Teaching data management system
By employing a multimodal fusion architecture and intelligent hierarchical storage strategy in the teaching data management system, the problems of multimodal data collaboration and periodic awareness in educational data management have been solved, achieving efficient storage and analysis and improving the accuracy and reliability of teaching evaluation decisions and services.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-17
- Publication Date
- 2026-04-10
AI Technical Summary
Existing technologies in education data management suffer from problems such as lack of multimodal data collaboration, lag in periodic perception and response, and insufficient service agility, making it difficult to effectively support the efficient storage and analysis of multi-source heterogeneous education data.
A teaching data management system was designed, which adopts a multimodal fusion architecture and a microservice architecture. Combining the characteristics of data lakes and data warehouses, it achieves unified management and efficient analysis of multi-source heterogeneous data through intelligent hierarchical storage and dynamic migration strategies.
It enables efficient storage and analysis of multimodal data, supports the accuracy of teaching evaluation decisions and the high reliability of services, optimizes storage performance and cost, and adapts to dynamic load changes in teaching scenarios.
Smart Images
Figure CN121833656A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data management, and particularly relates to a teaching data management system. BACKGROUND
[0002] The digital transformation of education is profoundly reshaping the teaching management mode, one of the core driving forces of which is the explosive growth of multi-source heterogeneous education data, which covers structured (such as score sheets), semi-structured (such as teaching logs) and unstructured (such as teaching audio and video) and other forms, posing a serious challenge to traditional data architecture. The current mainstream solution has three limitations: 1. The lack of ACID transaction of data lake: although architectures such as Hadoop support low-cost storage of multi-modal data, they lack ACID guarantee, resulting in insufficient accuracy of teaching evaluation decisions based on such data; 2. Schema rigidity of data warehouse: although MPP warehouse has efficient transaction processing and analysis capabilities, the rigid constraints of its Schema make it difficult to adapt to dynamically generated teaching resources (such as real-time learning behavior data), which can easily cause data redundancy storage problems; 3. Lack of dynamic response to periodic load: teaching scenarios have significant strong periodic characteristics (for example, data access peaks during exam week / low data access during holidays), and existing static storage strategies cannot dynamically perceive and respond to such load fluctuations, which restricts resource utilization efficiency and service response capability.
[0003] Lake-warehouse integration, as a new data management paradigm, combines the flexibility of data lake (such as low-cost storage of massive unstructured teaching videos) with the efficiency of data warehouse (such as ACID transaction support for accurate score analysis), providing a new path for education data management. However, the application of lake-warehouse integration paradigm to the education field still faces three challenges: first, the lack of multi-modal data collaboration: structured score data and unstructured classroom videos are often stored separately, hindering cross-modal data fusion and correlation analysis; second, periodic perception and response lag: relying on static and unified storage strategies makes it difficult to respond in real time to dynamic needs driven by school calendar events (such as high-frequency access to test paper library resources before exams); third, insufficient service agility: the high coupling degree of functional modules under the monolithic architecture makes it difficult to quickly support agile development and deployment of innovative applications such as personalized learning path recommendations, for which a teaching data management system is proposed. SUMMARY
[0004] The purpose of the present application is to solve the problems in the prior art and propose a teaching data management system.
[0005] A teaching data management system comprises: Data integration layer: unified data specification description through standardized metadata management framework; multi-modal fusion architecture is used to integrate multi-source heterogeneous data; a unified data integration engine is designed, and modular open source tool chain is used to realize dynamic standardized collection, cleaning and conversion of full or incremental data; Data platform layer: fusion of multi-modal storage of data lake and ACID transaction characteristics of data warehouse, design of education lake-warehouse integrated data platform, including: Intelligent hierarchical storage module: for data storage; Development governance integration module: for data management; Hierarchical teaching data warehouse module: for data application; Data service layer: encapsulating data processing and analysis functions based on microservice architecture, using Dubbo RPC+ZooKeeper registration center to realize unified scheduling of education data services, using RESTful API to standardize encapsulation of core functions as independent microservices, supporting upper-layer teaching application calls, and using a dual-center governance model to realize data service management.
[0006] Preferably, in the data integration layer, the standardized metadata management framework integrates dynamic Schema evolution and data bloodline tracking technology; the multi-source heterogeneous data includes structured data, semi-structured data and unstructured data; For structured data, realize real-time synchronization of database through DataX or Sqoop tools, and combine Kafka to guarantee millisecond-level transmission and transaction consistency; For semi-structured data, use Kafka, LogStash or canal to dynamically parse time series data, automatically generate Schema through regular or XPath and convert it into structured data; For unstructured data, use AI-driven feature extraction technology to convert unstructured content into vectorized feature labels, and establish a unified label library for management.
[0007] Preferably, the intelligent hierarchical storage module is used to build an education data intelligent hierarchical storage architecture that integrates the multi-modal storage capability of the data lake and the ACID transaction characteristics of the data warehouse, specifically including: (1) Unified data management: unified storage of structured and unstructured data based on Delta Lake or Hudi, and optimization of multi-dimensional combined query performance through Z-Order index; (2) Intelligent hierarchical storage unit: dynamically scheduling hot, warm and cold three-layer storage; (3) Cross-modal integration: establish mapping association between video metadata and learning behavior logs.
[0008] Preferably, the development governance integration module is used to integrate the development governance integration DataOps concept, design a serverless lake warehouse architecture to realize the integration of development governance, specifically including: (1) Metadata management: build metadata bloodline link based on Apache Atlas; (2) Quality rule library management: AI algorithm driven generated quality rule library, realize quality control in development stage; (3) Scheduling management: through DolphinScheduler to coordinate Kubernetes to realize ETL task scheduling and resource dynamic scheduling; (4) Compliance and security management: data desensitization, data encryption and audit module are embedded into ETL data cleaning task, realizing sensitive data desensitization and encryption.
[0009] Preferably, the hierarchical teaching data warehouse module is used to build an ODS-DW-ADS three-layer architecture, store business detail data through DeltaLake or Hudi table, store and manage database basic configuration through MySQL, support high-frequency access through Redis cache server, and realize the standardized management of teaching data from collection, cleaning to analysis; The ODS is the original data layer, which stores unmodified raw data; The DW is the data warehouse layer, which stores the integrated theme wide table; The ADS is the application service layer, which stores lightweight data for application.
[0010] Preferably, the intelligent hierarchical storage unit specifically includes: Dynamic grading strategy: based on data access heat gradient and business importance weight, the data is accurately allocated to NVMe SSD, HDD or object storage and other differentiated storage layers; Dual-drive adaptive engine: periodic driving uses LSTM time series prediction to identify load changes triggered by calendar events, multi-modal driving associates multi-modal data such as video-log through metadata tags, and realizes dynamic migration data; Intelligent hierarchical storage mechanism engine: through multi-modal feature fusion analysis of data attributes, based on heat decision function, automatically trigger hierarchical storage strategy and dynamic migration strategy, form a closed-loop management mechanism.
[0011] Preferably, the heat decision function is specifically: (1) Heat decision function definition Through Fourier series modeling of teaching period, embedding event factor response to examination week fluctuation, separating semester trend and short-term disturbance, accurately analyzing education time series data characteristics, based on prediction results , fusion real-time access Cross-system association Feature, build decision function, formula as follows:
[0012] Wherein, The heat decision value of the time window, The number of data access in the time window, Cross-system association weight, The teaching cycle influence coefficient generated by the Prophet model, The adjustment coefficient and satisfy ; (2) Teaching cycle coefficient calculation Input data: historical access time series , school calendar event label, Prophet model decomposition as follows:
[0013] Wherein, The growth trend item, The periodic term, The school calendar event impact item, The noise term; Periodic term is quantified as follows:
[0014] Wherein, The access volume predicted by the Prophet model at time t, normalized processing ensures ; (3) System association degree calculation Cross-system entity association weight based on knowledge graph, as follows:
[0015] Wherein, Associated business system entity, Association identifier, PageRank value reflects the importance of entity in the cross-system knowledge graph.
[0016] Preferably, the dynamic migration strategy is specifically: (1) Migration trigger rule Migration rule based on time series feature data heat value Hot layer migration: >0.5, data is migrated to high-performance NVMe SSD storage pool in real time; Cold layer degradation: <0.1, data is compressed and archived to low-cost object storage; Warm layer buffer: 0.1≤ ≤0.5, maintain HDD storage layer; (2) Elastic resource scheduling Predictive-driven elasticity: Based on the Prophet time series prediction algorithm, the historical access load is analyzed to extract the teaching cycle characteristics of semesters / weeks / lessons, and the HPA (Horizontal Pod Autoscaler) dynamically adjusts the number of Pod instances to realize dynamic resource matching. Priority and isolation mechanism: The Kubernetes priority scheduling strategy ensures the real-time performance of hot layer migration tasks, and the heterogeneous computing resource isolation avoids transmission congestion. Anti-interference migration control: Design anti-jitter dynamic window mechanism to resist short-term fluctuations, and hot layer to warm layer migration needs to be continuous for 24 hours <0.5; Warm layer to hot layer real-time migration responds to teaching emergencies, and workflow orchestration is based on ApacheDolphinScheduler to build a directed acyclic graph DAG.
[0017] Preferably, the dual-center governance model includes a service management center and a load management center; The service management center realizes the automatic registration, discovery and monitoring of services, and ensures the transparency of service calling through interface definition and method signature; The load management center adopts dynamic weight adjustment combined with routing rule configuration strategy to realize fine traffic control, supports gray release and flow limiting and fusing, and guarantees service stability.
[0018] Compared with the prior art, the present application has the following advantages: 1. The intelligent lake warehouse storage mechanism driven by the teaching cycle is based on the present application, the periodicity of the teaching activity is analyzed and predicted, the heterogeneous storage architecture of hot layer, warm layer and cold layer is adopted, the data is accurately allocated to different storage layers, and the performance and efficiency of the storage system are optimized, and the bidirectional optimization of performance improvement and cost reduction is realized.
[0019] 2. The system of the present application integrates the multi-modal storage capability of the data lake and the ACID transaction characteristics of the data warehouse, not only supports the low-cost storage of massive unstructured teaching videos, but also ensures the accuracy of the teaching evaluation decision based on the ACID transaction, provides a powerful infrastructure for education data analysis, and the service layer guarantees the high reliability and high availability of the service through the dual-center load balancing and fusing and limiting mechanism. BRIEF DESCRIPTION OF DRAWINGS
[0020] Figure 1 The teaching data management system architecture of the present application.
[0021] Figure 2 The flowchart of the data integration layer in the present application.
[0022] Figure 3 The figure of the architecture of the intelligent layered storage unit in the application.
[0023] Figure 4 The flow chart of the development and management integrated module in the application.
[0024] Figure 5 The flow chart of the data service layer in the application. DETAILED DESCRIPTION
[0025] In order to make the technical means, creative features, purposes and effects of the application easy to understand, the application is further described below in conjunction with specific embodiments.
[0026] Referring to Figure 1 The figure of the architecture of the intelligent layered storage unit in the application. Data integration layer: unified description of data specifications through standardized metadata management framework (covering column name, data type, order, length and format); integration of multi-source heterogeneous data using multi-modal fusion architecture to eliminate data islands; design of unified data integration engine to realize dynamic standardized collection, cleaning and conversion of full or incremental data using modular open source tool chain, ensuring efficient access and interconnection of multi-source heterogeneous data; Data platform layer: fusion of multi-modal storage of data lake and ACID transaction characteristics of data warehouse to design an integrated data platform of education lake and warehouse, including: Intelligent layered storage module: for data storage; Development and management integrated module: for data management; Layered teaching data warehouse module: for data application; Data service layer: encapsulation of data processing and analysis functions based on microservice architecture, realization of unified scheduling of education data services using Dubbo RPC+ZooKeeper registration center, standardized encapsulation of core functions as independent microservices through RESTful API, support for upper-layer teaching application calls, and realization of data service management using a dual-center governance model.
[0027] As Figure 2 shown in the figure, in the data integration layer, dynamic Schema evolution and data bloodline tracking technology are integrated in the standardized metadata management framework; the multi-source heterogeneous data includes structured data, semi-structured data and unstructured data; For structured data such as student performance tables, database real-time synchronization is realized through DataX or Sqoop tools, and millisecond-level transmission and transaction consistency are guaranteed through Kafka; For semi-structured data such as XML, text, and logs, use Kafka, LogStash, or canal to dynamically parse time series data, automatically generate Schema through regular expressions or XPath, and convert to structured data; For unstructured data such as audio, pictures, and videos, use AI-driven feature extraction technology to convert unstructured content into vectorized feature labels and establish a unified label library for management.
[0028] The intelligent hierarchical storage module is used to build an educational data intelligent hierarchical storage architecture that integrates the multi-modal storage capabilities of the data lake and the ACID transaction characteristics of the data warehouse, specifically including: (1) Unified data management: based on Delta Lake or Hudi to realize unified storage of structured and unstructured data, and optimize multi-dimensional combined query performance such as "course-semester-class" through Z-Order indexing; (2) Intelligent hierarchical storage unit: dynamically scheduling hot, warm, and cold three-tier storage, significantly reducing the cost of educational data storage; (3) Cross-modal integration: establish a mapping relationship between video metadata and learning behavior logs to solve the problem of heterogeneous data islands.
[0029] As shown in Figure 3 The intelligent hierarchical storage unit specifically includes: Dynamic grading strategy: based on data access heat gradient and business importance weight, accurately allocate data to NVMe SSD, HDD, or object storage, etc. differentiated storage layer; Dual-drive adaptive engine: periodic driving uses LSTM time series prediction to identify calendar event triggered load changes, multi-modal driving associates multi-modal data such as video-logs through metadata tags, and realizes dynamic migration data; Among them, the original pool stores unclassified basic data; the standardized pool is divided into four quadrants by value density (high / low) and data structure (structured / unstructured), and each quadrant implements a three-level storage strategy of hot-warm-cold to achieve dynamic balance of performance and cost; Intelligent hierarchical storage mechanism engine: through multi-modal feature fusion analysis of data attributes, automatically trigger hierarchical storage strategy and dynamic migration strategy based on heat decision function, forming a closed-loop management mechanism.
[0030] By analyzing and predicting the periodicity of teaching activities, the performance and efficiency of the storage system are optimized, and the performance and cost are optimized.
[0031] The heat decision function specifically is: (1) Definition of heat decision function By Fourier series modeling teaching cycle, embedded event factor response test week fluctuations, separation of semester trends and short-term disturbance, accurate analysis of education time series data characteristics, based on the prediction results , fusion real-time access , cross-system correlation characteristics, construct decision function, the formula is as follows:
[0032] Among them, is the heat decision value of the time window, is the number of data access in the time window, is the cross-system correlation weight, is the teaching cycle influence coefficient generated by the Prophet model, is the adjustment coefficient and satisfies ; (2) Teaching cycle coefficient calculation Input data: historical access time series , calendar event label, Prophet model decomposition as follows:
[0033] Among them, is the growth trend item, is the cycle item, is the calendar event impact item, is the noise item; Periodic coefficient quantification as follows:
[0034] Among them, is the access volume at time t predicted by the Prophet model, and normalization processing ensures ; (3) System correlation degree calculation Based on the cross-system entity correlation weight of knowledge graph, as follows:
[0035] Among them, is the associated business system entity, association identifier, PageRank value reflects the importance of entity in the cross-system knowledge graph.
[0036] Hierarchical storage strategy adopts heterogeneous storage media collaborative architecture, through precise matching of data access characteristics and storage medium characteristics, realizes performance-cost pareto optimization, its core strategy design is as shown in the table 1: Table 1 Hierarchical storage strategy design
[0037] The dynamic migration strategy realizes the dynamic balance of three-level storage resources through the cooperation of the teaching cycle perception engine and the multi-modal heat decision model. The core mechanism is as follows: (1) Migration trigger rule Based on the heat value of the time sequence characteristic data Migration rule, hot layer migration: When >0.5, the data is migrated to the high-performance NVMe SSD storage pool in real time; cold layer degradation: When <0.1, the data is compressed and archived to the low-cost object storage; warm layer buffer: 0.1≤ ≤0.5, maintain the HDD storage layer; (2) Elastic resource scheduling Prediction-driven elastic scaling: based on the Prophet time series prediction algorithm to analyze the historical access load, extract the teaching cycle semester / week / time regularity characteristics, and dynamically scale the number of Pod instances through HPA (Horizontal Pod Autoscaler) to realize dynamic resource matching; Priority and isolation mechanism: Kubernetes priority scheduling strategy guarantees the real-time performance of hot layer migration tasks, and heterogeneous computing resource isolation avoids transmission congestion; Anti-interference migration control: design anti-jitter dynamic window mechanism to resist short-term fluctuations, and hot layer→warm layer migration needs to be continuous for 24 hours <0.5; warm layer→hot layer real-time migration responds to teaching emergencies, and workflow orchestration is based on Apache DolphinScheduler to build a directed acyclic graph DAG.
[0038] As shown in Figure 4 , the development governance integrated module is used to integrate the development governance integrated DataOps concept, and a serverless lake warehouse architecture is designed to realize the integration of development and governance. Specifically, it includes: (1) Metadata management: based on Apache Atlas to build metadata blood relationship link; (2) Quality rule library management: AI algorithm driven generated quality rule library (such as null value / unicity / consistency / integrity constraint, etc.), realizing quality control in the development stage; (3) Scheduling management: through DolphinScheduler to coordinate Kubernetes to realize ETL task orchestration and dynamic resource scheduling, and guarantee peak scaling; (4) Compliance and security management: data desensitization, data encryption and audit modules are embedded into ETL data cleaning tasks to realize education sensitive data desensitization encryption (such as student ID number).
[0039] The hierarchical teaching data warehouse module is used to build a three-tier architecture of ODS-DW-ADS. It stores detailed business data through Delta Lake or Hudi tables, manages the basic configuration of the database using MySQL, and supports high-frequency access using a Redis cache server, thereby achieving standardized management of the entire process of teaching data from collection, cleaning to analysis. The ODS is the raw data layer, which stores unmodified raw data (such as the original grade sheet of the academic affairs system and the original log of the LMS), and retains the "data traceability" capability. The DW is the data warehouse layer, which stores the integrated topic wide tables (such as the "student topic wide table" which includes grades, attendance, and video viewing time; and the "course topic wide table" which includes course information, teacher information, and student course selection rate). The ADS is an application service layer that stores lightweight data for applications (such as "current semester grade query results" and "class average grade statistics"), directly serving front-end applications (such as "teaching analysis" on the teacher's side and "learning analysis" on the student's side).
[0040] like Figure 5 As shown, the dual-center governance model includes a service management center and a load management center; The service management center enables automated service registration, discovery, and monitoring, and ensures the transparency of service calls through interface definitions and method signatures; The load management center uses dynamic weight adjustment combined with routing rule configuration strategies to achieve fine-grained traffic control, supports canary releases and rate limiting circuit breaking, and ensures service stability.
[0041] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or equivalent to the scope of this invention are included in this invention.
Claims
1. A teaching data management system characterized by: Comprise: Data integration layer: unified description of data specifications through standardized metadata management framework; Adopt multi-modal fusion architecture to integrate multi-source heterogeneous data; design unified data integration engine to realize dynamic standardization collection, cleaning and conversion of full or incremental data using modular open source tool chain; Data platform layer: integrate multi-modal storage of data lake and ACID transaction characteristics of data warehouse to design education lake-warehouse integrated data platform, including: Intelligent layered storage module: for data storage; Development and governance integrated module: for data management; Layered teaching data warehouse module: for data application; Data service layer: encapsulate data processing and analysis functions based on microservice architecture, use Dubbo RPC+ZooKeeper registration center to realize unified scheduling of education data services, standardize core functions as independent microservices through RESTful API, support upper teaching application calling, and realize data service management using double center governance model.
2. The instructional data management system of claim 1, wherein: In the data integration layer, the standardized metadata management framework integrates dynamic Schema evolution and data bloodline tracking technology; the multi-source heterogeneous data includes structured data, semi-structured data and unstructured data; For structured data, realize database real-time synchronization through DataX or Sqoop tools, and combine Kafka to guarantee millisecond-level transmission and transaction consistency; For semi-structured data, use Kafka, LogStash or canal to dynamically parse time series data, and automatically generate Schema and convert to structured data through regular or XPath; For unstructured data, adopt AI-driven feature extraction technology to convert unstructured content into vectorized feature labels, and establish a unified label library for management.
3. The instructional data management system of claim 1, wherein: The intelligent layered storage module is used to build an education data intelligent layered storage architecture that integrates the multi-modal storage capability of data lake and the ACID transaction characteristics of data warehouse, specifically including: (1) Unified data management: realize unified storage of structured and unstructured data based on Delta Lake or Hudi, and optimize multi-dimensional combined query performance through Z-Order index; (2) Intelligent layered storage unit: dynamically schedule hot, warm and cold three-layer storage; (3) Cross-modal integration: establish mapping association between video metadata and learning behavior logs.
4. The instructional data management system of claim 1, wherein: The development and governance integrated module is used to integrate the development and governance integrated DataOps concept, and design serverless lake-warehouse architecture to realize development and governance integration, specifically including: (1) Metadata management: build metadata bloodline link based on Apache Atlas; (2) Quality rule library management: AI algorithm driven quality rule library generation realizes quality control in development stage; (3) Scheduling management: coordinate Kubernetes through DolphinScheduler to realize ETL task scheduling and resource dynamic scheduling; (4) Compliance and security management: data desensitization, data encryption and audit modules are embedded in ETL data cleaning tasks to realize education sensitive data desensitization and encryption.
5. The instructional data management system of claim 1, wherein: The hierarchical teaching data warehouse module is used to build an ODS-DW-ADS three-layer architecture, Delta Lake or Hudi table is used to store business detail data, MySQL is used to store and manage database basic configuration, and Redis cache server is used to support high-frequency access, so as to realize the standardized management of the whole process of teaching data from collection, cleaning to analysis. The ODS is the original data layer, and stores unmodified original data. The DW is the data warehouse layer, and stores the integrated theme wide table. The ADS is the application service layer, and stores lightweight data for applications.
6. The instructional data management system of claim 3, wherein: The intelligent hierarchical storage unit specifically comprises: Dynamic grading strategy: based on the data access heat gradient and the business importance weight, the data is accurately allocated to the differential storage layer such as NVMeSSD, HDD or object storage; Dual-drive adaptive engine: periodic driving adopts LSTM time series prediction to identify the load change triggered by calendar events, and multi-modal driving realizes the association of multi-modal data such as video-log through metadata tags, and realizes dynamic migration data; Intelligent hierarchical storage mechanism engine: through multi-modal feature fusion analysis of data attributes, the hierarchical storage strategy and dynamic migration strategy are automatically triggered based on the heat decision function, forming a closed-loop management mechanism.
7. A teaching data management system according to claim 6, wherein: The heat decision function is specifically: (1) Heat decision function definition By Fourier series modeling teaching cycle, embedded event factor response test week fluctuation, separation semester trend and short-term disturbance, accurate analysis of education time series data characteristics, based on the prediction results , fusion real-time access , cross-system association characteristics, construct decision function, the formula is as follows: wherein, is a hotness decision value for a time window, is a number of data accesses within a time window, is a cross-system association weight, is a teaching period influence coefficient generated by a Prophet model, is an adjustment coefficient and satisfies ; (2) Teaching cycle coefficient calculation Input data: historical visit time series , calendar event labels, the Prophet model decomposes as follows: wherein, is a growth trend term, is a periodic term, is a school calendar event impact term, is a noise term; The cycle quantity is quantified as follows: wherein, is the access volume at time t predicted by the Prophet model, normalized to ensure ; (3) System correlation calculation The cross-system entity correlation weight based on the knowledge graph is as follows: wherein, is an associated business system entity, an associated identifier, the PageRank value reflecting the importance of the entity in the cross-system knowledge graph.
8. The instructional data management system of claim 7, wherein: The dynamic migration strategy is specifically: (1) Migration trigger rule Migration rule based on time series feature data hotness value Hot tier migration: > 0.5, data real-time migration to high-performance NVMe SSD storage pool; cold tier degradation: < 0.1, data compression and archiving to low-cost object storage; warm tier buffer: 0.1 ≤ ≤ 0.5, maintain HDD storage tier; (2) Elastic resource scheduling Predictive-driven elastic scaling: based on the Prophet time series prediction algorithm to analyze historical access load, extract teaching cycle semester / week / time characteristics, dynamically scale the number of Pod instances through HPA (Horizontal Pod Autoscaler), and realize dynamic resource matching; Priority and isolation mechanism: Kubernetes priority scheduling strategy guarantees the real-time performance of hot layer migration tasks, and heterogeneous computing resource isolation avoids transmission congestion; Anti-interference migration control: design anti-shake dynamic window mechanism to resist short-term fluctuations, hot layer → warm layer migration needs 24 hours continuously <0.5; warm layer → hot layer real-time migration response to teaching emergencies, workflow orchestration is based on Apache DolphinScheduler to build a directed acyclic graph DAG.
9. The instructional data management system of claim 1, wherein: The dual-center governance model comprises a service management center and a load management center. The service management center realizes the automatic registration, discovery and monitoring of services, and ensures the transparency of service calling through interface definition and method signature; The load management center adopts dynamic weight adjustment combined with routing rule configuration strategy to realize fine control of traffic, supports gray release and flow limiting and fusing, and guarantees service stability.