Disease monitoring and early warning method based on multiple dimensions

Abstract

The invention discloses a disease monitoring and early warning method based on multiple dimensions, and relates to the technical field of data processing. Obtaining multi-dimensional data to generate a multi-dimensional data set; establishing a historical disease source network according to the multi-dimensional data set, and acquiring existing case data to establish a real-time disease source network; comparing the real-time disease source network with a plurality of historical disease source networks according to a hierarchical comparison principle to obtain the most similar historical disease source; marking the key area around each real-time mesh node according to a geographic information system technology to form punctuation data; continuously monitoring the punctuation data, and defining an abnormal period and an initial period; preliminarily determining disease source areas and suspected disease source nodes, and performing cross identification on infection sources; through integration of multi-dimensional data and a weighted fusion algorithm, the accuracy and robustness of disease early warning are improved, the mode difference with existing diseases is accurately recognized, prevention and control measures are made based on the most similar historical disease source network, and the effects of accurate recognition and targeted prevention and control are achieved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a disease monitoring and early warning method based on multiple dimensions. Background Art

[0002] With the rapid development of smart healthcare, the role of monitoring and early warning systems in the field of epidemiology has become increasingly important. Traditional disease monitoring methods rely on manual data collection and analysis, which is not only inefficient but also prone to false warnings due to information delays or errors. In order to overcome these limitations, smart healthcare monitoring and early warning systems have been proposed to achieve automated monitoring and timely early warning of disease transmission.

[0003] The Chinese invention patent with application number 202311679085.6 provides an intelligent medical monitoring and early warning method based on disease diagnosis data. It uses the epidemic investigation trend thermal image blocks of the first x-1 case epidemic investigation events to determine the trend thermal matching vector for the x-th case epidemic investigation event in the epidemic investigation trend thermal image stream. Based on the feature differences between P trend thermal embedding vectors and the trend thermal matching vectors of the x-th case epidemic investigation event, the epidemic investigation trend thermal image block of the x-th case epidemic investigation event is determined to obtain the judgment result.

[0004] In the prior art patent, new case epidemiological events are identified and compared based on the target knowledge vector generated by existing cases and historical data, and the judgment results are obtained for prevention and control, which solves the problem of delay in traditional manual data reporting. However, the accuracy of the judgment results needs to be based on a large amount of existing case data, which makes it impossible to effectively monitor and warn of infectious diseases in the early stage, delaying the prevention and control time. Summary of the invention

[0005] This application solves the problem that the existing technology cannot effectively monitor and warn infectious diseases in the early stages of the outbreak by providing a multi-dimensional disease monitoring and early warning method, and achieves the technical effect of accurately identifying infectious diseases in the early stages of the outbreak for monitoring and early warning.

[0006] The present application provides a multi-dimensional disease monitoring and early warning method, the method comprising: S100: Acquire multidimensional data to generate a multidimensional data set, wherein the multidimensional data set includes a plurality of multidimensional feature vectors; S200: Establishing a historical pathogenic source network based on a multidimensional data set, acquiring existing case data to establish a real-time pathogenic source network, wherein the historical pathogenic source network uses fused feature vectors of multiple cases as historical mesh nodes, and uses disease change differences as edges to connect them; the real-time pathogenic source network uses multidimensional feature vectors of a single case as real-time mesh nodes, and uses change differences of different cases as edges to connect them; the fused feature vector is generated by fusing the multidimensional feature vectors of multiple cases; S300: Compare the real-time pathogen source network with multiple historical pathogen source networks according to the hierarchical comparison principle, obtain the most similar historical pathogen sources, set corresponding prevention and control measures, and conduct monitoring and early warning; S400: Mark key areas around each real-time mesh node based on geographic information system technology to form punctuation data, assign a unique identifier to each punctuation data, and record its basic information; continuously monitor the punctuation data, and define the abnormal period and the initial period; preliminarily determine the source area and suspected source nodes, and cross-identify the source of infection.

[0007] Furthermore, the hierarchical comparison principle is to compare the real-time pathogenic source network with multiple historical pathogenic source networks one by one through different levels, calculate the comprehensive similarity, and select the historical pathogenic source network with the highest comprehensive similarity.

[0008] Furthermore, the hierarchical comparison principle also includes: sorting the historical pathogen source network in descending order according to the comprehensive similarity, pre-setting a similarity threshold, performing spatiotemporal overlap analysis on the historical pathogen source network above the similarity threshold and the real-time pathogen source network to obtain several suspected paths of the predecessor key nodes, screening the suspected paths according to a pre-set conditional filtering mechanism, and calculating the difference between multiple suspected paths and actual propagation paths; calculating the estimated path from the terminal propagation node to the next real-time mesh node based on the difference value, and presetting prevention and control measures.

[0009] Furthermore, the conditional filtering mechanism includes mandatory constraints and flexible evaluation conditions. All suspected paths are preliminarily screened according to the mandatory constraints. The suspected paths after the preliminary screening are evaluated in sections according to the flexible evaluation conditions, and an evaluation score is given. The evaluation score is multiplied by its corresponding weight to obtain a weighted score for the condition. All weighted scores are added together to obtain the final weight of the suspected path. The mandatory constraints include time consistency, regional rationality and type matching; the flexible evaluation conditions include quantity similarity, speed consistency and reference to other factors.

[0010] Furthermore, the spatiotemporal overlap analysis regards the real-time pathogenic source network and the historical pathogenic source network as a graph structure, selects the most similar real-time mesh nodes and historical mesh nodes as comparison starting nodes, uses a graph matching algorithm to perform overlap comparison based on the comparison starting nodes, and identifies the overlapping parts between the real-time pathogenic source network and the historical pathogenic source network.

[0011] Further, the predecessor key node is the penultimate real-time mesh node of the real-time pathogen network, and the terminal propagation node is the last real-time mesh node of the real-time pathogen network.

[0012] Furthermore, the suspected path refers to the development path from multiple historical mesh nodes overlapping with the predecessor key node to the next historical mesh node; the actual propagation path is the propagation change path between the predecessor key node and the terminal propagation node, and the difference value is the comprehensive calculated value of multiple suspected paths and the actual propagation path.

[0013] Furthermore, the abnormal period refers to an abnormal situation in punctuation data monitoring within a short period of time, and the initial period refers to a situation in which continuous abnormalities occur over a period of time and case data increases; monitoring rules are set for the abnormal period and the initial period respectively.

[0014] Furthermore, step S400 also includes: performing a pinch-off screening on the area between multiple different real-time mesh nodes around the suspected disease source node, wherein the pinch-off screening is to calculate the distance between different real-time mesh nodes and the distance between the punctuation data between the real-time mesh nodes to form multiple radial areas for secondary screening; identifying the overlapping areas of multiple radial areas, calculating the intersection points of adjacent radial areas, connecting the intersection points to form an overlapping area, and performing a deep screening on the overlapping area; and finally determining the disease area and the source of the disease.

[0015] Furthermore, the secondary screening is to conduct an initial screening of the real-time mesh nodes in the radiation area, and evaluate the real-time mesh nodes after the initial screening according to the predetermined node risk level evaluation index to obtain high-risk nodes and high-risk areas; the deep screening is to use a clustering algorithm to cluster the real-time mesh nodes in the overlapping area, and divide them into different groups according to the correlation and similarity between the real-time mesh nodes, and apply association rule mining technology to identify the potential connections between different groups, and finally determine the source of the disease and the actual transmission mode.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: Through the integration of multi-dimensional data and weighted fusion algorithms, the accuracy and robustness of disease warnings have been improved. The construction of real-time and historical pathogen source networks can monitor the spread of diseases in real time and accurately identify the differences in patterns from existing diseases. Prevention and control measures can be formulated based on the most similar historical pathogen source networks, achieving the effects of accurate identification and targeted prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the overall process of the multi-dimensional disease monitoring and early warning method in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0020] Embodiment 1: Figure 1 As shown, a multi-dimensional disease monitoring and early warning method comprises: S100: Acquire multidimensional data to generate a multidimensional data set, wherein the multidimensional data set includes a plurality of multidimensional feature vectors.

[0021] In some embodiments, the multi-dimensional data is acquired based on the health and medical big data platform and the municipal government exchange and sharing platform. The acquired multi-dimensional data uses a machine learning algorithm to automatically identify and correct data errors, improve the efficiency and accuracy of data cleaning, and dynamically adjust the standardization method according to the data characteristics to ensure the consistency and comparability of the data at different times and in different situations. Key indicators related to the disease are extracted from the data, and multi-source data are combined to extract multi-dimensional features such as incidence rate, transmission speed, public attention, and meteorological factors. An early warning model based on multi-source data fusion is constructed, and a weighted fusion algorithm is used to improve the accuracy and robustness of the early warning.

[0022] In some embodiments, the multidimensional data set includes medical treatment data, personnel information and drug data, Internet data, disease monitoring subject library data information, etc. The Internet data is the network search data, discussion data, meteorological data, etc. generated in social media, and the disease monitoring subject library data information refers to syndrome monitoring, fever clinic monitoring, risk population monitoring, drug monitoring and disease epidemic intensity monitoring data.

[0023] In some embodiments, statistical analysis is performed based on the number of visits to fever clinics, main symptoms, epidemiological history and other data, and the control chart is drawn in accordance with the control chart-related algorithm in a combination of average values ​​and control charts, and abnormal points are identified in the statistical charts through control rules. At the same time, intelligent early warning analysis models are generated according to different symptoms, different moving average methods and different rules to achieve real-time monitoring of mass and high-incidence populations and symptoms. Early warning of infectious diseases and epidemic intensity monitoring are achieved through drug retail information monitoring of pharmacies. Analysis is carried out on the shared daily sales data of medical insurance drugs in retail pharmacies, specifically for the daily sales of cold and fever medicines, cough medicines, antidiarrhea medicines, and anti-inflammatory drugs. The EARS-3Cs early warning model is used to monitor and warn based on the sales of the previous 7 days. At the same time, the amount of TOP drugs prescribed in hospital prescription drugs is monitored, focusing on the ranking and dosage trends of hormones, anticoagulants, and antibiotics.

[0024] In some embodiments, the multidimensional dataset includes several multidimensional feature vectors, which are vectors containing multiple features (or attributes), which together describe multiple aspects of an entity or event. For infectious disease monitoring and early warning systems, multidimensional feature vectors contain various information related to disease transmission, such as incidence, transmission speed, public attention, meteorological factors, etc. For example, a multidimensional feature vector containing the following features is constructed: date xxx; region xxx; incidence xxx; transmission speed xxx (relative value); public attention xxx (number of mentions on social media); average temperature xxx; relative humidity xxx; cold medicine sales xxx; number of hospital visits on the same day xxx; TOP drug prescription volume xxx; In this example, the multidimensional feature vector contains 11 features, each of which has a corresponding feature value, and these features together describe the disease development status of the region on a specific date.

[0025] S200: Establish a historical pathogenicity network based on a multidimensional data set, and obtain existing case data to establish a real-time pathogenicity network, wherein the historical pathogenicity network uses fused feature vectors of multiple cases as historical mesh nodes, and uses disease change differences as edges to connect them; the real-time pathogenicity network uses multidimensional feature vectors of single cases as real-time mesh nodes, and uses change differences between different cases as edges to connect them.

[0026] In some embodiments, the historical pathogenicity network is a network constructed based on historical case data, which is used to analyze the historical patterns and trends of disease transmission, including a complete record of all disease transmission processes from start to end that have occurred. Relevant case data, including basic information, diagnosis information, treatment information, etc. of patients, are collected from the historical case database. The collected data are cleaned and standardized to ensure data quality and consistency. Features related to disease transmission, such as patient location, onset time, contact history, etc., are extracted from the cleaned data. Based on the extracted features, a historical pathogenicity network is constructed using graph theory or network science methods, and each disease generates a corresponding historical pathogenicity network.

[0027] In the historical pathogenic network, the fused feature vectors of multiple cases are used as historical mesh nodes. The fused feature vector is generated by fusing the multidimensional feature vectors of multiple cases. Assuming that there are multidimensional feature vectors of n cases, each vector contains m features, the fused feature vector F can be expressed as:

[0028] in, Represents the jth eigenvalue of the ith case. The average value of each feature in the multidimensional feature vectors of multiple cases is obtained and combined to obtain a fused feature vector. The historical mesh node represents the comprehensive characteristic status of multiple cases at a specific time node. Under this node, the comprehensive characteristic vector obtained according to the relevant disease conditions of multiple cases is connected with the disease change difference as an edge. The disease change difference refers to the existence of one or more change differences in the comprehensive characteristic vectors of two historical mesh nodes. The change difference can be determined by calculating the Euclidean distance, cosine similarity and other metrics between the two fused characteristic vectors. The historical pathogenic source network is analyzed to identify the key nodes and paths of disease transmission and discover potential transmission laws and patterns. In the real-time pathogenic source network, the information characteristic vector of a single case is used as a real-time mesh node, and the change difference of different cases is used as an edge for connection. A real-time mesh node represents an actual case data information. The actual case data information should at least include the date, location, onset time, contact history, basic disease information, medical records, adverse reactions, etc. The connection between different real-time mesh nodes is the edge, which represents the change difference between the two. The change difference can also be determined by calculating the similarity or distance between two multidimensional feature vectors. For example, the difference between two feature vectors is calculated using the Euclidean distance:

[0029] in, and denote eigenvectors 1 and 2 respectively, and represents the jth eigenvalue in the two vectors respectively, and m represents the total number of features. The historical mesh node and the real-time mesh node are both connected with a plurality of edges.

[0030] S300: Compare the real-time pathogen source network with multiple historical pathogen source networks according to the layered comparison principle to obtain the most similar historical pathogen sources, set corresponding prevention and control measures, and conduct monitoring and early warning.

[0031] In the prior art, several multidimensional feature vectors are obtained based on the historical pathogenic network, and the multiple multidimensional feature vectors are directly obtained and compared with the case feature vectors generated by the current newly identified case to obtain feature differences. Feature differences refer to the differences in each feature dimension between different data entities. The most similar known disease pattern is selected based on the different differences, and then the discrimination results are obtained for prevention and control.

[0032] The hierarchical comparison principle is to compare the real-time pathogen source network with multiple historical pathogen source networks one by one through different levels, calculate the comprehensive similarity, and select the historical pathogen source network with the highest similarity.

[0033] In some embodiments, the settings are divided into a time layer, a regional layer, and a feature layer to ensure that the data of all historical pathogenic source networks are complete and standardized. From the time layer: according to the timestamp of the real-time pathogenic source network, a reasonable time window (such as the past year, half a year, or a specific season) is determined, and the historical pathogenic source network within the time window is selected as a preliminary candidate set, and the historical pathogenic source network with the same or similar disease type as the real-time pathogenic source network is further screened; from the regional layer: according to the region where the real-time pathogenic source network is located, the geographical region is divided into several sub-regions. In the preliminary candidate set, the historical pathogenic source network with the same or similar geographical features as the region where the real-time pathogenic source network is located is retained; from the feature layer: according to the needs of disease monitoring and early warning, key features are selected for comparison, such as incidence rate, transmission speed, public attention, meteorological factors, etc. For each feature, the cosine similarity can be used to calculate the similarity between the real-time pathogenic source network and the candidate historical pathogenic source network. Assuming that the feature vector of the real-time pathogenic source network is , the characteristic vector of the historical disease source network is , the directional similarity of two vectors in multidimensional space The calculation formula is as follows:

[0034] Considering the similarity of all features, a weighted average method is used to obtain a comprehensive similarity score. According to the comprehensive similarity score, the candidate historical pathogenic source networks are sorted, and the historical pathogenic source network with the highest similarity is selected as the most similar historical pathogenic source. In-depth analysis of the disease transmission pattern and trend of the most similar historical pathogenic source network. Based on the analysis results and combined with the current actual situation, targeted prevention and control measures are formulated.

[0035] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application improves the accuracy and robustness of disease warning through the integration of multi-dimensional data and weighted fusion algorithms. The construction of real-time and historical pathogen source networks can monitor the spread of diseases in real time and accurately identify the differences in patterns from existing diseases. Prevention and control measures are formulated based on the most similar historical pathogen source networks, achieving the effects of accurate identification and targeted prevention and control.

[0036] Embodiment 2: In embodiment 1, the accuracy and reliability of disease warning are improved by constructing a multi-dimensional feature vector and adopting a weighted fusion algorithm. This embodiment makes further improvements on the basis of the above embodiment.

[0037] The hierarchical comparison principle also includes: sorting the historical pathogen source network in descending order according to the comprehensive similarity, pre-setting a similarity threshold, performing spatiotemporal overlap analysis on the historical pathogen source network that is higher than the similarity threshold and the real-time pathogen source network, obtaining several suspected paths of the predecessor key nodes, screening the suspected paths according to a pre-set conditional filtering mechanism, and calculating the difference between multiple suspected paths and actual propagation paths; calculating the estimated path from the terminal propagation node to the next real-time mesh node based on the difference value, and presetting prevention and control measures.

[0038] In some embodiments, a plurality of historical pathogenic source networks after hierarchical comparison are obtained according to the above content, and are sorted in descending order according to the comprehensive similarity of each historical pathogenic source network. A similarity threshold is pre-set according to historical experience and experimental data, and a spatiotemporal overlap analysis is performed on the historical pathogenic source network with a similarity threshold higher than the similarity threshold and the real-time pathogenic source network. The spatiotemporal overlap analysis is to regard the real-time pathogenic source network and the historical pathogenic source network as graph structures, and select the most similar real-time mesh nodes and historical mesh nodes as comparison starting nodes. The similarity calculation here is calculated based on the specific vector data of the real-time mesh nodes and the historical mesh nodes, including analysis and comparison of different levels (time layer, regional layer and feature layer), and the most similar one is selected. In this embodiment, the real-time mesh node is the first node in the real-time pathogenic source network, but the selected historical mesh node is not necessarily the first node in the historical pathogenic source network, and it only needs to be selected according to the similarity; a graph matching algorithm is used to perform overlap comparison based on the comparison starting node to identify the overlapping part between the real-time pathogenic source network and the historical pathogenic source network.

[0039] In some embodiments, the method for setting the similarity threshold is: select multiple historical pathogenic source networks. Since the disease information of the historical pathogenic source network is known information data, the same method is used to judge the similarity between the multiple historical pathogenic source networks to obtain the comprehensive similarity of the historical pathogenic source networks. The appropriate similarity threshold is set through the actual existing similarity calculation to ensure that the historical pathogenic source network with effective comparison can be selected. In order to ensure the analysis efficiency of the historical pathogenic source network and the real-time pathogenic source network, the number of historical pathogenic source networks above the similarity threshold is calculated. If the number is large, for example, it exceeds the preset maximum selection number, a dynamic adjustment method is adopted, for example, the first 25% of the total number is selected; the setting of the maximum selection number also adopts the above-mentioned similarity threshold method, and the number is set by using the analysis of the historical pathogenic source network to ensure the comparison effect while improving the comparison efficiency.

[0040] In some embodiments, the suspected path refers to the development path from multiple historical mesh nodes overlapping with the predecessor key node to the next historical mesh node; the actual propagation path is the propagation change path between the predecessor key node and the terminal propagation node, and the difference value is the comprehensive calculation value of multiple suspected paths and the actual propagation path; the predecessor key node is the second to last real-time mesh node of the real-time pathogen network, and the terminal propagation node is the last real-time mesh node of the real-time pathogen network. Assume that the actual development path of the real-time pathogen network is ,in represents the i-th node on the path, and the suspected path obtained by the historical pathogenic source network is ,in represents the first suspected path, Represents the i-th node on the suspected path. The difference value is calculated based on the node similarity method, and the node similarity function is defined as , indicating a node and The similarity between the two nodes, the path difference value D is defined as the negative average of the similarities of the corresponding nodes, and the calculation formula is as follows: ,in The value range of is (0, 1), which means completely dissimilar to completely identical. The smaller the difference value D is, the more similar the two paths are.

[0041] In some embodiments, the difference value threshold is preset. , used to judge the similarity between the suspected path and the actual path, and select the path with a difference value less than The path is taken as a candidate path. For each candidate path, check the next node after the corresponding terminal propagation node. , obtain the change trend of the previous and next nodes, and estimate the next change trend of the real-time pathogen network.

[0042] Another method for screening suspected paths is also provided in the present application, using a conditional screening method, setting a conditional filtering mechanism, the conditional filtering mechanism includes mandatory constraints and elastic evaluation conditions, the mandatory constraints include time consistency, regional rationality and type matching; the elastic evaluation conditions include quantity similarity, speed consistency and reference to other factors; all suspected paths are preliminarily screened according to the mandatory constraints, and the suspected paths after the preliminary screening are evaluated in sections according to the elastic evaluation conditions, and an evaluation score is given, and the evaluation score is multiplied by its corresponding weight to obtain the weighted score of the condition, and all weighted scores are added to obtain the final weight of the suspected path; the mandatory constraints represent the basic principles and laws that cannot be violated in the analysis of disease transmission, which are based on scientific research and actual data, and must be strictly followed to ensure the accuracy and reliability of the analysis results. For example, time consistency, geographical location rationality and virus type matching are all mandatory constraints; the elastic evaluation condition is an auxiliary indicator used to further evaluate the rationality and possibility of the suspected path on the basis of the satisfaction of the mandatory constraints, providing additional information, which can more comprehensively understand the potential patterns and factors of disease transmission. For example, the similarity of case numbers, consistency of transmission speed, and environmental factors can all be considered as elasticity evaluation conditions. Assuming that the suspected path is P, the set of mandatory constraints is , the elastic evaluation condition set is , and the corresponding weight settings are and , the condition scores are and ; The final weight calculation formula is:

[0043] in represents the weighted sum of the mandatory constraints, Represents the sum of weighted scores of the resilience assessment criteria.

[0044] In some embodiments, an estimated path from the terminal propagation node to the next real-time mesh node is calculated based on the difference value, and prevention and control measures are preset. The estimated path is an estimate of the next node of the current real-time pathogen network (i.e., a new case or a new symptom change of the current case), and prevention and control means and related measures are implemented based on the estimated results.

[0045] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application uses spatiotemporal overlap analysis and path difference value calculation to more accurately identify historical pathogen networks similar to real-time pathogen networks, thereby improving the accuracy of disease warnings; based on estimated paths and conditional filtering mechanisms, it can more accurately formulate prevention and control measures and improve prevention and control effects; by setting similarity thresholds and maximum selection numbers, it reduces invalid comparisons, improves analysis efficiency, and achieves the effect of accurately judging disease transmission patterns and making accurate predictions for early prevention.

[0046] Embodiment 3: In the above embodiment, the identification and monitoring of the disease is realized, and the transmission mode and onset characteristics of the disease are determined. This embodiment makes further improvements on the basis of the above embodiment.

[0047] The method also includes: S400: marking key areas around each real-time mesh node according to geographic information system technology to form punctuation data, assigning a unique identifier to each punctuation data, and recording its basic information; continuously monitoring the punctuation data, and defining abnormal periods and initial periods; preliminarily determining the source area and suspected source nodes, and cross-identifying the source of infection.

[0048] In some embodiments, the key areas refer to drug stores, clinics, hospitals, etc. around the cases. The basic information of each punctuation data includes information data such as the geographical location, type, scale, number of mobile personnel, etc. The distance selection of the punctuation data should be based on the comprehensive consideration of the actual possibility of disease transmission and monitoring efficiency. In the above embodiments, the current disease pattern monitoring and development path estimation are given, which is still based on the data foundation of the cases. When there are few case data, basic monitoring and prevention should be the main focus. In this embodiment, the key areas are marked to form punctuation data, and the occurrence of abnormal situations is discovered in time according to the real-time monitoring of the punctuation data. The abnormal period refers to the stage in which the punctuation data monitoring shows abnormal conditions in a short period of time, but the emergence of infectious diseases has not yet been confirmed. Abnormal phenomena include a surge in drug sales and an increase in cases of specific symptoms. According to specific experiments, a short period of time refers to a continuous abnormality in punctuation data within a week, which is often regarded as the early stage of the disease. The initial stage refers to a continuous abnormality in a period of time and an increase in case data. The initial stage is determined when the abnormality lasts for a period of time, which is about one to three months, and is accompanied by a significant increase in case data, which basically determines that the infectious disease has appeared, such as the emergence of a large number of cases with the same symptoms and a surge in the number of hospital admissions. The method for distinguishing between the abnormal period and the initial stage is to set a monitoring threshold. According to historical data and expert experience, the monitoring threshold is set for different types of punctuation data (such as the growth rate of drug sales, the number of cases, etc.); real-time monitoring and comparison, by real-time monitoring of the punctuation data, and comparing it with the set threshold, to distinguish the abnormal period from the initial stage, the sales volume of specific drugs and the number of cases at normal times are set as the safe period, and the distinction is made by comparing the differences; combined with the monitoring of multiple punctuation data, a comprehensive analysis is conducted to improve the accuracy of the distinction.

[0049] In some embodiments, monitoring rules are set for the abnormal period and the initial period respectively. For abnormal period monitoring, the monitoring frequency of punctuation data should be strengthened, real-time attention should be paid to the development and changes of abnormal phenomena, preliminary investigation and analysis should be conducted on abnormal phenomena, emergency preparations should be made, and the monitoring state of the initial period should be ready at any time; for initial period monitoring, the monitoring frequency should be further increased to ensure the real-time and accuracy of the data, and key monitoring should be carried out on suspected pathogen source nodes. Combined with the monitoring methods of the above-mentioned embodiments one and two, the characteristics and transmission mode of the current disease should be determined according to the historical pathogen source network, the development path of the next stage should be estimated, and further precise prevention and control should be carried out.

[0050] In some embodiments, the preliminary determination of the pathogenic area and suspected pathogenic node is set for the initial outbreak period. During the initial outbreak period, according to the monitoring of the punctuation data, combined with the data of the real-time mesh nodes, the trend and path of disease transmission are analyzed, and the pattern and characteristics of the disease are determined in combination with the technical solution implemented above, and the prevention and control strategy is set in a targeted manner. Through GIS technology, the punctuation data and the real-time mesh nodes are spatially superimposed and analyzed to identify the pathogenic area and suspected pathogenic node. The pathogenic area usually refers to the area where the disease is most concentrated, and the suspected pathogenic node may be the place where the disease first appeared; combined with the data information of the historical pathogenic network, the suspected pathogenic node is further analyzed, and by comparing the monitoring data at different time points, the transmission pattern and law of the infectious disease are identified, and the multi-dimensional information such as geographical location, population mobility, and environmental factors are comprehensively analyzed to cross-identify the source of infection.

[0051] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application uses GIS technology to mark key areas and conduct continuous monitoring, which can more accurately capture early signs of disease transmission and improve monitoring accuracy and efficiency; by defining abnormal periods and initial stages, it can timely identify different stages of disease transmission and achieve early warning and pre-emptive prevention and control effects; through GIS spatial overlay analysis and multi-dimensional information comprehensive analysis, it can achieve the effect of accurately locating disease source areas and suspected disease source nodes.

[0052] Embodiment 4: In the above embodiment, punctuation data is set based on multi-dimensional data information for real-time monitoring, which can timely detect disease abnormalities and then carry out prevention and control. This embodiment makes further improvements on the basis of the above embodiment.

[0053] Step S400 also includes: performing a pinch-off screening on the area between multiple different real-time mesh nodes around the suspected disease source node, wherein the pinch-off screening is to calculate the distance between different real-time mesh nodes and the distance between the punctuation data of the real-time mesh nodes to form multiple radial areas for secondary screening; identifying the overlapping areas of multiple radial areas, calculating the intersections of adjacent radial areas, connecting the intersections to form overlapping areas, and performing in-depth screening on the overlapping areas; and finally determining the disease area and the source of the disease.

[0054] In some embodiments, based on the third embodiment, the key area has been marked and punctuation data has been formed, and its basic information has been recorded. At the same time, in the mesh structure, the suspected pathogen source node and multiple different mesh nodes around it are determined, and the punctuation data between the suspected pathogen source node and the remaining nodes are calculated, and the distance between the suspected pathogen source node and the punctuation data is calculated in turn, and the distance between the suspected pathogen source node and the punctuation data is used as the radius to form a radial area, and the radial area is screened twice; the overlapping area of ​​multiple radial areas is identified, the intersection of adjacent radial areas is calculated, and the intersection is connected to form an overlapping area, and the overlapping area is deeply screened; for example, the suspected pathogen source node and the node There are multiple pharmacies in the area between a and pharmacies. When obtaining pharmacies that sell drugs for related diseases, suspected pharmacies 1, 2, 3, and 4 are found. The distance between the suspected source node and pharmacy 1 is calculated, and pharmacy 1 is used as the origin and the distance is used as the radius for searching to form radial area 1. The mesh nodes in the radial area are screened twice. Similarly, the distance between pharmacy 1 and pharmacy 2 is calculated, and pharmacy 2 is used as the origin and the distance is used as the radius for searching to form radial area 2. The intersection area of ​​the front and rear radial area 1 and radial area 2 is set as the overlapping area, and the intersection points on both sides of the overlapping area are connected. The data information in the overlapping area is deeply screened to finally determine the source of the disease.

[0055] In some embodiments, the secondary screening is to perform a preliminary screening of the real-time mesh nodes in the radiation zone. The preliminary screening is to exclude the real-time mesh nodes that are irrelevant to the disease or have little impact. The real-time mesh nodes after the preliminary screening are evaluated according to the predetermined node risk level evaluation index to obtain high-risk nodes and high-risk areas; the deep screening is to use a clustering algorithm to cluster the real-time mesh nodes in the overlapping area, and divide them into different groups according to the correlation and similarity between the real-time mesh nodes. The association rule mining technology is used to identify the potential connection between different groups, and finally determine the source of the disease and the actual mode of transmission, which helps to identify nodes with similar characteristics, which may be the key path of disease transmission. The secondary screening is mainly based on the weighted scoring method to evaluate the risk level of the node, which is an extension of the preliminary screening, while the deep screening pays more attention to the correlation and similarity between the nodes, as well as the trend of data changes. It is a more in-depth analysis based on the secondary screening. The deep screening is based on the LSTM neural network to evaluate the data change trend. The time series data of each node in the overlapping area (such as the number of daily cases, population flow data, etc.) are preprocessed, including data cleaning, normalization, etc., to determine the number of layers of the LSTM network and the number of neurons in each layer, select the activation function (such as ReLU, Sigmoid, etc.) and the loss function (such as mean square error MSE), use the training data to train the LSTM model, optimize the model parameters through the back propagation algorithm, use the trained LSTM model to predict the test data, evaluate the prediction performance of the model, analyze the change trend of each node data according to the prediction results, and identify the nodes with abnormal changes or trend reversals; the calculation formula is:

[0056] in Represents the hidden state at time step t, which is a vector containing the internal state information of the LSTM unit at the current time step; Represents the activation function, and different activation functions are used for calculation according to different actual situations; The weight matrix representing the hidden state to the hidden state determines how the hidden state of the previous time step affects the hidden state of the current time step; represents the hidden state of the previous time step; The weight matrix representing the input data to the hidden state determines how the input data at the current time step affects the hidden state; The input data at time step t is a vector containing the external information received by the LSTM unit at the current time step; Represents the bias term, which is used to adjust the calculation of the hidden state.

[0057] In some embodiments, multi-dimensional data information (such as geographic location, population flow, environmental factors, etc.) is combined with mesh node data analysis to set a screening mode. In overlapping areas, the source of the disease is quickly located through data comparison and trend analysis. Based on the located source of the disease, the disease area is locked. Combined with GIS technology, targeted prevention and control measures are formulated, such as resource allocation, isolation area setting, personnel flow control, etc., to continuously monitor the prevention and control effects, and adjust the prevention and control strategies in a timely manner according to the monitoring results.

[0058] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: This application uses cut-off screening, secondary screening and deep screening, combined with data change trend evaluation of LSTM neural network, to more accurately locate the source and transmission path of the disease; combined with multi-dimensional data information and GIS technology, it can formulate more accurate and effective prevention and control measures, improve prevention and control efficiency, and achieve the effect of accurately locating the source of the disease and the source area.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-dimensional disease monitoring and early warning method, characterized in that: The method comprises: S100: Acquire multidimensional data to generate a multidimensional data set, wherein the multidimensional data set includes a plurality of multidimensional feature vectors; S200: Establishing a historical pathogenic source network based on a multidimensional data set, acquiring existing case data to establish a real-time pathogenic source network, wherein the historical pathogenic source network uses fused feature vectors of multiple cases as historical mesh nodes, and uses disease change differences as edges to connect them; the real-time pathogenic source network uses multidimensional feature vectors of a single case as real-time mesh nodes, and uses change differences of different cases as edges to connect them; the fused feature vector is generated by fusing the multidimensional feature vectors of multiple cases; S300: Compare the real-time pathogen source network with multiple historical pathogen source networks according to the hierarchical comparison principle, obtain the most similar historical pathogen sources, set corresponding prevention and control measures, and conduct monitoring and early warning; S400: Mark key areas around each real-time mesh node based on geographic information system technology to form punctuation data, assign a unique identifier to each punctuation data, and record its basic information; continuously monitor the punctuation data, and define the abnormal period and the initial period; preliminarily determine the source area and suspected source nodes, and cross-identify the source of infection.

2. The multi-dimensional disease monitoring and early warning method according to claim 1, characterized in that: The hierarchical comparison principle is to compare the real-time pathogen source network with multiple historical pathogen source networks one by one through different levels, calculate the comprehensive similarity, and select the historical pathogen source network with the highest comprehensive similarity.

3. The multi-dimensional disease monitoring and early warning method according to claim 1, characterized in that: The hierarchical comparison principle also includes: sorting the historical pathogen source network in descending order according to the comprehensive similarity, pre-setting a similarity threshold, performing spatiotemporal overlap analysis on the historical pathogen source network that is higher than the similarity threshold and the real-time pathogen source network, obtaining several suspected paths of the predecessor key nodes, screening the suspected paths according to a pre-set conditional filtering mechanism, and calculating the difference between multiple suspected paths and actual propagation paths; calculating the estimated path from the terminal propagation node to the next real-time mesh node based on the difference value, and presetting prevention and control measures.

4. The multi-dimensional disease monitoring and early warning method according to claim 3, characterized in that: The conditional filtering mechanism includes mandatory constraints and flexible evaluation conditions. All suspected paths are preliminarily screened according to the mandatory constraints. The suspected paths after the preliminary screening are evaluated in sections according to the flexible evaluation conditions, and evaluation scores are given. The evaluation scores are multiplied by their corresponding weights to obtain the weighted scores of the conditions. All weighted scores are added together to obtain the final weight of the suspected path. The mandatory constraints include time consistency, regional rationality and type matching. The flexible evaluation conditions include quantity similarity, speed consistency and reference to other factors.

5. The multi-dimensional disease monitoring and early warning method according to claim 3, characterized in that: The spatiotemporal overlap analysis regards the real-time pathogenicity network and the historical pathogenicity network as a graph structure, selects the most similar real-time mesh nodes and historical mesh nodes as comparison starting nodes, uses a graph matching algorithm to perform overlap comparison based on the comparison starting nodes, and identifies the overlapping parts between the real-time pathogenicity network and the historical pathogenicity network.

6. The multi-dimensional disease monitoring and early warning method according to claim 3, characterized in that: The predecessor key node is the penultimate real-time mesh node of the real-time pathogen network, and the terminal propagation node is the last real-time mesh node of the real-time pathogen network.

7. The multi-dimensional disease monitoring and early warning method according to claim 3, characterized in that: The suspected path refers to the development path from multiple historical mesh nodes overlapping with the predecessor key node to the next historical mesh node; the actual propagation path is the propagation change path between the predecessor key node and the terminal propagation node, and the difference value is the comprehensive calculated value of multiple suspected paths and the actual propagation path.

8. The multi-dimensional disease monitoring and early warning method according to claim 1, characterized in that: The abnormal period refers to abnormal conditions in punctuation data monitoring within a short period of time, and the initial period refers to continuous abnormal conditions within a period of time and an increase in case data; monitoring rules are set for the abnormal period and the initial period respectively.

9. The multi-dimensional disease monitoring and early warning method according to claim 1, characterized in that: Step S400 also includes: performing a pinch-off screening on the area between multiple different real-time mesh nodes around the suspected disease source node, wherein the pinch-off screening is to calculate the distance between different real-time mesh nodes and the distance between the punctuation data of the real-time mesh nodes to form multiple radial areas for secondary screening; identifying the overlapping areas of multiple radial areas, calculating the intersections of adjacent radial areas, connecting the intersections to form overlapping areas, and performing in-depth screening on the overlapping areas; and finally determining the disease area and the source of the disease.

10. The multi-dimensional disease monitoring and early warning method according to claim 9, characterized in that: The secondary screening is to perform a preliminary screening of the real-time mesh nodes in the radiation area, and to evaluate the real-time mesh nodes after the preliminary screening according to a predetermined node risk level evaluation index to obtain high-risk nodes and high-risk areas; The deep screening is to cluster the real-time mesh nodes in the overlapping area using a clustering algorithm, divide them into different groups according to the correlation and similarity between the real-time mesh nodes, apply association rule mining technology to identify the potential connections between different groups, and finally determine the source of the disease and the actual mode of transmission.

Images