Physiological world model-oriented full-modal physiological data acquisition method and system

Abstract

The application provides a physiological world model-oriented full-modal physiological data acquisition method and system. The method comprises the following steps: mapping multi-modal original data streams to a unified time coordinate to generate time-synchronized data streams; identifying key physiological events according to the time-synchronized data streams; according to the type of the key physiological events, controlling the corresponding acquisition device to switch from a baseline acquisition frequency to a high-precision frequency within an event window to obtain key window data; taking the sequence of the event windows as a reference, intercepting the time-synchronized data and the key window data within the event windows, and generating a multi-modal data set; based on the multi-modal data set, extracting distribution characteristics of physiological characteristic parameters, establishing an individual multi-day physiological baseline; extracting physiological response trajectories within the event windows and an observation period, determining state change results and recovery trajectories, and correlating to generate a human causal chain data set of key physiological events. The application can improve multi-modal data synchronization accuracy, key window acquisition integrity and continuity of human event response chains.

Description

Technical Field

[0001] This application relates to the field of physiological data acquisition and processing technology, and in particular to a method and system for acquiring full-modal physiological data oriented towards a physiological world model. Background Technology

[0002] With the development of wearable devices, remote health monitoring, and digital human body modeling technologies, various physiological monitoring devices and behavioral recording tools are increasingly being applied to daily health management. Current technologies typically utilize different devices to collect data such as heart rate, blood glucose, electrocardiogram, sleep, activity, and diet, and then aggregate and analyze this data through a data platform to achieve human condition monitoring and health risk assessment. However, different devices often use independent clocks, sampling frequencies, and data formats, making it difficult to accurately correlate multimodal data on the timeline. In particular, it is difficult to establish an effective correlation between high-frequency physiological waveforms and low-frequency state data within the same event range.

[0003] Meanwhile, existing data collection methods mostly employ fixed sampling frequencies, lacking mechanisms to dynamically adjust sampling accuracy based on key physiological events such as waking up, eating, exercise, and sleep transitions. This easily leads to data sparsity or missing data during critical transition phases. Furthermore, the relationship between behavioral events and physiological response data often relies on manual labeling, making it difficult to continuously reconstruct state changes and recovery processes after a behavior occurs. Individual physiological baselines are also mostly derived from single measurements or standard environments, failing to reflect the dynamic range of individuals under natural living conditions. Therefore, existing technologies struggle to generate high-quality datasets that are time-consistent, event-aligned, and capable of representing behavioral inputs, physiological responses, state changes, and recovery trajectories, thus hindering physiological world models and related digital applications of the human body. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method and system for acquiring multimodal physiological data based on a physiological world model, in order to solve the problems of difficulty in synchronizing multimodal data, insufficient acquisition accuracy in key windows, and incomplete human event response chains in the prior art.

[0005] A first aspect of this application provides a method for acquiring multimodal physiological data oriented towards a physiological world model, comprising: acquiring multimodal physiological and behavioral data of a target human body to generate a multimodal raw data stream; determining a reference clock, and mapping the multimodal raw data stream to a unified time coordinate based on the clock offset and drift of each acquisition device to generate a time-synchronized data stream; identifying key physiological events based on the time-synchronized data stream, determining event windows and generating an event window sequence; controlling the corresponding acquisition device to switch from a baseline acquisition frequency to a high-precision frequency within the event window according to the type of key physiological event to acquire key window data; extracting the time-synchronized data and key window data within the event window based on the event window sequence, and mapping them to a unified event time grid according to sampling attributes to generate an event-aligned multimodal dataset; extracting the distribution characteristics of physiological feature parameters based on the event-aligned multimodal dataset of multiple natural days to establish an individual multi-day physiological baseline; extracting physiological response trajectories within the event window and observation period based on key physiological events as time anchors, determining the state change results and recovery trajectories based on the differences between the physiological response trajectories and the individual multi-day physiological baseline, and generating a human causal chain dataset of key physiological events.

[0006] A second aspect of this application provides a multimodal physiological data acquisition system oriented towards a physiological world model, comprising: an acquisition module for acquiring multimodal physiological and behavioral data of a target human body and generating a multimodal raw data stream; a mapping module for determining a reference clock and mapping the multimodal raw data stream to a unified time coordinate based on the clock offset and drift of each acquisition device, generating a time-synchronized data stream; a determination module for identifying key physiological events based on the time-synchronized data stream, determining an event window, and generating an event window sequence; and a switching module for controlling the corresponding acquisition device to switch from a baseline acquisition frequency to a high-precision frequency within the event window according to the type of key physiological event, thereby acquiring key physiological data. The system comprises four modules: a key window data extraction module, a key window data extraction module, and a data generation module. The key window data extraction module extracts time-synchronized data and key window data within the event window sequence and maps them to a unified event time grid according to sampling attributes, generating an event-aligned multimodal dataset. The data generation module extracts the distribution characteristics of physiological parameters from the event-aligned multimodal dataset based on multiple natural days, establishing an individual's multi-day physiological baseline. The generation module uses key physiological events as time anchors to extract physiological response trajectories within the event window and observation period. Based on the differences between the physiological response trajectories and the individual's multi-day physiological baseline, it determines the state change results and recovery trajectories, and generates a human causal chain dataset of key physiological events.

[0007] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: By collecting multimodal physiological and behavioral data from the target human body, a multimodal raw data stream is generated. A reference clock is determined, and based on the clock offset and drift of each acquisition device, the multimodal raw data stream is mapped to a unified time coordinate to generate a time-synchronized data stream. Key physiological events are identified based on the time-synchronized data stream, event windows are determined, and event window sequences are generated. According to the type of key physiological event, the corresponding acquisition device is controlled to switch from the baseline acquisition frequency to a high-precision frequency within the event window to acquire key window data. Using the event window sequence as a reference, time-synchronized data and key window data within the event window are extracted and mapped to a unified event time grid according to sampling attributes to generate an event-aligned multimodal dataset. Based on the event-aligned multimodal dataset of multiple natural days, the distribution characteristics of physiological feature parameters are extracted to establish an individual multi-day physiological baseline. Using key physiological events as time anchors, physiological response trajectories within the event window and observation period are extracted. Based on the difference between the physiological response trajectory and the individual multi-day physiological baseline, the state change results and recovery trajectory are determined, and a human causal chain dataset of key physiological events is generated. This application can improve the synchronization accuracy of multimodal data, the completeness of key window acquisition, and the continuity of the human event response chain. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the full-modal physiological data acquisition method based on a physiological world model provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of the full-modal physiological data acquisition system based on the physiological world model provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0011] In existing technologies, wearable physiological monitoring devices, continuous biochemical indicator monitoring devices, electrocardiogram (ECG) acquisition devices, and behavioral recording tools are typically used to collect human physiological and behavioral data, which are then aggregated and analyzed through a data platform. However, because these various acquisition devices operate independently, they employ different clock references, sampling frequencies, and data formats, making it difficult to accurately correlate data from different modalities on a unified timeline. Furthermore, existing acquisition schemes often use fixed-frequency continuous acquisition, lacking dynamic sampling control mechanisms for key physiological events such as eating, exercise, waking up, and sleep transitions. This can easily lead to insufficient data accuracy or missing data during periods of rapid change in the body's state.

[0012] Therefore, existing technologies suffer from several problems, including difficulty in accurately synchronizing multimodal physiological and behavioral data, insufficient data collection completeness within key physiological event windows, and a lack of continuous correlation between behavioral events and physiological response processes. Furthermore, existing individual physiological baselines are typically established based on single measurement results or standardized environmental data, which fails to reflect the range of physiological fluctuations in individuals under natural living conditions. Consequently, it is difficult to accurately determine the magnitude of state changes and recovery processes caused by behavioral events, hindering the formation of dynamic human process data suitable for training physiological world models.

[0013] To address the aforementioned technical issues, this application collects multimodal physiological and behavioral data from the target human body. By determining a reference clock and the clock offsets and drifts of each acquisition device, the raw multimodal data streams generated by different devices are mapped to a unified time coordinate. Based on this, key physiological events are identified according to the time-synchronized data stream, and corresponding event windows are determined. According to the type of key physiological event, the relevant acquisition devices are controlled to switch from the baseline acquisition frequency to a high-precision frequency within the event window. Furthermore, using the event window sequence as a reference, time-synchronized data and key window data are truncated, and mapped to a unified event time grid according to the sampling attributes of different data, generating an event-aligned multimodal dataset.

[0014] This application also extracts the distribution characteristics of physiological feature parameters based on multimodal data aligned with events from multiple natural days, and establishes an individual multi-day physiological baseline that can characterize the intraday variation pattern and dynamic range of an individual; using key physiological events as time anchors, it extracts the physiological response trajectory within the event window and subsequent observation period, and determines the state change results and recovery trajectory based on the difference between the physiological response trajectory and the individual multi-day physiological baseline, thereby forming a human causal chain dataset.

[0015] Through the above technical solutions, this application can improve the time synchronization accuracy between different sampling frequencies and different data modalities, enhance the resolution and completeness of data acquisition within the key physiological event window, and improve the adaptability of individual physiological baselines to natural living conditions. Simultaneously, it can continuously characterize the physiological response, state changes, and recovery process after behavioral input, providing a time-unified, event-aligned, and dynamically information-rich data foundation for physiological world models, human digital twins, and intelligent health analysis.

[0016] The technical solution of this application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] Figure 1 This is a flowchart illustrating the full-modal physiological data acquisition method based on a physiological world model provided in this application embodiment. Figure 1 As shown, the method may specifically include: S101: Collect multimodal physiological and behavioral data of the target human body and generate multimodal raw data stream; S102, determine the reference clock, and based on the clock offset and drift of each acquisition device, map the multimodal raw data stream to a unified time coordinate to generate a time-synchronized data stream; S103, Identify key physiological events based on time-synchronized data stream, determine event windows, and generate event window sequences; S104, based on the type of key physiological event, control the corresponding acquisition device to switch from the baseline acquisition frequency to the high-precision frequency within the event window to acquire key window data; S105, based on the event window sequence, extracts the time synchronization data and key window data within the event window, and maps them to a unified event time grid according to the sampling attributes to generate an event-aligned multimodal dataset; S106, based on a multimodal dataset aligned with events from multiple natural days, extracts the distribution features of physiological characteristic parameters and establishes an individual's multi-day physiological baseline; S107 uses key physiological events as time anchors to extract physiological response trajectories within the event window and observation period. Based on the differences between the physiological response trajectories and the individual's multi-day physiological baseline, it determines the state change results and recovery trajectories, and generates a human causal chain dataset of key physiological events.

[0018] In some embodiments, multimodal physiological and behavioral data of the target human body are collected to generate a multimodal raw data stream, including: Multiple types of data acquisition devices are configured on the target human body for continuous sensing of physiological state, intermittent measurement of human body state and recording of behavioral input, and a correspondence between device identifiers and data modalities of each data acquisition device is established. Control multiple types of acquisition devices to acquire continuous physiological signals, periodic state data and behavioral event data according to their corresponding acquisition configurations; Associate the data output by each acquisition device with the device identifier, local acquisition time, and sampling attributes, and encapsulate it in a structured manner according to the corresponding data modality; The structured and encapsulated data is collected based on local acquisition time to generate a multimodal raw data stream.

[0019] Specifically, in one application scenario, the target human body is an adult user undergoing physiological state observation for seven consecutive natural days. Before the data collection begins, a wrist-based physiological data acquisition device integrating a photoplethysmography (PPG) sensor, a triaxial accelerometer, a skin temperature sensor, and an ambient light sensor is worn on the user's wrist. A continuous glucose monitoring patch is placed on the upper arm or abdomen, and a three-lead electrocardiogram (ECG) acquisition patch is attached to the chest. Body composition measurement equipment, a portable electroencephalogram (EEG) acquisition device, a behavior recording terminal, a diet and nutrition recording terminal, and a body state recording terminal are also configured.

[0020] The system assigns a unique device identifier to each acquisition device and records the device identifier, acquisition object, data mode, rated sampling frequency, data unit, communication method, and start / stop status in the device registry. For example, wrist devices correspond to pulse wave, activity intensity, skin temperature, and ambient light modes; continuous glucose monitoring patches correspond to interstitial fluid glucose modes; and ECG patches correspond to multi-lead ECG waveform modes.

[0021] Continuous sensing devices operate continuously at a low-burden baseline configuration. A wrist-worn device acquires pulse waves at 25 Hz and acceleration at 50 Hz, generating summaries of heart rate, activity intensity (including respiration and blood oxygenation), and skin temperature every minute. A continuous glucose monitoring patch outputs glucose concentration every 5 minutes. An ECG patch acquires ECG waveforms at preset intervals throughout the day, simultaneously recording lead status and electrode contact quality in the sampling log. A body composition measurement device performs measurements once each morning upon waking and before bedtime, outputting periodic data such as weight, body fat percentage, and body water percentage. A portable EEG acquisition device is activated before bedtime and during designated cognitive tasks to output EEG waveforms and electrode impedance information.

[0022] Behavioral input is obtained collaboratively by a behavior recording terminal, a diet and nutrition recording terminal, and a body status recording terminal. The behavior recording terminal records events such as waking up, eating, exercising, sleeping, and taking medication based on user actions or device detection results, and can save time-stamped images or short video indexes when events occur. The diet and nutrition recording terminal receives food categories, intake amounts, start and end times of meals, and the order of eating, generating meal records. The body status recording terminal receives subjective ratings such as fatigue level, hunger level, emotional state, and physical discomfort level according to preset prompt cycles. For exercise events not actively recorded by the user, the wrist device generates candidate behavioral events based on acceleration intensity and duration, which are then supplemented and confirmed by the behavior recording terminal.

[0023] Each time a device outputs data, it includes the device identifier, local acquisition time, sampling frequency, sampling duration, data unit, significant bits, and quality status. Continuous waveforms are encapsulated according to the device identifier, start time, sampling interval, and sample value sequence. Periodic status data is encapsulated according to the measurement time, parameter name, and parameter value. Behavioral event data is encapsulated according to the event type, start time, end time, source method, and confirmation status. For point-in-time images, only the file index, capture time, and behavior association identifier are saved in the data record to avoid directly writing large images to the continuous data channel.

[0024] The data aggregation end establishes a receiving buffer based on the local acquisition time carried by each data record and identifies the data source based on the device identifier. For devices with temporary network interruptions, data is first acquired locally and then retransmitted in the order of local acquisition time after the connection is restored. For duplicate data uploads, a unique data key is generated based on the device identifier, data modality, and local acquisition time, and deduplication is performed. After deduplication and integrity verification are completed, the data aggregation end writes continuous physiological signals, periodic state data, and behavioral event data into a unified multimodal raw data stream, retaining the original time and sampling attributes of each data point for subsequent clock correction, event recognition, and high-precision acquisition of critical windows.

[0025] Through the above implementation methods, continuous signals, periodic measurement results, and behavioral input records can be incorporated into a unified data entry point without changing the original acquisition characteristics of various devices. This preserves the temporal details of high-frequency waveforms as well as the original semantics of low-frequency state data and discrete events, providing a complete and traceable data foundation for subsequent multi-device time synchronization, event window recognition, and the construction of human response chains.

[0026] In some embodiments, a reference clock is determined, and based on the clock offset and drift of each acquisition device, the multimodal raw data stream is mapped to a unified time coordinate to generate a time-synchronized data stream, including: From all the acquisition devices, a reference acquisition device is determined, and the device clock of the reference acquisition device is determined as the reference clock. Obtain the time difference between the other acquisition devices and the reference acquisition device at multiple synchronization moments, and determine the clock offset and clock drift of the corresponding acquisition devices based on the time difference; Establish a time mapping relationship between the local time of each acquisition device and the unified time coordinate based on clock offset and clock drift; The time mapping relationship is updated according to the preset synchronization cycle, and the local acquisition time of each data in the multimodal raw data stream is corrected based on the updated time mapping relationship to generate a time synchronization data stream carrying a synchronization timestamp.

[0027] Specifically, in a seven-day continuous human physiological monitoring scenario, the data collection terminal establishes communication connections with wrist physiological data acquisition devices, continuous glucose monitoring patches, three-lead ECG patches, body composition measurement devices, EEG acquisition devices, and behavior recording terminals. The system prioritizes the wrist physiological data acquisition device, which is continuously online, has high clock stability, and can periodically communicate with the data collection terminal, as the reference acquisition device, and determines the built-in clock of this device as the reference clock. During the device registration phase, the device identifier, local clock accuracy, communication latency range, and most recent calibration time of each acquisition device are recorded as the basic parameters for subsequent time correction.

[0028] For other acquisition devices such as continuous glucose monitoring patches and three-lead ECG patches, the data aggregation terminal sends synchronization requests carrying the reference transmission time according to a preset synchronization procedure. After receiving the synchronization request, each acquisition device returns its local reception time, local transmission time, and device identifier. The data aggregation terminal then estimates the time difference between the device's local time and the reference time at the same synchronization moment by combining the request round-trip time delay. To reduce errors caused by wireless transmission fluctuations, multiple measurements are performed continuously in each synchronization round. Measurement results with significantly abnormal round-trip time delays are discarded, and the remaining time differences are weighted and statistically analyzed to obtain the clock offset corresponding to the current synchronization round.

[0029] The system repeats the above measurements at the start of monitoring, during operation, and at multiple subsequent synchronization points. For example, a continuous glucose monitoring patch is 1.8 seconds slower than the reference clock at the start of monitoring and 2.3 seconds slower after 6 hours of operation. The system calculates the clock drift rate based on the change in time difference and the interval between multiple synchronization points, which characterizes the cumulative deviation of the device's local clock over operating time. For high-frequency acquisition devices such as ECG patches, the synchronization cycle can be shortened; for low-frequency devices such as body composition measurement devices, synchronization can be performed at each measurement start to balance time accuracy and device power consumption.

[0030] For each acquisition device, the system establishes a mapping relationship between local time and a unified time coordinate based on the most recent valid synchronization time, clock offset, and clock drift rate. The synchronization timestamp of the data record is obtained by subtracting the corresponding clock offset from the local acquisition time and performing drift compensation according to the interval between the local acquisition time and the most recent synchronization time. For continuous waveform data, the system corrects the start time of the data segment and calculates the synchronization timestamp of each sampling point within the segment based on the sampling interval; for periodic state data and behavioral event data, the system corrects the measurement time, event start time, and event end time, respectively.

[0031] Synchronization parameters are updated according to a preset cycle corresponding to the device type. For example, wrist devices and ECG patches can be calibrated every 30 minutes, continuous glucose monitoring patches every 2 hours, and low-frequency measurement devices can be calibrated before each use. The system also saves the offset change trend of several consecutive synchronization rounds. When the drift rate changes stably within the allowable range, linear compensation is used. When the offset change is abrupt, the historical drift rate is stopped and the time mapping relationship is re-established. When a device restart, communication interruption recovery, clock jump, or synchronization error exceeding the threshold is detected, additional synchronization is immediately triggered, and the failed parameters are replaced with the newly obtained clock offset and clock drift rate. For data with insufficient synchronization quality, the system retains the original local acquisition time and adds a low-confidence time marker to avoid direct participation in high-precision event alignment.

[0032] Furthermore, after calibration, the data aggregation terminal associates a synchronization timestamp, original local acquisition time, synchronization parameter version, and synchronization quality level with each record in the multimodal raw data stream, and sorts them according to the synchronization timestamp to form a time-synchronized data stream with a unified time coordinate. This method corrects the initial clock deviation and operational drift of different acquisition devices, enabling millisecond-level ECG data, minute-level glucose data, periodic body component data, and discrete behavioral events to be accurately correlated on the same time axis, providing a reliable time basis for subsequent identification of key physiological events and alignment of multimodal data within the event window.

[0033] In some embodiments, identifying key physiological events based on time-synchronized data streams, determining event windows, and generating event window sequences include: Extract event recognition features that characterize changes in human activity state, behavioral input, and environmental state from time-synchronized data streams; The event identification features are matched with preset event identification rules, and candidate key physiological events are determined based on state transition conditions, the time of occurrence of the behavior, and the duration conditions. Based on the triggering time and event type of each candidate key physiological event, configure the corresponding window start time, window end time, and event confidence. Window boundary correction and event association are performed on candidate key physiological events that overlap or are correlated in time, and event window sequences carrying event type, window boundary and event confidence are generated in chronological order.

[0034] Specifically, in a human physiological monitoring scenario spanning seven consecutive natural days, the data analysis terminal reads data output from the wrist device, continuous glucose monitoring patch, ECG patch, diet and nutrition recording terminal, behavior recording terminal, and environmental sensing unit in sequence according to synchronized timestamps. The system analyzes data at a granular level of one minute, calculating activity intensity, posture changes, and rest duration from wrist acceleration data; extracting short-term trends from heart rate and heart rate variability data; extracting light levels and directions of change from ambient light data; and extracting behavior type, occurrence time, duration, and confirmation status from diet, exercise, sleep, and medication records, forming an event identification feature sequence corresponding to a unified time coordinate.

[0035] For wake-up events, the system detects the transition from a static to an active state within a preset morning time period. When the continuous static duration reaches more than 30 minutes, followed by an activity intensity exceeding the wake-up threshold for 5 minutes, and the posture changes from lying down to standing, the moment when the transition condition is first met is identified as a candidate wake-up trigger moment. For eating events, the system reads the start time of the meal generated by the diet and nutrition recording terminal, and verifies the authenticity of the event by combining the upload time of the time-point image, continuous hand movement characteristics, or user confirmation information. Meal records that pass the verification are identified as candidate eating events.

[0036] For exercise events, the system detects whether the activity intensity continuously exceeds the exercise threshold and confirms the exercise process by combining the increase in heart rate. When the activity intensity drops back to the normal level and continues for a preset duration, the starting time of the drop is determined as the candidate end time of the exercise. For bedtime events, the system detects the transition from an active state to a continuous static state during the nighttime period and determines the candidate bedtime trigger time when the ambient light is below the threshold and no subsequent significant activity occurs.

[0037] The system configures window boundaries according to event type. For example, the wake-up window can extend from 10 minutes before the trigger to 30 minutes after the trigger; the post-meal window can extend from the first bite of food to 120 minutes; the post-exercise window can extend from the end of exercise to 60 minutes thereafter; and the bedtime window can extend from 20 minutes before the start of a period of stillness to 30 minutes after sleep stabilizes. In practical applications, the bedtime window can also be the entire sleep cycle. The window duration can also be dynamically adjusted based on the duration of the behavior, historical response delays, and the availability of the data acquisition device, avoiding the omission of physiological changes that occur due to delays when using a fixed window.

[0038] Event confidence is determined by the degree of rule fulfillment and the consistency of multi-source evidence. Taking the wake-up event as an example, a higher confidence level is assigned when the activity state transition, posture change, and morning time period are all met; a lower confidence level is assigned when only enhanced activity is detected but posture evidence is lacking. For the eating event, the confidence level is higher when there are terminal records, image indexes, and user confirmations simultaneously than when there is only a single record source.

[0039] User confirmation refers to the process where, after detecting a meal record or receiving a picture of food consumption, the dietary nutrition recording terminal outputs an event confirmation prompt to the user, who then confirms or corrects whether the eating behavior occurred, the start and end times of the meal, and the meal information. The terminal records the user's confirmation result, confirmation time, and corrections, using this as evidence of the eating event. When the terminal record, picture index, and manual confirmation evidence match in time, the confidence level of the eating event is increased; when there is only a single record source or inconsistencies between different sources, the confidence level of the eating event is correspondingly decreased.

[0040] Post-exercise events are scored based on a comprehensive evaluation of activity intensity duration, heart rate response, and consistency of exercise recordings. Candidate events below a minimum threshold are temporarily stored as pending confirmation events and are not directly used to trigger high-precision data acquisition.

[0041] When multiple candidate events overlap in time, the system adjusts the window based on the semantic association and trigger sequence between the events. For example, if walking begins shortly after dinner, the post-dinner window and the post-exercise window can be retained as related windows, and their shared time periods are recorded separately. When the pre-sleep resting state is directly caused by the end of exercise, the system determines whether to shorten the post-exercise window or delay the pre-sleep window based on ambient light, subsequent activities, and sleep characteristics. For repeated candidate events of the same type, if the trigger time interval is less than the merging threshold and the event has not recovered to a stable state in between, they are merged into one event window, and the window boundaries are updated using the earliest trigger time and the latest end time.

[0042] Furthermore, after calibration, the system sorts the event windows according to their trigger times, assigns an event identifier to each event window, and associates the event type, trigger time, window start time, window end time, event confidence level, evidence source, and associated event identifier to generate an event window sequence. In this way, continuous physiological signals, discrete behavioral records, and environmental changes can be converted into event windows with clear time boundaries and confidence levels, providing a unified basis for subsequent switching of sampling frequencies according to event type and constructing physiological response trajectories after behavioral input.

[0043] In some embodiments, based on the type of key physiological event, the corresponding acquisition device is controlled to switch from the baseline acquisition frequency to a high-precision frequency within the event window to acquire key window data, including: Based on the key physiological event types, a matching acquisition strategy is used to determine the associated target data modalities and corresponding acquisition devices; Based on the window boundaries of the event window and the sampling attributes of the target data modality, the high-precision frequency and acquisition duration are determined, and a sampling switching command is generated. Send the sampling switching command to the corresponding acquisition device, so that the corresponding acquisition device can collect data at a high-precision frequency within the event window and add a synchronization timestamp; After the event window ends, control the corresponding acquisition device to return to the baseline acquisition frequency, and encapsulate the data acquired within the event window into key window data.

[0044] Specifically, in a seven-day continuous human physiological monitoring scenario, the sampling control terminal pre-establishes a strategy mapping between event types, target data modalities, acquisition devices, and sampling configurations. The wake-up event is associated with wrist pulse wave, acceleration, skin temperature, and multi-lead ECG data; the eating event is associated with continuous glucose, heart rate, heart rate variability, and dietary records; the post-exercise event is associated with ECG waveform, activity intensity, skin temperature (the thermic effect of food after eating causes a slight increase in body temperature of approximately 0.5 degrees Celsius), and continuous glucose data; and the bedtime event is associated with heart rate, heart rate variability, body movement, ambient light, and EEG data. The strategy mapping also records the baseline acquisition frequency, high-precision frequency, allowed duration, remaining device battery requirements, and data caching method for each modality.

[0045] When a post-meal window appears in the event window sequence, the sampling control terminal reads the post-meal acquisition strategy based on the event type and identifies the continuous glucose monitoring patch, wrist physiological data acquisition device, and ECG patch as the corresponding acquisition devices. Within the hardware support range, the continuous glucose monitoring patch adjusts its output from once every 5 minutes to once every 1 minute. The wrist device increases the pulse wave sampling frequency from 25 Hz to 100 Hz and shortens the heart rate and heart rate variability summary output interval from 1 minute to 10 seconds. The ECG patch switches from intermittent acquisition to continuous acquisition. For the post-exercise window, the system can increase the acceleration sampling frequency from 50 Hz to 100 Hz and maintain continuous ECG waveform acquisition to capture the decline after the heart rate peak. For the pre-sleep window, the system increases the acquisition density of body movement and heart rate variability data and initiates a preset pre-sleep EEG acquisition task when the EEG device is available.

[0046] The sampling control terminal reads the start and end times of the event window and the event confidence level, and determines the high-precision frequency and acquisition duration based on the original sampling attributes of each target data mode. For continuous waveform data, the high-precision frequency is determined according to the equipment's rated capacity and the target time resolution; for periodic output data, the time density is increased by shortening the output interval; for low-frequency measurement data, an additional directional measurement is added within the window instead of meaningless continuous frequency upsampling. When the event confidence level is low, the equipment power is insufficient, or the communication status is unstable, the system reduces the sampling frequency or reduces the number of participating devices according to preset degradation rules to avoid interruption of the acquisition task.

[0047] Subsequently, the sampling control terminal generates a sampling switching command containing an event identifier, device identifier, target modality, switching time, high-precision frequency, acquisition duration, and recovery conditions, and sends it to the corresponding acquisition device via the device communication channel. After the device returns configuration confirmation information, it enters high-precision acquisition state at the start time of the event window; if the command is issued later than the start time of the window, acquisition begins from the time the command takes effect, and the actual switching time is recorded. When acquiring data, each device calls the current time mapping relationship to convert the local acquisition time into a synchronization timestamp, and simultaneously associates the event identifier, sampling configuration version, and data quality status.

[0048] During the event window, the sampling control terminal periodically checks device response, buffer usage, and sampling integrity rate. When device overheating, battery level below a threshold, or data packet loss rate exceeding limits are detected, the high-precision frequency is promptly reduced, and the configuration change is recorded in the critical window data. When the event window ends and does not overlap with subsequent associated windows, the system sends a recovery command to each device, causing the device to return to the baseline sampling frequency. If the post-meal window overlaps with the post-exercise window, sampling continues according to the higher sampling requirement in the configuration corresponding to the two events until all associated windows end.

[0049] Furthermore, after data acquisition, the system aggregates data from each device within the window according to event identifiers, writes the actual sampling frequency, synchronization timestamp range, sampling integrity rate, configuration switching records, and quality identifiers into the window metadata, and encapsulates it into key window data according to data modality. Through this method, while controlling device power consumption and data load, the system can improve the temporal resolution and data integrity during periods of rapid change in human condition, providing a high-precision data foundation for subsequent event time grid alignment and physiological response trajectory extraction.

[0050] In some embodiments, based on the event window sequence, time synchronization data and key window data within the event window are extracted and mapped to a unified event time grid according to sampling attributes to generate an event-aligned multimodal dataset, including: Based on the window boundaries of each event window in the event window sequence, extract the corresponding data segments from the time synchronization data stream and the key window data; Based on the sampling frequency and timestamp density of each data segment, the corresponding data mapping method is determined, and a unified event time grid covering the window boundary is established. High-precision acquired data segments are interpolated and mapped according to a unified event time grid, while baseline acquired data segments retain their original timestamps and establish time-location associations. The mapped modal data fragments are combined according to the event window and associated with the corresponding key physiological events and sampling attributes to generate an event-aligned multimodal dataset.

[0051] Specifically, in a seven-day human physiological monitoring scenario, the data processing unit reads the event window sequence and parses the start time, end time, event type, and event confidence level of each window according to the event identifier. Taking a lunch event as an example, the start time of eating is 12:05, the end time is 12:25, and the post-meal observation range extends to 14:05. Based on this, the data processing unit extracts wrist heart rate, activity intensity, skin temperature, continuous glucose, and baseline ECG data for the corresponding time period from the time-synchronized data stream. It also extracts pulse wave, heart rate variability, and continuous ECG data obtained by increasing the sampling frequency within that time period from the key window data. During extraction, the device identifier, synchronization timestamp, actual sampling frequency, quality status, and sampling configuration version of each data segment are retained to avoid confusion between high-precision data and baseline data.

[0052] Wrist heart rate is calculated from the photoplethysmography (PPG) signal acquired by the wrist device. It mainly represents the number of pulses per unit time and its trend, and is suitable for reflecting heart rate levels over a long period of time with low workload, but it usually does not directly provide the waveform of cardiac electrical activity. Baseline ECG data is obtained by a multi-lead ECG patch in daily acquisition mode, including the ECG waveform, heart cycle, waveform morphology, temporal relationship between leads, and corresponding sampling quality information for each lead. It can further extract ECG characteristics such as heart rate, heart interval, P wave, QRS complex, and T wave duration and amplitude. Based on features such as premature heartbeat, compensatory pause, prolonged atrioventricular conduction time, abnormally widened QRS complex, or irregular heart rhythm, it can identify suspected premature beats, conduction block, and other abnormal ECG activity.

[0053] Continuous ECG data within the critical window refers to multi-lead ECG waveforms acquired by switching from intermittent to continuous high-frequency acquisition within event windows such as after meals and after exercise. This data is used to capture short-term heart rate changes before and after an event, as well as the timing, duration, and progression of abnormal heartbeats. The system uses wrist heart rate to characterize continuous heart rate trends, and uses baseline ECG data and continuous ECG data within the critical window to characterize cardiac electrical activity and suspected abnormal ECG events. Abnormal event identifiers, corresponding waveform segments, and event time locations are associated with the same event window.

[0054] The data processing unit statistically analyzes the sampling frequency and adjacent timestamp intervals of each data segment, dividing the data into high-frequency continuous data, periodic output data, and discrete event data. For 100 Hz pulse waves and continuous ECG waveforms, waveform-level time positions are established with a 10-millisecond basic interval; for heart rate and heart rate variability summaries output every 10 seconds, an analysis interval of 10 seconds is used; for high-precision glucose data output every 1 minute and baseline glucose data output every 5 minutes, a state analysis interval of 1 minute is used; for dietary records, subjective state records, and body composition measurements, event positions are established based on the actual occurrence time. This forms a unified event time grid covering 12:05 to 14:05, with the start time of eating set as a relative zero time, ensuring that each data point has both absolute synchronization time and relative event time.

[0055] For high-precision acquired data segments, the data processing end performs interpolation mapping according to the corresponding grid. For example, when there are valid sampling points between adjacent minutes in glucose data, the interpolation result of the grid position is calculated using adjacent sampling values ​​and time distance; linear interpolation is used for heart rate summaries, and local interpolation is only performed for continuous ECG and pulse waves when there are short-term missing values ​​and the missing length is less than the allowable threshold. If the missing range is too long or the data quality does not meet the requirements, no estimated value is generated, but a missing value is written at the corresponding grid position. For high-frequency waveforms, the original waveform is not compressed into low-frequency single values, but the original sampling sequence within the grid interval is retained, and the interval statistical results are extracted for subsequent trajectory analysis.

[0056] For baseline-acquired data segments, the data processing end maintains the original synchronization timestamps and does not force the generation of non-existent intermediate measurements. For example, glucose data collected every 5 minutes is time-correlated with the nearest grid position, and the time deviation between the sampling point and the grid position is recorded; body composition data is only correlated to the actual measurement time and is not extended to the entire postprandial window; behavioral event data has the start and end times of eating, food intake, and meal identifier written to the corresponding event nodes. When both baseline and high-precision data exist for the same modality, high-precision data is used to cover the corresponding time period first, while the baseline data index is retained to verify the numerical continuity before and after sampling switching.

[0057] Furthermore, after mapping, the data processing unit uses event identifiers as the primary index to combine data fragments from each modality according to a unified event time grid, forming post-meal event data units. Each data unit is associated with event type, window boundaries, relative time axis, data modality, sampling frequency, mapping method, missing status, and quality level. In cases where the post-meal window overlaps with the post-exercise window, data within the shared time period is stored only once, and associations are established with both event identifiers separately to avoid statistical bias caused by duplicate data. Multiple event data units are aggregated according to their occurrence time to generate an event-aligned multimodal dataset.

[0058] Through the above implementation methods, while preserving the original sampling characteristics of different modalities, millisecond-level electrocardiogram waveforms, second-level heart rate summaries, minute-level glucose data, and discrete behavioral records can be made to correspond to each other under the same event time coordinate, reducing information distortion caused by forced resampling, and providing a complete data foundation for subsequent establishment of individual multi-day physiological baselines, extraction of physiological response trajectories, and analysis of recovery processes.

[0059] In some embodiments, based on a multimodal dataset aligned with events from multiple natural life days, the distribution features of physiological characteristic parameters are extracted to establish an individual's multi-day physiological baseline, including: From the event-aligned multimodal dataset, multi-day data that meets the data integrity criteria are selected, and physiological feature parameters corresponding to each data modality are extracted. Physiological characteristic parameters are divided into time periods according to the natural day cycle and preset sliding windows to obtain feature samples corresponding to each sliding window; The distribution statistics of the feature samples are performed to determine the central tendency, dispersion, volatility and quantile intervals, and intraday variation curves are generated based on the statistical results of consecutive natural days. By correlating the distribution statistics and intraday variation curves with the corresponding time periods, the baseline range and temporal variation pattern of each physiological characteristic parameter are determined, and an individual's multi-day physiological baseline is established.

[0060] Specifically, in a seven-day human physiological monitoring scenario, the data processing unit first checks the multimodal dataset aligned with the event according to the calendar day and event identifier. Completeness criteria may include a minimum effective daily wear time of 20 hours, a continuous glucose recording rate of at least 90%, a missing heart rate data rate of at least 10%, complete sleep period coverage, and key event windows containing at least one continuous physiological data point and one behavioral data point. Calendar days that do not meet the criteria due to device detachment, communication interruption, or lack of user behavior recording are not directly included in the baseline statistics. Data with only partial missing data and a missing duration below the allowable threshold are included in the corresponding valid time period while retaining the missing identifier. After screening, for example, six complete calendar days are obtained from the seven calendar days as the individual baseline sample period.

[0061] The data processing unit extracts physiological characteristic parameters reflecting an individual's steady state and rhythm changes from various data modalities. Wrist data is used to extract data such as resting heart rate, active heart rate, daily activity level, skin temperature, and sleep-wake cycle characteristics. Continuous glucose data is used to extract fasting levels, pre-meal levels, post-meal peak levels, and intraday fluctuations. ECG results are used to extract heart rate intervals and calculate time-domain and frequency-domain parameters of heart rate variability. Sleep data is used to extract sleep onset time, sleep duration, sleep stage ratio, and number of awakenings. For heart rate variability, the standard deviation of normal heart rate intervals, root mean square of the difference between adjacent heart rate intervals, low-frequency power, high-frequency power, and the ratio of low-frequency to high-frequency power can be obtained, and associated with the corresponding data quality level.

[0062] Subsequently, a daily cycle was defined as the period from midnight to midnight of the following day, and a 30-minute sliding window was used to divide continuous physiological characteristics, with adjacent windows overlapping by 15 minutes. For data with clear state attributes, such as pre-meal, post-meal, post-exercise, and sleep, state windows were also set according to event type to avoid directly incorporating high heart rates after exercise into the resting heart rate baseline. Each sliding window aggregated feature values ​​from the same time period and similar states across multiple daily days, forming a window feature sample. For example, resting heart rate data from 6:30 AM to 7:00 AM over 14 daily days were used to form the morning window sample, and heart rate variability data from 11:00 PM to 11:30 PM were used to form the bedtime window sample.

[0063] For each set of window feature samples, the data processing unit calculates the mean and median to represent central tendency, the standard deviation and interquartile range to represent dispersion, and the coefficient of variation to represent relative volatility, and determines the quantile intervals corresponding to the 5th to 95th percentiles. For parameters susceptible to occasional anomalies, obvious outliers are first removed based on data quality and quantile boundaries, and then the statistical results are recalculated. Taking resting heart rate as an example, if the multi-day mean of the morning window is 62 beats per minute, the standard deviation is 4 beats per minute, and the 5th to 95th percentiles are 56 to 69 beats per minute, then this interval is used as the reference range for the corresponding time period. By connecting the statistical center values ​​of each sliding window in chronological order, intraday variation curves for heart rate, heart rate variability, glucose, sleep level, and activity level are generated.

[0064] Finally, the data processing unit correlates the distribution statistics of each window, intraday variation curves, status attributes, and the number of valid samples to form an individual multi-day physiological baseline organized by time period and physiological state. After newly collected data enters the corresponding time period, it is compared with the matched baseline range; when multiple consecutive sampling points exceed the baseline range, a baseline deviation marker is generated and associated with the current key physiological events to distinguish between normal event responses and non-event-related abnormal changes.

[0065] In some examples, when the number of valid samples is insufficient, the baseline for the corresponding time period is not output temporarily; instead, the data accumulation period is extended. When subsequent new daily data passes the integrity check, a rolling update method is used to correct the distribution parameters and intraday variation curves, and the baseline version and effective time are recorded. Through this method, an individual's multi-day physiological baseline can simultaneously reflect the statistical range, diurnal rhythm, and state differences under natural living conditions, providing an individualized reference for subsequently determining the amplitude of physiological responses, the direction of state changes, and recovery time.

[0066] In some embodiments, using key physiological events as time anchors, the physiological response trajectory within the event window and observation period is extracted, including: Based on the event type and window boundaries of key physiological events, the corresponding observation duration is determined, and the event trigger time is determined as the relative time reference. Extract event windows and physiological parameter sequences associated with key physiological events within the observation period from event-aligned multimodal datasets; The physiological parameter sequences are arranged temporally based on a relative time reference, and the continuity of abnormal sampling points and missing locations is processed. Extract the magnitude, direction, and rate of change of physiological parameter sequences relative to the pre-event state to generate physiological response trajectories corresponding to key physiological events.

[0067] Specifically, in a seven-day continuous human physiological monitoring scenario, the data processing terminal reads the event type, trigger time, window start time, and window end time of key physiological events according to event identifiers, and configures the corresponding observation duration based on the event type. For example, the observation period for a wake-up event is 30 to 60 minutes after triggering; for a post-meal event, it is 120 to 180 minutes after the start of eating; for a post-exercise event, it is 60 to 120 minutes after the end of exercise; and for a pre-sleep event, it can be observed continuously until the sleep state stabilizes. The observation duration can also be adjusted based on historical recovery time, device data integrity rate, and subsequent related events. When a new exercise, eating, or medication event occurs during the observation period, the system records the interference event identifier and decides whether to terminate the current observation period or retain the overlapping observation relationship based on the event interval.

[0068] Taking a lunch event as an example, the eating start time is 12:05 PM. The system sets 12:05 PM as the relative time zero point and defines the analysis range from 0 minutes before eating to 120 minutes after the first bite. Specifically, the 20 minutes before eating are used to determine the pre-event state, the eating window is used to characterize the behavioral input process, and the 150 minutes after eating are used to capture the response and decline changes in glucose, heart rate, and heart rate variability. For post-exercise events, the end time of exercise is set as the relative time zero point, and data from the end of exercise is extracted forward and extended backward until heart rate and activity intensity return to stability.

[0069] The data processing unit extracts physiological parameter sequences within the analysis scope from event-aligned multimodal datasets. For postprandial events, sequences of glucose concentration, heart rate, heart rate variability, activity intensity, and skin temperature can be extracted, along with behavioral attributes such as meal duration, intake, and nutritional composition. For post-exercise events, sequences of heart rate, heart rate intervals, activity intensity, and skin temperature corresponding to the electrocardiogram (ECG) waveform can be extracted. The heart rate interval sequence is formed by identifying the peak positions of adjacent R waves in continuous ECG waveforms, calculating the time interval between adjacent valid heartbeats, and arranging them according to their occurrence time. After removing intervals corresponding to abnormal heartbeats such as motion artifacts, missed beats, and premature beats, a normal sinus heart rate interval sequence can be further formed. The heart rate interval sequence is not equivalent to heart rate variability; rather, it is the underlying time-series data for calculating heart rate variability. The system performs statistical and spectral analysis on normal sinus interval sequences according to a preset analysis window, obtaining heart rate variability parameters such as the standard deviation of normal heart intervals, the root mean square of the difference between adjacent heart intervals, low-frequency power, high-frequency power, and the ratio of low-frequency to high-frequency power. These parameters are then used to generate a heart rate variability parameter sequence according to the window time. The heart rate sequence is used to characterize the heart rate decline process after exercise, the heart interval sequence is used to preserve the changes in heart rate over time, and the heart rate variability parameter sequence is used to characterize the autonomic nervous system regulation and recovery state. These three parameters, along with activity intensity and skin temperature sequences, are used to generate a post-exercise physiological response trajectory. Heart rate, body movement, ambient light, and EEG rhythm sequences can be extracted from pre-sleep events. Each parameter sequence is associated with a synchronization timestamp, relative event time, sampling frequency, quality level, and data source.

[0070] Subsequently, the system arranges the physiological parameter sequences in ascending order of relative time and checks the data quality. Abnormal heart rate points caused by motion artifacts in the pulse wave can be identified and removed by combining acceleration intensity and the amplitude of adjacent heart rate changes. Isolated abrupt changes in continuous glucose data can be corrected based on differences between adjacent sampled values ​​and sensor quality. For short-term missing data with a length below the allowable threshold, interpolation can be performed using adjacent valid sampled points; for continuous missing data exceeding the threshold, no estimated value is generated, but the missing segment is retained in the trajectory. For high-frequency ECG data, heart rate and interbeat interval features are first extracted at preset time intervals, and then a time correspondence is established with minute-level glucose data to avoid over-smoothing the original waveform directly.

[0071] The pre-event state can be determined by stable data within a preset interval before the trigger time. For example, if activity intensity is low and glucose fluctuations are within acceptable ranges within 20 minutes before a meal, the median of all parameters in that interval is taken as the pre-event state. The system compares each sampled value after the event with the pre-event state, calculates the direction and magnitude of change, and calculates the rate of change based on the difference and time interval between adjacent time points. Taking postprandial glucose as an example, if the pre-event state is 5.2 mmol / L and it reaches 8.1 mmol / L 45 minutes after eating, the increase is 2.9 mmol / L; simultaneously, the relative time to reach the peak, the pre-peak rise rate, and the post-peak decline rate are recorded. For heart rate, the post-event decline magnitude, the time of the lowest value, and the subsequent recovery rate can be recorded.

[0072] The data processing unit combines the pre-event state, direction of change, maximum amplitude of change, extreme moment, rate of change, and effective observation interval of each physiological parameter along a relative time axis to generate a physiological response trajectory corresponding to the key physiological event. The trajectory also associates event type, behavioral attributes, data quality, interfering events, and sampling configuration, enabling subsequent processing to distinguish between normal responses, delayed responses, and mixed responses influenced by other behaviors. Through the above implementation method, multimodal physiological data at different time scales can be converted into continuous dynamic trajectories revolving around the same event, providing comparable temporal data for further determining state change outcomes and recovery processes.

[0073] In some embodiments, the state change outcome and recovery trajectory are determined based on the difference between the physiological response trajectory and the individual's multi-day physiological baseline, and a human causal chain dataset of key physiological events is generated by association, including: Determine the initial state, extreme values, and post-event steady state from the physiological response trajectory; The steady state and extreme response values ​​after the event are compared with the individual's multi-day physiological baseline for the corresponding period to determine the direction and magnitude of change of each physiological characteristic parameter and generate state change results. The recovery time of each physiological characteristic parameter after the end of a key physiological event is detected when it re-enters the corresponding baseline range, and a recovery trajectory is generated based on the recovery time and the previous changes. Key physiological events, physiological response trajectories, state change results, and recovery trajectories are associated according to event identifiers to form causal chain records, and these causal chain records are aggregated to generate a human causal chain dataset.

[0074] Specifically, in a human physiological monitoring scenario spanning seven consecutive natural days, the data processing unit reads the corresponding physiological response trajectory for each key physiological event and identifies the trajectory status in three stages: before, during, and after the event. Taking the lunch event as an example, the system determines the statistical values ​​of physiological parameters where the activity intensity is stable and no other behavioral events occur within 20 minutes before eating as the initial response state, and determines the effective sampling state where glucose, heart rate, and heart rate variability deviate most from the initial response state before eating the first bite as the extreme response value.

[0075] When the rate of change of each physiological parameter is lower than the stability threshold for multiple consecutive sampling periods after the event, the statistical value of the corresponding interval is determined as the stable state after the event. If interfering events such as exercise, refeeding, or medication occur during the observation period, the time of interference is recorded in the trajectory, and the last stable interval before the interference occurs is taken as the valid analysis endpoint of the current event.

[0076] The data processing unit matches the baseline range of the individual's multi-day physiological baseline with data from the same time period and similar activity states based on the relative event time corresponding to each state. For example, if the lunch event occurs at 12:05, the system calls the baseline for glucose, heart rate, and heart rate variability corresponding to the resting state at noon, instead of using the baseline from the morning or during exercise. If the response's initial state has exceeded the baseline range, this situation is marked as a pre-event deviation to avoid attributing all existing abnormalities to the current event.

[0077] Subsequently, the extreme values ​​of the response and the steady state after the event were compared with the matched baseline center value and the baseline boundary, respectively, to determine the direction of increase, decrease or no significant change of the physiological characteristic parameters relative to the baseline, and to calculate the absolute change, the relative change ratio and the duration of the deviation from the baseline range.

[0078] For example, if glucose levels were 5.2 mmol / L before a lunch event, with a baseline range of 4.7 mmol / L to 5.8 mmol / L during the corresponding time period, and reached 8.1 mmol / L 45 minutes after the meal, then glucose levels are considered to be increasing, with the peak value increasing by 2.8 mmol / L relative to the baseline center. If heart rate increased from 62 beats per minute before the event to 76 beats per minute, and the root mean square difference of adjacent heartbeat intervals in heart rate variability decreased from 42 milliseconds to 28 milliseconds, then the increase in heart rate and the decrease in heart rate variability are recorded respectively. The system integrates the direction, magnitude, peak occurrence time, and deviation from the steady state of each parameter to generate the corresponding state change results for the event, and associates them with data quality level and interference event information.

[0079] Recovery testing begins at the end of the critical physiological event. For each physiological characteristic parameter, the system checks point-by-point whether the sampled value has re-entered the baseline range for the corresponding time period, requiring a continuous and preset stable duration to eliminate false positives caused by brief baseline crossings. For example, if glucose remains within the baseline range for 15 consecutive minutes and the rate of change is below a threshold, the moment it first enters this stable range is defined as the glucose recovery moment; if heart rate remains within the resting heart rate baseline range for 5 consecutive minutes, the heart rate recovery moment is determined; if heart rate variability recovers to the baseline range for three consecutive analysis windows, the heart rate variability recovery moment is determined. Parameters that have not recovered by the end of the observation period are recorded as incompletely recovered, and the observed duration and the degree of deviation at the end are saved.

[0080] The system collects data on the changes in each parameter from its extreme value back to the baseline range after the event, arranged in relative chronological order. It records the direction of decline, segmented rate of change, recovery time, and recovery duration, generating recovery sub-trajectories for each parameter. Further, it determines the overall recovery time based on the latest recovering parameter and combines the sub-trajectories into the recovery trajectory corresponding to the event. For example, if heart rate recovers in 35 minutes, heart rate variability in 70 minutes, and glucose in 125 minutes after lunch, the overall recovery time is determined to be 125 minutes, while retaining the different recovery rhythms of each parameter.

[0081] Finally, the data processing unit uses the event identifier as the primary key to associate the event type, occurrence time, and behavioral attributes of key physiological events with the physiological response trajectory, state change result, recovery trajectory, individual multi-day physiological baseline version, data quality, and interfering events, forming a causal chain record. The system collects causal chain records corresponding to multiple events such as waking up, eating, after exercise, and before sleeping, generating a human causal chain dataset. Through the above implementation method, the magnitude, direction, and recovery process of physiological changes after behavioral input can be converted into structured dynamic samples, improving the continuity, individual adaptability, and traceability of human state change processes, and providing a data foundation for physiological world models to learn the temporal mapping relationship between behavioral input and physiological response.

[0082] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of this application.

[0083] Figure 2 This is a schematic diagram of the structure of a full-modal physiological data acquisition system based on a physiological world model provided in an embodiment of this application. Figure 2 As shown, the system includes: The acquisition module 201 is used to acquire multimodal physiological and behavioral data of the target human body and generate multimodal raw data streams; The mapping module 202 is used to determine the reference clock and, based on the clock offset and drift of each acquisition device, map the multimodal raw data stream to a unified time coordinate to generate a time-synchronized data stream. The determination module 203 is used to identify key physiological events based on the time-synchronized data stream, determine the event window, and generate an event window sequence. The switching module 204 is used to control the corresponding acquisition device to switch from the baseline acquisition frequency to the high-precision frequency within the event window according to the type of key physiological event, so as to acquire key window data; The interception module 205 is used to intercept time synchronization data and key window data within the event window based on the event window sequence, and map them to a unified event time grid according to the sampling attributes to generate an event-aligned multimodal dataset. Module 206 is established to extract the distribution features of physiological characteristic parameters from a multimodal dataset aligned with multiple natural life days, and to establish an individual's multi-day physiological baseline. The generation module 207 is used to extract physiological response trajectories within the event window and observation period using key physiological events as time anchors, determine the state change results and recovery trajectory based on the differences between the physiological response trajectories and the individual's multi-day physiological baseline, and generate a human causal chain dataset of key physiological events.

[0084] In some embodiments, Figure 2 The acquisition module 201 is configured on the target human body with multiple acquisition devices for continuously sensing physiological state, intermittently measuring human state, and recording behavioral input, and establishes a correspondence between the device identifier and data modality of each acquisition device; controls the multiple acquisition devices to acquire continuous physiological signals, periodic state data, and behavioral event data according to the corresponding acquisition configuration; associates the data output by each acquisition device with device identifier, local acquisition time, and sampling attributes, and performs structured encapsulation according to the corresponding data modality; and aggregates the structured encapsulated data based on the local acquisition time to generate a multimodal raw data stream.

[0085] In some embodiments, Figure 2 The mapping module 202 determines the reference acquisition device from among the acquisition devices and sets the device clock of the reference acquisition device as the reference clock; it obtains the time difference between the other acquisition devices and the reference acquisition device at multiple synchronization moments, and determines the clock offset and clock drift of the corresponding acquisition devices based on the time difference; it establishes a time mapping relationship between the local time of each acquisition device and the unified time coordinate based on the clock offset and clock drift; it updates the time mapping relationship according to the preset synchronization cycle, and corrects the local acquisition time of each data in the multimodal raw data stream based on the updated time mapping relationship, generating a time synchronization data stream carrying a synchronization timestamp.

[0086] In some embodiments, Figure 2 The determination module 203 extracts event recognition features representing changes in human activity state, behavioral input, and environmental state from the time-synchronized data stream; matches the event recognition features with preset event recognition rules, and determines candidate key physiological events based on state transition conditions, behavior occurrence time, and duration conditions; configures corresponding window start time, window end time, and event confidence based on the trigger time and event type of each candidate key physiological event; performs window boundary correction and event association on candidate key physiological events that overlap or are related in time, and generates an event window sequence carrying event type, window boundary, and event confidence in chronological order.

[0087] In some embodiments, Figure 2 The switching module 204 matches the acquisition strategy according to the key physiological event type, determines the associated target data modality and corresponding acquisition device; based on the window boundary of the event window and the sampling attributes of the target data modality, it determines the high-precision frequency and acquisition duration, and generates a sampling switching instruction; it sends the sampling switching instruction to the corresponding acquisition device, so that the corresponding acquisition device acquires data at the high-precision frequency within the event window and adds a synchronization timestamp; after the event window ends, it controls the corresponding acquisition device to return to the baseline acquisition frequency, and encapsulates the data acquired within the event window into key window data.

[0088] In some embodiments, Figure 2 The interception module 205 extracts corresponding data segments from the time-synchronized data stream and key window data according to the window boundaries of each event window in the event window sequence; based on the sampling frequency and timestamp density of each data segment, it determines the corresponding data mapping method and establishes a unified event time grid covering the window boundaries; it performs interpolation mapping on the high-precision acquired data segments according to the unified event time grid, retains the original timestamps of the baseline acquired data segments and establishes time-location associations; it combines the mapped modal data segments according to the event windows and associates them with the corresponding key physiological events and sampling attributes to generate an event-aligned multimodal dataset.

[0089] In some embodiments, Figure 2 The establishment module 206 filters multi-day data that meets the data integrity conditions from the event-aligned multimodal dataset and extracts the physiological characteristic parameters corresponding to each data modality; divides the physiological characteristic parameters into time periods according to the natural day cycle and preset sliding window to obtain the characteristic samples corresponding to each sliding window; performs distribution statistics on the characteristic samples to determine the central tendency, dispersion, fluctuation degree and quantile interval, and generates intraday variation curves based on the statistical results of consecutive natural days; associates the distribution statistics results and intraday variation curves with the corresponding time periods to determine the baseline range and temporal variation pattern of each physiological characteristic parameter, and establishes an individual multi-day physiological baseline.

[0090] In some embodiments, Figure 2 The generation module 207 determines the corresponding observation duration based on the event type and window boundaries of the key physiological event, and determines the event trigger time as the relative time reference; it extracts the event window and the physiological parameter sequence associated with the key physiological event within the observation duration from the event-aligned multimodal dataset; it arranges the physiological parameter sequence in time sequence based on the relative time reference, and performs continuity processing on abnormal sampling points and missing positions; it extracts the change amplitude, change direction and change rate of the physiological parameter sequence relative to the pre-event state, and generates the physiological response trajectory of the corresponding key physiological event.

[0091] In some embodiments, Figure 2 The generation module 207 determines the initial state of the response, the extreme value of the response, and the stable state after the event from the physiological response trajectory; it compares the stable state after the event and the extreme value of the response with the individual's multi-day physiological baseline for the corresponding time period to determine the direction and magnitude of change of each physiological characteristic parameter and generate the state change result; it detects the recovery time when each physiological characteristic parameter re-enters the corresponding baseline range after the end of the key physiological event, and generates the recovery trajectory based on the recovery time and the previous change process; it associates the key physiological event, physiological response trajectory, state change result, and recovery trajectory according to the event identifier to form a causal chain record, and collects the causal chain records to generate a human causal chain dataset.

[0092] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0093] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various system embodiments described above.

[0094] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0095] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0096] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0097] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0098] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which may be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0099] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for acquiring full-modal physiological data oriented towards a physiological world model, characterized in that, include: Collect multimodal physiological and behavioral data of the target human body and generate multimodal raw data stream; A reference clock is determined, and based on the clock offset and drift of each acquisition device, the multimodal raw data stream is mapped to a unified time coordinate to generate a time-synchronized data stream; Key physiological events are identified based on the time-synchronized data stream; event windows are determined; and an event window sequence is generated. Based on the type of key physiological event, the corresponding acquisition device is controlled to switch from the baseline acquisition frequency to a high-precision frequency within the event window to acquire key window data. Based on the event window sequence, time synchronization data and key window data within the event window are extracted and mapped to a unified event time grid according to the sampling attributes to generate an event-aligned multimodal dataset. Based on the multimodal dataset aligned with the events of multiple natural life days, the distribution characteristics of physiological feature parameters are extracted to establish an individual multi-day physiological baseline. Using the key physiological events as time anchors, physiological response trajectories within the event window and observation period are extracted. Based on the differences between the physiological response trajectories and the individual's multi-day physiological baseline, the state change results and recovery trajectories are determined, and a human causal chain dataset of the key physiological events is generated.

2. The method according to claim 1, characterized in that, The process involves collecting multimodal physiological and behavioral data from the target human body, generating a multimodal raw data stream, including: Multiple types of data acquisition devices are configured on the target human body for continuous sensing of physiological state, intermittent measurement of human body state and recording of behavioral input, and a correspondence between device identifiers and data modalities of each data acquisition device is established. Control the various types of acquisition devices to acquire continuous physiological signals, periodic state data, and behavioral event data according to their corresponding acquisition configurations; Associate the data output by each acquisition device with the device identifier, local acquisition time, and sampling attributes, and encapsulate it in a structured manner according to the corresponding data modality; The multimodal raw data stream is generated by aggregating structured and encapsulated data based on the local acquisition time.

3. The method according to claim 1, characterized in that, The determination of the reference clock, based on the clock offset and drift of each acquisition device, maps the multimodal raw data stream to a unified time coordinate to generate a time-synchronized data stream, including: A reference acquisition device is determined from among the acquisition devices, and the device clock of the reference acquisition device is determined as the reference clock. Obtain the time difference between the other acquisition devices and the reference acquisition device at multiple synchronization moments, and determine the clock offset and clock drift of the corresponding acquisition devices based on the time difference; Establish a time mapping relationship between the local time of each acquisition device and the unified time coordinate based on the clock offset and clock drift; The time mapping relationship is updated according to a preset synchronization period, and the local acquisition time of each data in the multimodal raw data stream is corrected based on the updated time mapping relationship to generate a time synchronization data stream carrying a synchronization timestamp.

4. The method according to claim 1, characterized in that, The step of identifying key physiological events based on the time-synchronized data stream, determining the event window, and generating an event window sequence includes: Extract event recognition features that characterize changes in human activity state, behavioral input, and environmental state from the time-synchronized data stream; The event recognition features are matched with preset event recognition rules, and candidate key physiological events are determined based on state transition conditions, the time of occurrence of the behavior, and the duration conditions. Based on the triggering time and event type of each candidate key physiological event, configure the corresponding window start time, window end time, and event confidence. Window boundary correction and event association are performed on candidate key physiological events that overlap or are correlated in time, and the event window sequence carrying event type, window boundary and event confidence is generated in chronological order.

5. The method according to claim 1, characterized in that, The step of controlling the corresponding acquisition device to switch from the baseline acquisition frequency to a high-precision frequency within the event window, based on the type of the key physiological event, to acquire key window data includes: Based on the key physiological event types, a matching acquisition strategy is used to determine the associated target data modalities and corresponding acquisition devices. Based on the window boundaries of the event window and the sampling attributes of the target data modality, the high-precision frequency and acquisition duration are determined, and a sampling switching command is generated. The sampling switching command is sent to the corresponding acquisition device, so that the corresponding acquisition device can acquire data at a high-precision frequency within the event window and add a synchronization timestamp. After the event window ends, the corresponding acquisition device is controlled to return to the baseline acquisition frequency, and the data acquired within the event window is encapsulated as key window data.

6. The method according to claim 1, characterized in that, The process involves using the event window sequence as a reference, extracting time synchronization data and key window data within the event window, and mapping them to a unified event time grid according to sampling attributes to generate an event-aligned multimodal dataset, including: Based on the window boundaries of each event window in the event window sequence, extract the corresponding data segments from the time synchronization data stream and the key window data; Based on the sampling frequency and timestamp density of each data segment, the corresponding data mapping method is determined, and a unified event time grid covering the window boundary is established. The high-precision acquired data segments are interpolated and mapped according to the unified event time grid, and the original timestamps of the baseline acquired data segments are preserved and time-location associations are established. The mapped modal data fragments are combined according to the event window and associated with the corresponding key physiological events and sampling attributes to generate the event-aligned multimodal dataset.

7. The method according to claim 1, characterized in that, The multimodal dataset aligned with events based on multiple natural days is used to extract the distribution features of physiological characteristic parameters and establish an individual multi-day physiological baseline, including: From the event-aligned multimodal dataset, multi-day data that meets the data integrity conditions are selected, and physiological feature parameters corresponding to each data modality are extracted; The physiological characteristic parameters are divided into time periods according to the natural day cycle and a preset sliding window to obtain the feature samples corresponding to each sliding window. The distribution statistics of the feature samples are performed to determine the central tendency, dispersion, volatility and quantile intervals, and intraday variation curves are generated based on the statistical results of consecutive natural days. By associating the distribution statistics and the intraday variation curve with the corresponding time period, the baseline range and temporal variation pattern of each physiological characteristic parameter are determined, and the individual's multi-day physiological baseline is established.

8. The method according to claim 1, characterized in that, The step of extracting the physiological response trajectory within the event window and observation period using the key physiological events as time anchors includes: Based on the event type and window boundaries of the key physiological events, the corresponding observation duration is determined, and the event trigger time is determined as the relative time reference. Extract the event window and the sequence of physiological parameters associated with key physiological events within the observation period from the event-aligned multimodal dataset; The physiological parameter sequence is arranged temporally based on the relative time reference, and the abnormal sampling points and missing positions are processed for continuity. Extract the magnitude, direction, and rate of change of the physiological parameter sequence relative to the pre-event state to generate the physiological response trajectory corresponding to the key physiological event.

9. The method according to claim 8, characterized in that, The process of determining the state change results and recovery trajectory based on the difference between the physiological response trajectory and the individual's multi-day physiological baseline, and generating a human causal chain dataset of the key physiological events, includes: The initial state, extreme values, and post-event steady state of the physiological response trajectory are determined from the physiological response trajectory. The post-event steady state and response extreme values ​​are compared with the individual's multi-day physiological baseline for the corresponding time period to determine the direction and magnitude of change of each physiological characteristic parameter, and the state change results are generated. The recovery time when each physiological characteristic parameter re-enters the corresponding baseline range after the critical physiological event is detected, and the recovery trajectory is generated based on the recovery time and the previous changes. The key physiological events, physiological response trajectories, state change results, and recovery trajectories are associated according to event identifiers to form causal chain records, and the causal chain records are aggregated to generate the human causal chain dataset.

10. A full-modal physiological data acquisition system oriented towards a physiological world model, characterized in that, include: The acquisition module is used to collect multimodal physiological and behavioral data of the target human body and generate multimodal raw data streams; The mapping module is used to determine the reference clock and, based on the clock offset and drift of each acquisition device, map the multimodal raw data stream to a unified time coordinate to generate a time-synchronized data stream. The determination module is used to identify key physiological events based on the time-synchronized data stream, determine the event window, and generate an event window sequence; The switching module is used to control the corresponding acquisition device to switch from the baseline acquisition frequency to a high-precision frequency within the event window according to the type of the key physiological event, so as to acquire key window data. The interception module is used to intercept time synchronization data and key window data within the event window based on the event window sequence, and map them to a unified event time grid according to the sampling attributes to generate an event-aligned multimodal dataset. A module is established to extract the distribution features of physiological characteristic parameters from a multimodal dataset aligned with the events over multiple natural days, and to establish an individual's multi-day physiological baseline. The generation module is used to extract the physiological response trajectory within the event window and observation period using the key physiological event as the time anchor point, determine the state change result and recovery trajectory based on the difference between the physiological response trajectory and the individual's multi-day physiological baseline, and generate a human causal chain dataset of the key physiological event.

Images