A multi-source health data dynamic early warning and collaborative management method and system

Abstract

This invention provides a method and system for dynamic early warning and collaborative management of multi-source health data, relating to the fields of data processing and collaborative management of information systems. The method performs semantic mapping and data quality verification on multi-source health data to obtain target health data and data quality scores; it calculates the dynamic baseline and fluctuation standard deviation of an individual to obtain an individual dynamic threshold; based on the target health data and the individual dynamic threshold, it compares and obtains indicator deviation feature data, and uses a nonlinear correlation adjustment algorithm for calculation and joint judgment to obtain dynamic early warning results; it uses smart contract execution task interlocking rules to obtain execution status data of cross-institutional collaborative tasks; based on the execution status data of cross-institutional collaborative tasks, it uses a hash algorithm to calculate and obtain a notarized hash value, which is then submitted to the blockchain for on-chain notarization. This invention solves the problems of inconsistent standards for heterogeneous data and the lack of automated mapping and rapid collaboration capabilities across network nodes.

Description

Technical Field

[0001] This invention relates to the field of data processing and collaborative management of information systems, and in particular to a method and system for dynamic early warning and collaborative management of multi-source health data. Background Technology

[0002] School health management involves data collaboration among three parties: medical institution physical examination and consultation records, school morning check-up and absence registration systems, and regional monitoring agency reporting systems. Currently, each of these three systems has accumulated massive amounts of data on the health status of individuals. However, in practical applications of system interconnection and data processing, the following three levels of underlying technical deficiencies exist: First, the lack of standardized heterogeneous data hinders automated mapping and rapid collaboration across network nodes. Medical institutions' HIS systems use HL7 V2.x format, school systems use a custom JSON structure, and monitoring agency systems use XMLSchema. These three systems suffer from heterogeneous data formats and incompatible network transmission protocols. Existing technologies lack deep semantic mapping methods across systems, data cleaning and alignment heavily rely on manual operations, resulting in time-consuming processing of individual records and a high error rate, leading to data silos between the three network nodes.

[0003] Second, the abnormal state triggering mechanism relies on static fixed thresholds, resulting in a high false alarm rate and an inability to identify spatiotemporal clustered data features. Existing computer monitoring systems generally use a single static threshold to determine data deviations. This mechanism cannot adapt to individual baseline data differences of the monitored objects, nor can it distinguish between data fluctuations caused by routine behaviors (such as after exercise) and data shifts caused by true abnormal states. Furthermore, current technologies lack algorithmic support for spatiotemporal joint computation of multi-source data, making it impossible to automatically identify early clustered abnormal data features, leading to system response delays.

[0004] Third, cross-agency collaborative scheduling relies on unstructured communication networks, resulting in significant delays in instruction flow and the inability to prevent tampering and trace system operation logs. When abnormal data characteristics require collaborative handling by multiple network nodes, existing mechanisms largely depend on the transmission of unstructured information through conventional communication software. The inconsistent formats of operation logs and the discrepancies in system timestamps among different agency nodes necessitate manual splicing of records from multiple servers for event tracing, which is extremely time-consuming and fails to establish a complete trust chain. Furthermore, the existing system's on-site personnel scheduling algorithm only considers a single variable, failing to integrate data such as terminal network connection readiness status, physical location coordinates, and professional qualification identifiers for comprehensive computing power and resource allocation, leading to low scheduling response efficiency. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a method and system for dynamic early warning and collaborative management of multi-source health data. These solutions address the following issues: First, heterogeneous data standards are not unified, resulting in a lack of automated mapping and rapid collaboration capabilities across network nodes; second, the abnormal state triggering mechanism relies on static fixed thresholds, leading to a high false alarm rate and an inability to automatically identify clustered data characteristics; and third, cross-agency collaborative scheduling relies on unstructured communication networks, resulting in significant instruction flow delays and the inability to prevent tampering and trace operation logs.

[0006] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a method for dynamic early warning and collaborative management of multi-source health data, comprising: S1: Based on multi-source health data, semantic mapping and data quality verification are performed using feature extraction and similarity calculation algorithms to obtain unified standard target health data and corresponding data quality scores; S2: Based on target health data, calculate the individual's dynamic baseline and standard deviation of fluctuation using an exponentially weighted moving average algorithm; correct the standard deviation of fluctuation based on data quality scores, and combine the dynamic baseline with the corrected standard deviation of fluctuation to obtain the individual's dynamic threshold; S3: Based on target health data and individual dynamic thresholds, deviation feature data of indicators is obtained by comparison; based on indicator deviation feature data and target health data, a comprehensive risk score is calculated using a nonlinear correlation adjustment algorithm; simultaneously, based on target health data, a spatiotemporal scanning statistical algorithm is used to obtain a group clustering detection index; based on the comprehensive risk score and the group clustering detection index, a joint judgment is made to obtain a dynamic early warning result; S4: Based on the dynamic early warning results, the execution status data of cross-organizational collaborative tasks is obtained by using smart contract execution task interlock rules and combining them with task routing and scheduling algorithms. S5: Based on the execution status data of cross-institutional collaborative tasks, a hash algorithm is used to calculate and obtain the evidence storage hash value, which is then submitted to the blockchain for on-chain evidence storage, thus completing the cross-institutional task collaborative management.

[0007] Further, S1 includes: Based on multi-source health data, a biomedical text pre-trained encoder is used to extract deep semantic feature vectors from health event record texts; The cosine similarity algorithm is used to calculate the similarity score between deep semantic feature vectors, and a mapping relationship is established between the similarity score and the judgment threshold. When there are conflicts in health data from multiple sources within the same preset time window, the data conflict arbitration unit is used to adjust the weight of the health data by combining the basic credibility weight of the source and the time decay factor, and the credible data is determined according to the hierarchical arbitration rules. Based on the mapping relationship and reliable data, a multi-dimensional weighted algorithm is used to calculate the target health data and the corresponding data quality score.

[0008] Furthermore, the calculation of an individual's dynamic baseline and standard deviation of fluctuation based on target health data using an exponentially weighted moving average algorithm includes: Based on the time and event information in the target health data, the dynamic smoothing coefficient adjustment algorithm is used to calculate the post-physical education effect decay factor, the morning and afternoon check-up time adjustment factor, and the seasonal prevalence adjustment factor. Based on the post-physical education effect attenuation factor, the adjustment factor for morning and afternoon health check times, and the seasonal prevalence adjustment factor, a comprehensive smoothing coefficient that dynamically changes over time is calculated. Based on the comprehensive smoothing coefficient, the target health data is calculated using an exponential weighting method to obtain the individual's dynamic baseline and standard deviation of fluctuation.

[0009] Furthermore, the expression for the dynamic baseline of the individual is: ; ; ; ; in, This represents the dynamic baseline of an individual at the current moment. This represents the overall smoothing coefficient at the current moment. This represents the observed value at the current moment. This represents the dynamic baseline of an individual at the previous moment. Represents the basic smoothing coefficient. This represents the attenuation factor of the post-physical education effect. This indicates the adjustment factor for the morning or afternoon health check period. Indicates the seasonal popularity adjustment factor. Indicates the maximum attenuation. Indicates the decay rate. Indicates the time remaining until the end of the most recent physical education class. Indicates adjustment of intensity. Represents the seasonal coefficient; where, , , and These are the system's preset parameters. , , and All data were obtained through time and event information from the target health data.

[0010] Furthermore, the comprehensive risk score calculated using a nonlinear correlation adjustment algorithm based on indicator deviation feature data includes: Based on the indicator deviation feature data, a basic risk score is obtained by using a linear weighted algorithm. When target health data from different sources are closely clustered within a preset time threshold, a time clustering reward is calculated based on the time information in the target health data. When there is a conflict in the determination of target health data from different sources, a mutual exclusion penalty term is calculated based on the semantic difference of the target health data. When the conclusions of target health data from three or more sources are consistent, a source consistency reward is calculated based on the number of sources of target health data. Based on the basic risk score, a comprehensive risk score is calculated by combining time aggregation reward items, evidence mutual exclusion penalty items, and source consistency reward items.

[0011] Furthermore, the expression for the comprehensive risk score is: ; ; ; ; in, This represents the overall risk score. Indicates the basic risk score. This indicates a time-based reward item. This indicates a penalty for mutually exclusive evidence. This indicates a reward item for consistency of origin. This represents the time-gathering reward coefficient. Indicates the rate of decay over time. This represents the difference in timestamps for anomalous events from different data sources. Indicates the conflict penalty coefficient. Indicates a conflict indicator variable. Indicates the factor that amplifies the degree of contradiction. Indicates the degree of semantic difference between the two sides of a contradiction. This represents the consistency reward coefficient. Indicates saturation rate, This indicates the number of sources providing valid evidence; among which, , , , , and These are the system's preset parameters. It is obtained by performing linear weighted calculation on the indicator deviation characteristic data. , , and All data were obtained through target health data.

[0012] Further, S4 includes: The risk level is determined based on the dynamic early warning results. Based on the risk level and the status feedback of the preceding task, the corresponding task interlocking rules are triggered using the smart contract state machine to obtain the activation status instruction of the subsequent task. When the activation status command triggers a collaborative task that requires personnel scheduling, the system's preset weight coefficients are dynamically adjusted based on the risk level to obtain the adjusted weight coefficients. Based on the adjusted weighting coefficients and online candidate data, a multi-dimensional weighted scoring scheduling algorithm is used to calculate the comprehensive scheduling score of the candidates, and cross-organizational collaborative tasks are issued according to the comprehensive scheduling score. By utilizing blockchain smart contracts to receive execution feedback from cross-institutional collaborative tasks, updating the status feedback of preceding tasks and the activation status instructions of subsequent tasks, the execution status data of cross-institutional collaborative tasks can be obtained.

[0013] Furthermore, the expression for the comprehensive scheduling score is: ; ; ; ; in, This represents the overall scheduling score for person i. Personnel Online status, Personnel The current task load rate, Indicates the degree of matching of epidemic prevention qualifications. Indicates the predictive coefficient for protective equipment. Indicates the urgency and suitability index of the task. , , , and These represent the corresponding weight coefficients. Personnel Does it have the first Qualifications Indicates whether the task requires the first... Qualifications Indicates the total number of qualification types. Personnel The current amount of protective equipment available for use. This indicates the minimum amount of supplies required to complete the task. , and These represent the weighting coefficients for historical response time, historical success rate, and historical experience, respectively. Personnel Historical average response time Personnel The success rate of completing historical missions. Personnel The number of times one has participated in similar emergency missions; among them, , , , , , , and The parameters are preset by the system, except , , , , , , and All external parameters are obtained from online candidate data.

[0014] This invention provides a multi-source health data dynamic early warning and collaborative management system, comprising: The scoring module is used to perform semantic mapping and data quality verification based on multi-source health data using feature extraction and similarity calculation algorithms, so as to obtain unified standard target health data and corresponding data quality scores. The correction module is used to calculate the dynamic baseline and standard deviation of fluctuation of an individual based on the target health data using an exponentially weighted moving average algorithm; correct the standard deviation of fluctuation based on the data quality score; and combine the dynamic baseline with the corrected standard deviation of fluctuation to obtain the individual dynamic threshold. The early warning module is used to compare target health data and individual dynamic thresholds to obtain indicator deviation feature data; based on the indicator deviation feature data and target health data, a nonlinear correlation adjustment algorithm is used to calculate a comprehensive risk score; simultaneously, based on target health data, a spatiotemporal scanning statistical algorithm is used to obtain a group clustering detection index; and a dynamic early warning result is obtained by jointly judging the comprehensive risk score and the group clustering detection index. The execution module is used to obtain the execution status data of cross-organizational collaborative tasks based on the dynamic early warning results, using smart contract execution task interlock rules, and combining task routing and scheduling algorithms. The evidence storage module is used to calculate the evidence storage hash value based on the execution status data of cross-institutional collaborative tasks using a hash algorithm, and submit it to the blockchain for on-chain evidence storage, thereby completing the collaborative management of cross-institutional tasks.

[0015] The beneficial effects of this invention are as follows: This invention provides a method for dynamic early warning and collaborative management of multi-source health data. Based on multi-source health data, it utilizes feature extraction and similarity calculation algorithms for semantic mapping and data quality verification to obtain standardized target health data and corresponding data quality scores. This technical feature solves the problems of semantic inconsistency and data silos between heterogeneous systems, and improves the automation level of cross-network node data processing and the accuracy of heterogeneous data mapping.

[0016] Based on target health data, an exponentially weighted moving average algorithm is used to calculate the dynamic baseline and standard deviation of fluctuation for each individual, and then adjusted based on data quality scores to obtain the individual dynamic threshold. This technical feature enables the system to adaptively adapt to individual differences and fluctuations in underlying data characteristics, effectively reducing the high false alarm rate caused by using a single static fixed threshold.

[0017] After comparing individual abnormal indicators, a comprehensive risk score is calculated using a nonlinear correlation adjustment algorithm, while a spatiotemporal scanning statistical algorithm is used to calculate group clustering detection indicators. These are then combined to obtain a dynamic early warning result. This technology quantifies the spatiotemporal correlation effect between multi-source data, automatically identifies early clustering anomaly data characteristics, and improves the comprehensiveness and timeliness of the system's dynamic early warning.

[0018] Based on dynamic early warning results, the system utilizes smart contract execution task interlocking rules and combines them with task routing and scheduling algorithms to obtain execution status data for cross-organizational collaborative tasks. This technology eliminates the manual waiting time caused by unstructured cross-organizational communication, automates the flow of collaborative instructions across network nodes, and improves the rationality of computing resource scheduling and system response efficiency.

[0019] Based on the execution status data of cross-institutional collaborative tasks, a hash algorithm is used to calculate the notarized hash value and submit it to the blockchain for notarization. This technical feature ensures the non-repudiation and tamper-proof verification capabilities of cross-institutional task flow records, and enables rapid and reliable traceability of underlying operation logs. Attached Figure Description

[0020] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a schematic diagram of a multi-source health data dynamic early warning and collaborative management system according to some embodiments of this specification; Figure 2 This is an exemplary flowchart illustrating a method for dynamic early warning and collaborative management of multi-source health data, as shown in some embodiments of this specification. Detailed Implementation

[0021] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0022] The dynamic early warning results and comprehensive risk scores output by the method and system provided in this invention are only used as triggering conditions for cross-organizational task scheduling, communication resource allocation, and data physical node blocking within the computer system. This invention does not involve any clinical diagnostic actions directly targeting the human body, does not provide medical intervention or treatment suggestions, and all data processing results do not have medical diagnostic validity and do not constitute medical practice.

[0023] Example 1 Figure 1 This is a schematic diagram of a multi-source health data dynamic early warning and collaborative management system according to some embodiments of this specification.

[0024] In some embodiments, the multi-source health data dynamic early warning and collaborative management system may include a scoring module, used to perform semantic mapping and data quality verification based on multi-source health data using feature extraction and similarity calculation algorithms, to obtain unified standard target health data and corresponding data quality scores; a correction module, used to calculate an individual's dynamic baseline and standard deviation of fluctuation based on the target health data using an exponentially weighted moving average algorithm; correct the standard deviation of fluctuation based on the data quality score, and combine the dynamic baseline with the corrected standard deviation of fluctuation to obtain an individual dynamic threshold; and an early warning module, used to compare the target health data and the individual dynamic threshold to obtain indicator deviation feature data; and based on the indicators... A comprehensive risk score is calculated using a nonlinear correlation adjustment algorithm based on deviation feature data and target health data. Simultaneously, a group clustering detection index is obtained using a spatiotemporal scanning statistical algorithm based on the target health data. A dynamic early warning result is obtained by jointly judging the comprehensive risk score and the group clustering detection index. The execution module is used to obtain the execution status data of cross-institutional collaborative tasks by using smart contracts to execute task interlocking rules based on the dynamic early warning result and combining them with a task routing and scheduling algorithm. The evidence storage module is used to calculate the evidence storage hash value based on the execution status data of cross-institutional collaborative tasks using a hash algorithm and submit it to the blockchain for on-chain evidence storage, thus completing the cross-institutional task collaborative management.

[0025] In some embodiments, the scoring module further includes a data conflict arbitration unit. This data conflict arbitration unit is used to detect whether there are data conflicts from multiple sources for the same indicator item of the same monitored object within a preset time window when acquiring multi-source health data. When a conflict exists, the system dynamically adjusts the weights of each data source based on preset source reliability weights and time decay factors, and outputs unique comprehensive reliable data to the subsequent processes of the scoring module according to the hierarchical arbitration rules for numerical or classification-based conflicts, thereby avoiding data overlay and logical contradictions between heterogeneous network nodes.

[0026] In some embodiments, the system further includes an edge computing gateway deployed in the local network environment of each organization. The edge computing gateway communicates with the system's cloud server and is internally configured with a time-series database and a local inference engine. The edge computing gateway monitors network connectivity readiness via a heartbeat mechanism; when a network interruption is detected, it automatically switches to offline degradation mode, independently determining whether indicators deviate from characteristic data using locally cached dynamic threshold model parameters, and generating a degradation warning signal to be pushed to the local management terminal; when the network recovers, the edge computing gateway uploads the cached labeled data from the offline period to the cloud server for incremental synchronization according to the time series, and requests the data conflict arbitration unit to handle potential synchronization conflicts.

[0027] In some embodiments, a multi-source health data dynamic early warning and collaborative management system can be used to execute a multi-source health data dynamic early warning and collaborative management method, including: S1: Based on multi-source health data, semantic mapping and data quality verification are performed using feature extraction and similarity calculation algorithms to obtain unified standard target health data and corresponding data quality scores; S2: Based on the target health data, the dynamic baseline and standard deviation of fluctuation for an individual are calculated using an exponentially weighted moving average algorithm; the standard deviation of fluctuation is corrected based on the data quality score, and the dynamic baseline and the corrected standard deviation of fluctuation are combined to obtain the individual dynamic threshold; S3: Based on the target health data and the individual dynamic threshold, the indicator bias is compared to obtain the index bias. The system first obtains characteristic data; then, based on the deviation of the indicator from the characteristic data and the target health data, a comprehensive risk score is calculated using a nonlinear correlation adjustment algorithm; simultaneously, based on the target health data, a spatiotemporal scanning statistical algorithm is used to obtain a group clustering detection indicator; a dynamic early warning result is obtained by jointly judging the comprehensive risk score and the group clustering detection indicator; S4: Based on the dynamic early warning result, the system uses smart contract execution task interlocking rules and combines them with a task routing and scheduling algorithm to obtain the execution status data of cross-institutional collaborative tasks; S5: Based on the execution status data of cross-institutional collaborative tasks, a hash algorithm is used to calculate and obtain the notarization hash value, which is then submitted to the blockchain for on-chain notarization, completing the cross-institutional task collaborative management.

[0028] In some embodiments of this specification, the processor utilizes a multi-source health data dynamic early warning and collaborative management system to execute a multi-source health data dynamic early warning and collaborative management method. In this way, firstly, the system adopts a highly modular architecture design, with clear boundaries for data flow between scoring, correction, early warning, execution, and evidence storage modules, effectively improving system throughput and underlying network resource allocation efficiency during large-scale concurrent processing of heterogeneous multi-source data. Secondly, the system integrates a data conflict arbitration unit, which automatically identifies and resolves time-series data conflicts through underlying computing power, ensuring the consistency and high availability of the underlying data chain during cross-system collaborative judgment. Thirdly, by introducing an edge computing gateway deployed in a local environment and an offline degradation architecture, the system can still independently maintain its core early warning function relying on local computing power even under extreme conditions of cloud network communication interruption, completely solving the underlying architectural defect of single-point network failure leading to complete system failure and ensuring the functional continuity of the data processing system. Example 2 Figure 2 This is an exemplary flowchart illustrating a method for dynamic early warning and collaborative management of multi-source health data, based on some embodiments of this specification. Figure 2 As shown, the process includes the following steps. In some embodiments, the process may be executed by a processor.

[0029] S1: Based on multi-source health data, semantic mapping and data quality verification are performed using feature extraction and similarity calculation algorithms to obtain standardized target health data and corresponding data quality scores.

[0030] Multi-source health data refers to raw health and related record information from different institutions or systems that has not undergone standardized processing. For example, multi-source health data may include physical examination records input from hospitals using the HL7 standard structure, JSON-structured data from schools containing basic student information and abnormal morning check-up records, and monitoring report data from disease control centers using an XML structure containing abnormal record report cards and status traceability survey results. Multi-source health data may include numerical data of abnormal body temperature, as well as non-numerical symptom record data, such as textual data of health events like diarrhea, vomiting, and rashes.

[0031] In some embodiments, the processor can use a multi-source heterogeneous data adapter to interface with the application programming interfaces of different institutional systems and acquire multi-source health data using a combination of polling and incremental fetching.

[0032] Target health data refers to standardized data packages formed by transforming, cleaning, and semantically mapping multi-source health data through a unified internal standard structure. For example, target health data may include diagnostic record data with unified health status classification codes after mapping, structured timestamp information data, and standardized numerical data of various physiological indicators after eliminating logical conflicts.

[0033] In some embodiments, the processor can use feature extraction and similarity calculation algorithms to perform heterogeneous protocol adaptation, semantic mapping, and multimodal data quality verification on the acquired multi-source health data, thereby obtaining target health data.

[0034] Data quality scoring refers to a quantitative score reflecting the reliability and usability of standardized target health data across multiple dimensions. For example, data quality scoring can include a comprehensive evaluation numerical data calculated by weighting the data integrity score, reasonableness score, time stability score, and mapping confidence output from the semantic mapping stage.

[0035] In some embodiments, the processor can obtain data quality score data by performing multi-dimensional verification statistics on multi-source health data and calculating the similarity score output by the semantic mapping algorithm.

[0036] In some embodiments, the expression for data quality scoring is: ; in, Indicates the data quality score. Indicates the completeness score. Indicates the consistency score. Indicates the reasonableness score. This represents the time stability score. Indicates the mapping confidence level. , , , and Let represent the weight coefficients of the corresponding dimensions, and let their sum be 1; where, , , , and These are the parameters preset by the system; all other parameters are obtained through multi-dimensional verification and calculation of multi-source health data.

[0037] In some embodiments, the similarity score is expressed as: ; in, Indicates the similarity score. This represents the deep semantic feature vector of the first-source health event record text. This represents the deep semantic feature vector of the second-source health event record text; where, and Both methods extract the text content from the target health data by inputting it into a pre-trained encoder.

[0038] In some embodiments, the processor can extract deep semantic feature vectors from health event record texts using a biomedical text pre-trained encoder based on multi-source health data; calculate similarity scores between deep semantic feature vectors using a cosine similarity algorithm; establish a mapping relationship between the similarity scores and a judgment threshold; when there are conflicts in health data from multiple sources within the same preset time window, use a data conflict arbitration unit to correct the weights of the health data by combining the source's basic credibility weight and the time decay factor, and determine credible data according to the hierarchical arbitration rules; based on the mapping relationship and credible data, use a multi-dimensional weighted algorithm to calculate the target health data and the corresponding data quality score.

[0039] A pre-trained encoder is a computational model that has been trained in advance on a large-scale biomedical text corpus to extract deep semantic features from text. For example, a pre-trained encoder may include a fixed BioBERT model that does not involve fine-tuning of model parameters, whose output layer contains data such as fixed-dimensional vectors representing the semantic features of specific medical terms.

[0040] In some embodiments, the processor can acquire pre-trained encoder data by loading a biomedical language model file pre-deployed in the system and fixing the parameters during the initialization phase.

[0041] Health event record text refers to unstructured or semi-structured text information from multi-source health data used to describe the disease status or physical examination conclusions of an individual. For example, health event record text may include specific disease description text data corresponding to ICD-10 codes in a hospital HIS system, or custom disease label description text data in a school health system.

[0042] In some embodiments, the processor can extract health event record text data from specific field contents of multi-source health data by parsing and extracting information from the multi-source health data.

[0043] Deep semantic feature vectors are numerical vectors that represent the deep semantic features of health event record text or medical tags in the biomedical semantic space. For example, deep semantic feature vectors can include 768-dimensional feature representation vector data extracted from the output layer of the BioBERT model.

[0044] In some embodiments, the processor can obtain deep semantic feature vector data by inputting the extracted health event record text data into a pre-trained encoder for computational processing.

[0045] Similarity score refers to a numerical result that quantifies the degree of consistency between two deep semantic feature vectors in semantic space. For example, similarity score can include specific numerical data between 0 and 1 obtained by calculating the cosine of the angle between the feature vector of a hospital health event record text and the feature vector of a disease label from a school in vector space.

[0046] In some embodiments, the processor can obtain similarity score data by performing a cosine similarity algorithm on the extracted deep semantic feature vectors between different systems.

[0047] A decision threshold is a reference boundary value used to distinguish different levels of semantic matching credibility and trigger corresponding processing procedures. For example, a decision threshold may include a fixed value of 0.85, set based on general medical semantic distance experience, to distinguish the boundary between automatic mapping and manual verification.

[0048] In some embodiments, the processor can obtain the judgment threshold data by reading a pre-configured configuration file of the system or by receiving an empirically set value input by the administrator based on biomedical semantic space features.

[0049] Mapping relationships refer to the corresponding association results of data fields or medical codes established between heterogeneous data sources after similarity comparison. For example, mapping relationships may include automatically matched connection data between hospital diagnosis codes and school system health status classification codes based on high similarity scores, or associated record data marked as low confidence and entering the manual review queue.

[0050] In some embodiments, the processor can establish and obtain mapping relationship data by comparing similarity scores with a decision threshold and making conditional decisions based on a hierarchical processing strategy.

[0051] Source reliability weight refers to the basic weight value that the system pre-sets for different data generating institutions, representing the initial reliability of their data. For example, source reliability weight may include a high weight value of 0.6 for hospitals, a medium weight value of 0.3 for disease control institutions, and a low weight value of 0.1 for schools.

[0052] In some embodiments, the processor can obtain source base credibility weight data by manual preset by the system administrator and by reading from the local system database or configuration file.

[0053] The time decay factor is a correction parameter that reflects the impact of data timeliness on credibility, used to reduce the influence of outdated data in conflict arbitration. For example, the time decay factor may include an adjustment coefficient value that decreases exponentially as the data acquisition time is longer than the current detection time, calculated according to a preset decay rate.

[0054] In some embodiments, the processor can obtain time decay factor data by extracting time information from multi-source health data and performing exponential function calculations in conjunction with a preset decay constant.

[0055] Credible data refers to the final retained data that has been screened or weighted and corrected by the system's arbitration mechanism when multiple data sources conflict within the same time window. For example, credible data may include the final body temperature value determined after weighted average calculation, or the final diagnostic classification conclusion data determined according to the principle of priority of evidence level.

[0056] In some embodiments, the processor can use a data conflict arbitration unit to modify the weight of the original data by combining the source base credibility weight and the time decay factor, and determine the credible data according to the hierarchical arbitration rules.

[0057] S2: Based on target health data, calculate the individual's dynamic baseline and standard deviation of fluctuation using an exponentially weighted moving average algorithm; correct the standard deviation of fluctuation based on data quality scores, and combine the dynamic baseline with the corrected standard deviation of fluctuation to obtain the individual's dynamic threshold.

[0058] A dynamic baseline is a reference value that reflects the expected normal level of an individual's physiological indicators at a specific point in time and can adapt to changing circumstances. For example, a dynamic baseline may include the current expected body temperature baseline value calculated by dynamically smoothing the data based on an individual's historical body temperature data and incorporating a comprehensive smoothing coefficient that varies with circumstances such as physical education classes, circadian rhythms, and seasons.

[0059] In some embodiments, the processor can obtain dynamic baseline data by performing time-series iterative calculations based on physiological observations in the target health data using an exponentially weighted moving average algorithm.

[0060] The standard deviation of fluctuation is a statistical dispersion parameter that measures the degree to which an individual's physiological indicators deviate from their dynamic baseline. It is used to define a reasonable range for normal fluctuations. For example, the standard deviation of fluctuation can include statistical values ​​of the allowable deviation range of an indicator, calculated based on the variance decay of an individual's historical data and adjusted for tolerance using the current data quality score.

[0061] In some embodiments, the processor can obtain the standard deviation of fluctuation data by calculating the variance of the historical observation sequence of the target health data and iteratively updating it in combination with the comprehensive smoothing coefficient of the current time.

[0062] In some embodiments, the expression for the corrected standard deviation of volatility is: ; in, This represents the corrected standard deviation of volatility. This represents the calculated initial standard deviation of the fluctuation. This represents the data quality score; where, It is obtained by performing time-series iterative calculations on the target health data. It is obtained through the data quality scoring formula.

[0063] In some embodiments, the processor can use a dynamic smoothing coefficient adjustment algorithm to calculate the post-physical education effect decay factor, the morning and afternoon health check time period adjustment factor, and the seasonal prevalence adjustment factor based on the time and event information in the target health data; calculate the comprehensive smoothing coefficient that changes dynamically over time based on the post-physical education effect decay factor, the morning and afternoon health check time period adjustment factor, and the seasonal prevalence adjustment factor; and perform an exponential weighted calculation on the target health data based on the comprehensive smoothing coefficient to obtain the individual's dynamic baseline and standard deviation of fluctuation.

[0064] The post-physical education effect decay factor refers to a regulating parameter that quantifies the degree to which the increase in physiological indicators after physical exercise naturally diminishes over time. For example, the post-physical education effect decay factor may include dynamic decay coefficient data calculated based on the time immediately following the end of the most recent physical education class, showing a significant decrease in system sensitivity immediately after class, followed by a gradual recovery to the baseline smoothing level.

[0065] In some embodiments, the processor can obtain the physical education class end timestamp contained in the target health data, calculate the time difference with the current time, and perform an exponential decay algorithm to calculate the physical education class post-effect decay factor data.

[0066] The adjustment factor for morning and afternoon health checks refers to a sensitivity adjustment parameter set based on the differences in the human body's diurnal rhythm and the differences in the daily schedule of schools. For example, the adjustment factor for morning and afternoon health checks may include a smaller coefficient value of 0.75 to enhance smoothing and reduce interference from small fluctuations in the morning, and a larger coefficient value of 1.25 to accelerate tracking and capture abnormal fluctuations in the afternoon.

[0067] In some embodiments, the processor can obtain the morning and afternoon inspection time period adjustment factor data by comparing the timestamp of the current data with the system's preset detection time period division range for condition matching.

[0068] In some embodiments, in addition to the fixed time slots for morning and afternoon checks mentioned above, the system supports custom temporary testing time slots, such as testing boarding students before bedtime or temporary school-wide testing during public health emergencies. For temporary evening testing time slots, the system can configure specific time slot adjustment factors according to actual physiological rhythms, for example, setting it to 1.10, to adapt to changes in baseline body temperature at night. This configuration is triggered by matching the timestamp tags contained in multi-source health data with a preset time slot library.

[0069] The seasonal prevalence adjustment factor is a parameter that adjusts the sensitivity of the early warning baseline to changes in the prevalence trends of infectious diseases in different seasons. For example, the seasonal prevalence adjustment factor may include a higher coefficient value set during the epidemic season to improve system response, and a lower coefficient value set during the non-epidemic season to reduce false alarm rate.

[0070] In some embodiments, the processor can obtain seasonal epidemic adjustment factor data by reading the influenza epidemic calendar information published by the local disease control center, mapping it to the corresponding seasonal coefficient, and performing linear adjustment calculation.

[0071] The overall smoothing coefficient refers to a dynamically changing overall adjustment parameter used to control the smoothness of indicator data after incorporating various environmental and behavioral contextual factors. For example, the overall smoothing coefficient may include a dynamic product calculated by multiplying a fixed base smoothing coefficient by the aforementioned post-physical education effect attenuation factor, morning and afternoon check-up time adjustment factor, and seasonal prevalence adjustment factor.

[0072] In some embodiments, the processor can obtain comprehensive smoothing coefficient data by extracting the individual context adjustment factors calculated within the system and multiplying them together.

[0073] In some embodiments, the expression for the dynamic baseline of the individual is: ; ; ; ; in, This represents the dynamic baseline of an individual at the current moment. This represents the overall smoothing coefficient at the current moment. This represents the observed value at the current moment. This represents the dynamic baseline of an individual at the previous moment. Represents the basic smoothing coefficient. This represents the attenuation factor of the post-physical education effect. This indicates the adjustment factor for the morning or afternoon health check period. Indicates the seasonal popularity adjustment factor. Indicates the maximum attenuation. Indicates the decay rate. Indicates the time remaining until the end of the most recent physical education class. Indicates adjustment of intensity. Represents the seasonal coefficient; where, , , and These are the system's preset parameters. , , and All data were obtained through time and event information from the target health data.

[0074] Individual dynamic thresholds refer to upper and lower limit reference ranges for determining whether individual physiological indicators are abnormal, generated based on individual differences, current contextual dynamic baselines, and data quality. For example, individual dynamic thresholds may include the lower limit value obtained by subtracting twice the standard deviation of fluctuation after data quality score relaxation from the calculated individual dynamic baseline, and the upper limit value obtained by adding twice the standard deviation of fluctuation after correction.

[0075] In some embodiments, the processor can calculate individual dynamic threshold data by combining the calculated dynamic baseline with the fluctuation standard deviation corrected based on the data quality score.

[0076] S3: Based on target health data and individual dynamic thresholds, deviation feature data of indicators is obtained by comparison; based on indicator deviation feature data and target health data, a comprehensive risk score is calculated using a nonlinear correlation adjustment algorithm; simultaneously, based on target health data, a spatiotemporal scanning statistical algorithm is used to obtain a group clustering detection index; based on the comprehensive risk score and the group clustering detection index, a joint judgment is made to obtain a dynamic early warning result.

[0077] Indicator deviation feature data refers to event records or specific out-of-limit features where an individual's specific physiological observation values ​​exceed the normal range of the dynamic threshold calculated at the corresponding time. For example, indicator deviation feature data may include out-of-limit difference data generated when the current observed body temperature value is higher than the calculated upper limit of the individual's dynamic threshold, as well as system label data recording the abnormality type.

[0078] In some embodiments, the processor can obtain indicator deviation feature data by comparing the real-time observed values ​​in the target health data with the calculated individual dynamic threshold.

[0079] A comprehensive risk score is a non-linear total score that comprehensively quantifies the severity of health risk for an individual or a specific group by integrating the spatiotemporal correlations and corroborating evidence among multiple data sources. For example, a comprehensive risk score may include the base risk scores of abnormal indicators from various sources, plus additional scores from time aggregation, deducting deductions for contradictory conclusions, and adding saturation bonuses from consistent corroboration from multiple sources, resulting in a final calculated value.

[0080] In some embodiments, the processor can obtain comprehensive risk score data by performing comprehensive nonlinear equation calculations based on indicator deviation feature data and target health data using a nonlinear correlation adjustment algorithm.

[0081] In some embodiments, the processor can calculate a basic risk score based on indicator deviation feature data using a linear weighted algorithm; when target health data from different sources are closely clustered within a preset time threshold, a time clustering reward is calculated based on the time information in the target health data; when there is a judgment conflict between target health data from different sources, an evidence mutual exclusion penalty is calculated based on the semantic difference degree of the target health data; when the conclusions of target health data from three or more sources are consistent, a source consistency reward is calculated based on the number of sources of the target health data; and a comprehensive risk score is obtained by combining the basic risk score with the time clustering reward, the evidence mutual exclusion penalty, and the source consistency reward.

[0082] The basic risk score refers to a risk value obtained by making a preliminary assessment based directly on the deviation characteristics of indicators from each independent data source, without considering the non-linear correlation effects between data. For example, the basic risk score may include preliminary hazard scores calculated by simply linearly weighting and summing the deviation characteristics of indicators detected by hospitals, schools, and disease control centers, assigning initial preset weights to each.

[0083] In some embodiments, the processor can obtain basic risk score data by calculating the deviation feature data state values ​​of indicators from different sources using a linear weighted summation algorithm with preset weights.

[0084] The time information in target health data refers to data elements contained within a standardized data package that record the exact moment or time interval of a health event. For example, the time information in target health data may include precise timestamps recording the onset of abnormal symptoms, numerical data of the end time of a physical education class, and numerical data of the time difference when data from multiple sources are successively collected and stored by the system.

[0085] In some embodiments, the processor can obtain time information data from the target health data by parsing the standard structured fields of the target health data and extracting the corresponding timestamp tag content.

[0086] The time-clustering reward item refers to a score adjustment item used to reflect the synergistic early warning enhancement effect brought about by the successive and dense occurrence of health abnormal events from different sources within a short period of time. For example, the time-clustering reward item may include an exponential reward increase value calculated to enhance the basic risk when the time difference between abnormal records from hospitals and abnormal records from schools is extremely small and they are determined to be closely clustered.

[0087] In some embodiments, the processor can determine that target health data from different sources are clustered within a preset time threshold, and obtain time-clustered reward data by calculating an exponential decay function based on the extracted time information.

[0088] The semantic dissimilarity of target health data refers to a reference value that quantifies the degree of contradiction in diagnostic conclusions at a deep semantic level between health data from different sources that have conflicting conclusions. For example, the semantic dissimilarity of target health data can include the difference coefficient value calculated and extracted from consistent literature on underlying diagnostic methods to measure the distance between the two determinations when a hospital determines tuberculosis is positive while the CDC determines it is negative.

[0089] In some embodiments, the processor can obtain semantic difference data of the target health data by extracting the semantic similarity of the diagnostic conclusion texts of the conflicting parties from the semantic mapping engine and performing reverse transformation calculation.

[0090] The mutual evidence exclusion penalty is a deduction adjustment item used to reduce the risk of false alarms in the system when there is a significant conflict between the final conclusions from different data sources. For example, the mutual evidence exclusion penalty may include a negative adjustment value calculated by combining the conflict penalty coefficient and the contradiction amplification coefficient calculated by the system, which is used to deduct from the total risk score to reduce the confidence level.

[0091] In some embodiments, the processor can determine that there are diagnostic result conflicts in target health data from different sources, and obtain evidence mutual exclusion penalty term data by calculating a nonlinear penalty factor based on the aforementioned semantic difference degree.

[0092] The number of sources of target health data refers to the total number of system institutions that provide clear judgments or valid supporting evidence for the same monitored object within the same comprehensive early warning assessment cycle. For example, the number of sources of target health data can include the total number of independent institutions among hospitals, schools, and disease control centers that have determined the status to be abnormal and uploaded valid supporting materials.

[0093] In some embodiments, the processor can obtain the number of sources of the target health data by counting and statistically analyzing the different agency identification codes carried in the target health data participating in the current joint early warning determination.

[0094] Source consistency bonus items refer to bonus points that significantly enhance the confidence of a system's judgment when data from multiple institutional sources highly consistently point to the same anomalous risk. For example, source consistency bonus items may include exponentially increasing bonus values ​​triggered when three systems simultaneously reach a status confirmation judgment, which further boost the overall score.

[0095] In some embodiments, the processor can determine that the target health data from three or more sources are completely consistent, and obtain source consistency reward data by performing a nonlinear saturation function calculation based on the aforementioned calculated number of sources.

[0096] In some embodiments, the expression for the comprehensive risk score is: ; ; ; ; in, This represents the overall risk score. Indicates the basic risk score. This indicates a time-based reward item. This indicates a penalty for mutually exclusive evidence. This indicates a reward item for consistency of origin. This represents the time-gathering reward coefficient. Indicates the rate of decay over time. This represents the difference in timestamps for anomalous events from different data sources. Indicates the conflict penalty coefficient. Indicates a conflict indicator variable. Indicates the factor that amplifies the degree of contradiction. Indicates the degree of semantic difference between the two sides of a contradiction. This represents the consistency reward coefficient. Indicates saturation rate, This indicates the number of sources providing valid evidence; among which, , , , , and These are the system's preset parameters. It is obtained by performing linear weighted calculation on the indicator deviation characteristic data. , , and All data were obtained through target health data.

[0097] Clustering detection indicators refer to statistical test parameters used to quantify whether the actual distribution of abnormal records within a specific time and spatial window is significantly higher than historical expected levels. For example, clustering detection indicators may include spatiotemporal scan log-likelihood ratio data calculated based on a predefined class or grade spatial window to identify early infectious disease outbreak characteristics on campus. When a single symptom other than fever (such as diarrhea) appears in multi-source health data, although it does not trigger an individual dynamic body temperature threshold alarm, the system will still trigger a red alert if the frequency or clustering of semantic feature words related to diarrhea within the same spatial window (such as the same class) exceeds historical expected levels, in order to identify potential clustering risks of gastrointestinal infectious diseases.

[0098] In some embodiments, the processor can obtain population aggregation detection index data by iteratively calculating using the spatiotemporal scanning statistics algorithm formula based on the full distribution of target health data.

[0099] In some embodiments, the expression for the group aggregation detection index is: ; in, Indicators representing the detection of group clustering This indicates the actual number of exception records within the spatiotemporal window. Indicates the expected number of abnormal records. This represents the total number of monitored objects within the spatiotemporal window; among which, , and All of these methods involve statistically analyzing the distribution of target health data within a specific spatiotemporal range.

[0100] Dynamic early warning results refer to the warning status or label output after combining comprehensive risk scores at the individual level with group clustering risk indicators, which has guidance significance for graded handling. For example, dynamic early warning results may include yellow warning label data that triggers internal school attention, orange warning label data that triggers hospital follow-up tasks, and red emergency warning label data that triggers cross-institutional collaborative instructions across the entire chain.

[0101] In some embodiments, the processor can obtain dynamic early warning result data by inputting the comprehensive risk score value and the group aggregation detection index value into a preset joint classification judgment logic rule base for threshold comparison.

[0102] S4: Based on the dynamic early warning results, the execution status data of cross-organizational collaborative tasks is obtained by using smart contract execution task interlocking rules and combining them with task routing and scheduling algorithms.

[0103] A smart contract is a computer program code deployed on the underlying blockchain network architecture that contains preset rules for the automated change of cross-institutional task states and cannot be unilaterally tampered with. For example, a smart contract may include on-chain executable script code data that defines interlocking flow logic such as automatically triggering the school's physical node to change the task status from pending to executable upon receiving confirmation of the completion of a disease control status tracing investigation.

[0104] In some embodiments, the processor can obtain smart contract data by reading compiled and instantiated chaincode files deployed on blockchain nodes such as the Hyperledger Fabric consortium blockchain through the system management interface.

[0105] Task interlocking rules refer to the set of rule parameters set within a smart contract to regulate state dependencies, deadlock prevention logic, and sequential triggering conditions between task nodes of different institutions. For example, task interlocking rules may include logical binding rule configuration data stipulating that a subsequent corresponding task can only be allowed to update its execution status when the on-chain transaction ID of the preceding task has been verified by consensus and its status is marked as completed.

[0106] In some embodiments, the processor can extract task interlock rule data by parsing the predefined state machine transition mechanism file or configuration parameter text inside the smart contract.

[0107] Cross-agency collaborative tasks refer to specific emergency response matters involving multiple institutional nodes such as schools, hospitals, and disease control centers, requiring mutual cooperation and execution according to interlocked workflow processes. For example, cross-agency collaborative tasks may include a workflow sequence data of joint response tasks triggered by a red alert, encompassing multiple stages such as tracing and investigating the on-site status of disease control centers, physically blocking key populations within schools, and subsequent re-examination and confirmation at designated hospitals.

[0108] In some embodiments, the processor can automatically generate and distribute cross-agency collaborative task data to terminals of various agencies by determining the risk level based on the dynamic warning results using a workflow engine.

[0109] Execution status data refers to a comprehensive set of information used to record in real time the current stage of a cross-organizational collaborative task throughout its entire lifecycle, including feedback results from corresponding execution nodes. For example, execution status data may include character identifiers indicating whether a subtask is currently pending, suspended, or completed, as well as electronic signature signals confirming task completion from third-party systems.

[0110] In some embodiments, the processor can obtain execution status data by using blockchain smart contracts to receive task execution feedback operation records from each node terminal and update the status flags of the preceding and following tasks.

[0111] In some embodiments, the processor can determine the risk level based on the dynamic early warning results; based on the risk level and the status feedback of the preceding task, it can trigger the corresponding task interlocking rules using a smart contract state machine to obtain the activation status instruction of the subsequent task; when the activation status instruction triggers a collaborative task requiring personnel scheduling, it can dynamically adjust the system's preset weight coefficients based on the risk level to obtain the adjusted weight coefficients; based on the adjusted weight coefficients and online candidate personnel data, it can calculate the comprehensive scheduling score of the candidates using a multi-dimensional weighted scoring scheduling algorithm, and issue cross-organizational collaborative tasks based on the comprehensive scheduling score; it can use a blockchain smart contract to receive the execution feedback of the cross-organizational collaborative tasks, update the status feedback of the preceding task and the activation status instruction of the subsequent task, and obtain the execution status data of the cross-organizational collaborative tasks.

[0112] Risk level refers to a classification and grading identifier based on the risk score output by the system, used to directly guide the system's emergency response preemption mechanism and resource allocation intensity. For example, the risk level may include the highest emergency level classification identifier data generated by the system when the comprehensive risk score reaches or exceeds the 0.85 threshold, used to require the immediate activation of a preemptive emergency command.

[0113] In some embodiments, the processor can obtain risk level data by mapping and matching the specific value of the comprehensive risk score with the preset hierarchical decision boundary interval within the system based on the dynamic early warning results.

[0114] Status feedback for preceding tasks refers to the feedback of information regarding the current completion status of the initial planning and execution stages in a cross-agency collaborative workflow with sequential dependencies. For example, status feedback for preceding tasks may include a status confirmation message submitted by the CDC node confirming the completion of the epidemiological investigation task for suspected abnormal records, accompanied by the digital signature of the agency personnel and the operation execution timestamp data.

[0115] In some embodiments, the processor can obtain the status feedback data of the preceding task by receiving the business operation record submitted to the system interface by the receiving agency operation terminal and verified as legitimate by the blockchain smart contract node.

[0116] The activation state instruction for a post-task refers to the control signal automatically generated by the smart contract state machine after the preconditions are fully met and verified, used to wake up and allow subsequent execution operations. For example, the activation state instruction for a post-task may include control parameter data sent to the hospital registration system after the contract detects that the school's screening results have been successfully uploaded to the blockchain, changing the state of the follow-up appointment button for a specific student to be operable.

[0117] In some embodiments, the processor can obtain the activation status instruction data of the subsequent task by using the smart contract state machine logic to determine and trigger the corresponding task interlocking rules based on the determined risk level and the status feedback of the preceding task.

[0118] In some embodiments, the expression for the activation state instruction of the subsequent task is: ; in, Instructions indicating the activation status of subsequent tasks. This indicates the status feedback of the preceding task. This represents the on-chain transaction hash identifier of the preceding task. This represents the preset task interlock mapping function; where, The system's preset rule parameters, and All data is obtained through interaction with blockchain smart contract nodes.

[0119] Weighting coefficients refer to preset multiplier constants used in the multi-dimensional weighted scoring algorithm for personnel scheduling to dynamically adjust the proportion of different influencing factors such as personnel distance and workload in the final score. For example, weighting coefficients may include specific coefficient values ​​that emphasize the influence of distance under normal conditions, and coefficient adjustment values ​​that are automatically increased by the system under red high-risk alert conditions to emphasize the matching degree of personnel's professional epidemic prevention qualifications.

[0120] In some embodiments, the processor can obtain weight coefficient data by reading the system's underlying preset scheduling configuration file or by dynamically adjusting the default coefficient parameters according to the risk level determined in real time.

[0121] In some embodiments, the expression for the adjusted weighting coefficient is: ; ; in, This represents the adjusted weighting coefficient. Indicates the time decay factor. Indicates the decay rate. This indicates the time difference between the data collection time and the current time. This indicates the source's basic credibility weight; where, and These are the system's preset parameters. It is obtained by calculating time information from multi-source health data.

[0122] Online candidate personnel data refers to the collection of available personnel scheduling resources in the system that are currently connected to the network, have not been assigned exclusive tasks, and are capable of accepting orders. For example, online candidate personnel data may include the personnel's current GPS geographic location coordinates, a list of specific infectious disease prevention and control qualification certifications registered in the database, real-time data on the quantity of protective equipment such as N95 masks currently carried, and statistical data on the personnel's historical average response time for tasks.

[0123] In some embodiments, the processor can connect to the personnel status management terminals and inventory databases of participating institutions in real time through the system interface to aggregate and obtain online candidate data.

[0124] The comprehensive scheduling score refers to a quantitative ranking score of task suitability calculated by comprehensively considering multiple dimensions such as physical distance of candidates, professional qualification matching, available protective equipment, and historical performance. For example, the comprehensive scheduling score may include specific ranking scores calculated according to the scheduling algorithm to determine which online candidate has the highest priority to accept the current specific early warning task.

[0125] In some embodiments, the processor can obtain comprehensive scheduling score data by summing the weight coefficients adjusted for risk level and the online candidate data collected in real time using a multi-dimensional weighted scoring scheduling algorithm model.

[0126] In some embodiments, the expression for the comprehensive scheduling score is: ; ; ; ; in, This represents the overall scheduling score for person i. Personnel Online status, Personnel The current task load rate, Indicates the degree of matching of epidemic prevention qualifications. Indicates the predictive coefficient for protective equipment. Indicates the urgency and suitability index of the task. , , , and These represent the corresponding weight coefficients. Personnel Does it have the first Qualifications Indicates whether the task requires the first... Qualifications Indicates the total number of qualification types. Personnel The current amount of protective equipment available for use. This indicates the minimum amount of supplies required to complete the task. , and These represent the weighting coefficients for historical response time, historical success rate, and historical experience, respectively. Personnel Historical average response time Personnel The success rate of completing historical missions. Personnel The number of times one has participated in similar emergency missions; among them, , , , , , , and The parameters are preset by the system, except , , , , , , and All external parameters are obtained from online candidate data.

[0127] Execution feedback refers to the acknowledgment message submitted to the system by the specific executor node or dispatched personnel who assigned the collaborative task, regarding the actual operation process and final closed-loop result of the task handling. For example, execution feedback may include a joint confirmation signal message sent by the executor, containing electronic signatures of the three participating parties, a summary text of the conclusion of this task handling, and a standard handling completion timestamp.

[0128] In some embodiments, the processor can obtain execution feedback data by interfacing with the business management terminals deployed in each participating institution and receiving task completion operation log messages submitted by the task flow interface.

[0129] S5: Based on the execution status data of cross-institutional collaborative tasks, a hash algorithm is used to calculate and obtain the evidence storage hash value, which is then submitted to the blockchain for on-chain evidence storage, thus completing the cross-institutional task collaborative management.

[0130] A storage hash value refers to a fixed-length string generated by irreversibly performing a cryptographic digest calculation on critical business storage data and its associated metadata generated across institutional tasks. This string is used to verify the immutability of the original data. For example, a storage hash value may include a lightweight 32-byte SHA-256 message digest generated by combining a student's unique identifier hash character, a millisecond-level timestamp, a specific early warning risk score, an event trigger type, and an institutional identifier.

[0131] In some embodiments, the processor can obtain evidence hash value data by performing calculations on the extracted key field combinations based on operation records such as collected cross-agency collaborative task execution status data and using one-way hash algorithms such as SHA-256.

[0132] In some embodiments, to ensure the verifiability and stability of each algorithm model in practical applications, specific value ranges and preferred configurations for key constant parameters are provided in conjunction with the campus health management scenario: basic smoothing coefficient. The value range is [0.1, 0.4], with a preferred value of 0.2. Maximum attenuation amplitude. The value range is [0.3, 0.7], with a preferred value of 0.5. Attenuation rate. The value range is [0.5, 1.2], with a preferred value of 0.8, calculated based on the heart rate recovery constant after human exercise. Adjust the intensity. The value range is [0.2, 0.6], with a preferred value of 0.4. Time-aggregated reward coefficient. The value range is [0.2, 0.8], with a preferred value of 0.5. Time decay rate. The value range is [0.1, 0.5], with a preferred value of 0.3. Conflict penalty coefficient. The value range is [0.1, 0.5], with a preferred value of 0.3. This is the amplification factor for the degree of contradiction. The value range is [0.3, 0.8], with a preferred value of 0.5. Consistency reward coefficient. The value range is [0.2, 0.6], with a preferred value of 0.4. Saturation rate. Value range [0.5, 1.5], preferred value 1.0. Basic weight coefficient set: Online status weight $w_y$ preferred 0.10, load weight... Preferred score: 0.10; Qualification matching weight. Preferred weight 0.25, material forecast weight Optimal weight 0.15, emergency adaptation weight Optimal value: 0.15. Historical performance weight set: historical response time weight. Optimal score: 0.4, weighted by historical success rate. The preferred value is 0.4, with historical experience weighting. The preferred value is 0.2. The above weight parameters are fixed during system initialization. When the risk level changes, the system will dynamically scale the weights according to a preset strategy.

[0133] It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects may be any one or a combination of the above, or any other possible beneficial effects.

Claims

1. A method for dynamic early warning and collaborative management of multi-source health data, characterized in that, include: S1: Based on multi-source health data, semantic mapping and data quality verification are performed using feature extraction and similarity calculation algorithms to obtain unified standard target health data and corresponding data quality scores; S2: Based on target health data, calculate the individual's dynamic baseline and standard deviation of fluctuation using an exponentially weighted moving average algorithm; The standard deviation of fluctuation is corrected based on the data quality score, and the dynamic baseline is combined with the corrected standard deviation of fluctuation to obtain the individual dynamic threshold. S3: Based on target health data and individual dynamic thresholds, compare and obtain indicator deviation feature data; A comprehensive risk score is calculated based on indicator deviation feature data and target health data using a nonlinear correlation adjustment algorithm. Meanwhile, based on the target health data, spatiotemporal scanning statistical algorithms are used to obtain indicators for detecting group aggregation. Dynamic early warning results are obtained by combining comprehensive risk scores and clustering detection indicators. S4: Based on the dynamic early warning results, the execution status data of cross-organizational collaborative tasks is obtained by using smart contract execution task interlock rules and combining them with task routing and scheduling algorithms. S5: Based on the execution status data of cross-institutional collaborative tasks, a hash algorithm is used to calculate and obtain the evidence storage hash value, which is then submitted to the blockchain for on-chain evidence storage, thus completing the cross-institutional task collaborative management.

2. The method for dynamic early warning and collaborative management of multi-source health data according to claim 1, characterized in that, S1 includes: Based on multi-source health data, a biomedical text pre-trained encoder is used to extract deep semantic feature vectors from health event record texts; The cosine similarity algorithm is used to calculate the similarity score between deep semantic feature vectors, and a mapping relationship is established between the similarity score and the judgment threshold. When there are conflicts in health data from multiple sources within the same preset time window, the data conflict arbitration unit is used to adjust the weight of the health data by combining the basic credibility weight of the source and the time decay factor, and the credible data is determined according to the hierarchical arbitration rules. Based on the mapping relationship and reliable data, a multi-dimensional weighted algorithm is used to calculate the target health data and the corresponding data quality score.

3. The method for dynamic early warning and collaborative management of multi-source health data according to claim 1, characterized in that, The calculation of an individual's dynamic baseline and standard deviation of fluctuation based on target health data using an exponentially weighted moving average algorithm includes: Based on the time and event information in the target health data, a dynamic smoothing coefficient adjustment algorithm is used, such as calculating the post-physical education effect decay factor, the morning and afternoon health check time adjustment factor, and the seasonal prevalence adjustment factor. Based on the post-physical education effect attenuation factor, the adjustment factor for morning and afternoon health check times, and the seasonal prevalence adjustment factor, a comprehensive smoothing coefficient that dynamically changes over time is calculated. Based on the comprehensive smoothing coefficient, the target health data is calculated using an exponential weighting method to obtain the individual's dynamic baseline and standard deviation of fluctuation.

4. The method for dynamic early warning and collaborative management of multi-source health data according to claim 3, characterized in that, The expression for the dynamic baseline of the individual is: ； ； ； ； in, This represents the dynamic baseline of an individual at the current moment. This represents the overall smoothing coefficient at the current moment. This represents the observed value at the current moment. This represents the dynamic baseline of an individual at the previous moment. Represents the basic smoothing coefficient. This represents the attenuation factor of the post-physical education effect. This indicates the adjustment factor for the morning or afternoon health check period. Indicates the seasonal popularity adjustment factor. Indicates the maximum attenuation. Indicates the decay rate. Indicates the time remaining until the end of the most recent physical education class. Indicates adjustment of intensity. Represents the seasonal coefficient; where, , , and These are the system's preset parameters. , , and All data were obtained through time and event information from the target health data.

5. The method for dynamic early warning and collaborative management of multi-source health data according to claim 1, characterized in that, The comprehensive risk score, calculated using a nonlinear correlation adjustment algorithm based on indicator deviation feature data, includes: Based on the indicator deviation feature data, a basic risk score is obtained by using a linear weighted algorithm. When target health data from different sources are closely clustered within a preset time threshold, a time clustering reward is calculated based on the time information in the target health data. When there is a conflict in the determination of target health data from different sources, a mutual exclusion penalty term is calculated based on the semantic difference of the target health data. When the conclusions of target health data from three or more sources are consistent, a source consistency reward is calculated based on the number of sources of target health data. Based on the basic risk score, a comprehensive risk score is calculated by combining time aggregation reward items, evidence mutual exclusion penalty items, and source consistency reward items.

6. The method for dynamic early warning and collaborative management of multi-source health data according to claim 5, characterized in that, The expression for the comprehensive risk score is: ； ； ； ； in, This represents the overall risk score. Indicates the basic risk score. This indicates a time-based reward item. This indicates a penalty for mutually exclusive evidence. This indicates a reward item for consistency of origin. This represents the time-gathering reward coefficient. Indicates the decay rate over time. This represents the difference in timestamps for anomalous events from different data sources. Indicates the conflict penalty coefficient. Indicates a conflict indicator variable. Indicates the factor that amplifies the degree of contradiction. Indicates the degree of semantic difference between the two sides of a contradiction. This represents the consistency reward coefficient. Indicates saturation rate, This indicates the number of sources providing valid evidence; among which, , , , , and These are the system's preset parameters. It is obtained by performing linear weighted calculation on the indicator deviation characteristic data. , , and All data were obtained through target health data.

7. The method for dynamic early warning and collaborative management of multi-source health data according to claim 1, characterized in that, S4 includes: The risk level is determined based on the dynamic early warning results. Based on the risk level and the status feedback of the preceding task, the corresponding task interlocking rules are triggered using the smart contract state machine to obtain the activation status instruction of the subsequent task. When the activation status command triggers a collaborative task that requires personnel scheduling, the system's preset weight coefficients are dynamically adjusted based on the risk level to obtain the adjusted weight coefficients. Based on the adjusted weighting coefficients and online candidate data, a multi-dimensional weighted scoring scheduling algorithm is used to calculate the comprehensive scheduling score of the candidates, and cross-organizational collaborative tasks are issued according to the comprehensive scheduling score. By utilizing blockchain smart contracts to receive execution feedback from cross-institutional collaborative tasks, updating the status feedback of preceding tasks and the activation status instructions of subsequent tasks, the execution status data of cross-institutional collaborative tasks can be obtained.

8. The multi-source health data dynamic early warning and collaborative management method according to claim 7, characterized in that, The expression for the comprehensive scheduling score is: ； ； ； ； in, This represents the overall scheduling score for person i. Personnel Online status, Personnel The current task load rate, Indicates the degree of matching of epidemic prevention qualifications. Indicates the predictive coefficient for protective equipment. Indicates the urgency and suitability index of the task. , , , and These represent the corresponding weight coefficients. Personnel Does it have the first Qualifications Indicates whether the task requires the first... Qualifications Indicates the total number of qualification types. Personnel The current amount of protective equipment available for use. This indicates the minimum amount of supplies required to complete the task. , and These represent the weighting coefficients for historical response time, historical success rate, and historical experience, respectively. Personnel Historical average response time Personnel The success rate of completing historical missions. Personnel The number of times one has participated in similar emergency missions; among them, , , , , , , and The parameters are preset by the system, except , , , , , , and All external parameters are obtained from online candidate data.

9. A multi-source health data dynamic early warning and collaborative management system, used to execute the multi-source health data dynamic early warning and collaborative management method as described in any one of claims 1 to 8, characterized in that, include: The scoring module is used to perform semantic mapping and data quality verification based on multi-source health data using feature extraction and similarity calculation algorithms, so as to obtain unified standard target health data and corresponding data quality scores. The correction module is used to calculate an individual's dynamic baseline and standard deviation of fluctuation based on target health data using an exponentially weighted moving average algorithm. The standard deviation of fluctuation is corrected based on the data quality score, and the dynamic baseline is combined with the corrected standard deviation of fluctuation to obtain the individual dynamic threshold. The early warning module is used to compare target health data and individual dynamic thresholds to obtain indicator deviation feature data; A comprehensive risk score is calculated based on indicator deviation feature data and target health data using a nonlinear correlation adjustment algorithm. Meanwhile, based on the target health data, spatiotemporal scanning statistical algorithms are used to obtain indicators for detecting group aggregation. Dynamic early warning results are obtained by combining comprehensive risk scores and clustering detection indicators. The execution module is used to obtain the execution status data of cross-organizational collaborative tasks based on the dynamic early warning results, using smart contract execution task interlock rules, and combining task routing and scheduling algorithms. The evidence storage module is used to calculate the evidence storage hash value based on the execution status data of cross-institutional collaborative tasks using a hash algorithm, and submit it to the blockchain for on-chain evidence storage, thereby completing the collaborative management of cross-institutional tasks.

Images