Health information monitoring and management system based on multi-source data fusion analysis

Abstract

The invention relates to the technical field of health information monitoring and management, and discloses a health information monitoring and management system based on multi-source data fusion analysis, which comprises a physiological data acquisition unit, a fusion analysis engine, an intelligent decision management module and the like. The physiological data acquisition unit acquires a multi-source heterogeneous data stream, and a multi-layer fusion topology is constructed through preprocessing; the fusion analysis engine realizes data feature association and anomaly detection through feature association and mode recognition; and the intelligent decision management module generates a health state reference strategy and dynamically allocates data source weights. The real-time calibration module calibrates a signal time domain and adapts to an analysis frequency, the data weight optimization module evaluates an optimization strategy based on credibility, and the fault-tolerant processing module completes data verification and recovery in combination with the distributed cache unit. The system realizes efficient fusion, dynamic decision and reliable management of multi-source data, improves the accuracy of health monitoring and the robustness of the system, and is suitable for intelligent health management scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of health information monitoring and management, and in particular to a health information monitoring and management system based on multi-source data fusion analysis. Background Art

[0002] With the increasing popularity of smart medical devices and the increasing demand for health management, health monitoring and management based on multi-source data has become a hot topic in current research. Traditional health monitoring systems can usually only process physiological data of a single type or from a limited number of sources, making it difficult to meet the needs of complex health status assessment for multi-dimensional information fusion. For example, in existing technologies, the collection, transmission, and processing of multi-source heterogeneous data such as wearable devices, environmental sensors, and medical databases generally suffer from inconsistent sampling rates, incompatible data formats, and large timestamp deviations. These problems lead to inefficient data fusion and inaccurate feature extraction, which in turn affects the accuracy and real-time performance of health status assessments.

[0003] At the data preprocessing level, traditional systems often use fixed-parameter cleaning algorithms, which are unable to dynamically adjust processing strategies based on real-time data quality. For example, when signal integrity varies between different data sources, fixed cleaning rules may result in the loss of key features or the retention of invalid data, affecting the reliability of subsequent fusion analysis. Furthermore, the problem of asynchrony between multi-source data in the temporal dimension is particularly prominent. For example, it is difficult to directly correlate high-frequency physiological signals from wearable devices with low-frequency detection data from medical databases. Traditional time-domain alignment methods lack adaptability to the characteristics of dynamic data streams, resulting in the fused feature vector being unable to accurately reflect the true health status.

[0004] In terms of fusion analysis and decision management, existing systems have relatively limited feature association and pattern recognition capabilities and lack dynamic optimization mechanisms for multi-dimensional data association rules. For example, when abnormal physiological signals are detected, traditional systems struggle to quickly correlate environmental factors (such as temperature and humidity) with historical medical data, resulting in insufficient sensitivity and specificity in anomaly detection. Furthermore, decision modules typically employ static strategies, unable to dynamically adjust data source weights and monitoring priorities based on data credibility and monitoring requirements. This lacks effective fault tolerance and adaptive capabilities in the event of data conflicts or device failures.

[0005] In terms of system reliability and scalability, traditional health information management systems generally lack robust fault-tolerance and data caching mechanisms. Failures in a data source or processing node can easily lead to interruptions in data processing across the entire system or skewed results. Furthermore, as the number of connected devices and data types increases, traditional centralized data storage and processing architectures struggle to meet the performance demands of high-concurrency scenarios, resulting in poor scalability. Summary of the Invention

[0006] The purpose of the present invention is to provide a health information monitoring and management system based on multi-source data fusion analysis to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a health information monitoring and management system based on multi-source data fusion analysis, the system comprising:

[0008] Physiological data acquisition unit, fusion analysis engine and intelligent decision management module;

[0009] The physiological data acquisition unit includes a multi-source signal synchronization node and a heterogeneous data preprocessing network; the fusion analysis engine includes a feature association unit and a pattern recognition unit; the intelligent decision management module includes a dynamic decision node and a priority mapping unit; the multi-source signal synchronization node is used to obtain heterogeneous data streams from wearable devices, environmental sensors and medical databases in real time; according to the physiological signal sampling rate and data integrity index, a multi-layer data fusion topology is constructed through the heterogeneous data preprocessing network; the heterogeneous data preprocessing network is composed of N cleaning subnets based on data quality assessment in series; based on the real-time feature extraction capability of the N cleaning subnets, a health status benchmark strategy is generated through the dynamic decision node, and combined with the abnormal detection signal output by the pattern recognition unit, the monitoring task is dynamically assigned a data source weight through the priority mapping unit.

[0010] Preferably, the system further comprises a real-time calibration module and a data weight optimization module; the real-time calibration module comprises a signal compensation unit and a sampling rate adaptation unit;

[0011] According to the generated health status benchmark strategy and dynamic data source weight distribution results, the output signal of the physiological data acquisition unit is time-domain aligned and calibrated by the signal compensation unit, and the data parsing frequency of the fusion analysis engine is dynamically adjusted by the sampling rate adaptation unit; at the same time, the data credibility assessment network constructed by the data weight optimization module is used to perform real-time analysis of the multi-source data conflict rate, and the analysis results are fed back to the dynamic decision node and priority mapping unit to iteratively optimize the monitoring strategy.

[0012] Preferably, the signal compensation unit is composed of a time delay correction circuit corresponding to a cleaning subnet connected in series in a heterogeneous data preprocessing network; and the steps of constructing the multi-layer data fusion topology include:

[0013] Collect the original signal waveforms and data timestamp deviations of multiple source devices, input them into the sliding window filter configured in each cleaning subnet, generate signal synchronization compensation parameters and store them in the corresponding cleaning subnet;

[0014] Based on the synchronization compensation parameters stored in each layer of cleaning subnet and combined with the time continuity index of data flow, the sliding correlation coefficient algorithm is used to associate the multi-source data features with N cleaning subnets to form a dynamic feature fusion topology.

[0015] Preferably, the activation conditions of the cleaning subnet configured in the multi-source signal synchronization node include: the data source sampling rate is within a preset compatibility range, the feature retention rate corresponding to the cleaning subnet is greater than a threshold, and the data timestamp deviation is less than the allowed synchronization range;

[0016] The feature association unit is also configured with a multi-dimensional matching link; the multi-dimensional matching link includes a steady-state analysis sub-link and an abnormality correction sub-link; the steady-state analysis sub-link is generated based on the matching degree between the association rules of the fusion analysis engine and the health status benchmark strategy; the abnormality correction sub-link is constructed based on the association relationship between the instantaneous signal loss of the data acquisition unit and the monitored abnormal events.

[0017] Preferably, the dynamic decision node is configured with a collaborative decision model; the collaborative decision model includes a signal-feature mapping table corresponding to the physiological data acquisition unit and an abnormality detection parameter library of the fusion analysis engine; the steps of generating the health status benchmark strategy include:

[0018] According to the real-time signal quality index of the physiological data acquisition unit and the data source weight distribution results, the theoretical feature extraction capacity of each cleaning subnet in the steady-state analysis sublink is calculated;

[0019] The theoretical feature extraction capacity and real-time monitoring requirements are input into the collaborative decision-making model to generate a health status reference baseline. The reference baseline is then corrected for conflict rate through the anomaly correction sub-link to form a health status benchmark strategy.

[0020] Preferably, the parameter updating step of the collaborative decision-making model includes:

[0021] When it is detected that the data source sampling rate exceeds the preset compatibility range or the multi-source data conflict rate exceeds the threshold, the first optimization instruction is triggered, that is, the backup data source channel is started and the correlation dimension of the fusion analysis engine is adjusted;

[0022] If the data conflict rate has not returned to the allowable range after the execution of the first optimization instruction, the second optimization instruction is triggered, that is, switching to the anomaly correction sub-link through the multi-dimensional matching link, and reallocating the feature extraction weights of the data source based on the duration of the monitored abnormal event.

[0023] Preferably, the parameter updating step of the collaborative decision-making model further includes:

[0024] When the feature retention rate of a certain cleaning subnet in the physiological data acquisition unit is lower than the threshold, the third optimization instruction is triggered, that is, the data transmission channel of the subnet is blocked, and the corresponding signal flow is transferred to other cleaning subnets through the signal compensation unit;

[0025] During the execution of the third optimization instruction, if other cleaning subnets are detected to be overloaded, the fourth optimization instruction is triggered, that is, the data weight optimization module is called to downgrade the output strategy of the priority mapping unit and limit the monitoring authority of low-credibility data sources.

[0026] Preferably, the degradation processing logic of the data weight optimization module includes:

[0027] Construct a data source priority table based on the conflict rate index and signal synchronization stability parameters output by the data credibility assessment network;

[0028] When the fourth optimization instruction is triggered, the data source weight is dynamically downgraded according to the priority level table, and the parsing frequency of the fusion analysis engine is adjusted synchronously to adapt to the downgraded monitoring strategy.

[0029] Preferably, the system further includes a fault-tolerant processing module and a distributed cache unit;

[0030] The fault-tolerant processing module includes a redundancy check unit and an abnormal data recovery unit;

[0031] According to the conflict rate index and priority level table output by the data credibility assessment network, the redundancy check unit performs cross-validation on the cleaning subnet in the heterogeneous data preprocessing network, and generates data integrity repair instructions based on the verification results; when the data flow of a certain layer of the cleaning subnet is detected to be interrupted, the abnormal data recovery unit calls the pre-stored reference data copy in the distributed cache unit, reconstructs the feature vector of the missing signal segment and injects it into the dynamic feature fusion topology.

[0032] Preferably, the distributed cache unit is configured with a shard storage strategy and a consistency protocol link; the steps of constructing the shard storage strategy include:

[0033] Based on the signal sampling rates of multiple source devices and the data source weight distribution results, the real-time collected heterogeneous data stream is divided into M data blocks according to the time window;

[0034] M data blocks are backed up in distributed nodes through the consistency protocol link, and the cleaning subnet identifier corresponding to each data block is marked; when the abnormal data recovery instruction is triggered, the matching data block is extracted from the backup node according to the identifier for signal reconstruction.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] During data acquisition and preprocessing, multi-source signal synchronization nodes can acquire heterogeneous data streams from wearable devices, environmental sensors, and medical databases in real time, and construct a multi-layer data fusion topology through a heterogeneous data preprocessing network. This preprocessing network is composed of multiple cleaning subnets connected in series based on data quality assessment. It can dynamically adjust the cleaning strategy based on the physiological signal sampling rate and data integrity indicators, effectively solving problems such as inconsistent multi-source data sampling rates and large timestamp deviations. For example, signal synchronization compensation parameters are generated through a sliding window filter and stored in the corresponding cleaning subnet. Combined with a sliding correlation coefficient algorithm, dynamic correlation of multi-source data features is achieved, ensuring the accuracy and flexibility of data preprocessing and providing a high-quality data foundation for subsequent fusion analysis.

[0037] The fusion analysis engine achieves deep correlation of multi-dimensional data features and accurate detection of abnormal states through the collaborative work of the feature association unit and the pattern recognition unit. The multi-dimensional matching links configured by the feature association unit (including the steady-state analysis sub-link and the anomaly correction sub-link) can dynamically switch association rules according to the data state. In the steady-state case, the steady-state analysis sub-link is generated based on the matching degree of the fusion analysis engine's association rules and the health status baseline strategy to ensure the stable correlation of data features under normal conditions; when transient signal loss occurs or abnormal monitoring events occur, the anomaly correction sub-link is used to construct a correction link based on the association relationship between the data acquisition unit and the abnormal event, significantly improving the timeliness and accuracy of abnormal data processing. The anomaly detection signal output by the pattern recognition unit is combined with the dynamic decision node and priority mapping unit to realize dynamic data source weight allocation for monitoring tasks, enabling the system to flexibly adjust resource allocation based on real-time data quality and monitoring needs, improving the efficiency and reliability of fusion analysis.

[0038] The intelligent decision management module realizes the dynamic generation and optimization of the health status benchmark strategy through the linkage of dynamic decision nodes and priority mapping units. The collaborative decision model integrates the signal-feature mapping table of the physiological data acquisition unit and the anomaly detection parameter library of the fusion analysis engine. It can calculate the theoretical feature extraction capacity of each layer of the cleaning subnet based on the real-time signal quality index and the data source weight distribution results, and generate a health status reference baseline. The reference baseline is corrected for the conflict rate through the anomaly correction sub-link to form an accurate health status benchmark strategy. At the same time, the system has designed multi-level parameter update instructions (such as the first to fourth optimization instructions), which can automatically trigger corresponding optimization strategies for different scenarios such as abnormal data source sampling rate, excessively high data conflict rate, and low cleaning subnet feature retention rate, realizing adaptive update of the decision model and iterative optimization of the monitoring strategy, significantly improving the decision-making ability and robustness of the system in complex scenarios.

[0039] The real-time calibration module and data weight optimization module further enhance the system's data processing accuracy and resource management efficiency. The signal compensation unit uses a time delay correction circuit to achieve time domain alignment calibration of the output signal of the physiological data acquisition unit. The sampling rate adaptation unit dynamically adjusts the data parsing frequency of the fusion analysis engine to ensure the consistency of multi-source data in time and frequency dimensions. The data credibility assessment network analyzes the conflict rate of multi-source data in real time and feeds the results back to the decision module to form a closed-loop optimization mechanism. The degradation processing logic of the data weight optimization module constructs a data source priority level table based on the conflict rate index and signal synchronization stability parameters. When the system is overloaded or the data credibility is low, the data source weight and parsing frequency are dynamically adjusted, realizing the rational allocation of resources and flexible adjustment of monitoring strategies.

[0040] The design of the fault-tolerant processing module and distributed cache unit significantly improves the reliability and scalability of the system. The redundant check unit performs cross-validation on the cleaning subnet and generates data integrity repair instructions to ensure the accuracy of data processing. When the abnormal data recovery unit detects a data flow interruption, it calls the pre-stored baseline data copy in the distributed cache unit to reconstruct the feature vector of the missing signal segment, ensuring the continuity of the system in the event of equipment failure or data interruption. The sharding storage strategy and consistency protocol link of the distributed cache unit achieve efficient storage and rapid recovery of heterogeneous data streams, improving the system's processing capabilities and scalability for high-concurrency data, and can adapt to the future needs of multi-source device access and data volume growth. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a working principle diagram of the health information monitoring and management system based on multi-source data fusion analysis according to the present invention;

[0042] Figure 2 Flowchart generated for the health state baseline strategy;

[0043] Figure 3 Logic diagram for triggering the first and second optimization instructions for updating the parameters of the collaborative decision-making model;

[0044] Figure 4 Flowchart of the degradation processing logic of the data weight optimization module. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] See also Figures 1-4The present invention relates to a health information monitoring and management system based on multi-source data fusion analysis. Its core architecture includes a physiological data acquisition unit, a fusion analysis engine, and an intelligent decision management module. The specific implementation steps are as follows:

[0047] The physiological data acquisition unit includes a multi-source signal synchronization node and a heterogeneous data preprocessing network. The multi-source signal synchronization node acquires heterogeneous data streams from wearable devices, environmental sensors, and medical databases in real time. For example, it can collect multiple types of data, such as heart rate, body temperature, and ambient temperature and humidity. Based on physiological signal sampling rates and data integrity indicators, this node constructs a multi-layer data fusion topology through the heterogeneous data preprocessing network. The heterogeneous data preprocessing network is composed of N serially connected cleaning subnets based on data quality assessment, each of which performs data stream cleaning and feature extraction functions.

[0048] The fusion analysis engine includes a feature association unit and a pattern recognition unit. The feature association unit is responsible for performing association analysis on the features of the multi-source data processed by the cleaning subnet, while the pattern recognition unit performs anomaly detection on the associated feature data and outputs an anomaly detection signal.

[0049] The intelligent decision management module includes a dynamic decision node and a priority mapping unit. The dynamic decision node generates a health status baseline policy based on the real-time feature extraction capabilities of N cleaning subnets. The priority mapping unit combines the anomaly detection signals output by the pattern recognition unit to dynamically assign data source weights to monitoring tasks. For example, the weight of each data source is adjusted according to different anomaly levels to optimize monitoring resource allocation.

[0050] Example 1:

[0051] The real-time calibration module and the data weight optimization module work together during system operation to ensure the accuracy and consistency of multi-source data and the optimization of monitoring strategies. The real-time calibration module includes a signal compensation unit and a sampling rate adaptation unit, which calibrate data from the perspectives of time domain alignment and analytical frequency adaptation, respectively. The data weight optimization module, by constructing a data credibility assessment network, enables real-time analysis of multi-source data conflict rates and iterative optimization of monitoring strategies.

[0052] The core function of the real-time calibration module is to resolve the inconsistency issues of multi-source data in terms of time dimension and resolution frequency. The signal compensation unit, as a key component of the real-time calibration module, is composed of a time delay correction circuit corresponding to the cleaning subnet connected in series in the heterogeneous data preprocessing network. When the system collects heterogeneous data streams from multiple source devices, the output signals of different data sources (such as wearable devices, environmental sensors, and medical databases) often have time delay deviations due to differences in hardware characteristics, transmission paths, and processing procedures. For example, wearable devices may generate a delay of several milliseconds when transmitting data via Bluetooth, while interface calls to medical databases may generate varying degrees of delay due to network load. If these timestamp deviations are not calibrated, they will cause time misalignment in the fusion analysis of multi-source data, affecting the accuracy of feature association and anomaly detection.

[0053] The signal compensation unit operates as follows: After acquiring the raw data stream, the multi-source signal synchronization node first extracts the signal waveform and corresponding timestamp information from each data source. These signals, bearing timestamp deviations, are then fed into the cleaning subnets of each layer of the heterogeneous data preprocessing network. The delay correction circuit configured in each cleaning subnet generates corresponding synchronization compensation parameters based on the time delay characteristics of the input signal by calculating the time difference between adjacent signal samples. For example, for a wearable device's heart rate signal and an environmental sensor's temperature and humidity signals, the delay correction circuit can calculate the timestamp difference between the two and generate compensation parameters based on the magnitude and direction of the difference. These parameters are used to adjust the time base of subsequent signals, aligning signals from different data sources on the time axis. The generated synchronization compensation parameters are stored in the corresponding cleaning subnet for subsequent data processing, enabling real-time time domain calibration of the continuously input data stream.

[0054] The sampling rate adaptation unit addresses the mismatch between the fusion analysis engine and the sampling rates of multi-source data. The sampling rates of different data sources can vary significantly. For example, a wearable device may collect heart rate data at a higher sampling rate (e.g., 100Hz) to monitor physiological indicators in real time, while an environmental sensor may collect temperature and humidity data at a lower sampling rate (e.g., 1Hz) to reduce power consumption. If the fusion analysis engine analyzes data at a fixed rate, this may result in loss of detail in high-frequency data or redundant processing of low-frequency data. The sampling rate adaptation unit dynamically adjusts the fusion analysis engine's data analysis frequency by monitoring changes in the sampling rates of the data sources in real time. Specifically, when it detects an increase in the sampling rate of a data source, the sampling rate adaptation unit sends a command to the fusion analysis engine to increase the analysis frequency to match the data source's sampling rate, ensuring that every sample of the high-frequency data is effectively analyzed. When the sampling rate decreases, the analysis frequency is reduced accordingly, reducing computing resource consumption. This dynamic adjustment mechanism avoids data processing distortion caused by a fixed analysis frequency and improves the accuracy and efficiency of fusion analysis.

[0055] The data weight optimization module implements real-time analysis of multi-source data conflict rates by constructing a data credibility assessment network. The core of the data credibility assessment network is to establish a conflict assessment model for multi-source data. This model cross-validates data from different data sources based on dimensions such as temporal consistency, numerical rationality, and feature relevance. For example, when temperature data collected by a wearable device differs significantly from historical temperature data in a medical database at the same time point, the assessment network calculates the difference between the two as a conflict rate indicator. Combined with the data's timestamp deviation and feature retention rate, the assessment network determines whether the conflict is due to a data source failure, transmission error, or true physiological changes. The assessment network feeds the real-time calculated conflict rate indicator and credibility assessment results back to the dynamic decision node and priority mapping unit of the intelligent decision management module.

[0056] After receiving the feedback information, the dynamic decision node iteratively optimizes the health status benchmark strategy. The health status benchmark strategy is the core basis for the system to judge the user's health status. It generates feature extraction results and real-time monitoring requirements based on multi-source data. When the data credibility assessment network reports that the conflict rate of a certain data source has increased, the dynamic decision node will re-evaluate the importance of the data source in the current monitoring task and adjust its weight in the health status benchmark strategy. For example, if the conflict rate of temperature and humidity data of an environmental sensor continues to increase, it may indicate that the sensor has failed or there is interference in the environment. The dynamic decision node will reduce its weight in the environmental health indicator assessment and increase the weight of other reliable data sources to ensure the accuracy of the benchmark strategy.

[0057] The priority mapping unit dynamically allocates the weights of the data sources for the monitoring tasks based on the feedback from the data credibility assessment network. The priority mapping unit has a built-in priority algorithm that generates a corresponding weight coefficient for each data source based on the credibility level of the data source, the stability of the sampling rate, and the relevance to the monitoring task. When the data conflict rate increases, the priority mapping unit will automatically reduce the weight coefficient of the conflicting data source and reallocate the released monitoring resources to the high-credibility data source. For example, in a sleep monitoring task, if the conflict rate of the heart rate data of the wearable device increases, the priority mapping unit will reduce its weight and increase the weight of the body motion sensor data, compensating for the unreliability of a single data source through multi-dimensional data fusion.

[0058] There is a close synergistic relationship between the real-time calibration module and the data weight optimization module. The data calibration processing by the signal compensation unit and the sampling rate adaptation unit can reduce the time deviation and parsing error of multi-source data, thereby reducing the occurrence of data conflict rate and providing a more reliable evaluation basis for the data weight optimization module; and the data weight optimization module can guide the system to give priority to the use of data with good calibration effect and high credibility by adjusting the weight of the data source, further improving the processing efficiency of the real-time calibration module. For example, when a data source is calibrated by the signal compensation unit, its timestamp deviation is significantly reduced, the data credibility assessment network will increase the credibility level of the data source, and the priority mapping unit will increase its weight accordingly, making the system more dependent on the data of the data source for health status assessment, forming a virtuous circle.

[0059] During the system's actual operation, the real-time calibration module and data weight optimization module must continuously monitor the dynamic changes of multi-source data. The signal compensation unit's delay correction circuit must continuously update the synchronization compensation parameters based on the real-time input signal characteristics to adapt to changes in the data source hardware status or transmission environment. The sampling rate adaptation unit must monitor the sampling rate fluctuations of each data source in real time and adjust the analysis frequency of the fusion analysis engine within nanosecond timescales. The data credibility assessment network must calculate the conflict rate indicator with a millisecond response speed and feed the results back to the intelligent decision management module to ensure that the monitoring strategy is adjusted in sync with data changes.

[0060] Furthermore, the design of the real-time calibration module and data weight optimization module must consider the system's scalability and compatibility. For newly connected data sources, the signal compensation unit automatically identifies their time delay characteristics and generates corresponding synchronization compensation parameters. The sampling rate adaptation unit dynamically adapts to their sampling rate range. The data credibility assessment network trains new conflict assessment models based on historical data. This allows the system to support a wider range of data sources without requiring large-scale modifications, meeting the health information monitoring needs of different application scenarios.

[0061] Example 2:

[0062] The construction of a multi-layer data fusion topology and the activation conditions for cleaning subnets are key steps in achieving efficient multi-source data processing in this system. When building this multi-layer data fusion topology, the system acquires, filters, and correlates raw signals from multiple devices to form a dynamic fusion architecture. The activation conditions for cleaning subnets ensure high reliability of the subnets involved in data processing, taking into account factors such as sampling rate, feature retention rate, and timing deviation. The multi-dimensional matching links of the feature correlation unit further enhance the flexibility and accuracy of data correlation.

[0063] The construction of a multi-layer data fusion topology begins with the acquisition and preprocessing of raw signals from multiple source devices. The system uses multi-source signal synchronization nodes to acquire heterogeneous data streams from various sources, such as wearable devices, environmental sensors, and medical databases. These data streams contain multiple types of signals, including heart rate, body temperature, motion acceleration, and ambient temperature and humidity. During the acquisition process, the synchronization nodes simultaneously record the original waveforms and data timestamp deviations of each signal. For example, the difference between the timestamp of the heart rate signal collected by the wearable device and the timestamp of the historical data in the medical database, or the transmission delay of the temperature and humidity signals of the environmental sensors due to different deployment locations. These raw signals and timestamp information serve as basic input and are transmitted to the cleaning subnets of each layer of the heterogeneous data preprocessing network.

[0064] Each cleaning subnet layer is configured with a sliding window filter, which performs time-frequency domain analysis on the input raw signal, extracts the signal's characteristic parameters, and generates synchronization compensation parameters. The sliding window filter segments the signal into pre-set time windows (e.g., 1 second), calculating statistical features such as the signal's mean, variance, and frequency component within each window. It also compares the timestamp deviations of different data sources within the same time window. For example, for the heart rate signal from a wearable device and the temperature and humidity signals from an environmental sensor, the filter calculates the timestamp difference between the two within the same window and generates signal synchronization compensation parameters based on the distribution characteristics of the difference. These parameters are used to adjust the time base of subsequent signals and eliminate latency differences between different data sources. The generated synchronization compensation parameters are stored in the corresponding cleaning subnet for subsequent data processing, forming a layer-by-layer time alignment mechanism.

[0065] After generating and storing the synchronous compensation parameters, the system uses a sliding correlation coefficient algorithm to associate multi-source data features with N cleaning subnets based on the parameters of each cleaning subnet and the temporal continuity index of the data stream. The temporal continuity index measures the temporal stability of the data stream. For example, it measures whether the waveform of physiological signals conforms to normal physiological patterns (such as the tendency of heart rate to increase during exercise and decrease during rest) or whether environmental data conforms to physical laws (such as the gradual changes in temperature and humidity). The sliding correlation coefficient algorithm calculates the correlation coefficient between feature vectors from different data sources to determine whether they belong to the same feature cluster. For example, if the feature vectors of heart rate variability and acceleration show a high positive correlation across multiple consecutive time windows (such as both increasing simultaneously during exercise), the algorithm associates them with the same cleaning subnet for fusion processing. If the correlation is low (such as body temperature and ambient light intensity), they are assigned to different subnets. This dynamic association method creates a dynamic feature fusion topology that adapts to changing data features, allowing highly correlated features to be integrated within the same subnet, improving the efficiency and accuracy of feature extraction.

[0066] The activation conditions for the cleaning subnet are a key control mechanism for ensuring data processing quality. The multi-source signal synchronization node monitors the activation status of each cleaning subnet in real time. Only subnets that meet the following conditions will be triggered to participate in data processing: First, the data source sampling rate must be within a preset compatibility range. The preset compatibility range is pre-set based on the data source types and processing capabilities supported by the system. For example, for physiological data acquisition, the sampling rate compatibility range is set to 5Hz-200Hz. Below 5Hz may result in loss of signal details, while above 200Hz will increase computing resource consumption. When the sampling rate of a data source falls within this range, it indicates that its signal frequency meets the system processing requirements, and the corresponding cleaning subnet has the basis for activation.

[0067] The feature retention rate corresponding to the cleaned subnet must be greater than the threshold. The feature retention rate refers to the proportion of original data features retained after processing by the cleaned subnet, and the calculation formula is (number of features after cleaning / number of original features) × 100%. The threshold is set according to the monitoring accuracy requirements of different application scenarios. For example, in high-precision medical monitoring scenarios, the feature retention rate threshold can be set to 90% to ensure that only noise and redundant data are removed during the cleaning process, and the effective features are retained to the maximum extent. If the feature retention rate of a subnet is lower than the threshold, it means that its cleaning rules are too strict or there are improper parameter settings. At this time, the subnet will not be activated and the cleaning parameters need to be readjusted.

[0068] Data timestamp deviation must be within the allowed synchronization range. The allowed synchronization range is the maximum timestamp difference the system can tolerate, for example, 50 milliseconds. When timestamp deviations from multiple sources exceed this range, inconsistencies in the temporal dimension of the data will distort the fusion analysis results. The cleaning subnet is activated only when the data source timestamp deviations are within the allowed synchronization range, ensuring temporal comparability of the data involved in the fusion.

[0069] The multi-dimensional matching link configured by the feature association unit further enhances the system's adaptability to different monitoring states. The multi-dimensional matching link includes a steady-state analysis sub-link and an anomaly correction sub-link, each of which has different data association logic designed for normal monitoring scenarios and abnormal events, respectively.

[0070] The steady-state analysis sub-link is generated based on the matching degree between the association rules of the fusion analysis engine and the health status benchmark strategy. Association rules are logical relationships between multi-source data features pre-defined by the system. For example, heart rate and blood oxygen saturation are positively correlated under normal physiological conditions, and there is a linear mapping relationship between ambient temperature and body surface temperature. The health status benchmark strategy is a reference model set by the system based on the range of normal physiological indicators and environmental parameters. During the steady-state monitoring process, the feature association unit determines the rationality of the characteristics of each data source by matching the association rules with the benchmark strategy, and generates the corresponding steady-state analysis sub-link. For example, when the heart rate and blood oxygen saturation data collected by the wearable device are both within the normal range and the correlation between the two meets the preset rules, the steady-state analysis sub-link will guide the data to be fused through a conventional path to ensure the efficiency of the processing flow.

[0071] The anomaly correction sub-link is constructed based on the correlation between the instantaneous signal loss of the data acquisition unit and the monitoring abnormal events. Instantaneous signal loss may be caused by reasons such as device detachment and network interruption. Monitoring abnormal events include physiological indicators suddenly exceeding the normal range (such as a sudden increase in heart rate) or sudden changes in environmental parameters (such as rapid changes in temperature and humidity). When the system detects instantaneous signal loss or an abnormal event, the feature association unit automatically triggers the anomaly correction sub-link, which re-establishes the data association path by analyzing the duration of the signal loss and the characteristic pattern of the abnormal event. For example, if the wearable device causes a signal interruption of 5 seconds due to movement, the anomaly correction sub-link can call the historical data in the medical database, supplement the lost signal segment through the interpolation algorithm, and re-associate the heart rate and motion acceleration features to avoid monitoring misjudgments due to signal interruption.

[0072] During the dynamic adjustment of multi-layer data fusion topologies, the system must monitor the operating status of each cleaning subnet and the changing trends of data features in real time. The sliding window filter must dynamically update its synchronization compensation parameters based on the latest acquired signals to accommodate delay drift caused by factors such as device hardware aging and environmental interference. The sliding correlation coefficient algorithm must regularly recalculate the correlation of feature vectors to ensure that the dynamic feature fusion topology always reflects the true correlation of the current data. The activation status of the cleaning subnet must be switched in real time based on changes in the data source sampling rate, feature retention rate, and timing deviation to prevent invalid subnets from participating in data processing and wasting computing resources.

[0073] Furthermore, the design of multi-layer data fusion topologies must consider system scalability. For newly added data sources, the system can accommodate new signal characteristics by increasing the number of cleaning subnets or adjusting the filtering parameters of existing subnets. Multi-dimensional matching links can be expanded based on new association rules and abnormal event types, enabling the system to adapt to diverse health monitoring scenarios, such as sports health monitoring, chronic disease management, and environmental health early warning.

[0074] Example 3:

[0075] The collaborative decision-making model, which dynamically configures decision nodes, is the core component of the health monitoring system's intelligent decision-making. It generates and optimizes health status baseline policies by integrating the signal characteristics of physiological data acquisition units with the anomaly detection capabilities of the fusion analysis engine. This model, based on signal-feature mapping tables and an anomaly detection parameter library, enables dynamic assessment and prediction of health status by comparing real-time data with historical patterns.

[0076] The signal-feature mapping table is the foundational data structure of the collaborative decision-making model, recording the mapping relationship between physiological signals and characteristic parameters. During its construction, the system first categorizes the raw physiological signals collected by multiple devices, for example, categorizing heart rate signals, body temperature signals, and blood oxygen saturation signals. For each signal type, characteristic parameters are extracted through time-frequency domain analysis, such as the mean, standard deviation, and variability index for heart rate signals, and the diurnal fluctuation amplitude and trend rate of temperature signals. These characteristic parameters are organized into a mapping table, with each row corresponding to a specific physiological signal and each column corresponding to a characteristic parameter. Each element in the table represents the strength of the mapping relationship between the signal and the characteristic parameter. For example, the mapping strength between heart rate variability index and the functional status of the autonomic nervous system is strong, while the mapping strength with ambient temperature is weak. The mapping table is constructed based not only on prior medical knowledge but also through training and optimization using a machine learning algorithm on a large amount of historical data to ensure the accuracy and universality of the mapping relationship.

[0077] The anomaly detection parameter library stores various parameter thresholds and discrimination rules used to detect abnormal health conditions. Parameter thresholds are set based on medical clinical standards and statistical analysis results. For example, the normal range for heart rate is set at 60-100 beats per minute, and the normal range for body temperature is set at 36.0-37.5°C. The discrimination rules define how to determine abnormal conditions based on a combination of multiple characteristic parameters. For example, when the heart rate exceeds the threshold and the blood oxygen saturation is below the threshold, it is considered abnormal cardiopulmonary function. The anomaly detection parameter library adopts a hierarchical design, divided into a basic threshold layer, a personalized adjustment layer, and a dynamic adaptation layer. The basic threshold layer stores universal medical standard thresholds; the personalized adjustment layer modifies the basic thresholds based on individual differences such as the user's age, gender, and physical condition; and the dynamic adaptation layer adjusts the thresholds in real time based on the user's historical health data and current monitoring status, for example, appropriately raising the heart rate threshold during exercise.

[0078] The process of generating a health status baseline strategy begins with an assessment of the real-time signal quality metrics of the physiological data acquisition unit. Signal quality metrics include parameters such as signal strength, signal-to-noise ratio, and stability, which reflect the reliability of the data source and the accuracy of the data. For example, if the heart rate signal of a wearable device is disturbed by movement, its signal-to-noise ratio will decrease and its stability will deteriorate, resulting in a low signal quality metric. The system weights data sources based on the signal quality metric, assigning higher weights to sources with high signal quality and lower weights to sources with low signal quality. Simultaneously, the theoretical feature extraction capacity of each cleaning subnet in the steady-state analysis sublink is calculated based on the data source weight assignment results from the priority mapping unit. The theoretical feature extraction capacity represents the upper limit on the number and quality of features that can be effectively extracted by each subnet under the current data source weight assignment and signal quality conditions. This calculation takes into account factors such as the subnet's processing power, the complexity of the feature extraction algorithm, and data redundancy.

[0079] The calculated theoretical feature extraction capacity and real-time monitoring requirements are input into the collaborative decision-making model, which then analyzes the matching relationship between the two to generate a health status reference baseline. Real-time monitoring requirements are determined by the user's health goals and current monitoring tasks. For example, for patients with hypertension, monitoring requirements may focus on blood pressure fluctuations and cardiovascular function indicators; for sports enthusiasts, monitoring requirements may be more focused on exercise intensity and recovery status. Based on the monitoring requirements, the collaborative decision-making model selects relevant feature parameters from the signal-feature mapping table and, combined with the thresholds in the anomaly detection parameter library, constructs a health status reference baseline. The reference baseline represents the range of values ​​and change patterns that these feature parameters should exhibit under normal health conditions.

[0080] The generated reference baseline is corrected for its conflict rate through the anomaly correction sub-link. The conflict rate refers to the degree of inconsistency between the reference baseline and the actual collected data, and is calculated as the ratio of the number of data samples that do not conform to the reference baseline pattern to the total number of samples. When the conflict rate exceeds the preset threshold, it indicates that the reference baseline may be biased or the current health status has changed. The anomaly correction sub-link adjusts the reference baseline by analyzing the causes of the conflict, such as data source anomalies, environmental interference, or changes in the actual health status. For example, if the conflict rate of a data source is found to be continuously increasing, it may be that the data source is faulty. In this case, its weight in the reference baseline generation is reduced; if the conflict patterns of multiple data sources show consistent changes, it may reflect a change in the actual health status. In this case, the reference baseline is adaptively adjusted according to the rules in the anomaly detection parameter library, and ultimately a health status baseline strategy is formed.

[0081] The parameter update mechanism of the collaborative decision-making model ensures that the system can adapt to dynamically changing monitoring environments and health status. When it is detected that the data source sampling rate exceeds the preset compatibility range, it indicates that the working status of the data source is abnormal, which may affect the accuracy and real-time performance of the data. At this time, the system triggers the first optimization instruction and starts the backup data source channel. The backup data source channel is a pre-configured data source with similar monitoring functions. For example, when the heart rate monitoring sampling rate of the main wearable device is abnormal, it automatically switches to the backup heart rate sensor built into the smart watch. At the same time, the correlation dimension of the fusion analysis engine is adjusted to increase the analysis depth of the characteristic parameters related to the abnormal data source, such as strengthening the analysis of indirect indicators such as heart rate variability to make up for the information loss caused by the abnormality of the main data source.

[0082] If the data conflict rate still does not return to the allowable range after the execution of the first optimization instruction, it means that the problem may be more complicated, and the second optimization instruction is triggered at this time. The second optimization instruction switches to the abnormality correction sub-link through the multi-dimensional matching link, and reallocates the feature extraction weights of the data source based on the duration of the monitored abnormal event. The duration of the monitored abnormal event is an important indicator for judging the severity and stability of the abnormality. For example, a short-term heart rate abnormality may be caused by temporary exercise, while a long-term heart rate abnormality is more likely to reflect a health problem. The system dynamically adjusts the data source weight according to the duration of the abnormality. For abnormal events with a longer duration, the feature extraction weight of the data source related to the abnormality is increased. For example, when the heart rate is abnormal for a long time, the weight of the electrocardiogram monitoring device and the blood pressure sensor is increased to obtain more detailed heart function information. At the same time, the weight of the data source that is less related to the current abnormality is reduced to reduce the interference of irrelevant information and improve monitoring efficiency.

[0083] During the parameter update process, the collaborative decision-making model uses an incremental learning algorithm to dynamically optimize the signal-feature mapping table and anomaly detection parameter library. Incremental learning allows the model to continuously update its knowledge based on new data samples without retraining the entire dataset. For example, when the system detects a new abnormal pattern, it records the pattern and its corresponding characteristic parameter changes in the anomaly detection parameter library and updates the relevant discrimination rules. When a new mapping relationship between physiological signals and characteristic parameters is discovered, it is added to the signal-feature mapping table and the strength of the existing mapping relationship is adjusted. This dynamic optimization mechanism enables the collaborative decision-making model to continuously adapt to new health monitoring scenarios and individual differences, improving the accuracy and personalization of decisions.

[0084] The collaborative decision-making model also features a self-assessment mechanism to regularly review and correct its own decision accuracy. This self-assessment compares the model-generated health status baseline strategy with the diagnostic results of medical experts or gold-standard monitoring data to calculate the decision compliance rate. If the compliance rate falls below a preset threshold, the system initiates a model tuning process to analyze the causes of decision deviations, such as inappropriate feature selection and unreasonable threshold settings, and makes targeted adjustments to the signal-feature mapping table and anomaly detection parameter library. This closed-loop feedback mechanism ensures that the collaborative decision-making model can continuously learn and improve, continuously enhancing the effectiveness of health monitoring and management.

[0085] Example 4:

[0086] The collaborative decision-making model's parameter update mechanism is a core component of the system's ability to address data anomalies and ensure monitoring reliability. It dynamically adjusts data source status, subnet processing capabilities, and weight allocation strategies through multi-level optimization instructions. The third and fourth optimization instructions in this embodiment address abnormal feature retention rates and overload conditions in the cleaning subnet, respectively, by designing strategies such as shielding transmission channels, diverting signal flows, and performing downgrade processing, ensuring the system maintains stable operation under complex operating conditions.

[0087] When the feature retention rate of a cleaning subnet in a physiological data acquisition unit falls below a threshold, it indicates that the subnet's cleaning algorithm may be over-filtering valid data, or that hardware components are experiencing performance degradation. The feature retention rate is calculated by comparing data features before and after cleaning. For example, after noise reduction processing of a heart rate signal, if the loss of time-domain features (such as the standard deviation of the RR interval) or frequency-domain features (such as the proportion of low-frequency power) exceeds a preset threshold (e.g., 20%), the system determines that the subnet is processing abnormally. This triggers the third optimization instruction: first, the data transmission channel of the subnet is blocked to prevent it from further outputting potentially distorted data. Simultaneously, the signal compensation unit's delay correction circuit redirects the corresponding signal flow to another, functioning cleaning subnet. The signal flow transfer process adheres to load balancing principles, with the system prioritizing subnets that match the processing capabilities of the atomic network and are currently under less load. For example, heart rate signals processed by a subnet with a low feature retention rate may be transferred to another subnet on the same layer equipped with a heart rate signal optimization algorithm to ensure continuous and accurate data processing.

[0088] During the execution of the third optimization instruction, the system continuously monitors the load status of other cleaning subnets. The load status is comprehensively evaluated through indicators such as CPU occupancy, memory usage, and data processing delay. When it is detected that the load indicator of a subnet exceeds the overload threshold (such as the CPU occupancy rate exceeds 85% for 5 consecutive minutes), it indicates that it can no longer effectively undertake additional signal flows, and the fourth optimization instruction is triggered at this time. The fourth optimization instruction calls the data weight optimization module to downgrade the output strategy of the priority mapping unit. This process is based on the real-time analysis results of the data credibility assessment network and specifically includes the following steps:

[0089] The Data Credibility Assessment Network constructs a data source priority table based on the conflict rate and signal synchronization stability parameters of multi-source data. The conflict rate reflects the degree of consistency between different data sources, for example, the percentage of the difference in temperature data between a wearable device and a medical database within the normal fluctuation range. The signal synchronization stability parameter is measured by calculating the standard deviation of the signal timestamp deviation, with a smaller standard deviation indicating higher synchronization. The priority table uses a multi-level classification structure, categorizing data sources into three levels: "high confidence," "medium confidence," and "low confidence." High confidence data sources must meet the requirements of a conflict rate below 5% and a synchronization stability parameter of less than 10ms. These are typically devices that have undergone multiple calibrations and have a stable history (such as medical-grade heart rate monitors). Medium confidence data sources allow for a conflict rate between 5% and 15% and a synchronization stability parameter between 10 and 30ms. These are typically consumer-grade wearable devices. Low confidence data sources are devices with a conflict rate exceeding 15% or a synchronization stability parameter exceeding 30ms. These devices may have unreliable data due to hardware aging or environmental interference.

[0090] After the fourth optimization instruction is triggered, the system dynamically downgrades the weight of the data source according to the priority level table. The downgrade rule follows the principle of "protecting the main and giving up the secondary": the weight of the high-trust data source remains unchanged to ensure the monitoring accuracy of key health indicators (such as heart rate and blood oxygen); the weight of the medium-trust data source is reduced proportionally (such as from 30% to 20%) to reduce the processing pressure of non-critical data; the weight of the low-trust data source is reduced to the lowest priority (such as below 5%), and is only used as a backup when other data sources are unavailable. For example, in the sleep monitoring scenario, when the medium-trust wearable device subnet carrying body motion data is overloaded, the system reduces its weight and increases the data share of the high-trust sleep monitoring mattress, while limiting the monitoring permissions of the low-trust ambient light sensor to avoid frequent requests for computing resources.

[0091] Synchronously adjusting the parsing frequency of the fusion analysis engine to adapt to the downgraded monitoring strategy is a key step in the fourth optimization instruction. The adjustment of the parsing frequency is positively correlated with the weight of the data source: high-trust data sources carry critical data, so their parsing frequency is maintained at the highest level (such as 100Hz) to ensure real-time capture of subtle changes; the parsing frequency of medium-trust data sources is adjusted down accordingly as the weight decreases (such as from 50Hz to 30Hz), reducing the amount of calculation while ensuring basic monitoring needs; the parsing frequency of low-trust data sources is reduced to the lowest level (such as 10Hz), and only periodic status checks are performed. This hierarchical adjustment mechanism can effectively balance monitoring accuracy and system resource consumption. For example, in emergency monitoring scenarios, high-frequency parsing of electrocardiogram monitors (high-trust) is prioritized, while low-frequency parsing is used for ward environment monitoring equipment (medium-trust), freeing up computing resources for critical data processing.

[0092] The degradation processing logic of the data weight optimization module also needs to consider the real-time requirements of the monitoring task. For tasks with high real-time requirements (such as abnormal heart rate warning), the system maintains the independent operation of the processing links of high-trust data sources during the degradation process to avoid delays caused by shared resources; for tasks with lower real-time requirements (such as daily average activity statistics), it allows the merging of data from trusted data sources in processing to reduce the number of calculations. In addition, the degradation process is reversible. When the system detects that the subnet load has returned to the normal range (such as the CPU occupancy rate is less than 70% for 5 consecutive minutes), it automatically triggers the weight recovery mechanism, and gradually restores the original weight of each data source at a preset rate (such as an increase of 5% every 10 minutes), so that the monitoring strategy returns to the optimal state.

[0093] During the implementation of the third and fourth optimization instructions, the system records the timestamps, trigger conditions, and parameter changes of each operation through a distributed logging system, forming a complete operational audit trail. The audit trail can be used to trace the cause of anomalies. For example, by analyzing the logs of a subnet with a sudden drop in feature retention rate, problems such as loose hardware interfaces or algorithm parameter drift can be located, providing a basis for system maintenance. At the same time, operational audit data is input into the incremental learning module of the collaborative decision-making model to train the anomaly prediction model, identifying potential risks that may lead to subnet failure or overload in advance (such as abnormally high hardware temperatures and increased algorithm processing time), achieving a transition from passive response to active prevention.

[0094] In addition, the system features a manual intervention interface, allowing administrators to manually adjust the priority ranking table and downgrade policies in special circumstances. For example, if a trusted data source's credibility increases due to temporary calibration, administrators can use this interface to temporarily mark it as highly trusted, ensuring that its data retains a high priority during downgrade processing. This combination of manual intervention and automated mechanisms enables the system to handle both routine exception scenarios and adapt flexibly to special needs, enhancing overall robustness.

[0095] Example 5:

[0096] The fault-tolerant processing module works in conjunction with the distributed cache unit to form a complete data reliability assurance system. Through the redundancy check unit and abnormal data recovery unit, the fault-tolerant processing module verifies the integrity of the cleaning subnet data in real time and quickly recovers data flow interruptions. The distributed cache unit ensures efficient and consistent data backup through sharding storage strategies and consistency protocol links, providing fundamental support for fault-tolerant processing.

[0097] The redundant check unit of the fault-tolerant processing module performs cross-validation on the cleaning subnet in the heterogeneous data preprocessing network based on the conflict rate index and priority level table output by the data credibility assessment network. The core idea of ​​cross-validation is to use different subnets to independently process data from the same data source and compare the consistency of the processing results. The specific process is: for each data source, the data stream is simultaneously input into at least two different cleaning subnets in the heterogeneous data preprocessing network (called the master check subnet and the slave check subnet). The two subnets use different cleaning algorithms or parameter configurations (such as different filter window sizes, noise reduction thresholds) to process the data. The redundant check unit periodically (such as every 100 milliseconds) extracts the output results of the master and slave subnets and calculates the difference in characteristic parameters between the two. The formula for calculating the difference value is:

[0098]

[0099] Where D represents the difference value of the characteristic parameters, n is the number of characteristic parameters, and F 主i The i-th characteristic parameter value output by the main verification subnet, F 从i is the i-th characteristic parameter value output from the verification subnet. When the difference value D exceeds the preset integrity threshold, it indicates that the processing result of at least one subnet is abnormal. The redundancy check unit generates a data integrity repair instruction, triggering reprocessing or parameter adjustment of the abnormal subnet.

[0100] The abnormal data recovery unit is activated when it detects a data flow interruption in a certain layer of the cleaning subnet. Its core function is to call the pre-stored reference data copy in the distributed cache unit to reconstruct the feature vector of the missing signal segment. The detection of data flow interruption is achieved by monitoring the continuity of the output signal of the cleaning subnet. For example, when the subnet does not output valid data within a preset time window (such as 5 seconds), it is determined to be a data flow interruption. At this time, the abnormal data recovery unit retrieves the corresponding reference data copy from the distributed cache unit based on the data source type and signal characteristics processed by the subnet. The reference data copy is a complete data snapshot stored at a fixed period (such as every minute) during normal operation of the system, containing information such as signal waveform, timestamp and characteristic parameters.

[0101] The distributed cache unit's sharding storage strategy is based on the signal sampling rates of multiple source devices and the weight distribution of data sources. It divides the real-time heterogeneous data stream into M data blocks according to time windows. The length of the time window is determined by the sampling rate. The higher the sampling rate, the shorter the time window, ensuring that each data block contains a reasonable amount of data. For example, for a heart rate signal with a sampling rate of 100Hz, the time window is set to 1 second, and each data block contains 100 sampling points; for an ambient temperature and humidity signal with a sampling rate of 1Hz, the time window is set to 60 seconds, and each data block contains 60 sampling points. Each data block is assigned a unique identifier that contains information such as the data source type, the time window start time, and the cleaning subnet number, facilitating rapid retrieval and matching.

[0102] Distributed node backup of M data blocks via a consistency protocol link is a key component of the sharded storage strategy. The consistency protocol link uses a replication protocol commonly used in distributed systems (such as RAFT or a variant of the Paxos algorithm) to ensure that each data block is stored on at least two different physical nodes. During the backup process, the system tags each data block with a corresponding cleaning subnet identifier. This identifier is bound to the hardware address or logical number of the cleaning subnet and is used to determine the processing path of the data block during data recovery. For example, if a data block is marked as "Subnet 3", it will be automatically injected into the Layer 3 cleaning subnet of the heterogeneous data preprocessing network for processing during recovery.

[0103] When an abnormal data recovery instruction is triggered, the abnormal data recovery unit extracts all backup data blocks marked for the subnet where the data flow was interrupted from the distributed cache unit based on the identifier of the subnet where the data flow was interrupted. First, the data blocks adjacent to the interruption time are filtered out based on the time continuity requirement to ensure that the reconstructed signal segment seamlessly connects with the previous and next data. Then, an interpolation algorithm (such as linear interpolation or spline interpolation) is used to fill in the missing signal samples to generate a complete signal waveform. Finally, the reconstructed signal segment is converted into a feature vector and injected into the dynamic feature fusion topology to replace the output data of the interrupted subnet, maintaining continuous monitoring of the system.

[0104] The coordination between the fault-tolerant processing module and the distributed cache unit must meet strict timing requirements. The cross-validation cycle of the redundant check unit must be synchronized with the data backup cycle of the distributed cache unit to ensure that the benchmark data used for verification is consistent in time with the currently processed data. For example, if the distributed cache unit backs up data every 1 second, the verification cycle of the redundant check unit is also set to 1 second, and the latest backup data is used as a reference benchmark. In addition, the system has designed a cache data elimination mechanism. For old data blocks that exceed the preset storage time (such as 7 days), they are automatically deleted to free up storage space, while retaining long-term trend data of key health indicators (such as historical averages of heart rate and body temperature) for long-term health analysis.

[0105] In terms of hardware implementation, the distributed cache unit utilizes a distributed storage architecture, consisting of a cluster of multiple storage nodes connected by a high-speed network. Each node is equipped with solid-state storage media to meet the performance requirements of high-frequency data read and write operations. The fault-tolerant processing module is deployed on an independent control node and communicates with the storage nodes and scrubbing subnet via a message queue, ensuring rapid response in emergency situations such as data flow interruptions.

[0106] The fault-tolerant processing module verifies data integrity through redundancy checks and leverages the distributed cache unit's sharded storage and backup mechanisms for rapid data recovery. These two mechanisms combine to form a data reliability assurance chain for the system. A difference value calculation formula provides a quantitative basis for redundancy checks, while sharded storage and consistency protocols ensure the availability and consistency of backup data. A dynamic reconstruction mechanism ensures the continuity of the monitoring process. This system effectively addresses abnormal situations such as cleaning subnet failures and data flow interruptions, enhancing system stability and reliability in complex environments.

[0107] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0108] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A health information monitoring and management system based on multi-source data fusion analysis, characterized in that: include: Physiological data acquisition unit, fusion analysis engine and intelligent decision management module; The physiological data acquisition unit includes a multi-source signal synchronization node and a heterogeneous data pre-processing network; The fusion analysis engine includes a feature association unit and a pattern recognition unit; The intelligent decision management module includes a dynamic decision node and a priority mapping unit; The multi-source signal synchronization node is used to obtain heterogeneous data streams from wearable devices, environmental sensors and medical databases in real time; Based on the physiological signal sampling rate and data integrity indicators, a multi-layer data fusion topology is constructed through the heterogeneous data preprocessing network; the heterogeneous data preprocessing network is composed of N cleaning subnets based on data quality assessment connected in series; based on the real-time feature extraction capabilities of the N cleaning subnets, a health status benchmark strategy is generated through a dynamic decision node, and combined with the abnormal detection signal output by the pattern recognition unit, the monitoring task is dynamically weighted for the data source through the priority mapping unit.

2. The health information monitoring and management system based on multi-source data fusion analysis according to claim 1, characterized in that: The system also includes a real-time calibration module and a data weight optimization module; The real-time calibration module includes a signal compensation unit and a sampling rate adaptation unit; According to the generated health status benchmark strategy and dynamic data source weight distribution results, the output signal of the physiological data acquisition unit is time-domain aligned and calibrated by the signal compensation unit, and the data parsing frequency of the fusion analysis engine is dynamically adjusted by the sampling rate adaptation unit; at the same time, the data credibility assessment network constructed by the data weight optimization module is used to perform real-time analysis of the multi-source data conflict rate, and the analysis results are fed back to the dynamic decision node and priority mapping unit to iteratively optimize the monitoring strategy.

3. The health information monitoring and management system based on multi-source data fusion analysis according to claim 2, characterized in that: The signal compensation unit is composed of a delay correction circuit corresponding to a cleaning subnet connected in series in a heterogeneous data preprocessing network; the steps of constructing the multi-layer data fusion topology include: Collect the original signal waveforms and data timestamp deviations of multiple source devices, input them into the sliding window filter configured in each cleaning subnet, generate signal synchronization compensation parameters and store them in the corresponding cleaning subnet; Based on the synchronization compensation parameters stored in each layer of cleaning subnet and combined with the time continuity index of data flow, the sliding correlation coefficient algorithm is used to associate the multi-source data features with N cleaning subnets to form a dynamic feature fusion topology.

4. The health information monitoring and management system based on multi-source data fusion analysis according to claim 3, characterized in that: The activation conditions of the cleaning subnet configured in the multi-source signal synchronization node include: the data source sampling rate is within the preset compatibility range, the feature retention rate corresponding to the cleaning subnet is greater than the threshold, and the data timestamp deviation is less than the allowed synchronization range; The feature association unit is also configured with a multi-dimensional matching link; the multi-dimensional matching link includes a steady-state analysis sub-link and an abnormality correction sub-link; the steady-state analysis sub-link is generated based on the matching degree between the association rules of the fusion analysis engine and the health status benchmark strategy; the abnormality correction sub-link is constructed based on the association relationship between the instantaneous signal loss of the data acquisition unit and the monitored abnormal events.

5. The health information monitoring and management system based on multi-source data fusion analysis according to claim 4, characterized in that: The dynamic decision node is configured with a collaborative decision model; The collaborative decision-making model includes a signal-feature mapping table corresponding to the physiological data acquisition unit and an abnormality detection parameter library of the fusion analysis engine; the steps of generating the health status benchmark strategy include: According to the real-time signal quality index of the physiological data acquisition unit and the data source weight distribution results, the theoretical feature extraction capacity of each cleaning subnet in the steady-state analysis sublink is calculated; The theoretical feature extraction capacity and real-time monitoring requirements are input into the collaborative decision-making model to generate a health status reference baseline. The reference baseline is then corrected for conflict rate through the anomaly correction sub-link to form a health status benchmark strategy.

6. The health information monitoring and management system based on multi-source data fusion analysis according to claim 5, characterized in that: The parameter updating step of the collaborative decision-making model includes: When it is detected that the data source sampling rate exceeds the preset compatibility range or the multi-source data conflict rate exceeds the threshold, the first optimization instruction is triggered, that is, the backup data source channel is started and the correlation dimension of the fusion analysis engine is adjusted; If the data conflict rate has not returned to the allowable range after the execution of the first optimization instruction, the second optimization instruction is triggered, that is, switching to the anomaly correction sub-link through the multi-dimensional matching link, and reallocating the feature extraction weights of the data source based on the duration of the monitored abnormal event.

7. The health information monitoring and management system based on multi-source data fusion analysis according to claim 6, characterized in that: The parameter updating step of the collaborative decision-making model further includes: When the feature retention rate of a certain cleaning subnet in the physiological data acquisition unit is lower than the threshold, the third optimization instruction is triggered, that is, the data transmission channel of the subnet is blocked, and the corresponding signal flow is transferred to other cleaning subnets through the signal compensation unit; During the execution of the third optimization instruction, if other cleaning subnets are detected to be overloaded, the fourth optimization instruction is triggered, that is, the data weight optimization module is called to downgrade the output strategy of the priority mapping unit and limit the monitoring authority of low-credibility data sources.

8. The health information monitoring and management system based on multi-source data fusion analysis according to claim 7, characterized in that: The degradation processing logic of the data weight optimization module includes: Construct a data source priority table based on the conflict rate index and signal synchronization stability parameters output by the data credibility assessment network; When the fourth optimization instruction is triggered, the data source weight is dynamically downgraded according to the priority level table, and the parsing frequency of the fusion analysis engine is adjusted synchronously to adapt to the downgraded monitoring strategy.

9. The health information monitoring and management system based on multi-source data fusion analysis according to claim 1, characterized in that: The system also includes a fault-tolerant processing module and a distributed cache unit; The fault-tolerant processing module includes a redundancy check unit and an abnormal data recovery unit; According to the conflict rate index and priority level table output by the data credibility assessment network, the redundancy check unit performs cross-validation on the cleaning subnet in the heterogeneous data preprocessing network, and generates data integrity repair instructions based on the verification results; when the data flow of a certain layer of the cleaning subnet is detected to be interrupted, the abnormal data recovery unit calls the pre-stored reference data copy in the distributed cache unit, reconstructs the feature vector of the missing signal segment and injects it into the dynamic feature fusion topology.

10. The health information monitoring and management system based on multi-source data fusion analysis according to claim 9, characterized in that: The distributed cache unit is configured with a shard storage strategy and a consistency protocol link; The steps for constructing the shard storage strategy include: Based on the signal sampling rates of multiple source devices and the data source weight distribution results, the real-time collected heterogeneous data stream is divided into M data blocks according to the time window; M data blocks are backed up in distributed nodes through the consistency protocol link, and the cleaning subnet identifier corresponding to each data block is marked; when the abnormal data recovery instruction is triggered, the matching data block is extracted from the backup node according to the identifier for signal reconstruction.

Images