receives task acknowledgments, generates updated sample entries, updates individual baseline parameters based on these updated sample entries, and updates the plan generation model, including: The target nursing plan is expanded into a task queue, which includes a task identifier, a target status identifier, a distribution time tag, an execution window, and a receipt judgment condition. Receive task receipts from the task queue and generate updated sample entries. The updated sample entries include the target care plan identifier, task receipt set, sleep state unit summary, rehabilitation data frame sequence summary, and actual execution status sequence. The stable slice set is reconstructed based on the updated sample entries, and the individual baseline parameters are updated; the scheme generation model is incrementally trained based on the updated sample entries to update the model parameters.

8. A remote nursing analysis system for rehabilitation status assessment according to claim 5, characterized in that: The trajectory prediction records generated by the scheme generation module include: The set of candidate nursing solutions is expanded according to the nursing state machine according to the control cycle, and the start and end time series of the state is generated based on the execution window and duration of the task node. Map the transition conditions of the triggering edge and the backtracking edge to a sequence of triggering events and write them into the trajectory prediction record; Based on resource constraints, the resource usage segments of task nodes in each control cycle are analyzed and a resource usage sequence is generated; Write the state identifiers of the nursing state machine into the state sequence in chronological order and align them with the start and end time sequences of the states to output trajectory prediction records.

9. A remote nursing analysis system for rehabilitation status assessment according to claim 7, characterized in that: The closed-loop update module's process of receiving task feedback also includes: Create a receipt index for each task identifier in the task queue. The task receipt includes a receipt timestamp, completion flag, timeout flag, and exception flag. Sort the task receipts by their issuance time label and establish a task status mapping with the target status identifier of the nursing status machine. The effective interval of the task status is determined based on the execution window and the receipt timestamp, and the actual execution status sequence is generated. Write the actual execution state sequence into the updated sample entry.

10. A remote nursing analysis system for rehabilitation status assessment according to claim 7, characterized in that: The closed-loop update module generates a model based on the update scheme for the updated sample entries, including: The model input entries are constructed by extracting sleep state unit summaries, individual baseline parameters, and rehabilitation data frame sequence summaries from the updated sample entries. The monitoring items are constructed by extracting target care plan identifiers, task receipt sets, and actual execution status sequences from the updated sample items. A sorting label is generated for the candidate nursing plan set based on the task feedback set. The sorting label is determined by the completion mark ratio, the number of timeout marks, the number of abnormal marks, and the number of backtracking edge triggers. Incremental training is performed on the scheme generation model based on the model input entries and sorting labels to update the model parameters.

Images