Regional medical big data-oriented infectious disease space-time transmission monitoring and early warning method

Abstract

The invention relates to the technical field of digital medical treatment, in particular to a regional medical big data-oriented infectious disease spatio-temporal propagation monitoring and early warning method, which comprises the following steps of: acquiring multi-source data based on a data source, performing data integration to obtain continuous spatio-temporal data, and constructing a behavior chain database according to the continuous spatio-temporal data; dividing monitoring areas and counting crowd density corresponding to each monitoring area in combination with the behavior chain database; analyzing the behavior chain database to extract key knowledge data associated with the monitoring area, converting the key knowledge data into knowledge triples, and storing a plurality of knowledge triples in a graph database; traversing the graph database to obtain crowd behavior knowledge so as to construct a behavior graph database, predicting change characteristics of crowd density based on a neural network learning model, and analyzing propagation risks based on the change characteristics in combination with the behavior database so as to carry out corresponding early warning. The method has the effect of improving the accuracy and timeliness of infectious disease transmission monitoring and early warning.

Description

Technical Field

[0001] The present application relates to the technical field of digital medicine, and in particular to a method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data. Background Art

[0002] Smart healthcare uses advanced information technology and data analysis methods to promote the modernization of the medical and health field, especially in infectious disease monitoring and early warning, and has been widely used in recent years.

[0003] In traditional infectious disease monitoring and prevention systems, rule-based expert systems or simple statistical methods are usually relied on to analyze the spread of infectious diseases. These systems generally build simple prediction models based on historical data, or judge the current spread through regular population sampling and monitoring reports.

[0004] Traditional methods have great limitations in data collection and processing. First, the data source is single and can only rely on data provided by hospitals or public health departments. There is a lack of comprehensive analysis of multi-dimensional information such as patient flow, gathering areas, and crowd behavior. In addition, traditional models often lack sufficient flexibility and accuracy when dealing with complex spatiotemporal dynamic changes, and cannot respond to the rapidly changing infectious environment in real time. At the same time, many traditional systems rely too much on empirical judgments and find it difficult to achieve data-driven accurate predictions, resulting in the timeliness and accuracy of their warnings failing to meet the needs of emergency response. Summary of the invention

[0005] In order to improve the accuracy and timeliness of infectious disease transmission monitoring and early warning, the present application provides an infectious disease spatiotemporal transmission monitoring and early warning method for regional medical big data.

[0006] This application provides a method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data, which adopts the following technical solutions:

[0007] A method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data, comprising the following steps:

[0008] Acquire multi-source data based on data sources and integrate the data to obtain continuous spatiotemporal data, and construct a behavior chain database based on the continuous spatiotemporal data, wherein the behavior chain database is composed of behavior chains corresponding to a number of patients independently;

[0009] Divide the monitoring areas and count the crowd density corresponding to each monitoring area in combination with the behavior chain database;

[0010] Analyzing the behavior chain database to extract key knowledge data associated with the monitoring area and converting the key knowledge data into knowledge triples, and storing a number of the knowledge triples in a graph database;

[0011] The graph database is traversed to obtain crowd behavior knowledge to construct a behavior graph database, the change characteristics of the crowd density are predicted based on a neural network learning model, and the transmission risk is analyzed based on the change characteristics combined with the behavior database to make corresponding warnings.

[0012] In some embodiments, acquiring multi-source data based on data sources and integrating the data to obtain continuous spatiotemporal data, and constructing a behavior chain database based on the continuous spatiotemporal data include the following steps:

[0013] Obtain hospital data tables and extract patient information and their associated visit information and sort them by time to generate a continuous visit record;

[0014] Acquire the travel data associated with the patient information, extract the residence information corresponding to the stay time exceeding the preset time based on the travel data to construct a residence chain, and extract the travel information corresponding to the stay time exceeding the preset time based on the travel data to construct a travel chain;

[0015] When the sum of the stay times corresponding to the residence chain and the travel chain is greater than a preset time threshold, the residence chain and the travel chain are merged to form the behavior chain associated with the patient information, wherein the behavior chain includes behavior location information and stay time;

[0016] The behavior chains corresponding to the patient information are integrated into a database to obtain the behavior chain database.

[0017] In some embodiments, dividing the monitoring areas and counting the crowd density corresponding to each monitoring area in combination with the behavior chain database includes the following steps:

[0018] Dividing each area based on a division type option to obtain a number of monitoring areas of corresponding results, wherein the division type option includes an area option, a function option, and a representative feature option;

[0019] Based on the mapping of each behavior location information stored in the behavior chain database with the monitoring area to calculate the number of overlapping patients, the crowd density corresponding to each monitoring area in each time period is generated in combination with the corresponding stay time, wherein the calculation formula is: ,

[0020] in, It is represented by the population density of the monitoring area s in the time period t, It is represented by the number of overlapping patients corresponding to the monitoring area s within the time period t, Characterized by the area of ​​the monitoring area s.

[0021] In some of the embodiments, after counting the crowd density corresponding to each monitoring area, the following steps are also included:

[0022] Determine whether the crowd density is greater than a preset density, and if so, mark the monitored area as a high-risk area.

[0023] In some embodiments, analyzing the behavior chain database to extract key knowledge data associated with the monitoring area and converting it into knowledge triples includes the following steps:

[0024] Generate a plurality of triple structures corresponding to each of the monitoring areas based on the behavior chain database, wherein the triple structures include a head entity, a relationship, and a tail entity;

[0025] Extract the patient information in the behavior chain database and correspond it to the head entity, extract the behavior location information belonging to the monitoring area and correspond it to the tail entity, extract the behavior action and correspond it to the relationship;

[0026] Determine whether the head entity and the tail entity exist in a namespace;

[0027] If it exists, the existing naming rules are used to assign the identifier. If it does not exist, a new namespace is created and the identifier is assigned based on the naming rules. The identifier is used to identify and describe the associated object.

[0028] In some embodiments, storing the plurality of knowledge triples in a graph database comprises the following steps:

[0029] Storing the head entity and the tail entity in nodes of the graph database, and storing the relationship in edges of the graph database;

[0030] An expression relationship between the associated nodes and the edges is defined, wherein the expression relationship is characterized by the strength of the relationship between the nodes and the edges. Specifically, ,

[0031] in, It is characterized by the strength of the relationship from node i to node j. Represented as the attributes of node i related to relation m, It is represented as the attribute weight from relationship m to node j, and n is the number of relationships involved.

[0032] In some of the embodiments, storing the plurality of knowledge triples in a graph database further includes the following steps:

[0033] Obtaining the number of the knowledge triples;

[0034] If the number is less than the preset value, a triple table structure is created, the head entity is stored corresponding to the subject, the relationship is stored corresponding to the predicate, the tail entity is stored corresponding to the object, and a joint index is established for the subject and the object through B+Tree and stored;

[0035] If the number is greater than the preset value, the column increment or row increment in the triple table is recorded, and the row index and column index are stored using the minimum bit width integer type to achieve compression, and the columns or rows in the knowledge triples whose relationship strength between the node and the edge exceeds the preset value are allocated to the same shard for storage.

[0036] In some embodiments, traversing the graph database to obtain crowd behavior knowledge to construct a behavior graph database includes the following steps:

[0037] Traversing the nodes in the graph database layer by layer through a queue structure to obtain the shortest behavior path and the shortest space-time path between the nodes to complete a breadth-first search;

[0038] Traversing the indirect paths between the nodes through recursion or stack structure to obtain cross-region behavior associations and exploring continuous behavior sequences based on the stay time to obtain long-term behavior associations to complete depth-first exploration;

[0039] The search results of the breadth-first search and the depth-first exploration are integrated into crowd behavior knowledge to construct the behavior graph database.

[0040] In some embodiments, predicting the change characteristics of the crowd density based on a neural network learning model, and analyzing the transmission risk based on the change characteristics in combination with the behavior database to make corresponding warnings include the following steps:

[0041] In the behavior graph database, several behavior triplets are selected and used as training samples to train the first neural network model, the crowd density corresponding to the monitoring area at the current time is input into the trained first neural network model, and the predicted potential crowd density is obtained based on the output result. Specifically, ,

[0042] in, Characterized as the forecast period The potential population density, Characterized by the density of the crowd at the current time, is an external influencing factor associated with the monitoring area, Characterized as model parameters of the first neural network model;

[0043] Obtain the density spatial distribution corresponding to each monitoring area in the historical data and convert it into an image sample, use a number of the image samples to train a second neural network model, the second neural network model outputs a classification prediction result based on image analysis combined with the potential crowd density after obtaining the input crowd density distribution map, and the classification prediction result includes potential high risk and potential low risk;

[0044] When the output potential crowd density exceeds a preset value or the output is a potential high risk, an early warning is triggered.

[0045] In some of the embodiments, when predicting the changing characteristics of the crowd density based on the neural network learning model, the following steps are also included:

[0046] High-risk patients are selected based on the population density, the behavior chains corresponding to the high-risk patients are obtained and matched with the behavior graph database, and iterative data are generated based on the matching results to iteratively update the model parameters of the first neural network model and the second neural network model.

[0047] The technical solution provided by the embodiment of the present application has the following technical effects:

[0048] (1) Realizing multi-source data fusion and synergy can effectively depict the spatiotemporal activity trajectory of patients and, combined with the gathering of people in regional places, provide a more accurate basis for the assessment and early warning of infectious disease spread;

[0049] (2) Accurate spatiotemporal dynamic monitoring and analysis: by building a patient behavior chain database and combining it with the division of regional locations, we can fully grasp the spatiotemporal dynamics of patients and changes in regional population density;

[0050] (3) Efficient crowd density prediction and early warning. The neural network based on time series can predict the changes in regional crowd density in real time and dynamically, determine whether the warning threshold of infectious disease outbreak has been reached, identify the crowd gathering trend in different regions and time periods, and accurately assess potential risks.

[0051] (4) Improve the scientificity and accuracy of prevention and control decisions: Through the reasoning function of the knowledge graph and the predictive ability of the neural network learning model, it is possible to accurately simulate the spatiotemporal dynamic migration of the population and provide a scientific basis for the transmission path of infectious diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the steps of a method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data provided in an embodiment of the present application.

[0053] Figure 2 It is a flow chart of S100 in the embodiment of the present application.

[0054] Figure 3 It is a flow chart of S200 in the embodiment of the present application.

[0055] Figure 4 It is a flowchart of S300 in an embodiment of the present application.

[0056] Figure 5 It is a flowchart of S400 in the embodiment of the present application. DETAILED DESCRIPTION

[0057] To more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. However, it should be understood by those of ordinary skill in the art that the present application can be implemented without these details. In some cases, in order to avoid unnecessary descriptions that make various aspects of the present application obscure, well-known methods, processes, systems, components and / or circuits that have been described at a higher level will not be described in detail. For those of ordinary skill in the art, it is obvious that various changes can be made to the embodiments disclosed in the present application, and without departing from the principles and scope of the present application, the general principles defined in the present application can be applied to other embodiments and application scenarios. Therefore, the present application is not limited to the embodiments shown, but conforms to the broadest scope consistent with the scope claimed for protection of the present application.

[0058] It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in each embodiment of the present invention described below can be combined with each other as long as there is no conflict between them.

[0059] In the description of this application, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used to distinguish the technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0060] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples.

[0061] like Figure 1 As shown, the embodiment of the present application discloses a method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data, comprising the following steps:

[0062] S100, acquiring multi-source data based on data sources and integrating the data to obtain continuous spatiotemporal data, and constructing a behavior chain database according to the continuous spatiotemporal data, wherein the behavior chain database is composed of behavior chains corresponding to a number of patients independently.

[0063] First, multi-source data related to the patient are obtained from multiple data sources. At the same time, continuous time records are performed based on the time corresponding to each type of multi-source data of the patient, and continuous travel locations are obtained based on the behavior locations corresponding to each type of multi-source data of the patient. The above-mentioned time information and location information are analyzed and associated to obtain continuous spatiotemporal data.

[0064] A behavior chain corresponding to each independent patient is constructed by integrating continuous spatiotemporal data. The behavior chain is characterized by integrating multi-source data for subsequent analysis of the patient's movement trajectory.

[0065] The behavior chains of several patients are uniformly stored to obtain a behavior chain database.

[0066] S200, dividing monitoring areas and counting the crowd density corresponding to each monitoring area in combination with the behavior chain database.

[0067] The regions are divided into specific places within the regions, and the population density matrix corresponding to each monitoring area is statistically analyzed by time in the constructed behavior chain database. This is conducive to understanding the crowd gathering situation in different regions and different time periods, and providing basic data for infectious disease transmission risk assessment.

[0068] S300, analyzing the behavior chain database to extract key knowledge data associated with the monitoring area and converting them into knowledge triples, and storing a number of the knowledge triples in a graph database.

[0069] The behavior chain database is processed based on semantic analysis and grammatical concepts to construct a knowledge graph corresponding to regions and scenes, and a data model of crowd behavior knowledge is established based on this.

[0070] Analyze the patient behavior chain, identify the patient's mobile behavior data as structured knowledge triples and store them in a graph database that intuitively displays the relationship logic. This link effectively organizes and stores the behavioral knowledge of the crowd, facilitating subsequent reasoning and prediction.

[0071] S400, traversing the graph database to obtain crowd behavior knowledge to construct a behavior graph database, predicting the change characteristics of the crowd density based on a neural network learning model, and analyzing the transmission risk based on the change characteristics in combination with the behavior database to make corresponding warnings.

[0072] First, the corresponding traversal algorithm is used to traverse the data in the graph database to deeply explore the propagation path and chain and capture potential propagation risks. According to the traversal results, the knowledge triples of the graph database are converted into crowd behavior knowledge and stored again to construct a behavior graph database.

[0073] The data in the behavior graph database is used as historical data samples to train the deep learning prediction model. When new patient behavior data is obtained later, the knowledge triples of the corresponding behavior are input into the dynamic prediction model, and the crowd behavior graph database is used to simulate the dynamic spatiotemporal migration of the crowd to obtain the distribution of the spatiotemporal movement state of the crowd. At the same time, various parameters of the model are iterated in real time to refine the model accuracy to predict the regional population density distribution in the next period of time, and realize intelligent monitoring of the spatiotemporal spread of infectious diseases. At the same time, corresponding warnings are issued when a higher risk of infection is obtained after prediction.

[0074] Through the above method, multi-source data integration and synergy effects can be achieved, which can effectively depict the spatiotemporal activity trajectory of patients, and combined with the gathering of people in regional places, provide a more accurate basis for the assessment and early warning of infectious disease transmission; accurate spatiotemporal dynamic monitoring and analysis, by building a patient behavior chain database and combining the division of regional places, comprehensively grasp the spatiotemporal dynamics of patients and changes in regional population density; efficient population density prediction and early warning, based on time series neural networks, can predict changes in regional population density in real time and dynamically, and determine whether the warning threshold of infectious disease outbreaks has been reached, especially under complex spatiotemporal dynamics, the model can identify the trend of population gathering in different regions and time periods, and accurately assess potential risks; improve the scientificity and accuracy of prevention and control decisions: through the reasoning function of knowledge graphs and the predictive ability of neural network learning models, it can accurately simulate the spatiotemporal dynamic migration of people, and provide a scientific basis for the transmission path of infectious diseases.

[0075] like Figure 2 As shown, in other embodiments, multi-source data is acquired based on the data source and the data is integrated to obtain continuous spatiotemporal data, and a behavior chain database is constructed according to the continuous spatiotemporal data, including the following steps:

[0076] S110, obtaining a hospital data table and extracting patient information and its associated medical consultation information and sorting them by time to generate continuous medical consultation records.

[0077] First, by extracting patient information from data tables of different hospitals and performing data association based on "outpatient number", each different disease type is regarded as an independent outpatient number. In this way, the specific information of each visit can be accurately tracked in the data table corresponding to the patient's medical record.

[0078] At the same time, in order to ensure the time sequence of medical information, the patient's medical information is sorted according to the time of medical treatment, and each patient's continuous medical records are obtained. These records specifically include clinic number, medical treatment time, diagnosis results, etc.

[0079] S120, obtaining the travel data associated with the patient information, extracting the residence information corresponding to the stay time exceeding the preset time based on the travel data to construct a residence chain, and extracting the travel information corresponding to the stay time exceeding the preset time based on the travel data to construct a travel chain.

[0080] The next step is to associate the patient's travel data to further construct the patient's residential and travel spatiotemporal information.

[0081] Specifically, the patient's travel data can provide the patient's continuous residence information and travel information. At the same time, each location information where there is effective behavior also corresponds to the time the patient stayed at these locations.

[0082] The travel data may be the travel information actively registered by the patient through APP, mini-programs, online notebooks, etc. It may also include the travel information analyzed by associating with online data such as ticket data, subway ticket data, shopping receipt data, and dining data generated by the patient.

[0083] The location information and the length of stay are combined to construct a chain. Specifically, if the patient stays in a certain location for more than 0.5 days, it is considered that the patient has stayed in that location for too long, and these locations will be included in the chain corresponding to the patient's residence or travel.

[0084] Then all places where the stay time exceeds 0.5 days are filtered and matched with chain nodes based on their type (residence / travel) and stay time to construct residence chain and travel chain.

[0085] S130, when the sum of the stay times corresponding to the residence chain and the travel chain is greater than a preset time threshold, the residence chain and the travel chain are merged to form the behavior chain associated with the patient information.

[0086] For the patient's residential chain and travel chain, when the sum of their stay time exceeds the preset threshold, it is considered that the patient's movement between the two locations has a high spatiotemporal correlation, so the two chains can be merged into an overall patient mobility behavior chain.

[0087] The behavior chain includes behavior location information and residence time. At the same time, in order to further accurately construct the patient's movement trajectory, the following formula is used to express the calculation method of residence time: ,

[0088] It is represented by the total length of time a patient stays at a location. It is represented as the timestamp of the patient's i-th recorded trip. Characterized by the number of patient stay records, - It is characterized by the patient's stay time between the i-th recording location and the i+1-th recording location.

[0089] Among them, the dwell time specifically corresponds to the type of behavior, such as medical consultation time, shopping time, meal time, class time, etc.

[0090] Through the above formula, the patient's stay time in different locations can be effectively calculated to determine whether the conditions for merging the residence chain and travel chain are met.

[0091] S140, integrating and storing a plurality of behavior chains corresponding to the patient information to obtain the behavior chain database.

[0092] The behavior chains of several patients are systematically stored and managed to build a behavior chain database, which contains the spatiotemporal trajectory information of the patients. This information can not only provide basic data for subsequent infectious disease transmission analysis, but also provide key input for subsequent model training and prediction.

[0093] The patient's behavior chain can help us identify potential sources of infection and transmission paths, and thus provide strong support for the prediction and early warning of the spatiotemporal spread of infectious diseases. In the subsequent knowledge graph construction, population density statistics, and crowd status prediction, the patient's behavior chain data will serve as an important spatiotemporal information basis to guide the model to make accurate spatiotemporal dynamic predictions. Therefore, the data integration and behavior chain construction at this stage are not only the first step in the monitoring and early warning of infectious disease transmission, but also the core data source for subsequent analysis.

[0094] like Figure 3 As shown, in other embodiments, dividing the monitoring areas and counting the crowd density corresponding to each monitoring area in combination with the behavior chain database includes the following steps:

[0095] S210, dividing each area based on the division type option to obtain a number of monitoring areas of corresponding results.

[0096] First, the area is divided into multiple sub-areas or places as monitoring areas. At the same time, in order to ensure the analysis of crowd gathering at a more detailed level, different division type standards can be used in different scenarios.

[0097] Among them, the division type options include area options, function options, and representative feature options.

[0098] The area division is characterized by not considering the specific building type in each area, but dividing the entire area into several monitoring areas of equal area only by area.

[0099] Functional division is characterized by division based on the overall functional role of the building complex, such as commercial area, school area, residential area, park area, etc.

[0100] The representative feature division is characterized by division based on specific representative features of the building complex, such as shopping malls, hospitals, squares, subway stations, etc.

[0101] S220, mapping the behavior information stored in the behavior chain database with the monitoring area to calculate the number of overlapping patients, and generating the crowd density corresponding to each monitoring area in each time period in combination with the corresponding stay time.

[0102] The number of patients staying in different places during a certain period of time is analyzed based on the constructed patient behavior chain database. Specifically, after a monitoring area is selected, the name of the monitoring area is mapped to the behavior chain database to analyze the number of matching behavior chains to calculate the clustering density corresponding to the number of overlapping patients.

[0103] The calculation formula is: ,

[0104] in, It is represented by the population density of the monitoring area s in the time period t, It is represented by the number of overlapping patients corresponding to the monitoring area s within the time period t, Characterized by the area of ​​the monitoring area s.

[0105] That is, the number of patients (or other people) staying in the monitoring area during a specific period of time is calculated, and by calculating the patient density per unit area, the crowd gathering situation in different places at different time periods can be analyzed.

[0106] In some other embodiments, after counting the crowd density corresponding to each monitoring area, the following steps are also included:

[0107] S230, determining whether the crowd density is greater than a preset density, and if so, marking the monitoring area as a high-risk area.

[0108] Furthermore, when it is calculated that the population density in a certain monitoring area during a specific period of time is greater than the preset density, the monitoring area is marked as a high-risk area. If the population density in the monitoring area is not greater than the preset density, regular records are made and statistical calculations continue.

[0109] The significance of calculating population density is that it can provide key data support for subsequent infectious disease transmission risk assessment. For example, when the population density in a certain area is high during peak hours (8-9 a.m., 11-12 p.m., and 5-6 p.m.), it means that the transmission risk in that area is also relatively high. Through this density-based analysis method, high-risk areas can be more clearly identified, further guiding the allocation of resources and the formulation of prevention and control strategies. In model prediction, based on this data, the changing trend of the population in the future can be accurately estimated, especially in densely populated areas, and potential outbreak points can be predicted in advance.

[0110] In addition, in the statistics of crowd density, we also need to consider the spatiotemporal variation characteristics of patient behavior. Since the activities of patients are affected by many factors, the changes in their behavior chains in different time periods and different regions are nonlinear. Therefore, through the statistical analysis of crowd density, we can more accurately identify the gathering trend of people in specific areas and provide a basis for subsequent spatiotemporal transmission models. By continuously updating the patient's behavior chain data and constantly adjusting and optimizing the statistical model, we can accurately predict the population density in different time periods and regions, thereby providing important decision-making basis for the monitoring and early warning of the spatiotemporal transmission of infectious diseases.

[0111] like Figure 4 As shown, in other embodiments, analyzing the behavior chain database to extract key knowledge data associated with the monitoring area and converting it into knowledge triples includes the following steps:

[0112] S310: Generate a plurality of triplet structures corresponding to each of the monitoring areas based on the behavior chain database.

[0113] In order to achieve a structured and systematic expression of the patient's mobile behavior and provide strong support for subsequent reasoning and prediction, it is first necessary to disassemble and analyze the behavior chain database through semantic analysis and the use of grammatical concepts, extract the knowledge involved in the regional places, scenes, and crowd behaviors, and convert them into structured knowledge triples. The structure of the triple includes head entity, relationship, and tail entity.

[0114] S311, extract the patient information in the behavior chain database and correspond it to the head entity, extract the behavior location information belonging to the monitoring area and correspond it to the tail entity, extract the behavior action and correspond it to the relationship.

[0115] The head entity and tail entity represent specific elements related to patient behavior, while the relationship is used to describe the association between these entities.

[0116] Specifically, taking the patient's stay in a certain place as an example, assuming that patient A is shopping in mall X, then the corresponding knowledge triple structure is (patient A, shopping, mall X). This method expresses the knowledge of crowd behavior very intuitively and can be efficiently stored and queried through the graph database.

[0117] S312, determine whether there is a namespace for the head entity and the tail entity.

[0118] S313: If it exists, the existing naming rule is used to assign the identifier; if it does not exist, a new namespace is created and the identifier is assigned based on the naming rule, and the identifier is used to identify and describe the associated object.

[0119] At the same time, in order to ensure that entities of different fields and types can be accurately distinguished, each head entity and tail entity will be accompanied by an identifier corresponding to the namespace, and elements of the same type will be distinguished by slashes and real numbers. For example, the identifier of the head entity may be "Medical / Patient A", and the tail entity may be "Functional Place / Shopping Mall X". This method can effectively avoid confusion between different fields and entity types.

[0120] At the same time, before setting the identifier, you first need to determine whether there is a namespace for the head entity and the tail entity. The namespace represents whether there is space for the entity to add an identifier. If the namespace corresponding to it exists, it can be used directly according to the existing naming rules. If the namespace does not exist, create a new namespace and assign identifiers according to the corresponding naming rules.

[0121] In some other embodiments, storing the plurality of knowledge triples in a graph database comprises the following steps:

[0122] S320, storing the head entity and the tail entity in the nodes of the graph database, and storing the relationship in the edges of the graph database.

[0123] Graph databases are ideal for storing the above-mentioned knowledge triples. Graph databases contain nodes and edges, head entities and tail entities are stored in the nodes of the graph database, and relationships are stored in the edges of the graph database.

[0124] Nodes contain the relevant attributes of entities and identifiers used for description, while edges contain the relationship types between entities and pointers to entities.

[0125] S321, defining an expression relationship between the associated nodes and the edges, wherein the expression relationship is characterized by the strength of the relationship between the nodes and the edges, specifically, ,

[0126] in, It is characterized by the strength of the relationship from node i to node j. Represented as the attributes of node i related to relation m, It is represented as the attribute weight from relationship m to node j, and n is the number of relationships involved.

[0127] Through the above formula, the strength of the relationship between two nodes can be evaluated, so as to understand the reasoning process and help perform efficient path reasoning and knowledge query in the graph database.

[0128] When there is a strong relationship strength between nodes, it is characterized by a high degree of association between the two nodes, and the possibility of an outbreak in the transmission path of the infectious disease is higher.

[0129] In the subsequent intelligent monitoring of the spatiotemporal spread of infectious diseases, with the help of the entity relationship network in the knowledge graph, the model can quickly obtain, analyze and infer the spatiotemporal behavior characteristics of the crowd, and then predict the potential risk of transmission. In addition, the query and reasoning capabilities of the graph database enable us to conduct complex infectious disease transmission simulations and optimize decisions from different dimensions. This not only improves the efficiency of data utilization, but also provides an accurate prediction basis for infectious disease prevention and control strategies.

[0130] In some other embodiments, storing the plurality of knowledge triples in a graph database further includes the following steps:

[0131] S330, obtaining the number of the knowledge triples.

[0132] In order to optimize storage performance and query efficiency, different storage methods need to be selected according to the number of knowledge triples.

[0133] S331, if the number is less than the preset value, create a triple table structure, store the head entity corresponding to the subject, the relationship corresponding to the predicate, and the tail entity corresponding to the object, and establish a joint index for the subject and the object through B+Tree and store them.

[0134] If the number of knowledge triples is small, they are stored based on a simple indexing method. Specifically, a triple table structure of (subject, predicate, object) is adopted, and a joint index is established for the subject and object through B+Tree.

[0135] B+Tree is a multi-way balanced search tree. Each node can have multiple child nodes. The root node and internal nodes do not contain data and are only used for indexing. All data is stored in leaf nodes.

[0136] Each head entity and tail entity is combined with the subject and object through B+Tree to obtain the index of the table structure for storage.

[0137] S332: If the number is greater than a preset value, record the column increment or row increment in the triple table, and use the minimum bit width integer type to store the row index and column index to achieve compression, and allocate the columns or rows in the knowledge triples whose relationship strength between the node and the edge exceeds the preset value to the same shard for storage.

[0138] If the number of knowledge triples is large, then the simple index storage method will lead to a large amount of redundant duplicate data on the one hand, and it will not be convenient for accurate search on the other hand.

[0139] Therefore, more complex compression and sharding techniques are needed to improve storage and query speed. Specifically, record row / column increments instead of absolute values ​​(e.g., use variable-length integers to store the difference in row numbers of adjacent elements, and use the minimum bit width integer type to store row and column indexes (e.g., uint16_t stores matrices with a maximum size of 65535).

[0140] Allocate columns or rows with strong relationships to the same shard, and determine the storage node by (col_index % shard_num), which is suitable for column traversal scenarios.

[0141] Through the above method, while reducing the storage space as much as possible, several knowledge triples are stored in fragments, and several knowledge triples with high correlation are stored in a unified manner to facilitate subsequent searches.

[0142] like Figure 5 As shown, in some other embodiments, traversing the graph database to obtain crowd behavior knowledge to construct a behavior graph database includes the following steps:

[0143] S410, traversing the nodes in the graph database layer by layer through a queue structure to obtain the shortest behavior path and the shortest space-time path between the nodes to complete a breadth-first search.

[0144] Using the BFS (Breadth First Search) traversal method to implement hierarchical traversal can effectively capture the shortest path from one node to other nodes. The shortest path specifically includes the shortest behavioral path and the shortest space-time path. The shortest behavioral path is characterized by the shortest path of crowd migration and diffusion, so as to identify key propagation nodes. The shortest space-time path is characterized by analyzing the behavior of a specific area at a specific time in the movement trajectory of the crowd, such as subway transfers, emergency evacuations, etc.

[0145] S420, traversing the indirect paths between the nodes through recursion or stack structure to obtain cross-region behavior associations and exploring continuous behavior sequences based on the stay time to obtain long-term behavior associations to complete depth-first exploration.

[0146] DFS (Depth-First Search) is used for deep traversal to help better explore the relationships between different areas and potential transmission chains.

[0147] Specifically, depth-first search deeply explores branch paths through recursion or stack structures, such as analyzing the indirect path selection of people from nodes across regions, revealing indirect transmission mechanisms.

[0148] Track the patient's same or similar behavioral sequences for multiple consecutive days to obtain fixed long-term behaviors.

[0149] Secondly, DFS can also connect loops and branches in the graph database to help discover periodic gathering times, such as weekly markets and festivals. It can also analyze hierarchical interactions within specific groups, such as families, communities, and cities, through the attributes and relationships between nodes.

[0150] S430, integrating the search results of the breadth-first search and the depth-first exploration into crowd behavior knowledge to construct the behavior graph database.

[0151] BFS extracts immediate, short-distance interactions, and DFS supplements long-term, cross-regional behavioral associations to form a complete spatiotemporal feature matrix, which is used as crowd behavior knowledge to construct a behavior graph database.

[0152] In other embodiments, predicting the change characteristics of the crowd density based on a neural network learning model, and analyzing the transmission risk based on the change characteristics in combination with the behavior database to make corresponding warnings, includes the following steps:

[0153] S440, selecting a number of behavior triplets in the behavior graph database and using them as training samples to train the first neural network model, inputting the crowd density corresponding to the monitoring area at the current time into the trained first neural network model and obtaining the predicted potential crowd density based on the output result.

[0154] Once the behavior graph database is built, the crowd behavior knowledge stored in it can be used as historical data as a good sample for training neural network models. The dynamic prediction model based on time series can calculate the changing characteristics of the crowd density in the sub-region and associate it with the population size. Based on historical data, these changing characteristics can help establish warning thresholds and predict the changing trend of population density.

[0155] In the embodiment of the present application, the first neural network model based on the time series prediction method is a recursive neural network (RNN) or a long short-term memory network (LSTM) model. Such models can capture long-term dependencies in time series data, and by training historical data samples, the first neural network model can predict changes in regional population density within a certain period of time in the future.

[0156] Specifically, in order to improve the accuracy of the prediction, the following formula is used to describe the change prediction of crowd density: ,

[0157] in, Characterized as the forecast period The potential population density, Characterized by the density of the crowd at the current time, is an external influencing factor associated with the monitoring area, The model parameters of the first neural network model are characterized, and the model parameters are obtained by training with a number of historical data samples obtained from the behavior graph database.

[0158] Through this formula, the model can predict future density changes based on current and historical density data and external factors, and provide a basis for whether to trigger an early warning.

[0159] S450, obtaining the density spatial distribution corresponding to each of the monitoring areas in the historical data and converting it into image samples, using a number of the image samples to train a second neural network model, after obtaining the input crowd density distribution map, the second neural network model outputs a classification prediction result based on image analysis combined with the potential crowd density, and the classification prediction result includes potential high risk and potential low risk.

[0160] In this application, in addition to time series dynamic prediction, migration prediction is also performed by using a convolutional neural network (CNN) as the second neural network model.

[0161] Convolutional neural networks are good at processing image data, and in this solution, they are used to classify and predict spatial distribution image samples of crowd density. By converting the population density distribution in the area into image samples to train the second neural network model, the model can classify different spatial distribution patterns and predict the risk of infectious disease transmission in different scenarios.

[0162] Specifically, the second neural network model can distinguish the different transmission patterns of potential high-risk areas and potential risk areas after migration prediction in the future, so that after training it can determine which monitored areas are more likely to become potential outbreak points.

[0163] S460, triggering an early warning when the output potential crowd density exceeds a preset value or is output as a potential high risk.

[0164] When two different neural network models are fused and judged, if the model outputs that the potential population density exceeds the preset value or the output is a potential high-risk area, an early warning is generated to remind the area that there is a greater risk of infection.

[0165] The time series model provides the long-term trend of density changes, while the convolutional neural network helps to classify the spatial distribution in detail, thus achieving accurate monitoring and early warning of the spatiotemporal spread of infectious diseases. This multi-model collaborative approach not only enhances the accuracy of the infectious disease early warning system, but also provides a scientific basis for the formulation of actual prevention and control measures.

[0166] In some other embodiments, when predicting the changing characteristics of the crowd density based on the neural network learning model, the following steps are also included:

[0167] S470, selecting high-risk patients based on the population density, obtaining the behavior chain corresponding to the high-risk patients and matching it with the behavior graph database, and generating iterative data based on the matching result to iteratively update the model parameters of the first neural network model and the second neural network model.

[0168] During the prediction process, the model will continuously select high-risk patients based on the population density of the monitored area and match their behavior chains with the crowd behavior graph database. This process makes the patient's activities highly correlated with the regional population density data through the comparison of knowledge triples, thereby providing more accurate predictions. After a series of iterative operations, the model can continuously update the patient's movement trajectory, graph database, and crowd density matrix, and predict the regional population density distribution in the next period in real time. The iteration of this prediction process not only improves the accuracy of spatiotemporal propagation predictions, but also enhances the real-time and sensitivity of the early warning system.

[0169] The implementation principle is:

[0170] The realization of multi-source data integration and synergy can effectively depict the spatiotemporal activity trajectory of patients, and combined with the gathering of people in regional places, provide a more accurate basis for the assessment and early warning of infectious disease transmission; accurate spatiotemporal dynamic monitoring and analysis, by building a patient behavior chain database and combining the division of regional places, comprehensively grasp the spatiotemporal dynamics of patients and changes in regional population density; efficient population density prediction and early warning, based on time series neural networks, can predict changes in regional population density in real time and dynamically, and determine whether the early warning threshold of infectious disease outbreaks has been reached, especially under complex spatiotemporal dynamics, the model can identify the trend of population gathering in different regions and time periods, and accurately assess potential risks; improve the scientificity and accuracy of prevention and control decisions: through the reasoning function of knowledge graphs and the predictive ability of neural network learning models, it can accurately simulate the spatiotemporal dynamic migration of people, and provide a scientific basis for the transmission path of infectious diseases.

[0171] It should be understood that although the steps in the flowchart of the accompanying drawings are shown in sequence according to the instructions of the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise clearly stated in this document, the execution of these steps is not strictly limited in order and can be performed in other orders.

[0172] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data, characterized in that: The following steps are involved: Acquire multi-source data based on data sources and integrate the data to obtain continuous spatiotemporal data, and construct a behavior chain database based on the continuous spatiotemporal data, wherein the behavior chain database is composed of behavior chains corresponding to a number of patients independently; Divide the monitoring areas and count the crowd density corresponding to each monitoring area in combination with the behavior chain database; Analyzing the behavior chain database to extract key knowledge data associated with the monitoring area and converting the key knowledge data into knowledge triples, and storing a number of the knowledge triples in a graph database; The graph database is traversed to obtain crowd behavior knowledge to construct a behavior graph database, the change characteristics of the crowd density are predicted based on a neural network learning model, and the transmission risk is analyzed based on the change characteristics combined with the behavior database to make corresponding warnings.

2. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 1 is characterized in that: Acquiring multi-source data based on data sources and integrating the data to obtain continuous spatiotemporal data, and constructing a behavior chain database based on the continuous spatiotemporal data, including the following steps: Obtain hospital data tables and extract patient information and their associated visit information and sort them by time to generate a continuous visit record; Acquire the travel data associated with the patient information, extract the residence information corresponding to the stay time exceeding the preset time based on the travel data to construct a residence chain, and extract the travel information corresponding to the stay time exceeding the preset time based on the travel data to construct a travel chain; When the sum of the stay times corresponding to the residence chain and the travel chain is greater than a preset time threshold, the residence chain and the travel chain are merged to form the behavior chain associated with the patient information, wherein the behavior chain includes behavior location information and stay time; The behavior chains corresponding to the patient information are integrated into a database to obtain the behavior chain database.

3. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 2 is characterized in that: Dividing the monitoring areas and counting the crowd density corresponding to each monitoring area in combination with the behavior chain database includes the following steps: Dividing each area based on a division type option to obtain a number of monitoring areas of corresponding results, wherein the division type option includes an area option, a function option, and a representative feature option; Based on the mapping of each behavior location information stored in the behavior chain database with the monitoring area to calculate the number of overlapping patients, the crowd density corresponding to each monitoring area in each time period is generated in combination with the corresponding stay time, wherein the calculation formula is: ， in, It is represented by the population density of the monitoring area s in the time period t, It is represented by the number of overlapping patients corresponding to the monitoring area s within the time period t, Characterized by the area of ​​the monitoring area s.

4. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 3 is characterized in that: After counting the crowd density corresponding to each monitoring area, the following steps are also included: Determine whether the crowd density is greater than a preset density, and if so, mark the monitored area as a high-risk area.

5. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 2 is characterized in that: Analyzing the behavior chain database to extract key knowledge data associated with the monitoring area and converting it into knowledge triples includes the following steps: Generate a plurality of triple structures corresponding to each of the monitoring areas based on the behavior chain database, wherein the triple structures include a head entity, a relationship, and a tail entity; Extract the patient information in the behavior chain database and correspond it to the head entity, extract the behavior location information belonging to the monitoring area and correspond it to the tail entity, extract the behavior action and correspond it to the relationship; Determine whether the head entity and the tail entity exist in a namespace; If it exists, the existing naming rules are used to assign the identifier. If it does not exist, a new namespace is created and the identifier is assigned based on the naming rules. The identifier is used to identify and describe the associated object.

6. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 5 is characterized in that: Storing the plurality of knowledge triples in a graph database comprises the following steps: Storing the head entity and the tail entity in nodes of the graph database, and storing the relationship in edges of the graph database; An expression relationship between the associated nodes and the edges is defined, wherein the expression relationship is characterized by the strength of the relationship between the nodes and the edges. Specifically, ， in, It is characterized by the strength of the relationship from node i to node j. Represented as the attributes of node i related to relation m, It is represented as the attribute weight from relationship m to node j, and n is the number of relationships involved.

7. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 6 is characterized in that: Storing the plurality of knowledge triples in a graph database further comprises the following steps: Obtaining the number of the knowledge triples; If the number is less than the preset value, a triple table structure is created, the head entity is stored corresponding to the subject, the relationship is stored corresponding to the predicate, the tail entity is stored corresponding to the object, and a joint index is established for the subject and the object through B+Tree and stored; If the number is greater than the preset value, the column increment or row increment in the triple table is recorded, and the row index and column index are stored using the minimum bit width integer type to achieve compression, and the columns or rows in the knowledge triples whose relationship strength between the node and the edge exceeds the preset value are allocated to the same shard for storage.

8. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 6 is characterized in that: Traversing the graph database to obtain crowd behavior knowledge to build a behavior graph database includes the following steps: Traversing the nodes in the graph database layer by layer through a queue structure to obtain the shortest behavior path and the shortest space-time path between the nodes to complete a breadth-first search; Traversing the indirect paths between the nodes through recursion or stack structure to obtain cross-region behavior associations and exploring continuous behavior sequences based on the stay time to obtain long-term behavior associations to complete depth-first exploration; The search results of the breadth-first search and the depth-first exploration are integrated into crowd behavior knowledge to construct the behavior graph database.

9. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 8, characterized in that: Predicting the change characteristics of the crowd density based on the neural network learning model, and analyzing the transmission risk based on the change characteristics combined with the behavior database to make corresponding warnings, including the following steps: In the behavior graph database, several behavior triplets are selected and used as training samples to train the first neural network model, the crowd density corresponding to the monitoring area at the current time is input into the trained first neural network model, and the predicted potential crowd density is obtained based on the output result. Specifically, ， in, Characterized as the forecast period The potential population density, Characterized by the density of the crowd at the current time, is an external influencing factor associated with the monitoring area, Characterized as model parameters of the first neural network model; Obtain the density spatial distribution corresponding to each monitoring area in the historical data and convert it into an image sample, use a number of the image samples to train a second neural network model, the second neural network model outputs a classification prediction result based on image analysis combined with the potential crowd density after obtaining the input crowd density distribution map, and the classification prediction result includes potential high risk and potential low risk; When the output potential crowd density exceeds a preset value or the output is a potential high risk, an early warning is triggered.

10. The method for monitoring and early warning of the spatiotemporal spread of infectious diseases based on regional medical big data according to claim 9, characterized in that: When predicting the changing characteristics of the crowd density based on the neural network learning model, the following steps are also included: High-risk patients are selected based on the population density, the behavior chains corresponding to the high-risk patients are obtained and matched with the behavior graph database, and iterative data are generated based on the matching results to iteratively update the model parameters of the first neural network model and the second neural network model.

Images