Health data disaster recovery backup method and system

Abstract

The invention relates to the technical field of data disaster recovery backup, and discloses a health data disaster recovery backup method and system, and the method comprises the steps: respectively deploying a heterogeneous agent and a sensing agent in three preset data centers; performing protocol conversion and priority distribution on the medical data based on the heterogeneous proxy to obtain a priority sequence, and executing consistency verification based on the priority sequence to obtain a verification result; performing fault prediction based on the data center performance indexes collected by the sensing intelligent agent to obtain a fault prediction value; according to the verification result and the fault prediction value, generating a fault arbitration result, and when the fault arbitration result is that a fault is detected, executing a switching operation of the data center to obtain a disaster recovery backup result, the application can identify a potential fault risk in advance and trigger a preventive measure, can adaptively adjust a disaster recovery response scheme according to a system state, and improves the system reliability. A traditional static rule engine is replaced, and intelligent and autonomous management of the disaster recovery backup system is achieved.

Description

Technical Field

[0001] This application relates to the field of data disaster recovery and backup technology, and in particular to a method and system for disaster recovery and backup of health data. Background Technology

[0002] With the deepening of medical informatization, various medical institutions have widely deployed heterogeneous medical information systems such as HIS (Hospital Information System), PACS (Picture Archiving and Communication System), and LIS (Laboratory Information System). These systems use different data formats and communication protocols, such as HL7, DICOM, and XML, generating massive amounts of medical data. Medical data has special attributes such as high sensitivity, extremely high real-time requirements, and strict business continuity requirements. Any data loss or system interruption may directly threaten the patient's life safety. Therefore, data disaster recovery and backup systems have far higher technical requirements than general information systems. Existing medical data disaster recovery and backup technologies mainly adopt the traditional dual-center master-slave replication architecture, which performs backup through a simple data synchronization mechanism. However, this dual-center architecture has the risk of single point of failure. When the main data center fails, problems are likely to occur, and data consistency cannot be guaranteed. Summary of the Invention

[0003] This application provides a method and system for disaster recovery backup of health data. This application can identify potential failure risks in advance and trigger preventive measures, and can adaptively adjust the disaster recovery response plan according to the system status. It replaces the traditional static rule engine and realizes intelligent and autonomous management of the disaster recovery backup system.

[0004] The first aspect of this application provides a method for disaster recovery backup of health data, the method comprising: Heterogeneous agents and perceptual agents are deployed in three pre-designed data centers respectively; Based on the heterogeneous agent, medical data is converted using a protocol to obtain standardized data; The standardized data is prioritized to obtain a priority sequence, and a consistency verification is performed based on the priority sequence to obtain the verification result. Fault prediction is performed based on the data center performance indicators collected by the sensing agent to obtain the fault prediction value. A fault arbitration result is generated based on the verification result and the fault prediction value. When the fault arbitration result indicates that a fault has been detected, a data center switchover operation is performed to obtain a disaster recovery backup result.

[0005] In conjunction with the first aspect, in a first implementation of the first aspect of the present invention, the deployment of heterogeneous agents and perceptual agents in three preset data centers respectively includes: A ring topology network is configured to connect the three preset data centers and a bidirectional communication link is established between each data center to obtain a data center network. The three data centers include a primary data center, a backup data center, and an arbitration data center. Based on the data center network, heterogeneous agents are deployed in each data center, and the heterogeneous agents are used to support the HL7 protocol parser; The heterogeneous agent is connected to the HIS system, PACS system, and LIS system. The data interface type and data format specification of each medical system are automatically identified through the medical system discovery protocol to obtain the medical system connection mapping table. Based on the medical system connection mapping table, a perceptual agent is deployed in each data center.

[0006] In conjunction with the first aspect, in a second implementation of the first aspect of the present invention, the step of connecting the heterogeneous agent with the HIS system, PACS system, and LIS system, and automatically identifying the data interface type and data format specifications of each medical system through a medical system discovery protocol to obtain a medical system connection mapping table includes: Based on the heterogeneous agent, the medical system discovery protocol is initiated to scan the network environment of each data center and identify the HIS system, PACS system and LIS system in the data center network through port detection and service response, so as to obtain a list of medical system discovery. The protocol type of each medical system in the medical system discovery list is identified to obtain system protocol feature information, and data interface connections are made to each medical system according to the system protocol feature information to obtain system connection channels; Based on the system connection channel, the data format specifications of each medical system are parsed, the HL7 message segment structure, DICOM tag information, and XML node hierarchy are extracted, and a medical system connection mapping table is constructed based on the HL7 message segment structure, the DICOM tag information, and the XML node hierarchy.

[0007] In conjunction with the first aspect, in a third implementation of the first aspect of the present invention, the step of performing protocol conversion on medical data based on the heterogeneous proxy to obtain standardized data includes: Based on the medical system connection mapping table, an HL7 message format parser, a DICOM image data converter, and an XML structured data processor are integrated into the heterogeneous agent to obtain a multi-protocol heterogeneous agent; The multi-protocol heterogeneous agent is used to identify the format of medical data from the HIS, PACS, and LIS systems to obtain protocol matching results. Based on the protocol matching result, the corresponding parser of the multi-protocol heterogeneous proxy is driven to perform data structure parsing to obtain structured parsed data; Based on the structured parsed data, the multi-protocol heterogeneous agent performs medical semantic transformation to obtain standardized data.

[0008] In conjunction with the first aspect, in a fourth implementation of the first aspect of the present invention, the step of prioritizing the standardized data to obtain a priority sequence, and performing consistency verification based on the priority sequence to obtain a verification result, includes: Priority identifiers are obtained by assigning priority based on the urgency of medical services in the standardized data, including vital sign monitoring data, emergency test results, general medical records, and historical archive data. Data stream scheduling is performed based on the priority identifier to obtain a priority sequence; The standardized data is verified according to the priority sequence to obtain consensus verification data; Based on the consensus verification data, a medical business logic check is performed to obtain the verification results.

[0009] In conjunction with the first aspect, in a fifth implementation of the first aspect of the present invention, the step of verifying the standardized data according to the priority sequence to obtain consensus verification data includes: Based on the data order in the priority sequence, the primary data center is set as the master node for the pre-preparation phase, and a data proposal message containing medical data content, timestamp, and digital signature is sent to the backup data center and the arbitration data center through the primary data center. According to the data proposal message, the preparation phase is verified in the backup data center and the arbitration data center. The received standardized data is checked for format correctness, timestamp validity and digital signature integrity and a preparation confirmation message is sent. Once more than two-thirds of the data centers have received the prepared confirmation messages, the submission phase operation is triggered, and a submission confirmation result is obtained. Based on the submitted confirmation results, a medical business continuity constraint check is performed to obtain consensus verification data.

[0010] In conjunction with the first aspect, in the sixth implementation of the first aspect of the present invention, the step of performing fault prediction based on the data center performance indicators collected by the sensing agent to obtain a fault prediction value includes: The main data center, the backup data center, and the arbitration data center are monitored by the performance indicator monitoring module of the sensing agent to obtain data center performance indicators. The performance indicator monitoring module includes a system resource collector, a network status analyzer, and a service load evaluator. The data center performance indicators are input into the time-series pattern analysis module of the sensing agent for deep learning to obtain a time-series feature vector. The time-series pattern analysis module includes an LSTM neural network, which contains an input layer, three LSTM units, and an output layer. The temporal feature vector is input into the fault probability calculation module of the perceptual agent for risk assessment to obtain the fault prediction value. The fault probability calculation module includes a probability prediction network and a threshold decision unit.

[0011] In conjunction with the first aspect, in the seventh implementation of the first aspect of the present invention, the step of generating a fault arbitration result based on the verification result and the fault prediction value, and performing a data center switching operation when the fault arbitration result indicates that a fault has been detected, to obtain a disaster recovery backup result, includes: The verification results and the fault prediction values ​​are input into the arbitration data center. The version number, timestamp, and checksum information of each data object in the primary data center, the backup data center, and the arbitration data center are compared through the global consistency status table to obtain the consistency check results. Based on the consistency check results, the arbitration algorithm of the three data centers is executed to obtain the fault arbitration result; Based on the fault arbitration result, the reinforcement learning decision-making module of the perceptual agent is used to select a switching strategy to obtain an intelligent switching strategy. The intelligent switching strategy drives the corresponding data center to perform fault switching operations and monitors the RTO recovery time target and RPO recovery point target in real time during the switching process to obtain disaster recovery backup results.

[0012] In conjunction with the first aspect, in the eighth implementation of the first aspect of the present invention, the step of selecting a switching strategy using the reinforcement learning decision module of the perceptual agent based on the fault arbitration result to obtain an intelligent switching strategy includes: Based on the fault arbitration result, the data center performance indicators are input into the reinforcement learning decision module for state encoding to obtain the data center state vector; Based on the data center state vector, executable actions are identified through the action space definer of the reinforcement learning decision module to obtain a set of candidate actions; The data center state vector and the candidate action set are input into the Q-learning algorithm to calculate the action value, resulting in an action value matrix; The medical service priority is weighted according to the action value matrix, and the optimal switching action is selected using an ε-greedy strategy. Based on the optimal switching action, the strategy parameters are configured to generate an intelligent switching strategy.

[0013] A second aspect of this application provides a health data disaster recovery backup system, the health data disaster recovery backup system comprising: The deployment module is used to deploy heterogeneous agents and sensing agents in three pre-defined data centers respectively; The protocol conversion module is used to convert medical data according to the heterogeneous agent to obtain standardized data. The consistency verification module is used to assign priorities to the standardized data to obtain a priority sequence, and perform consistency verification based on the priority sequence to obtain the verification result. The fault prediction module is used to predict faults based on the data center performance indicators collected by the sensing agent, and obtain the fault prediction value. The switching operation module is used to generate a fault arbitration result based on the verification result and the fault prediction value. When the fault arbitration result indicates that a fault has been detected, the data center switching operation is performed to obtain a disaster recovery backup result.

[0014] Compared with existing technologies, this application has the following beneficial effects: By deploying heterogeneous agents and perceptual intelligent agents, the technical level of the medical data disaster recovery and backup system is significantly improved. The heterogeneous agent integrates multi-protocol adaptation interfaces, realizing unified conversion and medical semantic perception processing of heterogeneous medical systems such as HIS, PACS, and LIS, solving the problem of poor compatibility in traditional solutions. The perceptual intelligent agent utilizes LSTM neural networks for fault prediction, achieving a technological leap from passive response to proactive prevention. The three-data center ring topology completely eliminates the risk of single-point failures and split-brain problems, significantly improving system reliability. The Byzantine fault-tolerant consensus algorithm provides tamper-proof data security, meeting the stringent requirements of the medical industry. The reinforcement learning decision module achieves intelligent disaster recovery management through the Q-learning algorithm, replacing the traditional static rule engine. The medical business priority mechanism ensures the continuity of critical medical services, effectively protecting patient safety. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] The structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0017] Figure 1 This is a flowchart illustrating the health data disaster recovery backup method provided in an embodiment of the present invention; Figure 2 This is a schematic block diagram of the structure of the health data disaster recovery backup system provided in the embodiments of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this application specification and the appended claims refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations. See also Figure 1 One embodiment of the health data disaster recovery backup method in this application includes: Step 100: Deploy heterogeneous agents and sensing agents in the three preset data centers respectively; It is understood that the executing entity of this application can be a health data disaster recovery backup system, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0022] Specifically, a ring topology network connection configuration is performed on the three preset data centers. A logical closed-loop structure is constructed to achieve bidirectional redundant communication links between the primary data center, backup data center, and arbitration data center, forming a stable data center network structure with fault recovery capabilities. This network structure can maintain interconnection and data access to the remaining nodes even if any node fails. Heterogeneous proxy modules with multi-protocol adaptation capabilities are deployed within the data centers. Each heterogeneous proxy embeds an HL7 protocol parser, enabling correct format parsing and field recognition when dealing with structured data transmitted from various medical business systems, and supporting compatibility between multiple HL7 V2.x and V3.0 versions. By binding communication links between the heterogeneous proxies and various medical information systems, including HIS, PACS, and LIS systems, access configuration for these systems is achieved. Based on the medical system discovery protocol, system probe operations are proactively initiated to automatically identify the data interface types, data format specifications, and message structure information opened by each system, generating a medical system connection mapping table containing system identifiers, protocol types, and data structure definitions. Based on the medical system connection mapping table, the initial installation and adaptation of the sensing agents in each data center are guided, so that each sensing agent can accurately identify and associate the data source type of its service when it starts up, and at the same time carry out tasks such as collecting system performance parameters, monitoring status and sensing data quality.

[0023] Step 200: Perform protocol conversion on the medical data based on the heterogeneous agent to obtain standardized data; Specifically, based on the medical system connection mapping table, sub-module components with multi-protocol processing capabilities are integrated into the heterogeneous agents deployed in various data centers. These include an HL7 message format parser for processing structured patient information, a DICOM image data converter for processing medical images, and an XML structured data processor for extracting complex nested structural information, thereby constructing a multi-protocol heterogeneous agent. The multi-protocol heterogeneous agent automatically connects to the interface channels of medical information systems such as HIS, PACS, and LIS, receiving data streams from each system in real time. It performs protocol identification and structure judgment operations on each input data, obtaining protocol matching results including data type, version specifications, and field structure. Based on the protocol matching results, the heterogeneous agent dynamically loads the corresponding protocol parser module and performs syntax-level structure parsing operations on the received medical data content, extracting key elements such as field tags, hierarchical relationships, timestamps, diagnostic codes, and image identifiers to form machine-readable structured parsed data. The multi-protocol heterogeneous agent calls its built-in medical semantic conversion logic module, which combines ontology mapping rules, medical terminology dictionary and disease code conversion table to perform cross-system semantic alignment and unified standardization processing of terminology, code and label information in structured parsed data, thereby completing the data mapping and conversion from raw heterogeneous medical data to a unified expression format and obtaining standardized data.

[0024] Step 300: Prioritize the standardized data to obtain a priority sequence, and perform consistency verification based on the priority sequence to obtain the verification results; It should be noted that the urgency of medical information is identified based on the different types of medical information contained in the standardized data. Vital signs monitoring data, directly related to the patient's current vital signs, is classified as the highest urgency level. Emergency test results, due to their direct support for emergency decision-making, are assigned high priority. Routine medical records, important in daily medical practice, are given medium priority. Historical archived data, as long-term archival material, is marked as the lowest priority. This forms a set of priority markers reflecting the urgency of medical services. Based on a preset dynamic scheduling strategy, resources are allocated to all standardized data to be transmitted synchronously. High-priority data occupies a larger proportion of network bandwidth and storage I / O resources to ensure the real-time transmission and integrity of critical medical data, forming a data flow priority sequence arranged according to urgency. A consistency verification process is triggered based on the priority sequence. In the verification phase, an improved Byzantine fault-tolerant algorithm is used to prioritize format verification, timestamp verification, and signature validity verification for high-priority data, ensuring the integrity, correctness, and source reliability of critical medical data during transmission, generating a set of preliminary consensus verification data. Based on consensus verification data, medical business logic checks are performed, automatically analyzing the temporal relationships, dependency paths, and logical integrity between various data. For example, test reports must be generated after test requests, and prescription drugs must be issued based on clear diagnostic results. If there are abnormal records that violate medical business processes, the abnormal records are marked as business inconsistencies and are refused to be written. The output is a data verification result that has both semantic consistency assurance and technical verification compliance.

[0025] Step 400: Based on the data center performance indicators collected by the sensing agent, perform fault prediction to obtain fault prediction values; Specifically, the performance indicator monitoring module, integrated into the sensing agents deployed in the main data center, backup data center, and arbitration data center, continuously monitors the operational status of each data center. The performance indicator monitoring module consists of a system resource collector, a network status analyzer, and a business load evaluator. The system resource collector acquires key resource parameters such as CPU utilization, memory usage, disk read / write speed, and database connection count in real time. The network status analyzer monitors communication quality indicators such as network latency, packet loss rate, and link jitter. The business load evaluator assesses the current medical business processing pressure through task queue length, request response time, and interface concurrency, forming a multi-source performance indicator data set containing multiple dimensions. Performance index data is input into the temporal pattern analysis module of the perceptual agent for temporal modeling and deep learning. The temporal pattern analysis module integrates a set of LSTM neural network structures based on Long Short-Term Memory (LSTM) characteristics. Specifically, it includes an input layer for receiving multi-dimensional performance data sequences, three stacked LSTM unit layers for extracting long-term and short-term dependencies of system state changes, and generating temporal feature vectors reflecting the system's evolutionary trend. The output layer abstracts the learned implicit patterns into intermediate representation vectors suitable for prediction. The temporal feature vectors are then input into the fault probability calculation module of the perceptual agent to perform risk assessment. The fault probability calculation module consists of a probability prediction network and a threshold determiner. The probability prediction network calculates the probability of system failure within a specific future time window (e.g., 15 minutes) based on learned abnormal behavior features. The threshold determiner dynamically compares and analyzes the probability values; when the probability value exceeds a preset threshold (e.g., 80%), it is automatically identified as a potentially high-risk state and a fault prediction value is generated and output.

[0026] Step 500: Generate a fault arbitration result based on the verification result and fault prediction value. When the fault arbitration result indicates that a fault has been detected, perform a data center switchover operation to obtain a disaster recovery backup result.

[0027] Specifically, the verification results output from the Byzantine fault-tolerant consistency verification mechanism and the fault prediction values ​​calculated by the fault prediction model in the perceptual agent are input into the arbitration data center. The arbitration data center then initiates a global consistency state comparison process. Using its internally maintained global consistency state table, it performs parallel comparisons of the version number, timestamp, and checksum recorded for each medical data object in the primary data center, backup data center, and arbitration data center's three storage environments. This generates a multi-dimensional consistency check result, used to identify version differences, redundancy conflicts, or potential tampering in each data center. Based on the consistency check result, the three-center arbitration algorithm is invoked. This algorithm integrates timestamp priority judgment logic, majority version identification mechanism, and business priority weight correction factor to comprehensively evaluate the legality and dominance of the current data state, generating a fault arbitration result. The fault arbitration result determines whether a real fault exists and indicates the fault's attribution node and corresponding impact scope. After the fault arbitration confirms the existence of a recoverable fault, the fault level and impact scope are input into the reinforcement learning decision-making module in the perceptual agent. This module, combining historical failover experience, current system load, and business continuity requirements, calculates the optimal intelligent failover strategy based on a policy update mechanism, explicitly selecting a service-level, system-level, or network-level failover mode. The failover control unit drives the target data center to perform the actual failover operation according to the intelligent failover strategy. Simultaneously, the system monitors the fulfillment of recovery time and recovery point targets in real time during the failover process, ensuring that data synchronization latency is controlled within 15 seconds and service recovery does not exceed 30 seconds, ultimately outputting disaster recovery backup results that meet high availability requirements.

[0028] In one specific embodiment, the process of performing step 100 may specifically include the following steps: Configure a ring topology network to connect the three preset data centers and establish bidirectional communication links between them to obtain a data center network. The three data centers include a primary data center, a backup data center, and an arbitration data center. Heterogeneous agents are deployed in each data center based on the data center network. These heterogeneous agents are used to support the HL7 protocol parser. The heterogeneous agent is connected to the HIS system, PACS system, and LIS system. The data interface type and data format specification of each medical system are automatically identified through the medical system discovery protocol to obtain the medical system connection mapping table. Based on the medical system connection mapping table, perceptual agents are deployed in each data center.

[0029] Specifically, a unified topology design was implemented for the network structure between the primary data center, backup data center, and arbitration data center. A ring topology architecture was adopted to configure bidirectional network links between the three points, forming redundant and mutually backup data center network channels. This ensures that even if one center fails, the remaining two centers can still maintain complete communication capabilities, enhancing the controllability of data flow and the flexibility of path switching. After completing the network topology, heterogeneous proxy modules with multi-protocol adaptation capabilities were deployed within each data center. These heterogeneous proxy modules include parser components supporting HL7 protocol versions 2.x and 3.0 to ensure the correct processing of message structures transmitted from various medical information systems, including various types of medical data such as patient basic information, examination requests, and diagnostic results. The heterogeneous agent establishes logical connection channels with the locally connected HIS, PACS, and LIS systems respectively. Through the medical system discovery protocol, it proactively probes and interacts with the connected systems, automatically identifying the open data interface types, message protocol specifications, encoding structures, and data field mapping relationships of each system. Based on the identification results, a unified medical system connection mapping table is constructed. This table uses the system identifier as an index and records the corresponding interface address, protocol type, data structure pattern, and access rules. After the medical system connection mapping table is constructed, a sensing agent is deployed in each data center according to the configuration information in the mapping table. This ensures that the sensing agent can associate with its monitored target data system during initialization and configure sensing strategies based on the protocol type, data characteristics, and load behavior of the connected medical system, thereby achieving comprehensive perception of system resource usage status, network performance, and business request load.

[0030] In one specific embodiment, the process of connecting the heterogeneous agent with the HIS system, PACS system, and LIS system, and automatically identifying the data interface type and data format specifications of each medical system through the medical system discovery protocol to obtain the medical system connection mapping table can specifically include the following steps: The heterogeneous agent-based medical system discovery protocol scans the network environment of each data center and identifies the HIS, PACS and LIS systems in the data center network through port probing and service response, thus obtaining a list of medical system discovery. The protocol type of each medical system in the medical system discovery list is identified to obtain system protocol feature information. Based on the system protocol feature information, data interface connections are made to each medical system to obtain system connection channels. Based on the system connection channel, the data format specifications of each medical system are parsed, the HL7 message segment structure, DICOM tag information, and XML node hierarchy are extracted, and a medical system connection mapping table is constructed based on the HL7 message segment structure, DICOM tag information, and XML node hierarchy.

[0031] Specifically, a predefined medical system discovery protocol is initiated in the heterogeneous agents deployed in various data centers. This protocol has the ability to automatically scan local area networks and subnets. It performs traversal detection of target service nodes in the network through active broadcasting and reverse listening mechanisms. During the traversal, it sequentially performs multi-port scanning tasks through TCP and UDP protocol stacks, focusing on commonly used communication ports of standard medical systems, such as TCP 2101 commonly used in HIS systems, DICOM C-STORE service port 104 used in PACS systems, and SOAP / XML interface ports 80 and 8080 of LIS systems. After detecting a responsive open port, the heterogeneous agent sends a handshake request or standard protocol detection command to the target service. Based on the returned data packet content, response format, field tags, or handshake characteristics, it determines the service type and system name, and generates an information list covering all detected medical systems, i.e., the medical system discovery list, which includes the IP address, port number, service response characteristics, and preliminary determination of system type for each system. By initiating characteristic interaction requests to each service node in the medical system discovery list, specific protocol handshake and data fragment analysis methods are adopted for different types of service nodes. For example, an HL7 V2.x format MSH segment initialization message is sent to suspected HIS system nodes and its response legality is judged; a DICOM C-ECHO command is sent to suspected PACS systems to check their DICOM protocol stack support; or a SOAP request payload is sent to LIS systems and their return of a structured XML format response is observed. This identifies the actual communication protocol type, version information, and encoding specifications used by each system, and extracts system protocol characteristic information such as protocol header fields, session identifiers, and structure tags. The heterogeneous agent constructs data access connection channels according to the matched protocol type, and uses the corresponding protocol stack to perform handshake, authentication, and data channel establishment operations on the target medical system to form a logically reachable system connection channel. Each connection channel corresponds to a stable medical information system data source interface. Based on the system connection channel, the heterogeneous agent performs in-depth analysis of the data format specifications flowing through the medical system. It extracts the segment-level structure information of HL7 messages from the HIS system, identifies key structures such as the message header fields of the MSH segment, the patient basic information fields of the PID segment, and the examination request parameters of the OBR segment; it extracts the DICOM image data tag structure from the PACS system, and parses tag information such as PatientID, StudyInstanceUID, Modality, and ImagePositionPatient; and it extracts the multi-level node hierarchy structure of XML format data from the LIS system, and identifies the correspondence between node paths, field levels, and attribute values.By unifying the abstraction and semantic mapping of different system protocol structures, a medical system connection mapping table is constructed in each data center. The medical system connection mapping table uses the system identifier as the main index and records the system type, communication protocol, data structure template, key field location and parsing rules.

[0032] In one specific embodiment, the process of performing step 200 may specifically include the following steps: Based on the medical system connection mapping table, an HL7 message format parser, a DICOM image data converter, and an XML structured data processor are integrated into a heterogeneous agent to obtain a multi-protocol heterogeneous agent; The format of medical data from HIS, PACS and LIS systems is identified by a multi-protocol heterogeneous agent to obtain protocol matching results. Based on the protocol matching results, the corresponding parser of the multi-protocol heterogeneous proxy is driven to perform data structure parsing to obtain structured parsed data; Based on structured parsing data, multi-protocol heterogeneous agents perform medical semantic transformation to obtain standardized data.

[0033] Specifically, a medical system connection mapping table serves as the unified system access benchmark. This table records the network addresses, protocol types, data structure characteristics, and interaction specifications of the HIS, PACS, and LIS systems in each data center. The heterogeneous proxy performs module initialization operations based on the medical system connection mapping table, loading parsing components corresponding to the identified protocol types as needed. These include an HL7 message format parser for processing structured diagnostic and treatment business information, a DICOM image data converter for parsing metadata and pixel content in medical image data, and an XML structured data processor for reading hierarchical data structures transmitted via SOAP or REST interfaces. This forms a multi-protocol heterogeneous proxy system architecture within the proxy, possessing protocol awareness and structure analysis capabilities. The system continuously receives raw business data streams from the HIS, PACS, and LIS medical systems through the network access channels of the multi-protocol heterogeneous proxy. It performs a format recognition process on the received data content, quickly determining the protocol type of the current data by comparing data header identifiers, message structure characteristics, and field arrangement patterns, and then generating a set of protocol matching results. Based on the protocol matching results, the multi-protocol heterogeneous agent routes the corresponding data to the corresponding parser module and drives the parser module to perform data structure parsing operations. For example, for structured text data identified as HL7 protocol, the HL7 message format parser reads key segments such as MSH, PID, OBR, and OBX line by line at the segment level, extracts timestamp, patient identifier, test item code, and result value fields, and verifies the legality and coding accuracy of the fields. For image data files identified as DICOM protocol, the DICOM image data converter reads the image header information according to the tag dictionary, extracts tag values ​​such as PatientName, StudyDate, Modality, SeriesDescription, and PixelData, and parses the pixel matrix format, compression method, and data dimension information. For XML protocol data, the XML structured data processor parses tags such as SOAP Envelope, Body, Message, and Data according to the node hierarchy, and uses XPath paths to quickly locate and extract specific medical attribute values. After the structured parsing of the data is completed, the multi-protocol heterogeneous agent enters the semantic conversion stage. It uses the internally embedded medical terminology dictionary, encoding mapping table and semantic ontology rules to normalize the heterogeneous codes and terms used in each system. For example, it performs cross-system semantic alignment and mapping unification of the ICD-10 disease codes used in the HIS system, the examination site codes used in the PACS system and the test item codes used in the LIS system to build a standardized data representation structure.

[0034] In one specific embodiment, the process of performing step 300 may specifically include the following steps: Priority labels are obtained by assigning priority based on the urgency of medical services using standardized data such as vital sign monitoring data, emergency test results, general medical records, and historical archive data. Data stream scheduling is performed based on priority identifiers to obtain a priority sequence; Standardized data is validated according to a priority sequence to obtain consensus validation data; The medical business logic is checked based on the consensus verification data, and the verification results are obtained.

[0035] Specifically, business type information is extracted from standardized medical data, and its data category is identified through content fields. Vital signs monitoring data, as continuous high-frequency data directly reflecting a patient's vital signs, is assigned the highest priority and marked as P0. Emergency laboratory results contain urgent judgments about disease status and have a decisive impact on the timeliness of clinical intervention; their business priority is P1. Routine medical records document daily outpatient and emergency procedures, routine medications, or follow-up visits, and have long-term diagnostic significance, but their urgency is lower than that of emergency situations; they are classified as P2. Historical archived data is mainly used for follow-up analysis, statistical assessment, or compliance auditing; it falls under the category of non-real-time data and belongs to the lowest priority, P3. Based on the classification criteria, each piece of standardized data is automatically assigned a corresponding priority label using business rule templates, forming a priority tag set. Data flow scheduling is performed based on priority identifiers. This process is jointly executed by the resource perception module and scheduling strategy module within the sensing intelligence body. By reading operational indicators such as network latency, bandwidth usage, and storage I / O load between data centers, the system dynamically calculates the resource allocation ratio for data of different priorities. 80% of the network bandwidth and cache resources are prioritized for P0 and P1 level data transmission channels, 15% for P2 level data, and the remaining 5% for P3 level archived data, forming an optimized priority data flow sequence. Simultaneously, the scheduling algorithm integrates a congestion control mechanism. When a link latency greater than 100 milliseconds or a packet loss rate higher than 0.1% is detected, the system automatically reduces the transmission rate of low-priority data and suspends its cache refresh frequency, freeing up transmission windows and I / O channels for high-priority data, thereby maintaining a highly available transmission path for core medical business data. Based on a priority sequence, a data consistency verification process is initiated. Through an improved Byzantine fault tolerance mechanism, the primary data center, backup data center, and arbitration data center perform three phases of verification operations on the scheduled data: pre-preparation, preparation, and submission. This primarily verifies the completeness of the data content, the validity of the timestamps, and the matching of the source signatures. Consensus verification data is generated when a majority of nodes have confirmed the data. This consensus verification data is then input into the medical business logic inspection module. This module constructs a logical constraint rule graph based on medical behavior process specifications. For example, test reports must be generated after the test request form, prescription drugs must be bound to an effective diagnostic code, and the time of test results cannot be earlier than the test request registration time. By verifying the consistency of time series, causal dependencies, and primary key relationships between data items in the logical graph, potential time series anomalies, logical conflicts, or business process errors are identified. All data items violating business rules are marked as anomalies, triggering an alarm process or data rollback mechanism. Data that meets all technical and business constraints is officially judged as having passed verification, thus forming the data verification result.

[0036] In one specific embodiment, the process of validating standardized data according to a priority sequence to obtain consensus verification data can specifically include the following steps: Based on the data order in the priority sequence, the primary data center is set as the master node for the pre-preparation phase. Data proposal messages containing medical data content, timestamps, and digital signatures are sent from the primary data center to the backup data center and the arbitration data center. Based on the data proposal message, the preparation phase verification is performed in the backup data center and the arbitration data center. The received standardized data is checked for format correctness, timestamp validity and digital signature integrity, and a preparation confirmation message is sent. Once more than two-thirds of the data centers have received preparation confirmation messages, the commit phase operation is triggered, and the commit confirmation result is obtained. Based on the submitted confirmation results, a medical business continuity constraint check is performed to obtain consensus verification data.

[0037] Specifically, in the data stream that has already been prioritized, the processing order of the medical data to be synchronized in the current batch is determined according to the priority sequence generated by the scheduling algorithm. The main data center is then dynamically set as the master node for this round of consensus process, and is responsible for initiating the pre-preparation phase of the Byzantine Fault Tolerance protocol. During the pre-preparation phase, the main data center sequentially encapsulates each piece of standardized medical data to be processed into a data proposal message. The data proposal message includes the original data content, the standardized structure, the data generation timestamp, and a digital signature encrypted with the main data center's private key. The main data center broadcasts the data proposal message to the backup data center and the arbitration data center via a bidirectional ring network connection, and records the unique identifier and sending status of each sent message. After receiving a data proposal message from the primary data center, the backup data center and the arbitration data center enter a preparation phase verification process. This process verifies whether the data structure conforms to registered standardized field specifications through a parsing module, such as HL7 field distribution, DICOM tag integrity, and XML node hierarchy compliance, ensuring data parsing compatibility at the format level. It also verifies the validity of the timestamp in the message, ensuring it is within the system's current time window and not earlier than the generation time of any dependent data to prevent rollback or duplicate synchronization. Finally, it verifies the integrity of the digital signature in the message by comparing the signature value with the data digest result using public key decryption to confirm that the data has not been tampered with during transmission. If all three verifications pass, the backup data center and the arbitration data center generate a preparation confirmation message and return it to the primary data center, marking the standardized data locally as "preparation successful." When the primary data center successfully collects preparation confirmation messages from more than two-thirds of the data centers (i.e., any two data centers) within the specified time window, thus satisfying the strong consistency consensus triggering condition proposed in the Byzantine fault tolerance protocol, the commit phase operation is initiated. The status of the data item is updated from "pre-commit" to "formal commit," and all data centers are notified to synchronously write medical data. Simultaneously, the commit time and write node version number are recorded, resulting in a commit confirmation result. Based on the commit confirmation result, a medical business continuity constraint check is performed. Semantic-level logical verification is executed based on the medical behavior dependency graph maintained internally by the system. This checks whether the submitted data violates the temporal relationships and business constraint rules in medical operations. For example, it ensures that each prescription data has valid diagnostic support, test reports must be generated after the corresponding application record, and nursing records are not earlier than the hospitalization registration time. By analyzing the logical relationship between the submitted data and historical records, business integrity is reviewed. Data that meets all business continuity constraints is marked as "consensus verification passed," ultimately forming a consensus verification dataset.

[0038] In one specific embodiment, the process of performing step 400 may specifically include the following steps: The main data center, backup data center, and arbitration data center are monitored by the performance indicator monitoring module of the sensing agent to obtain data center performance indicators. The performance indicator monitoring module includes a system resource collector, a network status analyzer, and a service load evaluator. The data center performance indicators are input into the time-series pattern analysis module of the perceptual agent for deep learning to obtain time-series feature vectors. The time-series pattern analysis module includes an LSTM neural network, which contains an input layer, three LSTM units, and an output layer. The temporal feature vector is input into the fault probability calculation module of the perceptual agent for risk assessment to obtain the fault prediction value. The fault probability calculation module includes a probability prediction network and a threshold decision unit.

[0039] Specifically, proactive sensing agents are deployed in the main data center, backup data center, and arbitration data center, respectively. The performance indicator monitoring module within these agents is activated. A collaborative monitoring unit, comprised of a system resource collector, a network status analyzer, and a service load estimator, performs fine-grained detection and real-time data collection of the data center's operational status from three perspectives: resource usage, network communication, and service processing. The system resource collector continuously acquires metrics such as CPU utilization, memory usage, disk I / O operation rate, thread pool usage, and database connection count at a preset time resolution. The network status analyzer periodically monitors link latency, packet loss rate, bandwidth utilization, and the number of communication anomalies. The service load estimator dynamically assesses the current service pressure level from dimensions such as request queue length, transaction processing time, and the number of concurrent services. These monitoring results are integrated into a multi-dimensional performance indicator dataset, which forms structured time-series data on a timeline, reflecting the dynamic evolution of the data center's status. The perceptual agent inputs a multi-dimensional performance index dataset into its internal time-series pattern analysis module. The time-series pattern analysis module embeds a deep learning structure with a Long Short-Term Memory (LSTM) neural network as its core. The neural network consists of an input layer, three LSTM units, and an output layer. The input layer receives the multi-dimensional index vector within the current time window. The three LSTM units are used to extract the coupling relationship between short-term fluctuations and long-term trends, capture the temporal dependence and evolution path between performance indicators, and perform state memory and time updates on the extracted dynamic features to construct a time-series feature vector that reflects the trend of system behavior. The temporal feature vector is input into the fault probability calculation module of the perceptual agent. The fault probability calculation module consists of a probability prediction network and a threshold decision-maker. The probability prediction network uses a feedforward structure combined with a Softmax or Sigmoid function to perform risk probability mapping on the input feature vector and outputs the probability value of an anomaly occurring in the data center within a specific future time window (such as 5 minutes or 15 minutes). The threshold decision-maker sets a critical threshold for fault probability based on historical training samples. When the predicted value exceeds the critical threshold, it is marked as a high-risk state and a corresponding fault prediction value is generated as output, which includes the probability value of the fault occurrence and a predicted fault type label, such as CPU resource saturation, disk write latency anomaly, or network communication congestion.

[0040] In one specific embodiment, the process of performing step 500 may specifically include the following steps: The verification results and fault prediction values ​​are input into the arbitration data center. The version number, timestamp, and checksum information of each data object in the primary data center, backup data center, and arbitration data center are compared through the global consistency status table to obtain the consistency check results. Based on the consistency check results, the arbitration algorithm of the three data centers is executed to obtain the fault arbitration result; Based on the fault arbitration result, the reinforcement learning decision-making module of the perceptual agent is used to select the switching strategy, and an intelligent switching strategy is obtained. The intelligent switching strategy drives the corresponding data center to perform failover operations and monitors the RTO recovery time target and RPO recovery point target in real time during the switching process to obtain disaster recovery backup results.

[0041] Specifically, the arbitration data center receives verification results from the consistency verification mechanism and fault prediction values ​​generated by the perception agent prediction module. The verification results are used to determine the synchronization accuracy of medical data across the three data centers, while the fault prediction values ​​are used to identify data center nodes that may experience current or future failures. The arbitration data center activates its internally maintained global consistency status table, which records the version number, generation timestamp, and checksum information of all standardized medical data objects in the primary, backup, and arbitration data centers. By comparing the status values ​​of the same data object in each center, the version consistency status and timing synchronization of each data object are calculated. If a data object has the same timestamp in the primary and backup data centers but a different checksum, it indicates a risk of data tampering or inconsistent writing. If a node's version number is significantly lagging, it means that the node's write has failed or data synchronization has been interrupted. These analysis results constitute the consistency check results and are marked in the status table in the form of structured tags. Based on the consistency check results, the multi-node disaster recovery arbitration algorithm is invoked. The arbitration algorithm comprehensively considers three factors: data version distribution, timestamp validity, and business priority weight. In particular, high-priority medical data (such as vital signs and emergency data) has priority in the arbitration process. If the main data center's data is complete and up-to-date, it maintains its master node role; otherwise, the backup or arbitration data center takes over the service. The output fault arbitration result indicates the data center where the fault occurred and designates the takeover node that should assume the master control role. The arbitration data center inputs the fault arbitration result into the reinforcement learning decision module of the perceptual agent. The reinforcement learning decision module performs policy optimization calculations on the current system state based on a preset state space, action space, and reward function. The state space includes the current load status, network connectivity, and resource utilization of each data center. The action space includes service-level, system-level, or network-level switching methods. The reward function takes into account the maintenance of business continuity, resource utilization efficiency, and service availability after the switch. The optimal switching strategy, i.e., the intelligent switching strategy, is generated through a deep Q-network (DQN) or policy gradient method. The intelligent switching strategy specifies the switching scope, the target switching center, and the required resource reconfiguration scheme. The arbitration data center drives specific execution modules to initiate switching operations based on intelligent switching strategies. These operations include multiple processes such as data service redirection, session state migration, and cache synchronization and reconstruction. During the entire switching operation, two key indicators are monitored in real time: Recovery Time Objective (RTO), which is the time required from fault identification to full service availability. By compressing the session migration process and warming up the backup node, the recovery time is ensured to be no more than 30 seconds. Recovery Point Objective (RPO), which is the time difference between the data integrity after the switch and the state before the fault, is controlled to be no more than 15 seconds through write-ahead logs, high-frequency synchronization, and data snapshot mechanisms.After the switchover is complete, a complete disaster recovery backup result is generated, including the source of the fault, the switchover target, the switchover method, and the recovery indicators.

[0042] In one specific embodiment, the process of selecting a switching strategy using the reinforcement learning decision module of the perceptual agent based on the fault arbitration result, and obtaining the intelligent switching strategy, can specifically include the following steps: Based on the fault arbitration result, the data center performance indicators are input into the reinforcement learning decision module for state encoding to obtain the data center state vector; Based on the data center state vector, the action space definer of the reinforcement learning decision module is used to identify executable actions and obtain a set of candidate actions. The data center state vector and candidate action set are input into the Q-learning algorithm to calculate the action value, resulting in the action value matrix; The medical business priorities are weighted according to the action value matrix, and the optimal switching action is selected using an ε-greedy strategy. Based on the optimal switching action, the strategy parameters are configured to generate an intelligent switching strategy.

[0043] Specifically, the arbitration data center receives the fault determination results output by the preceding fault arbitration module and digitally models the current system state by combining the real-time performance indicators of each data center. These performance indicators include CPU utilization, memory usage ratio, disk I / O rate, network latency, service request queuing length, and historical failure rate. The state encoder in the reinforcement learning decision module performs multi-dimensional combination, normalization, and temporal flattening operations on the above indicators to construct a data center state vector reflecting the overall system operation status. The data center state vector describes the current load and communication status of each node and embeds semantic information such as the fault impact range, node stability level, and service priority distribution. The data center state vector is input into the action space definer in the reinforcement learning module. The action space definer combines the resource bottleneck location, link reachability, and fault level characteristics reflected in the state vector to dynamically generate a set of feasible actions. These actions include disaster recovery solutions such as migrating some services from the primary data center to the backup data center, switching the network to redundant links, activating the read-only mode of the arbitration center, or temporarily upgrading it to the primary data center, forming a set of state-related candidate actions. The reinforcement learning module invokes the Q-learning algorithm, inputting the data center state vector and candidate action set into the Q-function calculation engine. It evaluates the expected benefit of each action in the current state through iterative Q-value calculation, generating an action value matrix containing all state-action pairs. This quantifies the ability of each executable action to improve overall disaster recovery performance in the current system state. Based on the action value matrix, the Q-value matrix is ​​weighted and adjusted according to the business priority information of the current medical data to be processed. Specifically, action paths involving P0 and P1 level data are given higher priority weights, while action paths affecting only P2 or P3 level data are subject to discount factors. This ensures that the switching decision fully responds to business urgency and data security requirements. The reinforcement learning module employs an ε-greedy strategy to select actions from the current action value matrix. The action with the largest Q-value is selected as the optimal switching action with a probability of 1-ε, and a sub-optimal action is randomly selected with a probability of ε to maintain strategy exploration capabilities and avoid getting trapped in local optima. Based on the optimal switching action and the target node state, strategy parameters are configured, including service migration path, connection persistence time, data snapshot synchronization frequency, and log replay window, outputting an intelligent switching strategy.

[0044] The above describes the health data disaster recovery backup method in the embodiments of this application. The following describes the health data disaster recovery backup system in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the health data disaster recovery backup system in this application includes: Deployment module 021 is used to deploy heterogeneous agents and sensing agents in three preset data centers respectively; Protocol conversion module 022 is used to convert medical data according to heterogeneous agents to obtain standardized data. The consistency verification module 023 is used to prioritize standardized data, obtain a priority sequence, and perform consistency verification based on the priority sequence to obtain the verification result. The fault prediction module 024 is used to predict faults based on the data center performance indicators collected by the sensing agent, and obtain the fault prediction value. The switching operation module 025 is used to generate a fault arbitration result based on the verification result and the fault prediction value. When the fault arbitration result is that a fault is detected, the data center switching operation is performed to obtain the disaster recovery backup result.

[0045] Through the collaborative efforts of the aforementioned components, this invention deploys a heterogeneous agent with multi-protocol adaptability interfaces, integrating an HL7 message format parser, a DICOM image data converter, and an XML structured data processor. This enables intelligent identification and unified conversion of data formats from different medical information systems, solving the technical challenge of poor compatibility between heterogeneous systems in traditional solutions and ensuring the semantic integrity and format consistency of medical data during disaster recovery backup. By introducing a perceptual intelligent agent, and through the collaborative work of a performance indicator monitoring module, a time-series pattern analysis module, and a fault probability calculation module, deep learning analysis of system performance data is performed using an LSTM neural network. This achieves a technological leap from traditional passive fault response to proactive fault prediction, enabling early identification of potential fault risks and triggering preventative measures. The adoption of a three-center ring topology architecture—a primary data center, a backup data center, and an arbitration data center—completely eliminates the single-point-of-failure risk and split-brain problem inherent in traditional dual-center solutions. The three-center data consistency arbitration algorithm ensures that the system can still operate normally even in the event of a single-center failure, significantly improving the reliability of the disaster recovery backup system. By introducing the Byzantine Fault Tolerance consensus algorithm, a three-phase verification mechanism of pre-preparation, preparation, and commit is employed. This not only tolerates node failures but also defends against malicious attacks and data tampering. Reinforcement learning technology is applied to disaster recovery decision-making. The optimal switching strategy is dynamically learned through the Q-learning algorithm, enabling adaptive adjustments to the disaster recovery response plan based on system status. This replaces the traditional static rule engine, achieving intelligent and autonomous management of the disaster recovery backup system. Addressing the unique characteristics of medical data, a priority allocation mechanism based on the urgency of medical services is established. Critical medical data such as vital sign monitoring data and emergency test results are allocated higher priority and resources, ensuring the continuity of critical medical services in the event of failures and effectively protecting patient safety. Through a medical terminology dictionary and disease coding mapping table, semantic alignment of professional terms across different medical systems is achieved. This not only completes data format conversion but also maintains the professional semantic integrity of medical data, avoiding the semantic loss of medical information caused by simple format conversion in traditional solutions.

[0046] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0047] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0048] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A health data disaster backup method, characterized by, The application relates to a fault arbitration method and device for a data center. The application comprises the following steps: Deploying heterogeneous agents and perception agents in three preset data centers respectively; Converting medical data based on the heterogeneous agents to obtain standardized data; Assigning priorities to the standardized data to obtain a priority sequence, and performing consistency verification based on the priority sequence to obtain a verification result; Performing fault prediction based on the data center performance indexes collected by the perception agents to obtain a fault prediction value; 2.The health data disaster backup method according to claim 1, characterized in that, Generating a fault arbitration result according to the verification result and the fault prediction value, and performing a switching operation of the data center when the fault arbitration result is a detected fault to obtain a disaster recovery backup result. The application comprises the following steps: Configuring a ring topology network connection of the three preset data centers and establishing a bidirectional communication link between the data centers to obtain a data center network, wherein the three data centers comprise a main data center, a backup data center and an arbitration data center; Deploying heterogeneous agents in each data center based on the data center network, wherein the heterogeneous agents are used for supporting an HL7 protocol parser; Connecting the heterogeneous agents with HIS systems, PACS systems and LIS systems, and automatically identifying the data interface types and data format specifications of the medical systems through a medical system discovery protocol to obtain a medical system connection mapping table; 3.The health data disaster backup method according to claim 2, characterized in that, Deploying perception agents in each data center based on the medical system connection mapping table. The application comprises the following steps: Starting the medical system discovery protocol based on the heterogeneous agents to scan the network environment of each data center and identify the HIS systems, PACS systems and LIS systems in the data center network through port detection and service response to obtain a medical system discovery list; Identifying the protocol types of each medical system in the medical system discovery list to obtain system protocol characteristic information, and connecting the data interfaces of each medical system according to the system protocol characteristic information to obtain a system connection channel; 4. The health data disaster backup method according to claim 3, characterized by, Analyzing the data format specifications of each medical system based on the system connection channel, extracting HL7 message segment structures, DICOM label information and XML node hierarchical relationships, and constructing a medical system connection mapping table based on the HL7 message segment structures, the DICOM label information and the XML node hierarchical relationships. The application comprises the following steps: Integrating an HL7 message format parser, a DICOM image data converter and an XML structured data processor in the heterogeneous agents based on the medical system connection mapping table to obtain a multi-protocol heterogeneous agent; Identifying the formats of the medical data of the HIS systems, the PACS systems and the LIS systems through the multi-protocol heterogeneous agent to obtain a protocol matching result; According to the protocol matching result, the corresponding parser of the multi-protocol heterogeneous agent is driven to perform data structure parsing, and structured parsing data is obtained; Based on the structured parsing data, the multi-protocol heterogeneous agent is driven to perform medical semantic conversion, and standardized data is obtained.

5. The health data disaster backup method according to claim 4, characterized by, The standardized data is assigned a priority, a priority sequence is obtained, and consistency verification is performed based on the priority sequence to obtain a verification result, including: Based on the medical business emergency degree of the vital sign monitoring data, emergency test results, general diagnosis and treatment records, and historical archive data in the standardized data, priority identification is performed to obtain priority identification; According to the priority identification, data flow scheduling is performed to obtain a priority sequence; According to the priority sequence, the standardized data is verified to obtain consensus verification data; Based on the consensus verification data, medical business logic checking is performed to obtain a verification result.

6. The health data disaster backup method according to claim 5, wherein, According to the priority sequence, the standardized data is verified to obtain consensus verification data, including: Based on the data sequence in the priority sequence, the master data center is set as a master node to perform a pre-preparation stage, and the master data center sends a data proposal message containing medical data content, a timestamp and a digital signature to the backup data center and the arbitration data center; According to the data proposal message, the backup data center and the arbitration data center perform a preparation stage verification, and the received standardized data is checked for format correctness, timestamp validity and digital signature integrity, and a preparation confirmation message is sent; When more than two-thirds of the data centers collect the preparation confirmation message, a commit phase operation is triggered to obtain a commit confirmation result; Based on the commit confirmation result, medical business continuity constraint checking is performed to obtain consensus verification data.

7. The health data disaster backup method according to claim 6, characterized by, The fault prediction value is obtained based on the data center performance index collected by the perception agent, including: The performance index monitoring module of the perception agent is used to monitor the master data center, the backup data center and the arbitration data center to obtain data center performance indexes, and the performance index monitoring module includes a system resource collector, a network state analyzer and a business load evaluator; The data center performance indexes are input into the time sequence mode analysis module of the perception agent for deep learning to obtain a time sequence feature vector, and the time sequence mode analysis module includes an LSTM neural network, which includes an input layer, three layers of LSTM units and an output layer; The time sequence feature vector is input into the fault probability calculation module of the perception agent for risk assessment to obtain a fault prediction value, and the fault probability calculation module includes a probability prediction network and a threshold determinator.

8. The health data disaster backup method according to claim 7, characterized by, According to the verification result and the fault prediction value, a fault arbitration result is generated, and when the fault arbitration result is a detected fault, a data center switching operation is performed to obtain a disaster recovery backup result, including: The verification result and the fault prediction value are input into the arbitration data center, a consistent check result is obtained by comparing the version number, timestamp and checksum information of each data object in the primary data center, the backup data center and the arbitration data center through a global consistent state table; An arbitration algorithm of the three data centers is executed according to the consistent check result, and a fault arbitration result is obtained; A switching strategy selection is performed by using a reinforcement learning decision module of the perception agent according to the fault arbitration result, and an intelligent switching strategy is obtained; The corresponding data center is driven to perform a fault switching operation according to the intelligent switching strategy, and the RTO recovery time target and the RPO recovery point target in the switching process are monitored in real time, and a disaster recovery backup result is obtained.

9. The health data disaster backup method according to claim 8, characterized by, The intelligent switching strategy is obtained by using the reinforcement learning decision module of the perception agent according to the fault arbitration result, and the intelligent switching strategy comprises: The data center performance index is input into the reinforcement learning decision module for state coding according to the fault arbitration result, and a data center state vector is obtained; Executable action recognition is performed by an action space definer of the reinforcement learning decision module based on the data center state vector, and a candidate action set is obtained; The data center state vector and the candidate action set are input into a Q-learning algorithm for action value calculation, and an action value matrix is obtained; The action value matrix is subjected to medical business priority weighting processing, and an optimal switching action is selected by using an ε-greedy strategy, and a strategy parameter configuration is performed based on the optimal switching action, and an intelligent switching strategy is generated.

10. A health data disaster backup system, characterized by, The health data disaster recovery backup system for executing the health data disaster recovery backup method according to any one of claims 1-9 comprises: A deployment module is configured to deploy heterogeneous agents and perception agents in the three preset data centers respectively; A protocol conversion module is configured to convert medical data into standardized data based on the heterogeneous agents; A consistency verification module is configured to assign priorities to the standardized data to obtain a priority sequence, and perform consistency verification based on the priority sequence to obtain a verification result; A fault prediction module is configured to perform fault prediction based on the data center performance index collected by the perception agent to obtain a fault prediction value; A switching operation module is configured to generate a fault arbitration result according to the verification result and the fault prediction value, and perform a switching operation of the data center when the fault arbitration result indicates that a fault is detected, to obtain a disaster recovery backup result.

Images