Intelligent auxiliary flow regulation method for whole flow regulation process

Abstract

The invention provides an intelligent auxiliary flow scheduling method for a whole flow scheduling process, and relates to the technical field of big data, and the method comprises the steps: S1, deploying an artificial intelligence computing power cluster at a center end; s2, registering a traffic scheduling task through a held mobile terminal by a traffic scheduling person during traffic scheduling, interacting with the central terminal in the traffic scheduling task execution process to call an artificial intelligence computing power cluster to perform remote identity verification and auxiliary traffic scheduling question and answer on a traffic scheduling object, and after the traffic scheduling task is finished, according to a generated traffic scheduling question and answer record, performing traffic scheduling on the traffic scheduling object according to the traffic scheduling question and answer record; and completing structured analysis and generating an epidemiological investigation report. The method has the beneficial effects that by deploying the artificial intelligence computing power cluster at the center end, unified management and application of the flow scheduling artificial intelligence models and reasonable scheduling of resources are realized, so that each flow scheduling artificial intelligence model can be integrated into the whole flow scheduling process, and the flow scheduling efficiency and accuracy are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to an intelligent assisted epidemic investigation method for the entire epidemic investigation process. Background Art

[0002] Epidemiological investigation (referred to as epidemiological investigation) is a key link in epidemic prevention and control, and its efficiency and accuracy are directly related to the effectiveness of epidemic control. Traditional epidemiological investigation relies on paper records and face-to-face interviews, which are inefficient and easily affected by human factors. The introduction of digital information collection technology has completely changed this situation. Various epidemiological investigation (referred to as epidemiological investigation) methods and systems have initially solved a series of business scenario problems in the epidemiological investigation process, such as: digital information collection (including manual data entry, audio collection and image collection, etc., to effectively improve information collection capabilities), data information management (through systematization to improve epidemiological investigation efficiency and data management capabilities), intelligent information analysis and reasoning (through data mining and artificial intelligence to improve information perception, accurate judgment and rapid traceability capabilities) and secure data sharing.

[0003] While existing technologies have made some progress in the field of intelligent epidemiological investigation, significant challenges remain in integrating these technologies into the entire epidemiological investigation process. For example, how can independent AI models for epidemiological investigation be integrated with business needs to address practical application challenges? Currently, there is a lack of an epidemiological investigation method that integrates AI models with the entire epidemiological investigation process. Summary of the Invention

[0004] In response to the problems existing in the prior art, the present invention provides an intelligent assisted epidemiological investigation method for the entire epidemiological investigation process, including:

[0005] Step S1: deploying an artificial intelligence computing cluster at the center;

[0006] In step S2, the epidemiologist registers the epidemiological investigation task through the mobile terminal he holds when conducting the epidemiological investigation, and interacts with the central terminal during the execution of the epidemiological investigation task to call the artificial intelligence computing power cluster to perform remote identity authentication on the epidemiological investigation object and assist in epidemiological investigation questions and answers, and generates an epidemiological investigation report based on the generated epidemiological investigation question and answer records after the epidemiological investigation task is completed.

[0007] Preferably, the artificial intelligence computing power cluster includes an identity authentication model;

[0008] In step S2, remote identity authentication of the flow survey object includes:

[0009] The epidemic investigator collects audio information and / or video information of the epidemic investigation object through the mobile terminal, and calls the identity authentication model to perform identity recognition based on the audio information and / or the video information to assist in verifying the identity information of the epidemic investigator.

[0010] Preferably, the artificial intelligence computing power cluster includes an intelligent auxiliary questioning model and a pre-built epidemic investigation question and answer knowledge graph, and the epidemic investigation question and answer knowledge graph is annotated with business scenarios and module elements;

[0011] In step S2, the auxiliary epidemiological investigation question and answer includes:

[0012] During the process of the epidemic investigator performing the epidemic investigation task, the mobile terminal calls the intelligent auxiliary questioning model to extract the personalized characteristics of the epidemic investigation object contained in the current epidemic investigation content, and dynamically matches the business scenarios and module elements in the epidemic investigation question and answer knowledge graph based on the personalized characteristics, and then displays the next round of questioning suggestions to the epidemic investigator based on the matched business scenarios and module elements.

[0013] Preferably, the business scenarios include scenarios for questions about Class A and Class B and Class C infectious diseases;

[0014] The module elements include basic personal information, place of residence, place of work, medical treatment and medication history, and activity trajectory survey.

[0015] Preferably, when the matched module elements are basic personal information, place of residence, place of work, medical treatment and medication, the intelligent auxiliary questioning model displays the next round of questioning suggestions to the epidemiologist according to the preset epidemiological survey question and answer template associated with the module elements;

[0016] When the matched module element is the activity trajectory survey, the intelligent auxiliary questioning model generates the next round of question suggestions based on the pre-deployed large language model.

[0017] Preferably, the artificial intelligence computing power cluster includes a real-time recording model for epidemic investigation information;

[0018] In step S2, the process of generating the epidemiological investigation question and answer record includes:

[0019] During the process of the epidemic investigator performing the epidemic investigation task, the mobile terminal calls the real-time recording model of epidemic investigation information to monitor the multiple rounds of conversation audio between the epidemic investigator and the epidemic investigation object in real time, and distinguishes the epidemic investigator's question audio and the epidemic investigation object's reply audio from the multiple rounds of conversation audio, and then converts the question audio and the reply audio into text for the epidemic investigator to modify and edit and generate the epidemic investigation question and answer record.

[0020] Preferably, the artificial intelligence computing power cluster includes an information credibility assessment model;

[0021] In step S2, when the flow investigation task is completed, the following steps are also included:

[0022] The mobile terminal calls the information credibility assessment model to perform mobile phone signaling spatiotemporal consistency and conversation internal context consistency assessment on the epidemic investigation question and answer record to obtain spatiotemporal consistency assessment results and context consistency assessment results, and when the spatiotemporal consistency assessment results or the context consistency assessment results indicate that the epidemic investigation personnel's reply information is unreliable, the epidemic investigation personnel are prompted to intervene to verify the information through confirmation questions, and at the same time provide a viewing and modification port for the epidemic investigation question and answer record.

[0023] Preferably, the artificial intelligence computing power cluster includes a structured parsing model for epidemic investigation elements and an automated writing model for epidemiological investigation reports;

[0024] In step S2, after the flow investigation task is completed, the following steps are also included:

[0025] The mobile terminal calls the epidemiological investigation element structured parsing model to extract key elements from the question and answer pairs in the epidemiological investigation question and answer record, then parses the extracted key elements and automatically stores them in the database, and displays the parsing results in the form of an epidemiological investigation report form on the front end for the epidemiologist to proofread and confirm;

[0026] After confirming that the flow survey report form is filled out correctly, the automatic writing model of the epidemiological investigation report is called to generate the epidemiological investigation report that meets the professional reporting requirements based on the preset epidemiological investigation report template.

[0027] Preferably, the epidemiological investigation task also includes a close contact discovery task associated with a facial image of a case subject, and when performing the close contact discovery task, the epidemiologist uploads the close contact discovery task to the central end for processing via the mobile end;

[0028] The central end is also connected to the public security system, and the artificial intelligence computing cluster includes a close contact detection model; the processing process of the central end includes:

[0029] The central end uses the facial image of the case subject and the multiple location surveillance video files provided by the public security system to call the close contact discovery model combined with the video clip positioning and multi-target tracking method to extract the mask wearing status, contact time points and key frame screenshots of the case subject and his / her contacts, and then uses the machine learning classification model to determine whether they are close contacts or general contacts. At the same time, the key frame screenshots are input into the auxiliary investigation identity interface of the public security system for facial recognition and body posture recognition to determine the identity of the contact and provide support for subsequent close contact screening;

[0030] The AI ​​computing cluster includes an intelligent epidemic analysis and tracing model; the central processing also includes:

[0031] The center adds a tracking and tracing event for the case object, adds relevant cases to the Class A and Class A controlled cases, and calls the epidemic intelligent analysis and tracking and tracing model to build a network map of the interpersonal relationships, whereabouts, and contact environment of the infected person based on the epidemiological survey reports of all relevant cases, to realize the analysis of the evolution mechanism of the epidemic spread, and to determine the possible patient zero of the case object while conducting close contact tracking through knowledge graph reasoning to realize tracking and tracing of the transmission chain analysis.

[0032] Preferably, the artificial intelligence computing power cluster includes a blockchain-based epidemic investigation data on-chain model and a blockchain-based epidemic investigation data verification model;

[0033] There are multiple central terminals, forming a blockchain network. The central terminal is also used for the initiator of the cooperative investigation to create a new cooperative investigation event, and call the blockchain-based epidemic investigation data chain model to store the structured information of the cooperative investigation event and the unstructured cooperative investigation attachments on the chain;

[0034] The coordinating party verifies the on-chain coordinating event and calls the blockchain-based flow investigation data verification model to perform asynchronous data verification on the coordinating event;

[0035] The coordinating party registers the corresponding epidemic investigation task according to the coordinating information, and after the epidemic investigation task is completed, calls the blockchain-based epidemic investigation data on-chain model to store the generated epidemiological investigation report on the chain as the coordinating feedback result;

[0036] And the initiator of the cooperative investigation can verify and view the data of the cooperative investigation feedback results associated with the cooperative investigation event stored on the chain.

[0037] The above technical solution has the following advantages or beneficial effects: by deploying an artificial intelligence computing power cluster at the center, unified management and application of epidemic investigation artificial intelligence models and rational scheduling of resources can be achieved, so that each epidemic investigation artificial intelligence model can be integrated into the entire epidemic investigation process, effectively improving the efficiency and accuracy of epidemic investigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a flow chart of an intelligent auxiliary flow investigation method for the entire flow investigation process in a preferred embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of a microservice architecture in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment, and other embodiments may also fall within the scope of the present invention as long as they conform to the gist of the present invention.

[0041] In a preferred embodiment of the present invention, based on the above problems existing in the prior art, an intelligent auxiliary flow adjustment method for the entire flow adjustment process is provided. Figure 1 Shown, including:

[0042] Step S1: deploying an artificial intelligence computing cluster at the center;

[0043] In step S2, the epidemiologist registers the epidemiological investigation task through the mobile terminal he holds when conducting the epidemiological investigation, and interacts with the central terminal during the execution of the epidemiological investigation task to call the artificial intelligence computing power cluster to perform remote identity authentication on the epidemiological investigation object and assist in epidemiological investigation questions and answers, and generates an epidemiological investigation report based on the generated epidemiological investigation question and answer records after the epidemiological investigation task is completed.

[0044] Specifically, in this embodiment, the above-mentioned artificial intelligence computing power cluster is formed by multiple epidemic investigation artificial intelligence models. The intelligent assisted epidemic investigation method of the present invention establishes a collaborative mode between the central end and the mobile end, wherein the mobile end is used for the intelligent collection of epidemic investigation object information, and instantly transmits the information to the central end. The computing power resources are called through the artificial intelligence service interface to complete data association analysis, key information extraction, credibility assessment, and intelligent analysis and reasoning of epidemic investigation information, etc., for epidemic investigators to consult and analyze in corresponding business scenarios. The central end preferably uses a microservice architecture to encapsulate the above-mentioned epidemic investigation artificial intelligence models to realize a rapidly deployable intelligent epidemic investigation product.

[0045] like Figure 2As shown in the figure, the microservice architecture is arranged from bottom to top in the basic platform layer 100, model support layer 200, data resource layer 300 and business layer 400, which can provide auxiliary support functions for the entire epidemic investigation process for two business scenarios: Class A (tube A) and multiple diseases (respiratory tract, intestinal tract, vectors).

[0046] Based on this, the above-mentioned data resource layer 300 preferably centrally deploys a storage cluster for massive flow survey data. The storage cluster here includes a distributed object storage collection, such as audio, video, pictures and files, as well as a relational database, a system database, a Class A management form database and a multi-disease form database to meet the storage needs of different objects.

[0047] The above-mentioned model support layer 200 preferably centrally deploys an artificial intelligence computing cluster, and multiple epidemic investigation artificial intelligence models include but are not limited to epidemic investigation object identity identification model, intelligent assisted questioning model, epidemic investigation information real-time recording model, information credibility assessment model, epidemic investigation element structured analysis model, epidemiological investigation report automatic writing model, close contact discovery model, epidemic situation intelligent analysis and tracking and tracing model, blockchain-based epidemic investigation data chain model, blockchain-based epidemic investigation data verification model and multi-source heterogeneous epidemic investigation model training system based on federated learning.

[0048] The above-mentioned business layer 400 is preferably connected to the mobile terminal to provide the mobile terminal with epidemic investigation task management services, epidemic investigation intelligent question and answer services, intelligent video analysis services and inter-provincial cooperation services, and is also used to provide federated learning services and user management services.

[0049] The above-mentioned basic platform layer 100 is preferably deployed with:

[0050] Epidemiological investigation management system is used to uniformly manage the entire process of epidemiological investigation.

[0051] Database management system, used for unified management of storage clusters.

[0052] Intelligent epidemic investigation artificial intelligence model management system is used to uniformly manage artificial intelligence computing power clusters.

[0053] Operating systems, information security systems, network and communication systems are used to manage information security and provide communication guarantees during the epidemic investigation process.

[0054] In summary, the central end provides services to the mobile end in the form of cloud services. To facilitate deployment and migration, the present invention preferably uses the containerization technology Docker for environmental isolation and management, and performs resource management and task scheduling based on the cluster management tool Kubernetes. The central end is used to perform fusion analysis on the aggregated data resource layer 300 flow survey data, realize the artificial intelligence model empowerment of the model support layer 200 of the entire flow survey process, and provide business-side flow survey intelligent question and answer, intelligent video analysis, inter-provincial joint investigation, federated learning and other functions.

[0055] Furthermore, the present invention can realize the integration of the epidemic investigation artificial intelligence model and business, and cover the entire epidemic investigation process. The specific epidemic investigation steps are as follows:

[0056] Step 1: Before the epidemiological investigation begins, the investigator registers the epidemiological investigation task through the mobile terminal.

[0057] Step 2: Conduct epidemiological investigation:

[0058] The epidemiologist will contact the subject of investigation by phone to conduct the investigation. A pop-up window will appear to remind you that the call from the subject of investigation is from the CDC, so please answer the call.

[0059] 1) Collect video and audio information of the subjects of the epidemic investigation

[0060] Specifically, the AI ​​computing power cluster includes an identity authentication model;

[0061] In step S2, remote identity verification of the epidemiological investigation subject includes:

[0062] The epidemic investigator collects audio information and / or video information of the epidemic investigation subject through the mobile terminal, and calls the identity authentication model to perform identity recognition based on the audio information and / or video information to assist in verifying the identity information of the epidemic investigator.

[0063] The above-mentioned identity authentication model can help epidemiologists conduct multimodal cross-verification of epidemiological investigation subjects through visual and voice dimensions, thereby assisting in verifying multiple attributes in the basic information of the epidemiological investigation subjects. The multiple attributes here include but are not limited to name, ID number, etc.

[0064] Among them, the above-mentioned identity authentication models include the ECAPA-TDNN model and the MTCNN model. In the telephone epidemic investigation scenario, the epidemic investigator can collect the real-time audio information of the epidemic investigation object through the mobile terminal, and click the "voiceprint verification" function to compare the audio with the sample in the voiceprint library of the epidemic investigation object through the ECAPA-TDNN model, calculate the similarity, and determine whether it passes the verification based on the set threshold. In the on-site epidemic investigation scenario, the epidemic investigator can record the on-site facial video of the epidemic investigation object and click the "face verification" function to detect and compare the face in the video with the image in the face library of the epidemic investigation object through the MTCNN model, calculate the similarity, and determine the verification result based on the threshold. In addition, the epidemic investigator can also conduct further verification by asking the epidemic investigation object's name, ID number and other basic information to complete the comprehensive identification of the epidemic investigation object.

[0065] 2) Officially enter the epidemiological investigation

[0066] Specifically, the AI ​​computing power cluster includes an intelligent assisted questioning model and a pre-built epidemic investigation question and answer knowledge graph, which is annotated with business scenarios and module elements;

[0067] In step S2, auxiliary epidemiological investigation questions and answers include:

[0068] When the epidemic investigator is performing the epidemic investigation task on the mobile terminal, the intelligent assisted questioning model is called to extract the personalized characteristics of the epidemic investigation object contained in the current epidemic investigation content, and dynamically matches the business scenarios and module elements in the epidemic investigation question and answer knowledge graph based on the personalized characteristics, and then displays the next round of question suggestions to the epidemic investigator based on the matched business scenarios and module elements.

[0069] In a preferred embodiment of the present invention, the business scenarios include a scenario for questioning about Class A and Class B and Class C infectious diseases;

[0070] The module elements include basic personal information, place of residence, place of work, medical treatment and medication, and activity trajectory survey.

[0071] In a preferred embodiment of the present invention, when the matched module elements are basic personal information, place of residence, place of work, medical treatment and medication, the intelligent auxiliary questioning model displays the next round of questioning suggestions to the epidemiologist according to the preset epidemiological survey question and answer template associated with the module elements;

[0072] When the matched module elements are for activity trajectory investigation, the intelligent assisted questioning model relies on the pre-deployed large language model to generate the next round of question suggestions.

[0073] Among them, it is preferred to pre-build an epidemic investigation question and answer knowledge graph covering two business scenarios of Class A and Class B and C (respiratory, intestinal, and vector-borne) infectious diseases, as well as five module elements (basic personal information, place of residence, place of work, onset of illness, medical treatment and medication, and activity trajectory investigation) based on existing epidemic investigation reports and conversation recording data, combined with the experience knowledge and information extraction technology of front-line epidemic investigators. The business scenarios and module elements are fully marked in the epidemic investigation question and answer knowledge graph.

[0074] When the epidemiologist contacts the epidemiological investigation subject by phone to perform the epidemiological investigation task, the intelligent assisted questioning model will integrate the personalized characteristics of the epidemiological investigation subject collected during the epidemiological investigation task registration (such as event type, pathogen type, personal basic information, certificate information, personnel affiliation, reporting source, etc.), and dynamically match business scenarios and module elements to ask precise questions.

[0075] The first four module elements are relatively fixed and are preferably deployed one by one through the establishment of epidemic investigation question and answer templates. The question elements of the activity trajectory survey are dynamically adjusted according to the specific circumstances of the activity venue (such as hospitals, entertainment venues, etc.). At this time, relying on the language generation ability, context understanding ability, thought chain reasoning ability, and extensive knowledge reserve of the large language model (GLM4 basic model), the prompt word project generates targeted suggestions for the next round of questions, thereby significantly improving the efficiency of epidemic investigation and the quality of question and answer.

[0076] In summary, by displaying the next round of question suggestions on the mobile terminal for the reference of the epidemiologists, intelligent epidemiological investigation can be completed through autonomous human-computer interaction, which solves the problem that the original process relies on professional epidemiologists, lacks epidemiological investigation skills and on-site handling experience, and is tedious and time-consuming. It improves the automation capability and realizes accurate epidemiological investigation.

[0077] 3) During the process of the epidemiologist asking questions and the epidemiological survey subjects responding

[0078] Specifically, the AI ​​computing power cluster includes a real-time recording model for epidemic investigation information;

[0079] In step S2, the process of generating the epidemiological investigation question and answer record includes:

[0080] When the epidemic investigator is performing the epidemic investigation task, the mobile terminal calls the real-time recording model of epidemic investigation information to monitor the multi-round conversation audio between the epidemic investigator and the epidemic investigation object in real time, and distinguishes the epidemic investigator's question audio and the epidemic investigation object's reply audio from the multi-round conversation audio, and then converts the question audio and reply audio into text for the epidemic investigator to modify and edit and generate an epidemic investigation question and answer record.

[0081] The FunASR model is the preferred model for real-time recording of epidemiological information. While conducting an epidemiological investigation, investigators can click the "Start Monitoring" button on their mobile device to capture multiple rounds of audio conversations between the investigator and the subject being investigated. The FunASR model then processes the audio. This model first distinguishes the speech of the investigator from the subject, then efficiently converts the speech to text, thereby recording epidemiological question and answer information in real time.

[0082] More specifically, the steps for calling the FunASR model to process audio are as follows:

[0083] Step 1: Use the FunASR model's voice activity detection module (FSMN-VAD) to extract valid speech segments from the audio.

[0084] Step 2: Use the FunASR model’s efficient embedding model ECAPA-TDNN to extract the speaker acoustic feature embedding of each speech segment;

[0085] Step 3: Introduce the cross-attention mechanism (CAM++) to strengthen the feature interaction between the reference and test speech, calculate similarity (such as cosine similarity), and determine whether they come from the same speaker based on a preset threshold;

[0086] Step 4: Map the judgment results back to the audio timeline to generate complete audio segments with speaker labels;

[0087] Step 5: The model also introduces SeACoParaformer hot word speech recognition technology to improve the recognition accuracy of proper nouns (such as place names and drug names) so as to convert the speech of epidemic investigators and subjects into text;

[0088] Step 6: Record the conversion results in the epidemiological investigation dialog box for further review and editing by the epidemiologist.

[0089] 4) Repeat 2) and 3) until all relevant information for the epidemiological investigation is obtained, completing contactless epidemiological investigation information collection. By incorporating the epidemiological investigation subject identification model, the intelligent auxiliary questioning model, and the real-time epidemiological investigation information recording model to assist in completing epidemiological investigation information collection, the contactless, comprehensive information collection capability can be effectively improved, minimizing the risk of infection during the epidemiological investigation process.

[0090] Specifically, the AI ​​computing power cluster includes an information credibility assessment model;

[0091] In step S2, when the flow investigation task is completed, it also includes:

[0092] The mobile terminal calls the information credibility assessment model to evaluate the spatiotemporal consistency of mobile phone signaling and the internal context consistency of the conversation on the epidemic investigation question and answer records to obtain the spatiotemporal consistency evaluation results and the context consistency evaluation results. When the spatiotemporal consistency evaluation results or the context consistency evaluation results indicate that the epidemic investigation personnel's reply information is unreliable, the epidemic investigation personnel are prompted to intervene to verify the information through confirmation questions, and at the same time provide a viewing and modification port for the epidemic investigation question and answer records.

[0093] The credibility assessment based on the spatiotemporal consistency of mobile phone signaling includes: judging the credibility based on the time and location information of the epidemiological survey subject's answer and the spatiotemporal data in the mobile phone signaling record. For time matching, it is judged whether the error between the answer time and the mobile phone signaling record time is within the range of ±1 hour; for location matching, a bigram-based word matching method is used to calculate the overlap of bigram phrases after word segmentation to evaluate the consistency between the answer location and the signaling record location;

[0094] Credibility assessment based on internal conversation context consistency: This assessment assesses credibility based on the consistency of responses from respondents to the same or similar questions within the same investigation. Using a large language model (GLM4 base model) and combined with annotation data from previous conversations, we employ instruction fine-tuning technology to assess contextual consistency of responses.

[0095] If both of the above credibility assessment results meet the requirements, the answer is deemed "credible"; otherwise, it is deemed "uncredible." If an answer is deemed "uncredible," the epidemic investigation personnel will be prompted to intervene and verify the information through confirmatory questions. The system also supports the review and modification of epidemic investigation question and answer records, completing the information credibility assessment and ensuring the authenticity and reliability of the epidemic investigation conversation content.

[0096] 5) After the epidemiologist completes the epidemiological investigation

[0097] Specifically, the AI ​​computing power cluster includes a structured parsing model for epidemic investigation elements and an automated writing model for epidemiological investigation reports;

[0098] In step S2, after the flow investigation task is completed, the following steps are also included:

[0099] The mobile terminal calls the epidemiological investigation element structured parsing model to extract key elements from the question and answer pairs in the epidemiological investigation question and answer records. The extracted key elements are then parsed and automatically stored in the database. The parsing results are displayed on the front end in the form of an epidemiological investigation report form for the epidemiologist to proofread and confirm.

[0100] After confirming that the epidemic investigation report form is filled out correctly, the automatic writing model of the epidemiological investigation report is called to generate an epidemiological investigation report that meets professional reporting requirements based on the preset epidemiological investigation report template.

[0101] Among them, the key to the processing of the structured parsing model of epidemic investigation elements is to use the questions asked by the epidemic investigators and the answers given by the epidemic investigation subjects in an epidemic investigation task as the input of the question and answer pair, and combine it with the predefined epidemic investigation question and answer template to identify the application scenarios and module elements (basic personal information, place of residence, place of work, onset of illness, medical treatment and medication, activity trajectory investigation), use the question intention recognition to determine the semantic target key elements of the question, and use the answer entity recognition to extract the corresponding epidemic information. The extracted key elements are then mapped to the corresponding table field values ​​in the form of JSON key-value pairs, and the back-end API is called to store the data in the database table. Finally, the question and answer pairs are converted into structured data and stored in the warehouse to provide support for subsequent queries and displays.

[0102] When generating an epidemiological survey report, first of all, you need to ensure that the structured data in the form is complete and accurate as the core input data source for generating the report. Based on the general framework of epidemiological surveys, a report template is designed that includes a fixed framework (such as personal basic information, onset and treatment conditions, epidemiological surveys, etc.) and dynamic paragraph interpolation points. By combining structured data (such as time, place, and symptom descriptions) with the natural language generation capabilities of large language models, the rigid structured data is organized into a fluent text description (such as: "On January 10, 2025, the epidemiological survey subject visited a supermarket and developed symptoms such as fever and cough."), and filled in the corresponding paragraphs of the template to generate a report.

[0103] In summary, by extracting and analyzing data on key elements through question-and-answer sessions and storing them in a database, and displaying the analysis results in an epidemic investigation report form, the automated extraction and management of epidemic investigation information can be achieved, greatly reducing the data entry time of epidemic investigators and providing data support for intelligent analysis of epidemic investigation data.

[0104] Step 3: Connect to the public security system surveillance video for intelligent video analysis and complete intelligent analysis and reasoning of epidemic investigation information:

[0105] Specifically, the epidemic investigation task also includes the close contact discovery task associated with the facial image of the case object. When performing the close contact discovery task, the epidemic investigator uploads the close contact discovery task to the central end for processing through the mobile end;

[0106] The center also connects to the public security system, and its AI computing cluster includes a close contact detection model. The center's processing includes:

[0107] Based on the facial images of the case subjects and multiple surveillance video files provided by the public security system, the center calls the close contact discovery model combined with video clip positioning and multi-target tracking methods to extract the mask wearing status, contact time points, and key frame screenshots of the case subjects and their contacts. Then, through machine learning classification model analysis, it determines whether they are close contacts or general contacts. At the same time, the key frame screenshots are input into the public security system's cooperative identity investigation interface for facial recognition and body posture recognition to determine the contact identity and provide support for subsequent close contact screening;

[0108] The AI ​​computing cluster includes models for intelligent epidemic analysis and tracing. The central processing also includes:

[0109] The tracking and tracing event of newly added case objects at the center is achieved by adding related cases to the Class A and A-controlled cases, and calling the epidemic intelligent analysis and tracking and tracing model to build a network map of the interpersonal relationships, whereabouts, and contact environment of the infected according to the epidemiological investigation reports of all related cases, to realize the analysis of the evolution mechanism of the epidemic spread, and to determine the possible patient zero of the case object while conducting close contact tracking through knowledge graph reasoning to realize tracking and tracing of the transmission chain analysis.

[0110] Specifically, the close contact detection model detects information such as the compliance of protective measures, safe distance, contact method, and duration of contact, extracting mask wearing status, contact time points, and key frame screenshots of the case subject and his / her contacts. Key frame screenshots contain image data of close contacts and epidemiological investigation subjects. The close contact detection model automatically identifies whether the case subject is a close contact or a general contact, effectively achieving accurate identification of epidemiological investigation subjects and their close contacts in complex scenarios.

[0111] Furthermore, the close contact discovery model preferably adopts semi-supervised video segmentation technology based on temporal consistency and context independence, effectively utilizing the temporal information between video frames to make up for the shortcomings of traditional target segmentation that only relies on static image training and evaluation; at the same time, the complex scene target detection model based on graph structure solves the problems of variable target scale, scarce target samples, increased types and target occlusion in complex scenes, significantly improving the detection accuracy and robustness.

[0112] After adding a new tracking and tracing event, add relevant cases from the Class A controlled cases to the event, click to generate a map, retrieve the structured data information from the epidemiological survey forms of all cases, and pass it into the epidemic intelligent analysis and tracking and tracing model. By constructing a network map of the infected person's interpersonal relationships, whereabouts, and contact environment, combined with traceability and tracking technology, the transmission chain analysis is completed.

[0113] Specifically, we first construct a map of personnel contacts, whereabouts, and contact environments, rely on dynamic time warping methods to analyze time series trends, cluster variables according to the evolution trend of abnormal states, and use time correlation analysis to realize the temporal discrimination of abnormal variables in each module, and construct an interpersonal relationship network of confirmed cases, close contacts, and secondary close contacts.

[0114] Provenance tracing technology uses Monte Carlo simulation to generate new data and annotates it to form a dataset. Combining the susceptible-infected-recovered model with a graph neural network, it estimates the probability of transmission caused by patient zero. Through dynamic message passing and Bayesian reasoning, it calculates the joint probability of different nodes being the source node and identifies the true source node (the higher the joint probability, the greater the probability that the node is the true source node).

[0115] This tracking technology uses knowledge graph reasoning to closely track connections. It employs a reinforcement learning algorithm to search for the optimal answer path within the knowledge graph and maximizes the probability of reaching the correct answer by selecting the decision sequence for relationship edges. To improve parsing accuracy, the algorithm introduces bidirectional path constraint search, making the reasoning path more precise and enabling flexible response to complex problems without the need for pre-training.

[0116] In summary, facial photos of case subjects can be collected or imported from the facial database through the epidemiological investigation task management synchronization task or the registration of new close contact discovery tasks. This can then be used to retrieve the corresponding public security system surveillance video files and invoke the close contact discovery model based on surveillance video to automatically identify close contacts and assist in guiding close contact screening. Faced with the ever-changing epidemiological investigation scenarios, the wide range of people involved, and the difficulty of tracing the source, this system can quickly identify close contacts, analyze the evolution of the epidemic spread, and track and trace the source, ensuring precise prevention and control of the situation.

[0117] Step 4: If the epidemiological investigation requires the cooperation of disease control units in other provinces, the coordinated investigation information and epidemic investigation data can be securely shared among the disease control units in multiple provinces that deploy the intelligent epidemic investigation auxiliary platform. Based on this, the artificial intelligence computing power cluster includes a blockchain-based epidemic investigation data on-chain model and a blockchain-based epidemic investigation data verification model;

[0118] There are multiple central ends to form a blockchain network. The central end is also used for the initiator of the cooperative investigation to create a new cooperative investigation event, and call the blockchain-based epidemic investigation data chain model to store the structured information of the cooperative investigation event and the unstructured cooperative investigation attachments on the chain; the encrypted epidemic investigation data is stored as blocks, each block contains information such as timestamp, data fingerprint and hash value of the previous block, forming a chain to ensure the encryption protection of the data, and can be shared and tracked in the distributed network.

[0119] This allows coordinating parties to verify the on-chain incidents and asynchronously validate the data of these incidents using a blockchain-based epidemic investigation data verification model. Leveraging the blockchain's immutability and consensus mechanism, the epidemic investigation data is verified to be consistent with the on-chain records. Once data enters the blockchain, all nodes can verify it, and the consensus mechanism confirms the validity of the data. If the data is tampered with or forged, the distributed nature of the blockchain will automatically detect and reject the tampering.

[0120] It also allows the coordinating parties to register corresponding epidemic investigation tasks based on the coordinating information. After the epidemic investigation task is completed, the blockchain-based epidemic investigation data chain model is called to store the generated epidemiological investigation report on the chain as the coordinating feedback result, ensuring that the feedback results are also protected and tracked by the blockchain.

[0121] It also allows the initiator of the cooperative investigation to verify and review the data of the cooperative investigation feedback results associated with the cooperative investigation events stored on the chain to ensure the completeness and accuracy of the feedback results.

[0122] Among them, the cooperative investigation function is carried out based on blockchain technology to realize the transmission of cooperative investigation information among the three places in the Yangtze River Delta where an integrated intelligent epidemic investigation assistance platform is deployed: the cooperative investigation summary information is uploaded to the chain and sent to the target area through the interface. After the other party completes the cooperative investigation, the summary of the cooperative investigation results will also be uploaded to the chain to ensure the safe sharing of information.

[0123] More specifically, the specific steps for implementing secure sharing of collaborative investigation information and epidemic investigation data among disease control units in multiple provinces deploying the intelligent epidemic investigation assistance platform are as follows:

[0124] 1) The party applying for joint investigation initiates the joint investigation application and publishes it on the chain: a new joint investigation event is created and the joint investigation summary information, joint investigation letter, etc. are uploaded to the chain for storage, and the release on the chain ensures the security and immutability of multi-source heterogeneous data.

[0125] 2) The co-investigating party receives and verifies the out-of-province co-investigation application: performs data verification on the received co-investigation event, and completes data asynchronous verification and integrity verification.

[0126] 3) The co-investigating party carries out the cooperative investigation and epidemiological investigation task: synchronize the cooperative investigation event to the epidemiological investigation task management module, complete the epidemiological investigation according to the above step 2 process, generate an epidemiological investigation report, and feedback the summary of the cooperative investigation results through the above-mentioned chain model interface.

[0127] 4) The party applying for the cooperative investigation reviews the cooperative investigation feedback: the data of the received cooperative investigation results is verified and the cooperative investigation feedback results are reviewed.

[0128] Step 5. Federated Learning Console

[0129] Specifically, the AI ​​computing power cluster includes a multi-source heterogeneous flow model training system based on federated learning;

[0130] It also involves pre-deploying a federated learning console, connecting to the corresponding central terminals of disease control units in multiple provinces. By invoking a multi-source heterogeneous epidemiological investigation model training system based on federated learning, federated learning is implemented to train the various models in the artificial intelligence computing cluster. Federated learning nodes are established in three locations in the Yangtze River Delta. Each computing node performs local model training without sharing original data, and only the updated model parameters are aggregated to the central server. This distributed learning method can effectively protect data privacy and security, reduce data transmission requirements, and reduce the computational burden on the central server, thereby improving system efficiency.

[0131] By building federated learning nodes between disease control units in multiple provinces, computing nodes can complete model training that follows secure computing protocols without sharing original data, thereby improving model accuracy.

[0132] The above description is only a preferred embodiment of the present invention and does not limit the implementation mode and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. An intelligent auxiliary flow investigation method for the entire flow investigation process, characterized in that: include: Step S1: deploying an artificial intelligence computing cluster at the center; In step S2, the epidemiologist registers the epidemiological investigation task through the mobile terminal he holds when conducting the epidemiological investigation, and interacts with the central terminal during the execution of the epidemiological investigation task to call the artificial intelligence computing power cluster to perform remote identity authentication on the epidemiological investigation object and assist in epidemiological investigation questions and answers, and generates an epidemiological investigation report based on the generated epidemiological investigation question and answer records after the epidemiological investigation task is completed.

2. The intelligent assisted flow adjustment method according to claim 1, characterized in that: The artificial intelligence computing power cluster includes an identity authentication model; In step S2, remote identity authentication of the flow survey object includes: The epidemic investigator collects audio information and / or video information of the epidemic investigation object through the mobile terminal, and calls the identity authentication model to perform identity recognition based on the audio information and / or the video information to assist in verifying the identity information of the epidemic investigator.

3. The intelligent assisted flow adjustment method according to claim 1, characterized in that: The artificial intelligence computing power cluster includes an intelligent auxiliary questioning model and a pre-built epidemic investigation question and answer knowledge graph, in which business scenarios and module elements are annotated; In step S2, the auxiliary epidemiological investigation question and answer includes: During the process of the epidemic investigator performing the epidemic investigation task, the mobile terminal calls the intelligent auxiliary questioning model to extract the personalized characteristics of the epidemic investigation object contained in the current epidemic investigation content, and dynamically matches the business scenarios and module elements in the epidemic investigation question and answer knowledge graph based on the personalized characteristics, and then displays the next round of questioning suggestions to the epidemic investigator based on the matched business scenarios and module elements.

4. The intelligent assisted flow adjustment method according to claim 3 is characterized in that: The business scenarios include the scenario of asking questions about Class A and Class B and Class C infectious diseases; The module elements include basic personal information, place of residence, place of work, medical treatment and medication history, and activity trajectory survey.

5. The intelligent assisted flow adjustment method according to claim 4 is characterized in that: When the matched module elements are basic personal information, place of residence, place of work, medical treatment and medication, the intelligent auxiliary questioning model displays the next round of questioning suggestions to the epidemiologist according to the preset epidemiological survey question and answer template associated with the module elements; When the matched module element is the activity trajectory survey, the intelligent auxiliary questioning model generates the next round of question suggestions based on the pre-deployed large language model.

6. The intelligent assisted flow adjustment method according to claim 1, characterized in that: The artificial intelligence computing power cluster includes a real-time recording model for epidemic investigation information; In step S2, the process of generating the epidemiological investigation question and answer record includes: During the process of the epidemic investigator performing the epidemic investigation task, the mobile terminal calls the real-time recording model of epidemic investigation information to monitor the multiple rounds of conversation audio between the epidemic investigator and the epidemic investigation object in real time, and distinguishes the epidemic investigator's question audio and the epidemic investigation object's reply audio from the multiple rounds of conversation audio, and then converts the question audio and the reply audio into text for the epidemic investigator to modify and edit and generate the epidemic investigation question and answer record.

7. The intelligent assisted flow adjustment method according to claim 1, characterized in that: The artificial intelligence computing power cluster includes an information credibility assessment model; In step S2, when the flow investigation task is completed, the following steps are also included: The mobile terminal calls the information credibility assessment model to perform mobile phone signaling spatiotemporal consistency and conversation internal context consistency assessment on the epidemic investigation question and answer record to obtain spatiotemporal consistency assessment results and context consistency assessment results, and when the spatiotemporal consistency assessment results or the context consistency assessment results indicate that the epidemic investigation personnel's reply information is unreliable, the epidemic investigation personnel are prompted to intervene to verify the information through confirmation questions, and at the same time provide a viewing and modification port for the epidemic investigation question and answer record.

8. The intelligent assisted flow adjustment method according to claim 1, characterized in that: The artificial intelligence computing power cluster includes a structured parsing model for epidemic investigation elements and an automated writing model for epidemiological investigation reports; In step S2, after the flow investigation task is completed, the following steps are also included: The mobile terminal calls the epidemiological investigation element structured parsing model to extract key elements from the question and answer pairs in the epidemiological investigation question and answer record, then parses the extracted key elements and automatically stores them in the database, and displays the parsing results in the form of an epidemiological investigation report form on the front end for the epidemiologist to proofread and confirm; After confirming that the flow survey report form is filled out correctly, the automatic writing model of the epidemiological investigation report is called to generate the epidemiological investigation report that meets the professional reporting requirements based on the preset epidemiological investigation report template.

9. The intelligent assisted flow adjustment method according to claim 1, characterized in that: The epidemiological investigation task also includes a close contact discovery task associated with a facial image of a case subject. When performing the close contact discovery task, the epidemiologist uploads the close contact discovery task to the central end for processing via the mobile end. The central end is also connected to the public security system, and the artificial intelligence computing cluster includes a close contact detection model; the processing process of the central end includes: The central end uses the facial image of the case subject and the multiple location surveillance video files provided by the public security system to call the close contact discovery model combined with the video clip positioning and multi-target tracking method to extract the mask wearing status, contact time points and key frame screenshots of the case subject and his / her contacts, and then uses the machine learning classification model to determine whether they are close contacts or general contacts. At the same time, the key frame screenshots are input into the auxiliary investigation identity interface of the public security system for facial recognition and body posture recognition to determine the identity of the contact and provide support for subsequent close contact screening; The AI ​​computing cluster includes an intelligent epidemic analysis and tracing model; the central processing also includes: The center adds a tracking and tracing event for the case object, adds relevant cases to the Class A and Class A controlled cases, and calls the epidemic intelligent analysis and tracking and tracing model to build a network map of the interpersonal relationships, whereabouts, and contact environment of the infected person based on the epidemiological survey reports of all relevant cases, to realize the analysis of the evolution mechanism of the epidemic spread, and to determine the possible patient zero of the case object while conducting close contact tracking through knowledge graph reasoning to realize tracking and tracing of the transmission chain analysis.

10. The intelligent assisted flow adjustment method according to claim 1, characterized in that: The artificial intelligence computing power cluster includes a blockchain-based epidemic investigation data on-chain model and a blockchain-based epidemic investigation data verification model; There are multiple central terminals, forming a blockchain network. The central terminal is also used for the initiator of the cooperative investigation to create a new cooperative investigation event, and call the blockchain-based epidemic investigation data chain model to store the structured information of the cooperative investigation event and the unstructured cooperative investigation attachments on the chain; The coordinating party verifies the on-chain coordinating event and calls the blockchain-based flow investigation data verification model to perform asynchronous data verification on the coordinating event; The coordinating party registers the corresponding epidemic investigation task according to the coordinating information, and after the epidemic investigation task is completed, calls the blockchain-based epidemic investigation data on-chain model to store the generated epidemiological investigation report on the chain as the coordinating feedback result; And the initiator of the cooperative investigation can verify and view the data of the cooperative investigation feedback results associated with the cooperative investigation event stored on the chain.

Images