Smart watch psychological stress prediction method and system based on context awareness

Abstract

The application belongs to the technical field of intelligent sensing, and particularly relates to a smartwatch psychological stress prediction method and system based on context awareness. The method comprises the following steps: acquiring heart rate data, acceleration signals and environmental sound signals and aligning them according to the same time window to obtain multi-source time series data; extracting body state fluctuation features, activity intensity features and environmental sound features, combining behavior habit information to correct the body state fluctuation features and obtaining state correction features; determining a stress evaluation result according to the similarity between the state correction features and historical scene features, and determining a stress source identification result according to historical feedback emotion labels when a preset stress condition is met; performing time series prediction based on the stress source identification result, and updating stress prediction parameters according to a user confirmation signal. The application can reduce stress misjudgment caused by activity and environmental interference, and improve the stability of stress source identification and trend prediction.

Description

Technical Field

[0001] This application belongs to the field of intelligent sensing technology, specifically relating to a method and system for predicting psychological stress in smartwatches based on context awareness. Background Technology

[0002] As wearable devices evolve into health management gateways, smartwatches have gradually shifted from activity tracking tools to terminals for continuous physiological monitoring and personal status recognition. Smart sensors continuously collect data such as heart rate, activity status, and ambient sound, providing a low-cost, real-time data foundation for assessing psychological stress levels. For the smartwatch industry, stress prediction capabilities directly impact the depth of the device's upgrade from "data display" to "health decision support," and also influence users' trust in wearable health functions and their willingness to use them long-term.

[0003] Existing technologies often rely on single or limited data such as heart rate, heart rate variability, or activity level for stress assessment. In real-world wearing environments, these data are easily influenced by external noise, physical activity, commuting conditions, exercise habits, and sleep patterns. For example, an increased heart rate in a noisy environment could stem from external stimuli or genuine psychological stress; heart rate fluctuations during walking or exercise may also be misinterpreted as increased stress. Current methods typically lack the integrated processing of the relationships between physical state, environmental conditions, behavioral habits, and historical feedback, making it difficult to distinguish between physiological changes caused by routine activities and stress-related changes. This leads to unclear identification of stress sources, unstable prediction results, and difficulty in continuously updating parameters to reflect individual user differences. Summary of the Invention

[0004] To address the above issues, this application provides a context-aware method and system for predicting psychological stress in smartwatches, which aims to at least solve the problem of accurately identifying the sources of psychological stress for smartwatch users and predicting stress changes under the combined influence of environmental sounds, physical activities, and user behavior habits.

[0005] To achieve the above objectives, the technical solution adopted in this application is as follows: Firstly, this application provides a context-aware method for predicting psychological stress in smartwatches, the method comprising: Acquire heart rate data, acceleration signals, and ambient sound signals, and align the heart rate data, acceleration signals, and ambient sound signals according to the same time window to obtain multi-source time series data; Based on multi-source time series data, physical state fluctuation features, activity intensity features, and environmental sound features are extracted, and physical state fluctuation features are corrected by combining behavioral habit information to obtain state correction features. The stress assessment results are determined based on the similarity between the state correction features and the historical scene features; When the stress assessment results meet the preset stress conditions, the stress source identification results are determined based on the historical feedback emotion tags. Based on the pressure source identification results, time series prediction is performed to obtain pressure fluctuation prediction results. User confirmation signals are received, and pressure prediction parameters are updated based on user confirmation signals.

[0006] In one possible implementation, heart rate data, acceleration signals, and ambient sound signals are aligned according to the same time window to obtain multi-source time series data, including: acquiring the acquisition timestamps of heart rate data, acceleration signals, and ambient sound signals respectively; merging the acquisition timestamps according to a preset time window; and associating heart rate data, acceleration signals, and ambient sound signals belonging to the same preset time window as the same time series record to obtain multi-source time series data.

[0007] In one possible implementation, body state fluctuation features, activity intensity features, and environmental sound features are extracted from multi-source time series data, including: determining the heart rate change rate and heart rate variability based on heart rate data to obtain body state fluctuation features; determining activity intensity and activity duration based on acceleration signals to obtain activity intensity features; and determining environmental noise intensity, interference duration, and frequency band energy distribution based on environmental sound signals to obtain environmental sound features.

[0008] In one possible implementation, the physical state fluctuation characteristics are modified by combining behavioral habit information to obtain the state-modified characteristics, including: determining the activity type based on activity intensity characteristics; determining the historical physical state fluctuation range corresponding to the activity type based on behavioral habit information; reducing the weight of the physical state fluctuation characteristics used to determine the stress assessment result when the physical state fluctuation characteristics are within the historical physical state fluctuation range; and modifying the physical state fluctuation characteristics based on the reduced weight to obtain the state-modified characteristics.

[0009] In one possible implementation, the state correction feature is obtained by combining behavioral habit information to correct the body state fluctuation characteristics. It also includes: determining the historical activity intensity range and historical body state fluctuation range corresponding to the same time window based on behavioral habit information; and obtaining the state correction feature based on the fluctuation part of the body state fluctuation characteristics that exceeds the historical body state fluctuation range when the activity intensity characteristics are within the historical activity intensity range and either the environmental noise intensity is greater than the environmental noise intensity threshold or the interference duration is greater than the interference duration threshold.

[0010] In one possible implementation, the stress assessment result is determined based on the similarity between the state-corrected features and the historical scene features, including: constructing a stress assessment feature vector based on the state-corrected features; representing at least one historical scene feature as a historical feature vector; calculating the feature vector distance between the stress assessment feature vector and the historical feature vector; determining the stress probability value based on the feature vector distance; and determining the stress assessment result based on the stress probability value.

[0011] In one possible implementation, the preset stress conditions include a stress probability value greater than a preset stress threshold; the preset similarity conditions include a corresponding feature vector distance less than or equal to a preset distance threshold; and the stress source identification result is determined based on historical feedback emotion tags, including: when the stress probability value is greater than the preset stress threshold, determining a target historical scene feature whose feature vector distance satisfies the preset similarity condition from at least one historical scene feature; obtaining the target historical feedback emotion tag corresponding to the target historical scene feature; determining the stress source category based on the target historical scene feature and the target historical feedback emotion tag, and using the stress source category as the stress source identification result.

[0012] In one possible implementation, time series prediction is performed based on the pressure source identification results to obtain pressure fluctuation prediction results, including: arranging the pressure source categories and pressure probability values ​​obtained in consecutive time windows in chronological order to obtain a pressure change sequence; performing time series prediction on the pressure change sequence to obtain the pressure probability change trend in the future time window, and using the pressure probability change trend as the pressure fluctuation prediction result.

[0013] In one possible implementation, the pressure prediction parameters include pressure assessment weights, a preset pressure threshold, and pressure source identification weights; a user confirmation signal is used to indicate the user's confirmation of the pressure source category and pressure fluctuation prediction results; updating the pressure prediction parameters based on the user confirmation signal includes: determining the prediction matching result based on the consistency between the user confirmation signal and the pressure source category and pressure fluctuation prediction results respectively; and updating the pressure assessment weights, the preset pressure threshold, and the pressure source identification weights if the prediction matching result does not meet the preset matching conditions.

[0014] Secondly, this application provides a context-aware smartwatch psychological stress prediction system for implementing a context-aware smartwatch psychological stress prediction method. The system includes: The data acquisition module is used to acquire heart rate data, acceleration signals, and ambient sound signals, and align the heart rate data, acceleration signals, and ambient sound signals according to the same time window to obtain multi-source time series data; The feature correction module is used to extract body state fluctuation features, activity intensity features and environmental sound features from multi-source time series data, and combine them with behavioral habit information to correct the body state fluctuation features and obtain state correction features. The stress assessment module is used to determine the stress assessment result based on the similarity between state correction features and historical scene features; The source identification module is used to determine the source of stress based on historical feedback emotion tags when the stress assessment results meet the preset stress conditions. The trend update module is used to perform time series prediction based on the pressure source identification results, obtain pressure fluctuation prediction results, receive user confirmation signals, and update pressure prediction parameters based on user confirmation signals.

[0015] Compared with existing technologies, the advantages and beneficial effects of this application are as follows: By aligning heart rate data, acceleration signals, and ambient sound signals according to the same time window, the data from different smart sensors are correlated under a unified time reference, avoiding incorrect matching between heart rate fluctuations, activity changes, and sound interference caused by different sampling frequencies and reporting delays.

[0016] By extracting physical state fluctuation features, activity intensity features, and environmental sound features from multi-source time series data, and combining them with behavioral habit information to correct the physical state fluctuation features, the system was able to distinguish between changes in physical state caused by routine activities and abnormal stress-related fluctuations, reducing misjudgments caused by judging stress status solely based on heart rate data.

[0017] By performing similarity analysis between state correction features and historical scene features, the matching between the current user state and similar historical scenes is achieved, so that the stress assessment results no longer rely on isolated real-time data, but are judged in combination with the historical patterns of individual users.

[0018] By introducing historical feedback emotion tags when the stress assessment results meet preset stress conditions, the identification of stress source categories is achieved, so that the prediction results can not only reflect the stress probability, but also give the possible sources of stress.

[0019] By performing time series prediction based on pressure source identification results and updating pressure prediction parameters using user confirmation signals, continuous prediction of pressure fluctuation trends and individualized parameter calibration are achieved, enabling the system to continuously adjust the judgment boundaries based on user feedback. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the method described in this application; Figure 2 This is a block diagram of the module combination of the system in this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solution, the present application will be described in detail below with reference to the embodiments. The description in this section is only exemplary and explanatory, and should not be used to limit the scope of protection of the present application in any way.

[0022] Context awareness refers to a processing method where, in addition to collecting single physiological signals, devices further interpret the meaning of current data by combining the user's environment, activity state, time patterns, and historical interaction feedback. For smartwatches, the heart rate data, acceleration signals, and ambient sound signals collected by smart sensors are not isolated data segments; the same elevated heart rate phenomenon has different meanings in motion, rest, noisy environments, or abnormal emotional states. Therefore, stress prediction cannot rely solely on a single type of real-time reading but should establish a correspondence between physical state, behavioral activities, environmental conditions, and historical scenarios within a unified time window. It should correct for fluctuations in physical state that may be caused by routine activities or environmental disturbances and further utilize historical feedback to identify stress sources. Based on this approach, this application constructs a complete processing flow around multi-source data synchronization, context feature correction, historical scene matching, stress source identification, and feedback parameter updates in smartwatches to achieve psychological stress prediction that better reflects the user's actual state.

[0023] like Figure 1 As shown, a context-aware smartwatch-based method for predicting psychological stress includes: Acquire heart rate data, acceleration signals, and ambient sound signals, and align the heart rate data, acceleration signals, and ambient sound signals according to the same time window to obtain multi-source time series data; After the smartwatch initiates the stress prediction task, it uses the heart rate detection unit to collect heart rate data, the inertial sensor to collect acceleration signals, and the microphone to collect ambient sound signals. Heart rate data reflects changes in the user's physical state, acceleration signals reflect activity intensity and posture changes, and ambient sound signals reflect noise interference in the user's environment. All three types of data are written to a local cache or the data cache of the accompanying terminal, retaining the acquisition time during writing. The processor reads the cached data according to a unified time window, associating the three types of data falling within the same time window to form a set of data records with the same time reference for each time window. This data record preserves the temporal correspondence between heart rate change segments, acceleration change segments, and ambient sound segments, serving as the data foundation for subsequent extraction of physical state fluctuation characteristics, activity intensity characteristics, and ambient sound characteristics.

[0024] Heart rate data, acceleration signals, and ambient sound signals are aligned according to the same time window to obtain multi-source time series data, including: acquiring the acquisition timestamps of heart rate data, acceleration signals, and ambient sound signals respectively; merging the acquisition timestamps according to a preset time window; and associating heart rate data, acceleration signals, and ambient sound signals belonging to the same preset time window into the same time series record to obtain multi-source time series data.

[0025] In one embodiment, when aligning heart rate data, acceleration signals, and ambient sound signals within the same time window, the merging is further defined by using the acquisition timestamp as the merging basis. The acquisition timestamp can be generated by the smartwatch system clock or written by the sensor driver layer during data reporting. The heart rate detection unit, inertial sensor, and microphone typically have different sampling frequencies; heart rate data is generated as low-frequency continuous readings or interval readings, acceleration signals are generated as higher-frequency continuous readings, and ambient sound signals are generated as short sound segments or sound energy sequences. After receiving the three types of data, the processor does not directly merge them according to their arrival order. Instead, it reads the acquisition timestamp carried by each data source and converts the acquisition timestamp to the same system time base, avoiding data mismatch caused by reporting delays from different sensors.

[0026] The length of the preset time window is set based on the real-time performance and data stability of the pressure prediction. For everyday wear scenarios, the preset time window can be set to a fixed window ranging from several seconds to tens of seconds. When the smartwatch processor has low computing power or the microphone sampling segment is long, the preset time window can be appropriately extended. When it is necessary to more quickly identify short-term noise stimuli or sudden heart rate changes, the preset time window can be shortened. The preset time window can also adopt a sliding window form, retaining some overlapping data between adjacent windows to reduce recognition errors caused by heart rate data or ambient sound signals being located at the window boundaries. The window length and sliding step size are written to the configuration file during system initialization and can be adjusted based on historical wear data and user feedback.

[0027] When the processor merges the acquired timestamps according to a preset time window, it assigns each heart rate data point, each acceleration signal segment, and each ambient sound signal segment to the time window containing the acquired timestamp. If multiple heart rate data points exist within a time window, the original reading sequence is retained and the number of samples is recorded. If a continuous acceleration signal exists within a time window, the triaxial acceleration sequence and sampling interval within that window are retained. If an ambient sound signal exists within a time window, the short-time energy, sound intensity record, or original sound feature cache of the sound segment is retained. For data whose acquired timestamps fall at the window boundaries, they can be grouped into adjacent windows according to the main duration of the data segment, or the feature summary of the boundary segment can be copied to the adjacent window. The specific method depends on the system's requirements for real-time performance and computational load.

[0028] After merging, the processor associates heart rate data, acceleration signals, and ambient sound signals belonging to the same preset time window into a single time series record. The time series record includes at least the window start time, window end time, heart rate data field, acceleration signal field, ambient sound signal field, and a data integrity flag. The data integrity flag records whether the three types of data within the window are complete. If the ambient sound signal is missing due to permission being disabled or microphone occupancy, it is marked as missing; if heart rate data cannot be collected due to loose clothing, it is marked as missing; if the acceleration signal is briefly interrupted, it is marked as incomplete activity data. For missing data records, the processor can fill in the gaps using stable readings from adjacent windows, or mark the window as a low-confidence window to reduce the data participation of that window in subsequent stress assessment phases.

[0029] Multi-source time series data consists of multiple time series records arranged chronologically. Each time series record maintains the temporal correspondence between the three types of signals and carries data integrity markers and window identifiers. When extracting features related to bodily state fluctuations, the processor reads heart rate change segments from the heart rate data field; when extracting activity intensity features, the processor reads activity change segments from the acceleration signal field; and when extracting environmental sound features, the processor reads sound intensity and related audio frequency band data from the environmental sound signal field. This recording structure ensures that subsequent feature extraction, behavioral habit correction, and stress source identification are all performed based on the same time reference, avoiding erroneous associations between heart rate changes, physical activity, and environmental sounds occurring at different times.

[0030] Based on multi-source time series data, physical state fluctuation features, activity intensity features, and environmental sound features are extracted, and physical state fluctuation features are corrected by combining behavioral habit information to obtain state correction features. After reading multi-source time-series data, the processor extracts heart rate, acceleration signal, and ambient sound signal fields one by one according to time windows. The heart rate field is used to generate body state fluctuation characteristics, the acceleration signal field to generate activity intensity characteristics, and the ambient sound signal field to generate ambient sound characteristics. The processor combines behavioral habit information to determine whether the body state fluctuations within the current time window can be explained by routine activities or existing lifestyle patterns, and adjusts the weight or fluctuation portion of the body state fluctuation characteristics in stress assessment based on the judgment result. The adjusted result retains body state changes that cannot be fully explained by routine activities and writes them as state correction features into the data record of the current time window for use in the stress assessment phase.

[0031] Based on multi-source time series data, we extract physical state fluctuation characteristics, activity intensity characteristics, and environmental sound characteristics, including: determining the heart rate change rate and heart rate variability based on heart rate data to obtain physical state fluctuation characteristics; determining activity intensity and activity duration based on acceleration signals to obtain activity intensity characteristics; and determining environmental noise intensity, interference duration, and frequency band energy distribution based on environmental sound signals to obtain environmental sound characteristics.

[0032] In one embodiment, bodily state fluctuation characteristics, activity intensity characteristics, and environmental sound characteristics are extracted separately from the original data fields within the same time window, ensuring that the three types of features maintain the same time boundaries. Heart rate data may include heart rate reading sequences, heartbeat interval sequences, or short-term statistical values ​​output by the smartwatch's heart rate detection unit. The processor performs anomaly checks on the heart rate reading sequences, removing invalid readings that significantly exceed the physiologically reasonable range or are caused by abnormal wearing conditions, and determines the heart rate variability and heart rate change rate based on the valid readings. The heart rate variability rate represents the magnitude of change in heart rate within the current time window relative to the previous time window or the initial reading within the window, while heart rate variability represents the dispersion of the heartbeat interval within that time window. If the heart rate detection unit only outputs the average heart rate, the processor may use the difference in average heart rate between adjacent time windows as the heart rate variability rate and use the fluctuation range of the average heart rate within consecutive windows as a substitute indicator for heart rate variability.

[0033] Acceleration signals can include a triaxial acceleration sequence. The processor performs gravity component separation or low-frequency smoothing on the triaxial acceleration sequence, calculates the changes in acceleration amplitude within a window, and determines the activity intensity based on the magnitude, duration, and rhythm of the changes. Activity intensity can be categorized into resting, mild, moderate, and vigorous activity levels, with the classification based on device factory settings, user historical motion data, or gait recognition models. The activity duration is determined by the number of windows that continuously meet the same activity intensity level, used to avoid short-term fluctuations being misjudged as stable activity. If the acceleration signal is briefly missing, the processor retains the data integrity flag of the current window and uses the activity status of adjacent windows as an auxiliary reference, without directly treating the missing window as a resting state.

[0034] Ambient sound signals can include raw sound segments, sound energy sequences, or sound features processed for local privacy. The processor does not need to identify specific semantic content; instead, it extracts ambient noise intensity, interference duration, and frequency band energy distribution. Ambient noise intensity can be represented by the short-time energy or equivalent sound level of a sound segment. Interference duration is determined by the time period exceeding the background sound level. Frequency band energy distribution is obtained through short-time frequency domain analysis of the sound segments. Frequency band energy distribution is used to distinguish between stable background sound, sudden high-intensity sound, and continuous narrow-band interference. After extracting these three types of features, the processor writes the body state fluctuation features, activity intensity features, and ambient sound features into the data record of the same time window, retaining feature source markers so that subsequent correction stages can identify the original data field and acquisition time range corresponding to each feature.

[0035] The physical state fluctuation characteristics are modified by combining behavioral habit information to obtain the state-modified characteristics, including: determining the activity type based on activity intensity characteristics; determining the historical physical state fluctuation range corresponding to the activity type based on behavioral habit information; reducing the weight of the physical state fluctuation characteristics in determining the stress assessment results when the physical state fluctuation characteristics are within the historical physical state fluctuation range; and modifying the physical state fluctuation characteristics based on the reduced weights to obtain the state-modified characteristics.

[0036] In one embodiment, when modifying the characteristics of bodily state fluctuations by incorporating behavioral habit information, the stress assessment weight is further reduced based on the historical bodily state fluctuation range corresponding to the activity type. Behavioral habit information originates from activity records, time period statistics, heart rate fluctuation records, and user feedback records generated by the smartwatch during historical wear. The processor determines the activity type corresponding to the current time window based on activity intensity characteristics. Activity types may include resting, walking, running, commuting, sleeping, and low-intensity activities during work. The determination of the activity type can employ threshold rules or an existing activity recognition model within the smartwatch. When using threshold rules, the processor determines the type based on acceleration amplitude, activity duration, and acceleration change rhythm. When using an activity recognition model, the processor calls the activity category output by the model and marks windows with output confidence levels lower than a preset category confidence threshold as having an uncertain activity type.

[0037] Once the activity type is determined, the processor retrieves the historical range of physical condition fluctuations corresponding to that activity type from behavioral habit information. This historical range is calculated based on the user's historical heart rate variability and heart rate change rate under the same activity type, and can use stable wearing records from the past few days or weeks as samples. The system removes samples with abnormal wearing, missing data, or those reported by the user as abnormal stress states, and then determines the historical range of physical condition fluctuations based on the remaining samples. The historical range of physical condition fluctuations can be represented by a quantile range or a reasonable deviation range from the mean, and the range boundaries are gradually updated as the user's historical data accumulates. For users with insufficient historical samples, the system uses the same age group or the device's default activity baseline as a temporary range, and switches to the individual's historical range once the user's data reaches a preset sample size.

[0038] When the body state fluctuation characteristics within the current time window fall within the historical range of body state fluctuations, it indicates that the change in body state is consistent with the user's historical patterns under this activity type. The processor reduces the weight of the body state fluctuation characteristics to determine the stress assessment result. The reduction magnitude is set based on the stability of the activity type and the number of historical samples. The more abundant the historical samples and the more stable the activity type identification, the greater the reduction magnitude can be. When the activity type is uncertain or the historical samples are insufficient, the reduction magnitude is kept within a smaller range to avoid excessively weakening the true stress fluctuations. The processor corrects the body state fluctuation characteristics based on the reduced weights, forming state-corrected characteristics. State-corrected characteristics retain the original heart rate change direction and duration, while recording the corrected participation weights, enabling the stress assessment stage to recognize that the body state changes in this window have been interpreted by behavioral habit information. If the current body state fluctuation characteristics are not within the historical range of body state fluctuations, the processor does not perform this weight reduction process, or only performs a slight correction based on data quality, and marks this window as one that needs further judgment in conjunction with environmental sound characteristics.

[0039] The process of combining behavioral habit information to correct the physical state fluctuation characteristics and obtain state correction characteristics also includes: determining the historical activity intensity range and historical physical state fluctuation range corresponding to the same time window based on behavioral habit information; and obtaining state correction characteristics based on the fluctuation portion of the physical state fluctuation characteristics that exceeds the historical physical state fluctuation range when the activity intensity characteristics are within the historical activity intensity range and either the environmental noise intensity is greater than the environmental noise intensity threshold or the interference duration is greater than the interference duration threshold.

[0040] In one embodiment, when correcting the body state fluctuation characteristics by incorporating behavioral habit information, the fluctuation portion of the body state fluctuation characteristics that exceeds the historical body state fluctuation range can be retained even when the activity intensity conforms to historical patterns but there is significant interference from ambient sound. The processor determines the historical activity intensity range and historical body state fluctuation range corresponding to the current time window based on behavioral habit information. The historical activity intensity range is obtained by statistically analyzing the user's acceleration amplitude, activity duration, and activity rhythm within the same time period or activity type, and is used to determine whether the current activity belongs to the user's normal activity state. The historical body state fluctuation range is obtained by statistically analyzing the heart rate change rate and heart rate variability within the same time period or activity type, and is used to determine whether the current body state change exceeds historical patterns.

[0041] The ambient noise intensity threshold and interference duration threshold are set by the system based on the device's microphone sensitivity, the background sound level of the environment, and historical sound records. The ambient noise intensity threshold determines whether the current ambient sound level has reached an intensity that may affect the user's physical condition. The interference duration threshold determines whether the sound interference has a sustained impact, rather than being a momentary, short-lived noise. The system can estimate the background sound level during low-interference periods when the user is wearing the device and set the sound intensity level above the background sound level by a certain margin as the ambient noise intensity threshold; the interference duration threshold can be set to the duration corresponding to one or more consecutive sound segments based on a preset time window length. If the background sound level varies significantly in different scenarios, the system can save thresholds separately for scenarios such as commuting, indoor rest, and outdoor activities.

[0042] If the current activity intensity characteristics are within the historical activity intensity range, it means that the current activity state can be explained by the user's historical behavioral patterns. If the ambient noise intensity is greater than the ambient noise intensity threshold, or the duration of the interference is greater than the interference duration threshold, the processor further checks whether the physical state fluctuation characteristics exceed the historical physical state fluctuation range. The excess portion can be the part of the heart rate variability exceeding the upper limit of the historical range, or the part of the heart rate variability deviating from the historical range. The processor uses this excess portion as the main content of the state correction characteristics and retains the interference markers corresponding to the ambient sound characteristics. When using this method, the physical state fluctuations caused by routine activities are absorbed by the historical range, while the fluctuations that cannot be explained within the activity intensity range and are accompanied by ambient sound interference are retained, enabling subsequent stress assessments to focus on the temporal correlation between environmental interference and abnormal physical states.

[0043] If the activity intensity characteristics are outside the historical activity intensity range, the processor prioritizes marking the current window as an abnormal activity window and reduces the influence of ambient sound characteristics on the state correction characteristics to avoid misattributing high heart rate changes during strenuous exercise to ambient sound. If neither the ambient noise intensity nor the duration of the interference exceeds the corresponding threshold, the processor does not use the ambient interference branch for correction, but instead performs conventional correction according to the historical body state fluctuation range corresponding to the activity type. If any type of data—heart rate, acceleration signal, or ambient sound signal—is missing, the processor reduces the confidence level of the state correction for that window based on data integrity and limits the window's influence on continuous trend judgment during the stress assessment phase.

[0044] The stress assessment results are determined based on the similarity between the state correction features and the historical scene features; The processor reads the state correction features corresponding to the current time window and retrieves historical scene features from historical data. Historical scene features are formed by activity types, ambient sound conditions, changes in physical condition, and user feedback records from past time windows, representing the user's historical state in different life scenarios. The processor places the current state correction features and historical scene features into the same feature space and calculates their similarity. Historical scenes with high similarity are used as the primary reference for judging the current stress state, while historical scenes with low similarity are not included or have a reduced degree of involvement. Based on the similarity calculation results, the processor determines the stress probability value and, combined with a preset stress threshold, generates a stress assessment result. The stress assessment result is written into the data record of the current time window, providing input for subsequent stress source identification.

[0045] The stress assessment result is determined based on the similarity between the state-corrected features and the historical scene features, including: constructing a stress assessment feature vector based on the state-corrected features; representing at least one historical scene feature as a historical feature vector; calculating the feature vector distance between the stress assessment feature vector and the historical feature vector; determining the stress probability value based on the feature vector distance; and determining the stress assessment result based on the stress probability value.

[0046] In one embodiment, when determining the stress assessment result based on the similarity between state-corrected features and historical scene features, the comparison is further limited to using a stress assessment feature vector and a historical feature vector. The processor constructs a stress assessment feature vector based on the state-corrected features. The stress assessment feature vector includes corrected heart rate change information, corrected heart rate variability information, activity intensity information, environmental noise intensity information, interference duration information, and data integrity markers. Each dimension is written into the vector in a preset order. Dimensions with missing data are not directly set as outliers, but are instead written with default placeholder values ​​and their weight is reduced in conjunction with the data integrity markers. The number of feature dimensions is determined by the system configuration. The configuration file records the name, source field, value range, and normalization method of each dimension to avoid direct comparison of data with different dimensions.

[0047] The processor selects at least one historical scene feature from historical data and represents each historical scene feature as a historical feature vector. Historical scene features are derived from a valid time window during the user's historical wearing process. A valid time window must simultaneously meet conditions such as data integrity, normal user wearing status, availability of historical feedback records, or reliable historical status labels. The historical feature vector and the stress assessment feature vector use the same dimensional order and normalization method. For users with a large amount of historical data, the processor can select only historical scene features that are close to the current time period, activity type, or ambient sound state to reduce the interference of irrelevant historical scenes on the stress assessment results.

[0048] The eigenvector distance between the stress assessment eigenvector and the historical eigenvector can be determined by the following formula: ; in, For the current time window and the first Feature vector distance between historical scenes The number of feature dimensions, For feature dimension index, For the first Weights of each feature dimension, The first eigenvector in the stress assessment feature vector 1 eigenvalue, For the first The th historical feature vector Each feature value represents a specific feature. The smaller the feature vector distance, the closer the current time window is to the corresponding historical scenario. The weights of the feature dimensions are set based on the reference value of each feature for stress assessment. Heart rate-related dimensions and fluctuation dimensions corrected for behavioral habits are usually given higher weights; dimensions with missing data, uncertain activity types, or insufficient environmental sound sampling quality have lower weights.

[0049] After the feature vector distance is calculated, the processor selects candidate historical scenes in ascending order of distance and generates a pressure probability value by considering whether historical pressure feedback exists in the candidate historical scenes. The pressure probability value can be determined by the following formula: ; ; in, This represents the probability value of stress within the current time window. A collection of candidate historical scenes, For the first The similarity weight of each candidate historical scene. For the first Historical stress markers correspond to candidate historical scenarios. These markers indicate whether the corresponding historical scenario has been reported by the user or recorded as a stress state by the system; a stress state is represented by a value of 1, and a non-stress state by a value of 0. The processor determines the stress assessment result based on the stress probability value. If the stress probability value is greater than a preset stress threshold, the current time window is marked as a stress-focused window; if the stress probability value is not greater than the preset stress threshold, the current time window is marked as a regular observation window. The preset stress threshold is set based on the accuracy of historical user feedback, false alarm tolerance, and the stability of continuous windows, and is updated after receiving subsequent user confirmation signals.

[0050] When the stress assessment results meet the preset stress conditions, the stress source identification results are determined based on the historical feedback emotion tags. The processor reads the stress assessment results for the current time window, confirming the stress probability value, candidate historical scenarios, and feature vector distance. When the stress probability value exceeds a preset stress threshold, the processor identifies the current time window as the source of stress and retrieves historical scenario features and historical feedback emotion tags from historical data. Historical scenario features represent changes in activity state, environmental sound state, and physical state within the historical time window, while historical feedback emotion tags represent the emotional state confirmed or recorded by the user in the corresponding historical scenario. The processor filters historical scenario features that are similar to the current state based on the feature vector distance and reads the associated historical feedback emotion tags. Combining the environmental, activity, and physical state information from the historical scenario features, the processor categorizes and matches the historical feedback emotion tags to determine the stress source category for the current time window. This stress source category is then written into the current data record, forming the stress source identification result, which is then used in the stress fluctuation prediction stage.

[0051] The preset stress conditions include a stress probability value greater than a preset stress threshold; the preset similarity conditions include a corresponding feature vector distance less than or equal to a preset distance threshold; the stress source identification result is determined based on historical feedback emotion tags, including: when the stress probability value is greater than the preset stress threshold, determining the target historical scene feature whose corresponding feature vector distance satisfies the preset similarity condition from at least one historical scene feature; obtaining the target historical feedback emotion tag corresponding to the target historical scene feature; determining the stress source category based on the target historical scene feature and the target historical feedback emotion tag, and using the stress source category as the stress source identification result.

[0052] In one embodiment, the stress source identification process is further limited to being jointly controlled by a preset stress threshold and a preset distance threshold. The preset stress threshold is used to determine whether the current time window needs to enter the stress source identification process. The preset stress threshold can be set according to the distribution of stress and non-stress states in the user's historical feedback, or it can use the system default value as the initial value. When the number of historical user feedbacks is insufficient, the processor uses the default threshold; when the number of historical feedbacks reaches the preset sample size, the processor readjusts the preset stress threshold according to the stress state window and non-stress state window confirmed by the user. The preset stress threshold should not be set too low to avoid frequent entry of physical state fluctuations caused by routine activities into the stress source identification process; nor should it be set too high to avoid filtering out continuous psychological stress in the early stages.

[0053] A preset distance threshold is used to determine whether there is sufficient similarity between historical scene features and the current state. The preset distance threshold can be determined based on the distribution of historical feature vector distances or configured according to the number of candidate historical scenes. When the pressure probability value is greater than the preset pressure threshold, the processor iterates through at least one historical scene feature obtained in the previous stage and reads the feature vector distance corresponding to each historical scene feature. Historical scene features whose feature vector distance is less than or equal to the preset distance threshold are identified as target historical scene features. If the number of target historical scene features meeting the condition exceeds a preset number, the processor can select the earlier target historical scene features in ascending order of feature vector distance; if no historical scene feature meets the preset distance threshold, the processor can use the historical scene feature with the smallest distance as a low-confidence reference and write a low-confidence marker in the current data record, thereby reducing the impact of this window in subsequent pressure fluctuation prediction stages.

[0054] The processor acquires the target historical feedback emotion tags corresponding to the target historical scene features. These tags can originate from user confirmations after historical alerts, manually recorded emotional states, emotional feedback records on the accompanying terminal, or selections saved by the smartwatch during historical interactions. The target historical feedback emotion tags must have the same time window as the target historical scene features or a confirmed association. If a target historical feedback emotion tag is missing, the processor does not directly generate a deterministic stress source category; instead, it marks the corresponding target historical scene feature as a sample with insufficient feedback and reduces its weight in category judgment. If multiple target historical feedback emotion tags conflict, the processor prioritizes tags with smaller feature vector distances, more recent feedback times, and more user confirmations as the basis for category judgment.

[0055] The stress source category is determined jointly based on the target's historical scene features and the target's historical feedback emotion labels. When the target's historical scene features show high environmental noise intensity and long duration of interference, and the target's historical feedback emotion labels indicate tension, irritability, or discomfort, the processor can classify the stress source category as environmental sound interference. When the target's historical scene features show high activity intensity and physical state fluctuations similar to historical activity states, the processor can classify the stress source category as activity load. When the target's historical scene features show low activity intensity but physical state fluctuations significantly deviate from historical ranges, and the target's historical feedback emotion labels indicate anxiety or stress, the processor can classify the stress source category as psychological stress-related. When the target's historical scene features simultaneously include environmental sound interference and activity deviations, the processor can classify the stress source category as a composite source category. The stress source category is written into the current time window as the stress source identification result, recording the category source, the number of referenced target historical scene features, the consistency of the target's historical feedback emotion labels, and low-confidence markers.

[0056] Based on the pressure source identification results, time series prediction is performed to obtain pressure fluctuation prediction results. User confirmation signals are received, and pressure prediction parameters are updated based on user confirmation signals.

[0057] The processor reads the pressure source identification results formed within consecutive time windows and extracts the pressure source category and pressure probability value corresponding to each time window. The pressure source category represents the possible source of the current pressure state, and the pressure probability value represents the likelihood of the current time window being under pressure. The processor arranges the above data according to chronological order to form a pressure change sequence. The pressure change sequence is input into a time series prediction model or a rule-based trend prediction unit, which outputs the pressure probability change trend within future time windows. After the prediction result is output, the smartwatch or accompanying terminal receives a user confirmation signal and compares the user confirmation signal with the pressure source category and pressure fluctuation prediction result. If the comparison result does not meet the preset matching conditions, the processor updates the pressure assessment weight, the preset pressure threshold, and the pressure source identification weight, so that the pressure prediction processing of subsequent time windows adopts the updated parameters.

[0058] Based on the pressure source identification results, time series prediction is performed to obtain pressure fluctuation prediction results, including: arranging the pressure source categories and pressure probability values ​​obtained in consecutive time windows in chronological order to obtain a pressure change sequence; performing time series prediction on the pressure change sequence to obtain the pressure probability change trend in the future time window, and using the pressure probability change trend as the pressure fluctuation prediction result.

[0059] In one embodiment, when performing time series prediction based on the pressure source identification results, the process is further limited to constructing a pressure change sequence according to continuous time windows. After completing the pressure source identification, the processor does not output the identification results of a single time window in isolation. Instead, it reads the pressure source category and pressure probability value within multiple consecutive time windows from the cache. The number of continuous time windows is set according to the prediction duration, sampling frequency, and processor computing power. When it is necessary to identify short-term pressure fluctuations, a smaller number of continuous time windows can be selected; when it is necessary to determine relatively stable trend changes, a larger number of continuous time windows can be selected. The number of windows is configured during system initialization and can be adjusted based on historical user feedback and false alarms.

[0060] The pressure change sequence is arranged chronologically, with each sequence position recording at least a time window identifier, pressure source category, pressure probability value, and data reliability marker. The pressure source category originates from the pressure source identification stage, the pressure probability value from the pressure assessment stage, and the data reliability marker indicates whether there is missing sensor data, uncertain activity type, insufficient historical feedback, or low-reliability historical scenario reference within that time window. If a pressure source identification is not triggered in a certain time window, the processor can mark the pressure source category of that time window as a regular observation category and retain the pressure probability value. If a low-reliability marker exists for a certain time window, the processor retains that time window when constructing the pressure change sequence, but reduces its participation in the predicted trend.

[0061] Time series forecasting can employ rule-based trend prediction units or lightweight time series forecasting models. When using rule-based trend prediction units, the processor determines the pressure probability change trend within future time windows based on the direction, duration, and changes in pressure source category of the pressure probability value within consecutive time windows. If the pressure probability value increases and the pressure source category remains consistent across multiple consecutive time windows, the processor marks the future time window as an upward trend; if the pressure probability value changes little within adjacent windows and the pressure source category remains stable, the processor marks the future time window as a stable trend; if the pressure probability value decreases continuously and there are many relief records corresponding to the same pressure source category in user historical feedback, the processor marks the future time window as a downward trend. When using lightweight time series forecasting models, the model input is a pressure change sequence, and the model output is the pressure probability change trend within one or more future time windows. The model can be trained on a compatible terminal or in the cloud, and a simplified inference structure can be deployed on the smartwatch to avoid excessive burden on the smartwatch's processor and battery.

[0062] The stress fluctuation prediction results include a future time window identifier, a trend in stress probability changes, the main stress source categories, and a prediction confidence marker. The prediction confidence marker is determined based on the data completeness of the input sequence, the stability of the stress source categories, and the number of historical feedback samples. If there are many missing input sequences, frequent changes in source categories, or insufficient historical feedback, the processor sets the prediction confidence marker to low confidence and limits the frequency of subsequent alarm triggers. The stress fluctuation prediction results are written to the current user status record and sent to the real-time feedback unit. The real-time feedback unit can prompt the user for confirmation through the smartwatch screen, vibration alerts, or the accompanying terminal interface. Before the user confirmation signal is returned, the stress fluctuation prediction results are only saved as predictions awaiting confirmation and are not directly used as new historical feedback sentiment tags.

[0063] The pressure prediction parameters include pressure assessment weights, preset pressure thresholds, and pressure source identification weights; user confirmation signals are used to indicate the user's confirmation of the pressure source category and pressure fluctuation prediction results; the pressure prediction parameters are updated based on the user confirmation signals, including: determining the prediction matching results based on the consistency between the user confirmation signals and the pressure source category and pressure fluctuation prediction results respectively; and updating the pressure assessment weights, preset pressure thresholds, and pressure source identification weights if the prediction matching results do not meet the preset matching conditions.

[0064] In one embodiment, when updating stress prediction parameters based on user confirmation signals, the triggering basis is further defined as the prediction matching result. The stress prediction parameters include stress assessment weights, a preset stress threshold, and stress source identification weights. The stress assessment weights are used to determine the degree of participation of physical state fluctuation characteristics, activity intensity characteristics, and environmental sound characteristics in stress probability calculation; the preset stress threshold is used to determine whether the stress probability value enters the stress source identification process; and the stress source identification weights are used to determine the degree of participation of target historical scene characteristics and target historical feedback emotion tags in stress source category judgment. These parameters use default configurations during system initialization, and are updated by the processor based on confirmation results as user historical data and confirmation records gradually increase.

[0065] User confirmation signals indicate the user's confirmation of the stress source category and stress fluctuation prediction results. These signals can originate from confirmation, denial, delayed confirmation, or emotion tag selection on the smartwatch, or from feedback operations on the paired terminal. Upon receiving a user confirmation signal, the processor reads the previously saved stress source category and stress fluctuation prediction results within the same time window and compares the consistency between the user confirmation signal and the stress source category, as well as the consistency between the user confirmation signal and the stress fluctuation prediction results. If the user confirms that the stress source matches the system-identified stress source category, and the user's subsequent stress state matches the predicted trend, the prediction matching result meets the preset matching conditions. If the user denies the stress source category, provides the opposite trend, or repeatedly ignores similar prediction prompts, the prediction matching result does not meet the preset matching conditions.

[0066] The preset matching conditions are set based on the number of user confirmations and feedback stability. For a single prediction result, explicit user confirmation can be considered a satisfied condition, and explicit denial a dissatisfied condition. For continuous prediction results, multiple adjacent time windows can be required to have consistent user confirmation signals with the pressure source category and pressure fluctuation prediction results. If a user confirmation signal is missing, the processor does not immediately determine it as a mismatch, but marks the current window as a non-feedback sample. Non-feedback samples can participate in trend caching, but are not used for parameter updates to avoid incorrect calibration due to user inaction. If the user confirmation signal is consistent with the pressure source category but inconsistent with the pressure fluctuation prediction result, the processor only updates the trend prediction-related parameters and does not change the pressure source identification weight. If the user confirmation signal is inconsistent with the pressure source category but the pressure fluctuation direction is consistent, the processor reduces the identification weight of the corresponding pressure source category while retaining the pressure assessment weight.

[0067] When the predicted matching result does not meet the preset matching conditions, the processor updates the stress prediction parameters according to the type of mismatch. If the system frequently classifies routine activities as stress states, the processor reduces the stress assessment weight of bodily state fluctuation features under the corresponding activity type and appropriately increases the preset stress threshold. If the system misses the stress state confirmed by the user, the processor increases the stress assessment weight of the corresponding bodily state fluctuation features or environmental sound features and appropriately decreases the preset stress threshold. If the stress source category is inconsistent with the user's confirmation result, the processor adjusts the stress source identification weight, so that historical feedback emotion tags and target historical scene features consistent with the user's confirmation result have a higher degree of participation in subsequent judgments. Each parameter update records the update time, trigger window, parameter value before update, and parameter value after update. If the update magnitude exceeds the system's allowed range, the processor adopts a phased update method to avoid a single abnormal feedback causing a significant deviation in subsequent prediction results.

[0068] like Figure 2 As shown, a context-aware smartwatch psychological stress prediction system is used to implement a context-aware smartwatch psychological stress prediction method. The system includes: The data acquisition module is used to acquire heart rate data, acceleration signals, and ambient sound signals. It aligns the heart rate data, acceleration signals, and ambient sound signals according to the same time window to obtain multi-source time series data. The data acquisition module consists of a heart rate detection unit, an inertial sensor, a microphone, a sensor interface circuit, a clock synchronization circuit, and a data buffer. It is used to collect heart rate data, acceleration signals, and ambient sound signals and write them to the buffer according to the same time window.

[0069] The feature correction module is used to extract body state fluctuation features, activity intensity features, and environmental sound features from multi-source time series data, and to correct the body state fluctuation features by combining behavioral habit information to obtain state-corrected features. The feature correction module consists of a processor, a memory, and a sensor data preprocessing circuit. It is used to read multi-source time series data, extract body state fluctuation features, activity intensity features, and environmental sound features, and call behavioral habit information in the memory to complete feature correction.

[0070] The stress assessment module is used to determine the stress assessment result based on the similarity between the state correction features and the historical scene features. The stress assessment module consists of a processor, a historical data storage unit and a feature comparison and operation unit. It is used to read the state correction features and historical scene features, calculate the similarity between the two, and generate the stress assessment result.

[0071] The source identification module is used to determine the source of stress based on historical feedback emotion tags when the stress assessment result meets the preset stress conditions. The source identification module consists of a processor, a historical feedback storage unit, and a category judgment unit. It is used to read historical feedback emotion tags and determine the source of stress when the stress assessment result meets the preset stress conditions.

[0072] The trend update module is used to perform time series forecasting based on the pressure source identification results, obtain pressure fluctuation forecast results, receive user confirmation signals, and update pressure forecast parameters based on the user confirmation signals. The trend update module consists of a processor, a time series calculation unit, a user interaction unit, and a parameter storage unit.

[0073] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A context-aware smartwatch-based method for predicting psychological stress, characterized in that, The method includes: Acquire heart rate data, acceleration signals, and ambient sound signals; align the heart rate data, acceleration signals, and ambient sound signals according to the same time window to obtain multi-source time series data; Based on the multi-source time series data, body state fluctuation features, activity intensity features, and environmental sound features are extracted, and the body state fluctuation features are corrected by combining behavioral habit information to obtain state correction features. The stress assessment result is determined based on the similarity between the state correction features and the historical scene features; When the stress assessment result meets the preset stress conditions, the stress source identification result is determined based on the historical feedback emotion tags. Based on the pressure source identification results, time series prediction is performed to obtain pressure fluctuation prediction results. A user confirmation signal is received, and the pressure prediction parameters are updated based on the user confirmation signal.

2. The method according to claim 1, characterized in that, The heart rate data, acceleration signal, and ambient sound signal are aligned within the same time window to obtain multi-source time series data, including: Acquire the acquisition timestamps of the heart rate data, the acceleration signal, and the ambient sound signal; The collected timestamps are merged according to a preset time window; The heart rate data, acceleration signal, and ambient sound signal belonging to the same preset time window are associated as the same time series record to obtain the multi-source time series data.

3. The method according to claim 1, characterized in that, The step of extracting body state fluctuation features, activity intensity features, and environmental sound features from the multi-source time series data includes: Based on the heart rate data, the heart rate change rate and heart rate variability are determined to obtain the body state fluctuation characteristics; The activity intensity and duration are determined based on the acceleration signal, thus obtaining the activity intensity characteristics; The ambient noise intensity, interference duration, and frequency band energy distribution are determined based on the ambient sound signal to obtain the ambient sound characteristics.

4. The method according to claim 3, characterized in that, The process of combining behavioral habit information to correct the body state fluctuation characteristics, resulting in state correction characteristics, includes: The activity type is determined based on the activity intensity characteristics; Determine the historical range of physical state fluctuations corresponding to the activity type based on the behavioral habit information; If the physical state fluctuation feature is within the range of historical physical state fluctuations, the weight of the physical state fluctuation feature in determining the stress assessment result is reduced. The state-corrected features are obtained by correcting the body state fluctuation features based on the reduced weights.

5. The method according to claim 3, characterized in that, The process of combining behavioral habit information to correct the body state fluctuation characteristics to obtain state correction characteristics further includes: Based on the behavioral habit information, determine the range of historical activity intensity and the range of historical physical state fluctuations corresponding to the same time window; If the activity intensity feature is within the range of the historical activity intensity, and either the environmental noise intensity is greater than the environmental noise intensity threshold or the interference duration is greater than the interference duration threshold, the state correction feature is obtained based on the fluctuation portion of the body state fluctuation feature that exceeds the range of the historical body state fluctuation.

6. The method according to claim 1, characterized in that, The step of determining the stress assessment result based on the similarity between the state correction features and historical scene features includes: Construct a stress assessment feature vector based on the state correction features; At least one of the historical scene features is represented as a historical feature vector; Calculate the eigenvector distance between the stress assessment eigenvector and the historical eigenvector, respectively; The pressure probability value is determined based on the feature vector distance, and the pressure assessment result is determined based on the pressure probability value.

7. The method according to claim 6, characterized in that, The preset pressure condition includes the pressure probability value being greater than a preset pressure threshold. The preset similarity condition includes the corresponding feature vector distance being less than or equal to a preset distance threshold; The determination of stress source identification results based on historical feedback emotion tags includes: If the pressure probability value is greater than the preset pressure threshold, determine the target historical scene feature whose feature vector distance satisfies the preset similarity condition from at least one of the historical scene features; Obtain the target historical feedback sentiment tags corresponding to the target historical scene features; The stress source category is determined based on the target historical scene features and the target historical feedback emotion tags, and the stress source category is used as the stress source identification result.

8. The method according to claim 7, characterized in that, The step of performing time series prediction based on the pressure source identification results to obtain pressure fluctuation prediction results includes: By arranging the pressure source categories and pressure probability values ​​obtained within consecutive time windows in chronological order, a pressure change sequence is obtained; Time series prediction is performed on the pressure change sequence to obtain the pressure probability change trend within the future time window, and the pressure probability change trend is used as the pressure fluctuation prediction result.

9. The method according to claim 8, characterized in that, The pressure prediction parameters include pressure assessment weights, the preset pressure threshold, and pressure source identification weights. The user confirmation signal is used to indicate the user's confirmation of the pressure source category and the pressure fluctuation prediction result; The step of updating the pressure prediction parameters based on the user confirmation signal includes: The prediction matching result is determined based on the consistency between the user confirmation signal and the pressure source category and the pressure fluctuation prediction result, respectively. If the predicted matching result does not meet the preset matching conditions, update the pressure assessment weight, the preset pressure threshold, and the pressure source identification weight.

10. A context-aware smartwatch psychological stress prediction system, used to implement the context-aware smartwatch psychological stress prediction method according to any one of claims 1 to 9, characterized in that, The system includes: The data acquisition module is used to acquire heart rate data, acceleration signals and ambient sound signals, and align the heart rate data, acceleration signals and ambient sound signals according to the same time window to obtain multi-source time series data; The feature correction module is used to extract body state fluctuation features, activity intensity features and environmental sound features from the multi-source time series data, and to correct the body state fluctuation features by combining behavioral habit information to obtain state correction features. The stress assessment module is used to determine the stress assessment result based on the similarity between the state correction features and the historical scene features; The source identification module is used to determine the source identification result of stress based on historical feedback emotion tags when the stress assessment result meets the preset stress conditions. The trend update module is used to perform time series prediction based on the pressure source identification results, obtain pressure fluctuation prediction results, receive user confirmation signals, and update pressure prediction parameters based on the user confirmation signals.

Images