Multi-source data fusion-based multi-point triggering intelligent early warning method for infectious diseases

Abstract

The invention discloses an infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion, and relates to the technical field of public health, and the method comprises the following steps: 1, collecting and integrating multi-source data, building data interfaces with all levels of medical institutions and grassroots public health units, collecting diagnosis and treatment data and abnormal health condition information, and storing the diagnosis and treatment data and abnormal health condition information; accessing online platform data, screening and removing invalid data, and constructing a regional diagnosis and treatment database to provide a comprehensive data basis for subsequent analysis; according to the invention, by integrating multi-source data, the accuracy and real-time performance of the infectious disease early warning system are significantly improved; according to the method, symptom description is preprocessed and vectorized by using a natural language processing technology, a similar mode between symptoms and examination data is deeply mined in combination with standardized examination data, so that the characteristics of infectious diseases are accurately extracted, a potential characteristic set of the infectious diseases is identified through clustering analysis and an association rule mining algorithm, and a characteristic knowledge base is constructed.

Description

Technical Field

[0001] The present invention relates to the field of public health technology, and in particular to an infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion. Background Art

[0002] In the current public health field, early warning and control of infectious diseases is a crucial link. Although existing infectious disease early warning systems are able to monitor and warn of epidemics to a certain extent, they generally have some limitations. These systems often rely on a single data source, such as hospital reports or laboratory test results, which limits their breadth and depth in data collection and analysis. In addition, traditional early warning systems are also insufficient in real-time, accuracy, and rapid response to emerging infectious diseases. These systems usually lack the ability to integrate multi-source data and cannot make full use of information from different channels, such as social media, online medical consultation platforms, etc., which can provide early signals about the epidemic.

[0003] With the rapid development of information technology, especially the application of big data and artificial intelligence technology, new possibilities have been provided for early warning of infectious diseases. Multi-source data fusion technology can integrate a large amount of data from medical institutions, public health units and online platforms, including diagnosis and treatment records, symptom descriptions, test results, etc. Through natural language processing (NLP) technology and data mining technology, valuable information can be extracted from these data to identify the potential characteristics and patterns of infectious diseases. However, how to effectively integrate and analyze these multi-source data and how to set reasonable early warning thresholds remain a technical challenge. In addition, the existing early warning system also lacks an effective mechanism for dynamically adjusting the warning threshold and feedback optimization, which limits the adaptability and long-term effectiveness of the early warning system.

[0004] In order to solve the above defects, a technical solution is now provided. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of the existing infectious disease early warning system, such as single data source, insufficient real-time performance, limited accuracy, and lack of dynamic adjustment and feedback optimization mechanism, and to propose an infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion.

[0006] The purpose of the present invention can be achieved through the following technical solutions: An infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion, comprising the following steps: S1: Multi-source data collection and integration, establishing data interfaces with medical institutions at all levels and primary public health units, collecting diagnosis and treatment data and abnormal health status information, accessing online platform data, and building a regional diagnosis and treatment database after screening and eliminating invalid data, providing a comprehensive data basis for subsequent analysis; S2: Analysis and extraction of infectious disease characteristics: using NLP technology to process symptom text data, combined with standardized examination data, mining similar symptom-examination data combination patterns through cluster analysis, and using association rule mining algorithms to build a feature knowledge base to extract infectious disease characteristics; S3: Setting and dynamic calibration of warning thresholds: setting initial warning thresholds based on regional characteristics and historical epidemic factors, introducing dynamic probability models, regularly updating thresholds based on multiple factors, and using grey correlation analysis and MCMC methods to ensure a balance between warning sensitivity and specificity; S4: Early warning execution and feedback optimization. When the data in a certain area reaches the early warning threshold, a feature portrait is generated and distributed to other areas for comparison. The comprehensive similarity is calculated based on Euclidean distance and cosine similarity to determine whether to issue an early warning. After receiving feedback, reinforcement learning is used to optimize the model parameters and thresholds to improve the effectiveness of early warning.

[0007] Furthermore, the execution steps of S1 are as follows: Establish a secure and stable data interface with the information systems of medical institutions at all levels, extract patient diagnosis and treatment records on a daily basis, obtain basic examination data including blood routine, biochemical examination, urine routine, as well as patient complaints and doctors' preliminary diagnosis conclusions, and classify and store them according to the patient's geographical area to build a regional diagnosis and treatment database; Data collection points should be set up in community health service stations, school clinics, and grassroots public health units in the health care rooms of enterprises and institutions. Medical staff should be arranged to regularly report abnormal health conditions of residents, students, and employees in the area. The focus should be on groups with the same symptoms. Information on the number of people involved, age distribution, and characteristics of the working and learning environment should be recorded and added to the regional health monitoring data set to fill the gap in grassroots health information. Access the data of online medical consultation platforms, analyze the heat map of keywords related to infectious disease symptoms in user consultation texts, sort out the regional aggregation of basic examination data uploaded by users, broaden the breadth of data acquisition, and capture potential epidemic signals of infectious diseases; Then filter the acquired data and eliminate invalid data.

[0008] Furthermore, the specific operation steps of screening the acquired data in S1 and eliminating invalid data are as follows: Screening is performed by feature quantity and information entropy, where feature quantity statistics are: Calculate the number of valid features contained in each patient's medical record, including the number of blood routine indicators, the number of biochemical examination indicators, and the number of recorded symptoms; set a threshold LK, and use the formula Get the number of features F, It is each feature; is an indicator function. When When ,on the contrary ; When F<LK, the diagnosis and treatment record is judged as single data and is eliminated; Information entropy calculation: For the symptom description text, calculate its information entropy to measure the information richness; the formula is , where m is the number of different symptoms, It is The probability of a symptom occurring; when the information entropy value H is lower than the preset threshold, it is judged that the symptom description is single, the diagnosis and treatment record is not effective enough, and it is eliminated.

[0009] Furthermore, the specific operation steps of S2 are as follows: Use natural language processing technology to pre-process the symptom description text data, first segment the words and remove stop words, and then convert the text into a computable vector form through the Word2Vec and GloVe word vector models; Combine the standardized and normalized numerical features of basic examination data and input them into the cluster analysis algorithm to mine similar symptom-examination data combination patterns hidden in historical cases and real-time new data, and identify infectious disease feature sets; Based on the Apriori algorithm for association rule mining, the co-occurrence relationship between symptoms and examination indicators was analyzed, and support and confidence thresholds were set to mine the co-occurrence relationship between symptoms and examination indicators, including headache duration exceeding 24 hours and platelet count less than 100×10 9 / L to enhance strong association rules for infectious disease risk, assist in refining and verifying infectious disease characteristics, and build a characteristic knowledge base; When analyzing the correlation between symptoms and examination indicators, strong association rules are mined by calculating support and confidence. The support reflects the frequency of occurrence of the combination of symptoms and examination indicators in the overall data. The confidence measures the possibility of the examination indicator appearing under the condition that the symptoms appear. Rules with high confidence are more reliable.

[0010] Furthermore, in the infectious disease feature analysis in S2, when the symptom text vector after natural language processing and the basic examination data numerical features are combined into data points for cluster analysis, the Euclidean distance formula is used to measure the similarity between these data points; The formula is as follows: , where: Represents the Euclidean distance between data points x and y, which is used to measure their similarity in the n-dimensional feature space; and are the values ​​of data points x and y on the kth feature dimension respectively: n is the dimension of the feature space, that is, the number of all features involved in the clustering analysis; By calculating the Euclidean distance, the DBSCAN clustering algorithm is used to cluster data points with similar symptom-examination data combination patterns according to the set distance threshold and minimum number of points. This helps to identify clusters representing infectious disease characteristics from case data and discover similar case groups hidden in the data.

[0011] Furthermore, the specific operation steps of S3 are as follows: Based on the demographic characteristics of each region, the allocation of medical resources, and the incidence of previous infectious diseases, a stratified and regional statistical analysis of the disease diagnosis and treatment data baseline was conducted. Combined with the review of local historical infectious disease epidemics and the summary of small-scale clustered unexplained disease events, initial warning thresholds were set for various characteristic indicators of infectious diseases. Introducing Bayesian networks and hidden Markov model dynamic probability models, taking real-time infectious disease prevention and control dynamics, seasonal climate variables, and environmental factors as external inputs, regularly updating warning thresholds to ensure a balance between warning sensitivity and specificity; When using Bayesian networks or hidden Markov models to dynamically calibrate warning thresholds and predict the development trend of infectious diseases, the unknown parameters in the model are estimated by the MCMC method, and the Metropolis-Hastings algorithm is used. , the proposed distribution , used to generate new candidate states , acceptance probability ,in is the target distribution; The MCMC method performs random sampling in the parameter space, accepts or rejects new sampling points based on the acceptance probability, and finally obtains the posterior distribution of the parameters; it helps to estimate the model parameters and improve the reliability of the prediction of the development trend of infectious diseases.

[0012] Furthermore, when considering the impact of multiple factors on the warning threshold in S3, the degree of correlation between each factor and the characteristic indicators of infectious diseases is determined by grey correlation analysis: Set the reference sequence , compare sequences , correlation coefficient ,in , is the resolution coefficient, ; By calculating the correlation , find out the key factors that affect the warning threshold beyond the preset standard, and provide a basis for the reasonable setting and dynamic adjustment of the warning threshold.

[0013] Furthermore, the process of S4 is as follows: When the comprehensive evaluation of the symptoms and examination data related to infectious diseases in a certain area reaches the warning threshold, the feature profile generation program is automatically started to summarize the complete feature profile of infectious diseases covering symptom details, characteristic intervals of examination indicators, and attributes of high-risk groups as a template for cross-regional screening and comparison; Relying on the computing power and storage advantages of the big data distributed computing framework, the template is distributed to data processing nodes in other regions, and local data screening and comparison work is carried out in parallel. Once the local data and the template match beyond the standard, the infectious disease warning signal of the region is triggered; the warning information is pushed through multiple channels such as mobile APP pop-up windows, government office system eye-catching reminders, and mobile phone text message emergency notifications, to convey detailed infectious disease information and ensure that the information reaches the front-line prevention and control personnel directly; The personnel at the receiving end of the warning carry out investigation and verification, isolation and control, and sample testing in accordance with the prevention and control disposal process, and feedback the disposal results. With the help of reinforcement learning algorithms, they optimize the feature analysis model parameters and calibrate the warning thresholds based on false alarms and missed alarms, and continuously iterate to improve the warning effectiveness.

[0014] Furthermore, the analysis process of the situation where the matching degree between the local data and the template exceeds the standard in S4 is as follows: For the local data to be screened, standardization is first performed to make different characteristics comparable; For numerical inspection indicators, the Z-score standardization method is used: ,in is the original data value, is the average value of the indicator in local historical data, is the standard deviation, which ensures that different inspection indicators are compared on the same scale through standardization methods; For symptom data, it is converted into vector form to facilitate subsequent calculations. At the same time, for the onset order of symptoms, different weights are given to symptoms in different positions to reflect their order information; The same standardization and vector construction operations are performed on the data in the complete characteristic portrait template of infectious diseases; the symptom details and the characteristic interval information of the inspection indicators in the template are converted into a computable vector form; for the symptom types, the vector is constructed in the manner of the symptom dictionary; for the characteristic interval of the inspection indicator, the midpoint value of the interval is taken for standardization and then included in the vector; The feature vector of the local data and the feature vector of the template data are calculated by Euclidean distance and cosine similarity to obtain two similarities, which are respectively denoted as Euclidean distance and cosine similarity , and assign weights to , after normalization, enter the following formula: To obtain the comprehensive similarity SDF, is the preset maximum value of the Euclidean distance; The calculated comprehensive similarity SDF is compared with the set 70% similarity threshold. When SDF ≥ 0.7, it is judged that the matching degree between the local data and the template exceeds the standard.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention significantly improves the accuracy and real-time performance of the infectious disease early warning system by integrating multi-source data, including medical institution diagnosis and treatment records, reports on abnormal health conditions of primary public health units, and user consultation data of online platforms. The present invention uses natural language processing (NLP) technology to pre-process and vectorize symptom descriptions. Combined with standardized examination data, the present invention can deeply mine similar patterns between symptoms and examination data, thereby accurately extracting the characteristics of infectious diseases. Through cluster analysis and association rule mining algorithms, the present invention can identify the potential feature sets of infectious diseases and construct a feature knowledge base, providing strong support for subsequent accurate diagnosis and early warning. (2) The present invention has a warning threshold setting and dynamic calibration mechanism. The initial warning threshold is set according to regional characteristics and historical epidemic factors, and a dynamic probability model is introduced to regularly update the threshold based on multiple factors. Grey correlation analysis and Markov chain Monte Carlo method are used to ensure the balance between warning sensitivity and specificity, thereby improving the reliability of the prediction of infectious disease development trends. This dynamic adjustment mechanism enables the warning system to adapt to changes in the epidemic situation, thereby improving the adaptability and long-term effectiveness of the warning. (3) When the data in a certain area reaches the warning threshold, the present invention can automatically start the feature profile generation program and distribute the template to other areas for comparison based on the big data distributed computing framework. Once the matching degree exceeds the standard, the warning signal is triggered immediately. This cross-regional feature profile comparison mechanism greatly improves the response speed of the warning, allowing prevention and control measures to be implemented more quickly. At the same time, by receiving feedback from front-line prevention and control personnel, the reinforcement learning algorithm is used to optimize the feature analysis model parameters and calibrate the warning threshold, thereby achieving continuous iteration to improve the warning efficiency. This feedback-based optimization mechanism enables the warning to continuously learn and improve, thereby improving long-term efficiency and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. 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 creative work are within the scope of protection of the present invention.

[0018] It should be understood that the terms “include” and “comprising” used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0019] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations.

[0020] like Figure 1 As shown, a multi-point triggering intelligent early warning method for infectious diseases based on multi-source data fusion includes the following steps: Step 1: Multi-source data collection and integration, establish data interfaces with medical institutions at all levels and primary public health units, collect diagnosis and treatment data and abnormal health status information, access online platform data, and build a regional diagnosis and treatment database after screening and eliminating invalid data to provide a comprehensive data basis for subsequent analysis; Establish a secure and stable data interface with the information systems of medical institutions at all levels (HIS, LIS, PACS), extract patient diagnosis and treatment records on a daily basis, obtain basic examination data including blood routine (various parameters of white blood cells, red blood cells, and platelets), biochemical tests (liver function ALT, AST, renal function CRE, BUN indicators), urine routine (urine protein, occult blood), as well as patient complaints (fever, cough, rash, diarrhea, etc., symptoms, duration, severity description), doctor's preliminary diagnosis conclusion (symptom name, confirmed disease), and classify and store them according to the patient's geographical area (accurate to the street and township level) to build a regional diagnosis and treatment database; Data collection points should be set up in community health service stations, school clinics, and grassroots public health units in the health care rooms of enterprises and institutions, and medical staff should be arranged to regularly (at least twice a week) report abnormal health conditions of residents, students, and employees in the area, focusing on groups with the same or similar symptoms, and recording the number of people involved, age distribution, and characteristics of the work and study environment, and supplementing them to the regional health monitoring data set to fill the gap in grassroots health information; Access the data of online medical consultation platforms, analyze the heat map of keywords related to infectious disease symptoms in user consultation texts (use natural language processing technology to extract expressions such as fever, fatigue, and "severe sore throat"), sort out the regional aggregation of basic examination data uploaded by users (some platforms allow uploading simple test report screenshots to analyze key values), broaden the breadth of data acquisition, and capture potential epidemic signals of infectious diseases; then filter the acquired data and eliminate invalid data; feature quantity statistics: calculate the number of valid features contained in each patient's diagnosis and treatment record, such as the number of blood routine indicators, the number of biochemical examination indicators, the number of recorded symptoms, etc.; set a threshold LK, and use the formula Get the number of features F, It is each feature; is an indicator function, if it exists (i.e. has value) ,otherwise ; When F<LK, the diagnosis and treatment record is judged as single data and is eliminated; Information entropy calculation: For the symptom description text, calculate its information entropy to measure the information richness; the formula is , where m is the number of different symptoms, It is The probability of a symptom occurring; when the information entropy value H is lower than the preset threshold, it is judged that the symptom description is single, the diagnosis and treatment record is not effective enough, and it is eliminated.

[0021] Step 2: Analyze and extract infectious disease characteristics. Use NLP technology to process symptom text data, combine with standardized examination data, mine similar symptom-examination data combination patterns through cluster analysis, and use association rule mining algorithms to build a feature knowledge base to accurately extract infectious disease characteristics. Preprocess the symptom description text data using natural language processing (NLP) techniques. First, perform word segmentation and remove stop words (such as meaningless words like "de" and "le"), and then convert the text into a computable vector form through Word2Vec and GloVe word vector models. Combine the numerical features after standardizing and normalizing the basic examination data, and input them together into a clustering analysis algorithm (select the DBSCAN algorithm to handle complex and irregular data distributions) to mine the hidden similar symptom-examination data combination patterns in historical cases and real-time newly added data, and identify possible infectious disease feature sets. In the analysis of infectious disease features, when combining the symptom text vectors after natural language processing with the numerical features of basic examination data into data points for clustering analysis, use the Euclidean distance formula to measure the similarity between these data points. , where: represents the Euclidean distance between data points x and y, which is used to measure their similarity in the n-dimensional feature space; and are the values of data points x and y in the k-th feature dimension respectively: n is the dimension of the feature space, that is, the total number of all features participating in the clustering analysis. By calculating the Euclidean distance, according to the set distance threshold and minimum number of points, use the DBSCAN clustering algorithm to cluster the data points with similar symptom-examination data combination patterns into classes. This helps to identify the clusters that may represent infectious disease features from the vast amount of mixed case data, discover those similar case groups hidden in the data, and thus more effectively refine the infectious disease feature set, providing strong support for subsequent accurate diagnosis and early warning. Based on an optimized version of the Apriori algorithm for association rule mining (improved for the high-dimensional and sparse characteristics of medical data), analyze the frequent co-occurrence relationship between symptoms and examination indicators, set thresholds for support (such as 10%) and confidence (such as 60%), and mine strong association rules such as "The risk of a new infectious disease increases when there is continuous headache (duration exceeding 24 hours) and the platelet count is lower than 100×10 9 / L". Assist in accurately refining and verifying infectious disease features and constructing a feature knowledge base. When analyzing the association relationship between symptoms and examination indicators, mine strong association rules by calculating support and confidence. Among them, support reflects the occurrence frequency of the combination of symptoms and examination indicators in the overall data, and only combinations with a support reaching a certain threshold have general significance to avoid mining overly rare combinations. Confidence measures the possibility of the examination indicator appearing under the condition that the symptom appears, and rules with high confidence are more reliable. Use these strong association rules to assist in refining and verifying infectious disease features and construct a richer and more accurate feature knowledge base.

[0022] Step 3: Setting and dynamic calibration of warning thresholds: setting the initial warning thresholds based on regional characteristics and historical epidemics, introducing a dynamic probability model, regularly updating the thresholds based on multiple factors, and using grey correlation analysis and MCMC methods to ensure a balance between warning sensitivity and specificity; Based on the demographic characteristics of each region (population size, density, age structure), medical resource allocation (number of hospitals, number of beds, ratio of medical staff), and previous infectious disease incidence factors, we statistically analyzed the baseline data of common disease diagnosis and treatment in stratified and zoned areas (divided into central urban areas, suburban fringe areas, remote suburban towns, etc.), and set initial warning thresholds for various characteristic indicators of infectious diseases in combination with the review of local historical infectious disease epidemics and the summary of small-scale clustered unexplained disease events; when considering the impact of multiple factors (demographic characteristics, medical resource allocation, prevention and control dynamics, seasonal climate, environmental factors, etc.) on the warning threshold, we determined the degree of correlation between each factor and the characteristic indicators of infectious diseases through grey correlation analysis, and set a reference sequence (such as characteristic indicator sequences of historical infectious disease outbreaks), comparative sequences (such as the characteristic indicator sequence currently monitored in real time), correlation coefficient ,in , is the resolution coefficient, with a value of 0.5, and the correlation By calculating the correlation , find out the key factors that have a greater impact on the warning threshold, and provide a basis for more reasonable setting and dynamic adjustment of the warning threshold; The Bayesian network and hidden Markov model dynamic probability models are introduced to use the real-time infectious disease prevention and control dynamics (the import of epidemics from surrounding areas, the intensity of local prevention and control measures such as the number of quarantined people, the coverage of nucleic acid testing), seasonal climate variables (high incidence of enteric infectious diseases in summer and prevalence of respiratory infectious diseases in winter), and environmental factors (the impact of air pollution index and water quality on the breeding of specific infectious diseases) as external inputs, and regularly (weekly) update the warning threshold to ensure a balance between warning sensitivity and specificity; during the influenza season, the respiratory symptom association threshold is appropriately relaxed, and the case determination criteria are strictly enforced during the tightening period of prevention and control; when using Bayesian networks or hidden Markov models to dynamically calibrate the warning threshold and predict the development trend of infectious diseases, the unknown parameters in the model are estimated by the MCMC method, and the Metropolis-Hastings algorithm is used to set the current state , the proposed distribution (used to generate new candidate states ), acceptance probability ,in It is the target distribution (such as the probability distribution of the development trend of infectious diseases). The MCMC method performs random sampling in the parameter space, accepts or rejects new sampling points according to the acceptance probability, and finally obtains the posterior distribution of the parameters. It helps to estimate the model parameters more accurately, improve the reliability of the prediction of the development trend of infectious diseases, and thus set the warning threshold more scientifically.

[0023] Step 4: Early warning execution and feedback optimization. When the data in a certain area reaches the early warning threshold, a feature profile is generated and distributed to other areas for comparison. The comprehensive similarity is calculated based on Euclidean distance, cosine similarity, etc. to determine whether to issue an early warning. After receiving feedback, reinforcement learning is used to optimize model parameters and thresholds to improve early warning efficiency. When the comprehensive evaluation of the symptoms and inspection data related to infectious diseases in a certain area reaches the warning threshold, the feature profile generation program is automatically started to summarize and summarize the complete feature profile of infectious diseases covering symptom details (symptom types, onset sequence, severity change curve), inspection index characteristic range (mean, standard deviation, abnormal fluctuation range of each key indicator), and high-risk population attributes (age range, occupational category, history of basic diseases) as a cross-regional screening and comparison template; relying on the computing power and storage advantages of big data distributed computing frameworks (such as Hadoop and Spark), the template is distributed to data processing nodes in other regions, and local data screening and comparison work is carried out in parallel. Once it is found that the local data matches the template beyond the standard; the analysis process of the exceeding standard is as follows: For the local data to be screened (including symptom data and examination data), standardization is first performed to make different characteristics comparable. For numerical examination indicators (such as various blood routine indicators, biochemical indicators, etc.), the Z-score standardization method is used: ,in is the original data value, is the average value of the indicator in local historical data, is the standard deviation. Through the standardization method, it is ensured that different inspection indicators are compared on the same scale to avoid the influence of different indicator dimensions on the similarity calculation. For symptom data, it is converted into vector form to facilitate subsequent calculations. At the same time, for the order of symptom onset, different weights can be given to symptoms in different positions to reflect their order information. The same standardization and vector construction operations are performed on the data in the complete characteristic portrait template of infectious diseases. The symptom details, inspection indicator characteristic intervals and other information in the template are converted into computable vector form. For symptom types, vectors are constructed in the form of symptom dictionaries. For inspection indicator characteristic intervals, the midpoint value of the interval (or other reasonable representative values ​​according to actual conditions) is taken for standardization and then included in the vector. The feature vector of the local data and the feature vector of the template data are calculated by Euclidean distance and cosine similarity to obtain two similarities, which are respectively denoted as Euclidean distance and cosine similarity , respectively assign weights , after normalization, enter the following formula: To obtain the comprehensive similarity SDF, is the maximum value of the pre-set Euclidean distance (determined based on data distribution and experience, and used to normalize the Euclidean distance to between 0 and 1); the calculated comprehensive similarity SDF is compared with the set 70% similarity threshold. When SDF ≥ 0.7, it is judged that the matching degree between the local data and the template exceeds the standard, and the infectious disease warning signal for the area is immediately triggered; the warning information is delivered through multiple channels such as mobile APP (exclusive for health department managers and medical institution leaders) pop-up window push, government office system eye-catching reminder, and mobile phone text message emergency notification, to convey detailed infectious disease information (outbreak source area, possible scope of impact, key features and key points), to ensure that the information reaches the front-line prevention and control personnel directly; The personnel at the receiving end of the warning carry out investigation and verification, isolation and control, and sample testing in accordance with the prevention and control disposal process, and feedback the disposal results to the early warning system. The system uses reinforcement learning algorithms to optimize feature analysis model parameters and calibrate warning thresholds based on false alarms and missed alarms, and continuously iterates to improve the effectiveness of early warnings.

[0024] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-point triggering intelligent early warning method for infectious diseases based on multi-source data fusion, characterized in that: The following steps are involved: S1: Multi-source data collection and integration, establishing data interfaces with medical institutions at all levels and primary public health units, collecting diagnosis and treatment data and abnormal health status information, accessing online platform data, and building a regional diagnosis and treatment database after screening and eliminating invalid data, providing a comprehensive data basis for subsequent analysis; S2: Analysis and extraction of infectious disease characteristics: using NLP technology to process symptom text data, combined with standardized examination data, mining similar symptom-examination data combination patterns through cluster analysis, and using association rule mining algorithms to build a feature knowledge base to extract infectious disease characteristics; S3: Setting and dynamic calibration of warning thresholds: setting initial warning thresholds based on regional characteristics and historical epidemic factors, introducing dynamic probability models, regularly updating thresholds based on multiple factors, and using grey correlation analysis and MCMC methods to ensure a balance between warning sensitivity and specificity; S4: Early warning execution and feedback optimization. When the data in a certain area reaches the early warning threshold, a feature portrait is generated and distributed to other areas for comparison. The comprehensive similarity is calculated based on Euclidean distance and cosine similarity to determine whether to issue an early warning. After receiving feedback, reinforcement learning is used to optimize the model parameters and thresholds to improve the effectiveness of early warning.

2. According to claim 1, a multi-point triggered intelligent early warning method for infectious diseases based on multi-source data fusion is characterized in that: The execution steps of S1 are as follows: Establish a secure and stable data interface with the information systems of medical institutions at all levels, extract patient diagnosis and treatment records on a daily basis, obtain basic examination data including blood routine, biochemical examination, urine routine, as well as patient complaints and doctors' preliminary diagnosis conclusions, and classify and store them according to the patient's geographical area to build a regional diagnosis and treatment database; Data collection points should be set up in community health service stations, school clinics, and grassroots public health units in the health care rooms of enterprises and institutions. Medical staff should be arranged to regularly report abnormal health conditions of residents, students, and employees in the area. The focus should be on groups with the same symptoms. Information on the number of people involved, age distribution, and characteristics of the working and learning environment should be recorded and added to the regional health monitoring data set to fill the gap in grassroots health information. Access the data of online medical consultation platforms, analyze the heat map of keywords related to infectious disease symptoms in user consultation texts, sort out the regional aggregation of basic examination data uploaded by users, broaden the breadth of data acquisition, and capture potential epidemic signals of infectious diseases; Then filter the acquired data and eliminate invalid data.

3. The infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion according to claim 2 is characterized in that: The specific operation steps of screening the acquired data and eliminating invalid data in S1 are as follows: Screening by feature quantity and information entropy, The number of features is counted as follows: Calculate the number of valid features contained in each patient's medical record, including the number of blood routine indicators, the number of biochemical examination indicators, and the number of recorded symptoms; set a threshold LK, and use the formula Get the number of features F, It is each feature; is an indicator function. When When ,on the contrary ; When F<LK, the medical record is judged as single data and is eliminated; Information entropy calculation: For the symptom description text, calculate its information entropy to measure the information richness; the formula is , where m is the number of different symptoms, It is The probability of a symptom occurring; when the information entropy value H is lower than the preset threshold, it is judged that the symptom description is single, the diagnosis and treatment record is not effective enough, and it is eliminated.

4. The infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion according to claim 1 is characterized in that: The specific operation steps of S2 are as follows: Use natural language processing technology to pre-process the symptom description text data, first segment the words and remove stop words, and then convert the text into a computable vector form through the Word2Vec and GloVe word vector models; Combine the standardized and normalized numerical features of basic examination data and input them into the cluster analysis algorithm to mine similar symptom-examination data combination patterns hidden in historical cases and real-time new data, and identify infectious disease feature sets; Based on the Apriori algorithm for association rule mining, the co-occurrence relationship between symptoms and examination indicators was analyzed, and support and confidence thresholds were set to mine the co-occurrence relationship between symptoms and examination indicators, including headache duration exceeding 24 hours and platelet count less than 100×10 9 / L to enhance strong association rules for infectious disease risk, assist in refining and verifying infectious disease characteristics, and build a characteristic knowledge base ； When analyzing the correlation between symptoms and examination indicators, strong association rules are mined by calculating support and confidence. The support reflects the frequency of occurrence of the combination of symptoms and examination indicators in the overall data. The confidence measures the possibility of the examination indicator appearing under the condition that the symptoms appear. Rules with high confidence are more reliable.

5. The infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion according to claim 4 is characterized in that: In the infectious disease feature analysis in S2, when the symptom text vector after natural language processing and the basic examination data numerical features are combined into data points for cluster analysis, the Euclidean distance formula is used to measure the similarity between these data points; The formula is as follows: , where: Represents the Euclidean distance between data points x and y, which is used to measure their similarity in the n-dimensional feature space; and are the values ​​of data points x and y on the kth feature dimension respectively: n is the dimension of the feature space, that is, the number of all features involved in the clustering analysis; By calculating the Euclidean distance, the DBSCAN clustering algorithm is used to cluster data points with similar symptom-examination data combination patterns according to the set distance threshold and minimum number of points. This helps to identify clusters representing infectious disease characteristics from case data and discover similar case groups hidden in the data.

6. The infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion according to claim 1 is characterized in that: The specific operation steps of S3 are as follows: Based on the demographic characteristics of each region, the allocation of medical resources, and the incidence of previous infectious diseases, a stratified and regional statistical analysis of the disease diagnosis and treatment data baseline was conducted. Combined with the review of local historical infectious disease epidemics and the summary of small-scale clustered unexplained disease events, initial warning thresholds were set for various characteristic indicators of infectious diseases. Introducing Bayesian networks and hidden Markov model dynamic probability models, taking real-time infectious disease prevention and control dynamics, seasonal climate variables, and environmental factors as external inputs, regularly updating warning thresholds to ensure a balance between warning sensitivity and specificity; When using Bayesian networks or hidden Markov models to dynamically calibrate warning thresholds and predict the development trend of infectious diseases, the unknown parameters in the model are estimated by the MCMC method, and the Metropolis-Hastings algorithm is used. , the proposed distribution , used to generate new candidate states , acceptance probability ,in is the target distribution; The MCMC method performs random sampling in the parameter space, accepts or rejects new sampling points based on the acceptance probability, and finally obtains the posterior distribution of the parameters; it helps to estimate the model parameters and improve the reliability of the prediction of the development trend of infectious diseases.

7. The infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion according to claim 6 is characterized in that: When considering the impact of multiple factors on the warning threshold in S3, the degree of correlation between each factor and the characteristic indicators of infectious diseases is determined by grey correlation analysis: Set the reference sequence , compare sequences , correlation coefficient ,in , is the resolution coefficient, ; By calculating the correlation , find out the key factors that affect the warning threshold beyond the preset standard, and provide a basis for the reasonable setting and dynamic adjustment of the warning threshold.

8. The infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion according to claim 1 is characterized in that: The process of S4 is as follows: When the comprehensive evaluation of the symptoms and examination data related to infectious diseases in a certain area reaches the warning threshold, the feature profile generation program is automatically started to summarize the complete feature profile of infectious diseases covering symptom details, characteristic intervals of examination indicators, and attributes of high-risk groups as a template for cross-regional screening and comparison; Relying on the computing power and storage advantages of the big data distributed computing framework, the template is distributed to data processing nodes in other regions, and local data screening and comparison work is carried out in parallel. Once the local data and the template match degree exceeds the standard, the infectious disease warning signal of the region is triggered; Early warning information is sent through multiple channels, such as mobile APP pop-up push, government office system prompts, and mobile SMS emergency notifications, to convey detailed infectious disease information and ensure that the information reaches frontline prevention and control personnel directly; The personnel at the receiving end of the warning carry out investigation and verification, isolation and control, and sample testing in accordance with the prevention and control disposal process, and feedback the disposal results. With the help of reinforcement learning algorithms, they optimize the feature analysis model parameters and calibrate the warning thresholds based on false alarms and missed alarms, and continuously iterate to improve the warning effectiveness.

9. The infectious disease multi-point triggering intelligent early warning method based on multi-source data fusion according to claim 8 is characterized in that: The analysis process of the situation where the matching degree between the local data and the template exceeds the standard in S4 is as follows: For the local data to be screened, standardization is first performed to make different characteristics comparable; For numerical inspection indicators, the Z-score standardization method is used: ,in is the original data value, is the average value of the indicator in local historical data, is the standard deviation, which ensures that different inspection indicators are compared on the same scale through standardization methods; For symptom data, it is converted into vector form to facilitate subsequent calculations. At the same time, for the onset order of symptoms, different weights are given to symptoms in different positions to reflect their order information; The same standardization and vector construction operations are performed on the data in the complete characteristic portrait template of infectious diseases; the symptom details and the characteristic interval information of the inspection indicators in the template are converted into a computable vector form; for the symptom types, the vector is constructed in the manner of the symptom dictionary; for the characteristic interval of the inspection indicator, the midpoint value of the interval is taken for standardization and then included in the vector; The feature vector of the local data and the feature vector of the template data are calculated by Euclidean distance and cosine similarity to obtain two similarities, which are respectively denoted as Euclidean distance and cosine similarity , and assign weights to , after normalization, enter the following formula: To obtain the comprehensive similarity SDF, is the preset maximum value of the Euclidean distance; The calculated comprehensive similarity SDF is compared with the set 70% similarity threshold. When SDF ≥ 0.7, it is judged that the matching degree between the local data and the template exceeds the standard.

Images