Intelligent resident health analysis method and system based on big data

Abstract

The invention discloses a resident health intelligent analysis method and system based on big data. Comprising the steps of obtaining historical data of a plurality of health indexes of residents, performing time alignment and cleaning, presetting a sliding window, calculating a short-term fluctuation degree, obtaining a trend baseline, and generating a health trajectory deviation trend in combination with the short-term fluctuation degree. The key life events are mapped into event impact vectors, and the event impact vectors and the health track deviation potential form a fusion deviation trend vector. A historical state vector is constructed based on historical data, short-term fluctuation, a health trajectory deviation trend and an event impact vector, a forgetting strategy is adopted to update the state vector, a time sequence prediction model is input to output an individual health risk, layered early warning is performed, and the accuracy and timeliness of health risk prediction are effectively improved. Through the technologies of event impact, state vectors, a time sequence model, layered early warning and the like, the accuracy and timeliness of health risk prediction are remarkably improved, and efficient closed-loop management of monitoring and intervention is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of big data, and specifically relates to a method and system for intelligent analysis of resident health based on big data. Background Art

[0002] Existing health risk assessments for residents often use cross-sectional methods based on single physical examinations or a small number of recent measurements. These methods can only interpret physiological status at a specific point in time, making it difficult to systematically utilize the evolutionary trajectories and cyclical, seasonal, and individualized rhythms of indicators such as blood sugar, blood pressure, and body mass index over longer time scales. Such methods ignore the inertial effects of temporal correlation and health status, and are prone to mistaking short-term noise for anomalies or slow deterioration for normal fluctuations, resulting in insufficient sensitivity and stability in risk identification. With the popularization of home monitoring and wearable devices, the problem of increasing data density while modeling remains at the cross-sectional level has become more prominent.

[0003] On the other hand, critical real-life events such as diagnosis and follow-up visits, surgery or hospitalization, medication initiation or adjustment, changes in exercise and dietary habits, and major stressors often lead to sudden or sustained shifts in health trajectories. Existing solutions generally lack a mechanism to jointly model these discrete events with continuous physiological indicators within a unified framework, making it difficult to capture baseline shifts, fluctuation amplification, and recovery processes before and after an event. Direct consequences of this include delayed early warning, inaccurate individual risk stratification, and reduced efficiency in resource allocation for follow-up and intervention. Summary of the Invention

[0004] To address the technical problems in existing health risk assessment schemes, which lack a mechanism to jointly model discrete events and continuous physiological indicators under a unified framework, making it difficult to accurately express the impact of events on health trajectories, thereby causing delayed early warning, inaccurate individual risk stratification, and reduced efficiency in follow-up and intervention resource allocation, the present invention provides the following technical solutions.

[0005] In the first aspect, a method for intelligent analysis of resident health based on big data includes: obtaining historical data of multiple health indicators of residents, and performing time alignment and cleaning; presetting a sliding window for historical data, and calculating the short-term volatility of each health indicator in the sliding window to characterize the short-term stability of each indicator, and obtaining the trend baseline of the historical data of each health indicator; generating a health trajectory deviation trend by combining the degree of deviation of the current health indicator of the resident from the trend baseline and the short-term volatility; mapping key life events into event impact vectors according to event types, and forming a fused deviation trend vector with multiple health trajectory deviations; constructing a historical state vector based on historical data, short-term volatility, health trajectory deviation trend and event impact vector; using a forgetting strategy to attenuate the historical state vector and then superimpose it with the fused deviation trend vector to update the state vector; inputting the sequence of historical state vectors into a time series prediction model in time sequence to output individual health risks and perform layered warnings.

[0006] Preferably, the time alignment includes time zone standardization, sampling frequency unification, resampling to a unified time grid, outlier quantile clipping, and window interpolation of missing segments, and adding confidence marks to the interpolated samples.

[0007] Preferably, the short-term volatility includes: Calculation is performed within a sliding window centered on the target time. The window width is adaptively set based on the indicator sampling frequency or business rules. Intraday and weekly fluctuations are obtained based on each preset scale window.

[0008] Preferably, the step of obtaining the health trajectory deviation trend includes: Based on the difference between the current health indicator and the trend baseline and combined with the amplification factor of short-term volatility, the health trajectory deviation potential of multiple indicators is combined to form a deviation potential vector.

[0009] Preferably, obtaining the event impact vector includes: Weighted mapping was performed based on event type, severity, and temporal proximity, including diagnosis or follow-up, surgery or hospitalization, medication initiation or adjustment, exercise, or dietary habits.

[0010] Preferably, the historical state vector includes: When updating the historical state vector, a forgetting factor is applied to the historical state to reflect the attenuation effect, and it is superimposed with the composite amount of the deviation potential vector and the event impact vector.

[0011] Preferably, the time series prediction model is a time series deep learning model of a long short-term memory network, and cross-validation and early stopping strategies are used to suppress overfitting, and weighted loss or resampling is used for correction.

[0012] The beneficial effects of the present invention are: The present invention solves the problems of inconsistent timestamps and sequence breaks through time alignment and cleaning, forming a usable continuous time series; short-term volatility overcomes the anomaly masking by the mean and highlights abnormal fluctuation signals; trend baseline uses long-term trajectory and seasonal laws to manifest the slowly deteriorating trend; deviation potential combines deviation and fluctuation to reduce the false positives and missed negatives of the threshold method; event impact incorporates key events such as confirmed surgery and medication into unified quantification; state vector and forgetting take into account historical memory and new information response to suppress noise interference; time series model models cross-time dependency and nonlinearity to improve the accuracy of risk prediction; layered warning linkage converts risk probability into executable actions, opening up a closed loop from monitoring to intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a method flow chart of steps S1 to S4 in the method for intelligent analysis of resident health based on big data in an embodiment of the present invention.

[0014] Figure 2 This is a structural block diagram of a resident health intelligent analysis system based on big data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0016] Reference Figure 1 The resident health intelligent analysis method based on big data includes steps S1 to S4, which are as follows: S1: Obtain the initial route planning results and order sites of the passenger vehicle, where the order sites are the coordinate points of the passenger vehicle, and construct graph structure data based on the coordinate points.

[0017] Collect multi-source raw data from channels such as electronic medical records, physical examination records, wearable devices and questionnaires, including multi-dimensional continuous indicators and discrete event sets; aggregate and deduplicate data by user ID and timestamp, and unify coding, units and field naming.

[0018] All timestamps are unified to the base time zone and the original time zone field is retained for auditing; the recorded times are sorted and verified to eliminate concurrent conflicts with the same second, forming a strictly monotonic timeline.

[0019] A unified time grid is set according to the sampling frequency of each data source, and resampling strategies such as linear interpolation or forward filling are implemented for multidimensional continuous indicators to align multiple variables on the same time axis; median filtering is used to smooth sudden high-frequency noise.

[0020] Quantile trimming is performed on obvious measurement errors and outliers and abnormal labels are recorded; window interpolation is performed on short missing segments, and model interpolation is performed on long missing segments and the confidence weight of the segment is reduced to ensure the integrity of the sequence while controlling error propagation.

[0021] Dictionary the discrete event set, unify the event type, severity and direction of action, establish a configurable mapping table from events to impact vectors and manage it in a versioned manner, supporting quantitative expression and traceability during subsequent fusion.

[0022] Desensitize and implement minimum authorization access control for personally identifiable information, enable audit logs and data lineage records, and ensure that subsequent training and inference processes are compliant and traceable.

[0023] S2: Preset a sliding window for historical data and calculate the short-term volatility of each health indicator within the sliding window to characterize the short-term stability of each indicator and obtain the trend baseline of the historical data of each health indicator; the degree of deviation between the current health indicators of residents and the trend baseline is combined with the short-term volatility to generate the health trajectory deviation trend.

[0024] A sliding window is preset for historical data, and the short-term volatility of each health indicator within the sliding window is calculated to characterize the short-term stability of each indicator.

[0025] Target time Build a center with a width of Sliding window , calculate the indicator within the window The short-term volatility is used to measure stability and to capture abnormal fluctuation signals: Specifically, short-term volatility satisfies the following relationship: ; Where, Indicates the health indicators at the target moment The short-term volatility of Represents a sliding window The number of data points in express, Indicates the Health indicators at time points The measured value of Indicates the Health indicators in the sliding window The average value within Indicates a time point In the sliding window Inside.

[0026] Trend baseline estimation includes: performing robust regression or smoothing (linear, polynomial, spline or kernel methods) on each indicator based on the complete historical series to obtain the long-term trend, and combining seasonal decomposition when necessary to separate the cyclical components and reduce the impact of short-term disturbances on trend estimation.

[0027] The generation of health trajectory deviation trend includes: combining the degree of deviation of residents' current health indicators from the trend baseline with short-term fluctuations to generate the health trajectory deviation trend.

[0028] ; Where, Indicates the health indicators at the target moment The health trajectory deviates from the trend, Indicates the health indicators at the target moment The actual measured value, Indicates the health indicators at the target moment The trend baseline value of represents the amplification factor, Indicates the health indicators at the target moment The short-term volatility of Indicates the The average short-term volatility of a health indicator.

[0029] Multi-index synthesis and scale alignment include: Composition vector , apply distribution normalization and robust scaling to indicators of different dimensions to ensure comparability across indicators; assign minimum weights to weak signal dimensions to prevent them from being completely overwhelmed.

[0030] S3: Map key life events into event impact vectors according to event types, and combine them with multiple health trajectory deviation potentials to form a fused deviation trend vector; construct a historical state vector based on historical data, short-term volatility, health trajectory deviation trend and event impact vectors.

[0031] Event impact vector synthesis includes: Key life events are mapped into event impact vectors according to event types and combined with multiple health trajectory deviation potentials to form a fused deviation trend vector.

[0032] The event type, severity and time proximity are retrieved or calculated from the mapping table. When there are multiple events at the same time, the equivalent impact is synthesized according to the importance and time weight to ensure that acute events have immediate impact and chronic events have continuous impact.

[0033] The state vector update includes: applying a forgetting factor to the historical state to reflect time decay, and then superimposing it with the current deviation potential and event shock to generate a new state representation, which is used as the input sequence of the downstream time series model.

[0034] Specifically, the state vector satisfies the following relationship: ; Where, Indicates target time The state vector of represents the forgetting factor, , Indicates the last moment The state vector of Indicates the current time The health trajectory deviates from the trend, Indicates the current time A collection of all key life events, Representing an event The event impact vector.

[0035] That is, when updating the resident health status vector When the forgetting factor The state vector at the previous moment Attenuation is performed to reflect the time decay effect; then the current health trajectory deviates from the trend , reflecting the deviation of the current health indicator from the trend baseline; finally, add the event impact vector of all key life events at the current moment This comprehensive process ensures that the health state vector can dynamically reflect the health status of residents, taking into account both historical information and new events.

[0036] Smoothing and hysteresis control include: in order to reduce the violent swing caused by short-term noise, the residents' health status vector Or its risk projection performs exponentially weighted moving average; responds immediately to rising signals and sets a hysteresis threshold for falling signals to avoid frequent repetitions.

[0037] Explainability records include: At the same time, it outputs the leading indicators, key events and contribution summaries, forming a decision-making chain of "indicator → deviation → event → status" for use in audit and intervention review.

[0038] S4: Use the forgetting strategy to decay the historical state vector and then superimpose it with the fusion deviation trend vector to update the state vector; input the sequence of historical state vectors into the time series prediction model according to the time sequence to output individual health risks and perform stratified warnings.

[0039] Training data preparation and division include: Construct samples for the input sequence, using the risk labels in the future window as supervisory signals; divide the training, validation, and test sets into user-based categories to avoid information leakage, and apply weighted losses or resampling to account for class imbalances. Weighted losses or resampling are applied to class imbalances to ensure balanced and effective model training.

[0040] Select LSTM, GRU, or an equivalent time series model as the base model, configure hyperparameters such as the number of layers, hidden units, learning rate, and dropout rate; use grid or Bayesian optimization to adjust the parameters on the validation set, and enable early stopping and weight decay to prevent overfitting.

[0041] Using sliding window sampling and mini-batch training, we monitor training and validation loss, AUC, recall, and calibration in real time. When calibration deviation is too large, we introduce temperature scaling or equivalent post-processing techniques to ensure model accuracy and reliability.

[0042] Maintaining the latest window in the online service , trigger or timed refresh of risk output; ensure that end-to-end latency meets business thresholds, and downsample non-critical indicators when resources are limited to ensure real-time performance.

[0043] Risk stratification and threshold setting include: setting multi-level thresholds based on the performance curve of the validation set, mapping risk probabilities into stratification levels; applying differentiated thresholds and alarm routing to highly sensitive populations or specific disease subgroups to achieve refined management.

[0044] Early warning linkage and intervention strategies include: linking stratified results with follow-up, retesting, medication review, or lifestyle intervention; recording execution and result feedback according to the strategy to form closed-loop data to support subsequent model and threshold retraining.

[0045] Perform incremental training regularly or in a triggered manner to compare the performance of the new and old models in offline and low-traffic grayscale environments. Complete version switching after meeting the preset threshold. Retain version archives of the model, dictionary, and threshold for retrospective use.

[0046] Continuously monitor service availability, latency, failure rate, and data drift; automatically issue alerts for abnormal fluctuations and switch to a safe degradation strategy; and retain audit logs and explanatory summaries throughout the entire chain to meet compliance requirements.

[0047] Establish fault-tolerant mechanisms and fallback models for data gaps, model unavailability, or external dependency failures to ensure that degraded but usable risk scores can still be provided during abnormal periods, and automatically switch back and fill in missing records after the service is restored.

[0048] It should be noted that those skilled in the art may make various modifications and improvements without departing from the scope of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be based on the appended claims.

[0049] The present invention also provides a resident health intelligent analysis system based on big data. Figure 2 As shown, the system includes a processor and a memory, and the memory stores computer program instructions. When the computer program instructions are executed by the processor, the method for intelligent analysis of resident health based on big data according to the first aspect of the present invention is implemented.

[0050] The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and therefore will not be described in detail here.

[0051] In the description of this specification, "multiple" and "several" mean at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0052] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. A method for intelligent analysis of residents' health based on big data, characterized by: include: Obtain historical data on multiple health indicators of residents and perform time alignment and cleaning; Preset a sliding window for historical data and calculate the short-term volatility of each health indicator within the sliding window to characterize the short-term stability of each indicator and obtain the trend baseline of the historical data of each health indicator; The deviation degree of residents' current health indicators from the trend baseline is combined with short-term fluctuations to generate the health trajectory deviation trend; Key life events are mapped into event impact vectors according to event types, and combined with multiple health trajectory deviation potentials to form a fused deviation trend vector; a historical state vector is constructed based on historical data, short-term volatility, health trajectory deviation trend, and event impact vectors; The forgetting strategy is used to decay the historical state vector and then superimpose it with the fusion deviation trend vector to update the state vector; the sequence of historical state vectors is input into the time series prediction model according to the time sequence to output individual health risks and perform stratified warnings.

2. The method for intelligent analysis of resident health based on big data according to claim 1 is characterized in that: The time alignment includes time zone standardization, sampling frequency unification, resampling to a unified time grid, outlier quantile clipping, and window interpolation of missing segments, and adding confidence marks to the interpolated samples.

3. The method for intelligent analysis of resident health based on big data according to claim 1, characterized in that: The short-term volatility includes: Calculation is performed within a sliding window centered on the target time. The window width is adaptively set based on the indicator sampling frequency or business rules. Intraday and weekly fluctuations are obtained based on each preset scale window.

4. The method for intelligent analysis of resident health based on big data according to claim 1, characterized in that: The steps of obtaining the health trajectory deviation trend include: Based on the difference between the current health indicator and the trend baseline and combined with the amplification factor of short-term volatility, the health trajectory deviation potential of multiple indicators is combined to form a deviation potential vector.

5. The method for intelligent analysis of resident health based on big data according to claim 1 is characterized in that: Obtaining the event impact vector includes: Weighted mapping was performed based on event type, severity, and temporal proximity, including diagnosis or follow-up, surgery or hospitalization, medication initiation or adjustment, exercise, or dietary habits.

6. The method for intelligent analysis of resident health based on big data according to claim 1, characterized in that: The historical state vector includes: When updating the historical state vector, a forgetting factor is applied to the historical state to reflect the attenuation effect, and it is superimposed with the composite amount of the deviation potential vector and the event impact vector.

7. The method for intelligent analysis of resident health based on big data according to claim 1, characterized in that: The time series prediction model is a time series deep learning model of a long short-term memory network, and uses cross-validation and early stopping strategies to suppress overfitting, and uses weighted loss or resampling for correction.

8. The resident health intelligent analysis system based on big data according to claim 1 is characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method for intelligent analysis of resident health based on big data according to any one of claims 1 to 7 is implemented.

Images