Multi-dimensional data visualization management system based on intelligent calculation platform

Abstract

The invention discloses a multi-dimensional data visualization management system based on an intelligent computing platform. The multi-dimensional data visualization management system comprises a cross-platform data interaction layer, a multi-scene visualization engine layer, a dynamic adaptation layer, a user interaction layer and a deployment operation and maintenance layer. According to the invention, by constructing a multi-scene visual template library and supporting dynamic adaptation of business scenes and development of user-defined components, visual full coverage of multiple scenes such as intelligent computing platform AI training and commercial computing power leasing is realized; by means of multi-protocol adaptation, heterogeneous hardware recognition and containerization deployment technologies, binding limitation of an x86 architecture, a Linux system and a Slurm scheduling system is broken through, and flexible adaptation and deployment across hardware, systems and cloud environments are supported; and meanwhile, a scene-based interface customization function, a template reuse function and an exclusive component development function are provided, so that personalized monitoring requirements of different intelligent calculation services are effectively met, and technical pain points of an existing system are comprehensively solved.

Description

Technical Field

[0001] This invention relates to the field of multidimensional data management technology, and in particular to a multidimensional data visualization management system based on an intelligent computing platform. Background Technology

[0002] As an infrastructure integrating artificial intelligence and high-performance computing, intelligent computing platforms have been widely used in various scenarios such as model training, scientific computing, and commercial computing power leasing. Their core requirement is to achieve visualized monitoring of heterogeneous resources, AI job scheduling, and multi-tenant resource management. However, existing intelligent computing visualization integration systems have significant shortcomings: 1. Limited scenario coverage: designed only for homogeneous resource management and scientific computing job monitoring in traditional supercomputing, unable to adapt to emerging scenarios such as AI training, heterogeneous resource scheduling, and hybrid cloud computing power management in intelligent computing platforms; 2. Poor cross-platform compatibility: often bound to specific hardware architectures (x86), operating systems (Linux), and scheduling systems (Slurm), making it difficult to connect with ARM architecture intelligent computing hardware, Windows Server intelligent computing nodes, and K8s containerized scheduling systems; 3. Insufficient visualization customization capabilities: lacking dedicated visualization templates and components for intelligent computing scenarios, unable to meet the personalized monitoring needs of different intelligent computing businesses. In light of the above, this application proposes a multi-dimensional data visualization management system based on an intelligent computing platform. Summary of the Invention

[0003] Based on the technical problems existing in the background technology, the present invention proposes a multi-dimensional data visualization management system based on an intelligent computing platform.

[0004] This invention proposes a multi-dimensional data visualization management system based on an intelligent computing platform, comprising a cross-platform data interaction layer, a multi-scenario visualization engine layer, a dynamic adaptation layer, a user interaction layer, and a deployment and maintenance layer. The cross-platform data interaction layer and the dynamic adaptation layer are connected through a standardized data interface, the multi-scenario visualization engine layer and the user interaction layer are connected through a Web front-end rendering interface, the user interaction layer and the deployment and maintenance layer are connected through an operation and maintenance control interface, and the dynamic adaptation layer and the multi-scenario visualization engine layer are connected through a Datart platform data access interface.

[0005] The cross-platform data interaction layer includes a protocol adaptation unit, a multi-source data access unit, a data format conversion unit, and a data cache synchronization unit.

[0006] The multi-scenario visualization engine layer integrates the open-source Datart data dashboard platform, constructs a scenario-based template library including AI training job monitoring, heterogeneous resource scheduling, multi-tenant resource metering, and hybrid cloud intelligent computing overview, and configures a custom visualization component development unit.

[0007] The dynamic adaptation layer includes a hardware architecture adaptation unit, a scheduling system adaptation unit, a business scenario adaptation unit, and an adaptation rule update unit.

[0008] The user interaction layer includes a cross-platform operation jump unit, a scenario-based customized interface, and a multi-terminal synchronization unit.

[0009] The deployment and operation layer uses Docker containerization to encapsulate the core modules of the system, and is configured with a container orchestration unit, a dynamic scaling unit, a cross-platform monitoring unit, and a one-click deployment unit.

[0010] Preferably, the protocol adaptation unit supports dynamic switching between multiple protocols such as SSH, RESTfulAPI, gRPC, and JDBC, and adapts to the communication specifications of different intelligent computing scheduling systems such as Slurm, K8s, and YARN; the multi-source data access unit can simultaneously connect to x86 / ARM heterogeneous hardware architectures, Linux / Windows Server operating systems, and intelligent computing nodes deployed in a hybrid public / private cloud environment. The multi-source data access unit is configured with a hardware architecture identification subunit, which automatically matches the corresponding communication protocol and data acquisition method by reading the CPU architecture identifier of the intelligent computing node. x86 nodes use the SSH protocol for data acquisition, and ARM nodes use the gRPC protocol for data acquisition.

[0011] The data format conversion unit, based on a preset intelligent computing data model, normalizes heterogeneous data such as JSON / XML / CSV into a standardized format; the data cache synchronization unit uses Redis distributed caching to save the collected data within 72 hours when the intelligent computing node is offline, and automatically retransmits it to the dynamic adaptation layer after the network is restored, ensuring cross-platform data integrity.

[0012] Preferably, the AI ​​training job monitoring template has a built-in anomaly detection submodule. By comparing the real-time loss function value of the AI ​​job with the historical average, a red warning sign is triggered when the deviation exceeds 20%, and an anomaly cause analysis pop-up window is displayed in the user interaction layer.

[0013] The component development unit supports users in developing dedicated components for GPU force heatmaps and AI job iteration progress curves based on WebGL / Canvas technology. After the component development is completed, it can be connected to the scenario-based template library through the Datart plugin interface.

[0014] The multi-scenario visualization engine layer can use the Datart time-series data rendering interface to draw real-time curves of AI job loss function values ​​and GPU memory usage indicators. When the indicators exceed preset thresholds, a visual warning sign is automatically triggered. The specific operation logic steps are as follows:

[0015] S101: The multi-scenario visualization engine layer receives AI job time-series data pushed by the dynamic adaptation layer through the Datart standardized data access interface. The data must conform to the preset intelligent computing data model specifications. The multi-scenario visualization engine layer performs legality verification on the received data and removes invalid data that is abnormal or missing.

[0016] S102: Write the validated data into the time-series database and build a three-level index structure to improve query efficiency. The first-level index is JobID + GPUID, used to distinguish data from different GPU devices in different jobs. The second-level index is t. collect Sort by collection time in ascending order, supports fast query within time range, and has a three-level index for indicator types to distinguish between "loss function" and "GPU memory" indicators, which facilitates classified query. After storage is completed, a data storage log is generated to record the data write time, data volume, and index building status for subsequent data traceability and troubleshooting.

[0017] S103: The multi-scenario visualization engine layer calls the Datart time-series data rendering interface to initialize curve drawing parameters according to the AI ​​job type;

[0018] S104: The specific steps for filtering the data to be drawn are as follows: Based on the sliding time window algorithm, the following steps are taken:

[0019] S1041: Determine window boundaries: Assume the sliding window duration is T = 300 seconds, and the current system timestamp is t. current Then the window start timestamp t start =t current -T×1000, window end timestamp t end =t current ;

[0020] S1042: Data Query: Filter by "JobID + GPUID" using the time-series database index, querying time ranges within [t]. start ,t end All time-series data within ];

[0021] S1043: Data Sorting: Sort the query results by t collect Sort in ascending order to ensure the temporal continuity of curve plotting;

[0022] S105: Based on the relative threshold and basic parameters in the received data, the "relative threshold" is converted into a directly comparable "absolute threshold" using a mathematical formula. The specific calculation is as follows: (1) Based on the historical mean of the loss function Compared with the relative warning threshold Thr loss,rel According to the formula Calculate the absolute threshold; (2) Based on the total GPU memory M gpu,totalCompared with the relative warning threshold Thr gpu,rel According to the formula Calculate the absolute threshold;

[0023] After the calculation is completed, the absolute threshold is bound to the corresponding job and GPU device and stored for subsequent real-time metric comparison.

[0024] S106: The multi-scene visualization engine layer passes the time-series data filtered in S104 to the Datart rendering center, and dynamically draws curves according to the following logic:

[0025] S1061: Coordinate Mapping: Mapping the timestamp t collect Convert L to x-axis coordinates current M gpu,current Mapped to the left and right y-axis coordinates respectively;

[0026] S1062: Vertex generation: Connect the data points in time sequence to generate curve vertices, and use the Bézier curve smoothing algorithm to optimize the curve transition effect and avoid the stuttering of the broken line.

[0027] S1063: Real-time update: Repeat the S104-S106 process every 1 second to refresh the sliding window data and update the curve to ensure the real-time display of the curve and the data collection.

[0028] S107: Compare the latest collected time-series data within the current window with the absolute threshold calculated in S105 to determine whether to trigger an alert: If L is satisfied... current >Thr loss,abs or L current If the value is less than 0, then the loss function alert status is set to Alert. loss =1, trigger an alert; otherwise, an alert will be issued. loss =0, no warning;

[0029] If M is satisfied gpu,current >Thr gpu,abs The GPU memory warning status will then be displayed as an Alert. gpu =1 triggers an alert; otherwise, an alert is issued. gpu =0, no warning;

[0030] At the same time, calculate the GPU memory utilization.

[0031] S108: If Alert loss =1, or Alert gpu=1. The multi-scenario visualization engine layer calls the Datart warning rendering sub-interface to generate corresponding visual warning labels. After the warning labels are generated, they are synchronized to the Web visualization portal of the user interaction layer through the Vue.js component communication interface to ensure that the data dashboard displays warning information in real time.

[0032] Preferably, the hardware architecture adaptation unit automatically adjusts the data collection granularity to the second level based on the resource characteristics of the CPU / GPU / AI chip, and supplements the corresponding hardware-specific display dimensions; the scheduling system adaptation unit has a built-in scheduling system feature library containing Slurm / K8s / YARN, and automatically matches the job status definition and resource indicator caliber in the feature library when accessing a new scheduling system, without the need for manual code modification; the adaptation rule update unit supports receiving user-defined adaptation rules through the API interface, and after updating, it is synchronized to the hardware architecture adaptation unit and the scheduling system adaptation unit in real time. The adaptation rule update unit supports users to import new adaptation rules through Excel, and automatically verifies the legality of the rules after import. After the verification is passed, the rule synchronization is completed within 10 seconds without restarting the system.

[0033] Preferably, the container orchestration unit is based on Kubernetes to achieve cross-environment deployment of physical machines / virtual machines / public clouds / private clouds; the dynamic scaling unit automatically adjusts the number of container instances for the visualization service according to the concurrent access volume; the cross-platform monitoring unit collects container CPU / memory usage and interface response time operation and maintenance data, pushes it to the multi-scenario visualization engine layer through the data interface, and generates system operation and maintenance visualization charts. The cross-platform monitoring unit is configured with an operation and maintenance alarm sub-module. When the container CPU usage exceeds 80% and the interface response time exceeds 500ms, it automatically triggers multi-channel alarms. The alarm content includes the abnormal module, the current value of the abnormal indicator, and the suggested handling solution.

[0034] The one-click deployment unit provides automated deployment scripts that include environment checks, dependency installation, and configuration initialization.

[0035] Preferably, the cross-platform operation jump unit enables one-click jump from the Datart dashboard to the Slurm job management interface and the K8s container console via a URL mapping interface; the scenario-based customization interface supports dragging and dropping visualization components, adjusting data display dimensions, and saving customized templates to the cloud. The scenario-based customization interface has a built-in template reuse submodule, which allows users to export customized templates as JSON format files. Other users can import the files and reuse the templates quickly after permission verification.

[0036] The multi-terminal synchronization unit is linked through user accounts to achieve real-time synchronization of customized templates and operation records between different terminals.

[0037] Preferably, the business scenario adaptation unit dynamically selects core display indicators based on the differences between scientific research computing and commercial computing power leasing scenarios. The specific logical steps are as follows:

[0038] S201: The business scenario adaptation unit obtains the current scenario type S through two methods: proactive configuration recognition and behavioral feature-assisted judgment. Proactive configuration recognition reads the scenario types preset by the intelligent computing platform administrator in the "scenario customization interface". When using behavioral feature-assisted judgment, if the scenario type is not proactively configured, the judgment is supplemented by analyzing the tenant's operational behavior characteristics over the past 30 days. If the tenant's operation includes "job script upload frequency ≥ 5 times / day" and "no billing query operation", it is judged as S=1, which is scientific research computing. If the tenant's operation includes "billing query frequency ≥ 3 times / day" and "core time recharge record ≥ 1", it is judged as S=2, which is commercial computing power leasing. After the judgment is completed, the current scenario type S is output as the basis for subsequent indicator screening.

[0039] S202: Based on the scenario type S determined in step S1, retrieve the corresponding scenario's indicator weight configuration W from the "Scene Feature Library". S,i If S = 1, call the weight set {W} 1,1 W 1,2 W 1,3 W 1,4 W 1,5 If S = 2, call the weight set {W}. 2,1 W 2,2 W 2,3 W 2,4 W 2,5 Meanwhile, tenants can customize and adjust weights through the user interaction layer. Custom weights must meet the condition that "the sum of all indicator weights is ≤ 5". After adjustment, the weight set W for the current scenario will be automatically updated. S,i ;

[0040] S203: Use the "weighted threshold screening method" to select the core display indicators under scenario S: I s,core ={I i |W S,i ≥T W ,I i ∈I}, where I represents the set of all indices traversed, and the weights W under scenario S are... S,i Greater than or equal to the filtering threshold T W Indicators, incorporated into the core display indicator set I S,core ;

[0041] S204: For the selected core indicators I S,core Sort by weight in descending order to determine display priority. The formula used is: Rank(I i =DescendingOrder(WS,i |I i ∈I S,core ), according to its weight W in scene S S,i Sort from largest to smallest;

[0042] S205: The business scenario adaptation unit will integrate "core indicator set I" S,core The "Display priority sorting results" option is pushed to the multi-scenario visualization engine layer through the Datart template configuration interface to achieve control over the visibility of indicators, adjustment of layout priority, and real-time synchronization updates.

[0043] S206: If the intelligent computing platform supports tenants switching business scenarios during use, the following steps must be performed:

[0044] S2061: Triggering Scene Switching and Resetting Weights: The tenant clicks the "Scene Switching" button in the user interaction layer and selects the target scene S. new The system automatically resets the indicator weight set to The tenant will be prompted that "The scene switch will update the visualization dashboard metrics. Do you confirm?". After confirmation, proceed to the next step.

[0045] S2062: Execute historical data adaptation and display: Call the historical data interface of the cross-platform data interaction layer to obtain the tenant's full historical indicator data for the past 7 days, and display it according to the new scenario S. new core indicator set Filter the corresponding indicator data from the historical data, redraw the visualization charts, and refresh the visualization dashboard after completing the historical data adaptation to ensure that the core indicators in the new scenario are displayed in a consistent manner with the historical data.

[0046] Preferably, the business scenario adaptation unit is further configured with a scenario dynamic switching submodule and a historical data adaptation submodule. The scenario dynamic switching submodule allows tenants to select the target scenario for scientific research computing or commercial computing power leasing in real time through the "Scenario Switching" function button in the user interaction layer. During switching, a secondary verification of the scenario type is automatically triggered. By comparing the matching degree between the tenant's current operation behavior and the characteristics of the target scenario, if the matching degree is ≥80%, the switching is completed directly. If the matching degree is <80%, a scenario switching risk warning pops up. After the tenant confirms, the indicator weight set of the current scenario is reset.

[0047] After a scenario switch, the historical data adaptation submodule obtains the tenant's full set of historical metrics data for the past 7 days through the historical data interface of the cross-platform data interaction layer, and then adapts it to the core metric set of the target scenario. According to the formula Filter the historical data required for the target scenario, and then redraw the historical trend charts using the Datart data re-rendering interface to ensure that the historical and real-time data of core indicators are displayed consistently after the scenario switch. To adapt to historical data, D total,i This is the complete historical data.

[0048] Compared with existing technologies, the beneficial effects of this invention are:

[0049] 1. By building a scenario-based template library that includes AI training job monitoring, heterogeneous resource scheduling, multi-tenant resource metering, and hybrid cloud intelligent computing overview, it covers the core business scenarios of the intelligent computing platform from AI model training to commercial computing power leasing. This solves the limitation of traditional systems that can only support traditional supercomputing scientific computing. Moreover, the business scenario adaptation unit can dynamically filter core display indicators and adjust display priorities according to the differences between scientific research computing and commercial computing power leasing scenarios. It also supports real-time scenario switching and historical data adaptation to meet the personalized monitoring needs of different tenants.

[0050] 2. Through the cross-platform data interaction layer, it supports dynamic switching of multiple protocols and can connect to x86 / ARM heterogeneous hardware, multiple operating systems and hybrid cloud intelligent computing nodes. The scheduling system adaptation unit has built-in feature libraries of mainstream intelligent computing scheduling systems. The deployment and operation layer adopts containerization and K8s orchestration technology to achieve one-click deployment across environments, which completely breaks the traditional system's binding restrictions on specific hardware, operating systems and scheduling systems, and greatly improves the system's deployment flexibility and portability.

[0051] 3. For AI training scenarios, the system can realize real-time curve plotting and accurate early warning of indicators such as loss function value and GPU memory usage. The AI ​​training monitoring template also has a built-in anomaly detection sub-module and a pop-up window to prompt the cause of the anomaly. The dynamic adaptation layer automatically optimizes the data collection and display dimensions, effectively solving the problem of untimely anomaly warning in traditional systems, greatly shortening the troubleshooting time for AI jobs, and improving the decision-making efficiency of intelligent computing resource management.

[0052] This invention achieves full visualization coverage of diverse scenarios such as AI training and commercial computing power leasing on intelligent computing platforms by constructing a multi-scenario visualization template library and supporting dynamic adaptation to business scenarios and custom component development. Leveraging multi-protocol adaptation, heterogeneous hardware identification, and containerized deployment technologies, it breaks the binding limitations of x86 architecture, Linux systems, and the Slurm scheduling system, supporting flexible adaptation and deployment across hardware, systems, and cloud environments. Simultaneously, it provides scenario-based customized interfaces, template reuse, and dedicated component development functions, effectively meeting the personalized monitoring needs of different intelligent computing businesses and comprehensively solving the technical pain points of existing systems. Attached Figure Description

[0053] Figure 1This is a block diagram of a multi-dimensional data visualization management system based on an intelligent computing platform proposed in this invention. Detailed Implementation

[0054] The present invention will be further explained below with reference to specific embodiments.

[0055] Example

[0056] Reference Figure 1 This embodiment proposes a multi-dimensional data visualization management system based on an intelligent computing platform, including a cross-platform data interaction layer, a multi-scenario visualization engine layer, a dynamic adaptation layer, a user interaction layer, and a deployment and maintenance layer. The cross-platform data interaction layer and the dynamic adaptation layer are connected through a standardized data interface, the multi-scenario visualization engine layer and the user interaction layer are connected through a Web front-end rendering interface, the user interaction layer and the deployment and maintenance layer are connected through an operation and maintenance control interface, and the dynamic adaptation layer and the multi-scenario visualization engine layer are connected through a Datart platform data access interface.

[0057] The cross-platform data interaction layer includes a protocol adaptation unit, a multi-source data access unit, a data format conversion unit, and a data cache synchronization unit.

[0058] The protocol adaptation unit supports dynamic switching between multiple protocols such as SSH, RESTfulAPI, gRPC, and JDBC, and adapts to the communication specifications of different intelligent computing scheduling systems such as Slurm, K8s, and YARN. The multi-source data access unit can simultaneously connect to heterogeneous hardware architectures of x86 / ARM, multiple operating systems such as Linux / Windows Server, and intelligent computing nodes deployed in a hybrid public / private cloud environment. The multi-source data access unit is equipped with a hardware architecture identification subunit, which automatically matches the corresponding communication protocol and data acquisition method by reading the CPU architecture identifier of the intelligent computing node. x86 nodes use the SSH protocol for data acquisition, and ARM nodes use the gRPC protocol for data acquisition.

[0059] The data format conversion unit, based on a preset intelligent computing data model, normalizes heterogeneous data such as JSON / XML / CSV into a standardized format; the data cache synchronization unit uses Redis distributed cache to save the collected data within 72 hours when the intelligent computing node is offline, and automatically re-transmits it to the dynamic adaptation layer after the network is restored, ensuring cross-platform data integrity.

[0060] The multi-scenario visualization engine layer integrates the open-source Datart data dashboard platform, builds a scenario-based template library that includes AI training job monitoring, heterogeneous resource scheduling, multi-tenant resource metering, and hybrid cloud intelligent computing overview, and configures custom visualization component development units.

[0061] The AI ​​training assignment monitoring template has a built-in anomaly detection submodule. By comparing the real-time loss function value of the AI ​​assignment with the historical average, a red warning sign is triggered when the deviation exceeds 20%, and an anomaly cause analysis pop-up window is displayed in the user interaction layer.

[0062] The component development unit supports users in developing dedicated components for GPU force heatmaps and AI job iteration progress curves based on WebGL / Canvas technology. After the component development is completed, it can be connected to the scenario-based template library through the Datart plugin interface.

[0063] The multi-scenario visualization engine layer can use the Datart time-series data rendering interface to draw real-time curves of AI job loss function values ​​and GPU memory usage indicators. When the indicators exceed preset thresholds, a visual warning sign is automatically triggered. The specific operation logic steps are as follows:

[0064] S101: The multi-scenario visualization engine layer receives AI job time-series data pushed by the dynamic adaptation layer through the Datart standardized data access interface. The data must conform to the preset intelligent computing data model specifications. The multi-scenario visualization engine layer performs legality verification on the received data and removes invalid data that is abnormal or missing.

[0065] S102: Write the validated data into the time-series database and build a three-level index structure to improve query efficiency. The first-level index is JobID + GPUID, used to distinguish data from different GPU devices in different jobs. The second-level index is t. collect Sort by collection time in ascending order, supports fast query within time range, and has a three-level index for indicator types to distinguish between "loss function" and "GPU memory" indicators, which facilitates classified query. After storage is completed, a data storage log is generated to record the data write time, data volume, and index building status for subsequent data traceability and troubleshooting.

[0066] S103: The multi-scenario visualization engine layer calls the Datart time-series data rendering interface to initialize curve drawing parameters according to the AI ​​job type;

[0067] S104: The specific steps for filtering the data to be drawn are as follows: Based on the sliding time window algorithm, the following steps are taken:

[0068] S1041: Determine window boundaries: Assume the sliding window duration is T = 300 seconds, and the current system timestamp is t. current Then the window start timestamp t start =t current -T×1000, window end timestamp t end =t current ;

[0069] S1042: Data Query: Filter by "JobID + GPUID" using the time-series database index, querying time ranges within [t]. start ,t end All time-series data within ];

[0070] S1043: Data Sorting: Sort the query results by t collect Sort in ascending order to ensure the temporal continuity of curve plotting;

[0071] S105: Based on the relative threshold and basic parameters in the received data, the "relative threshold" is converted into a directly comparable "absolute threshold" using a mathematical formula. The specific calculation is as follows: (1) Based on the historical mean of the loss function Compared with the relative warning threshold Thr loss,rel According to the formula Calculate the absolute threshold; (2) Based on the total GPU memory M gpu,total Compared with the relative warning threshold Thr gpu,rel According to the formula Calculate the absolute threshold;

[0072] After the calculation is completed, the absolute threshold is bound to the corresponding job and GPU device and stored for subsequent real-time metric comparison.

[0073] S106: The multi-scene visualization engine layer passes the time-series data filtered in S104 to the Datart rendering center, and dynamically draws curves according to the following logic:

[0074] S1061: Coordinate Mapping: Mapping the timestamp t collect Convert L to x-axis coordinates current M gpu,current Mapped to the left and right y-axis coordinates respectively;

[0075] S1062: Vertex generation: Connect the data points in time sequence to generate curve vertices, and use the Bézier curve smoothing algorithm to optimize the curve transition effect and avoid the stuttering of the broken line.

[0076] S1063: Real-time update: Repeat the S104-S106 process every 1 second to refresh the sliding window data and update the curve to ensure the real-time display of the curve and the data collection.

[0077] S107: Compare the latest collected time-series data within the current window with the absolute threshold calculated in S105 to determine whether to trigger an alert: If L is satisfied... current >Thr loss,abs or L current If the value is less than 0, then the loss function alert status is set to Alert. loss =1, trigger an alert; otherwise, an alert will be issued. loss=0, no warning;

[0078] If M is satisfied gpu,current >Thr gpu,abs The GPU memory warning status will then be displayed as an Alert. gpu =1 triggers an alert; otherwise, an alert is issued. gpu =0, no warning;

[0079] At the same time, calculate the GPU memory utilization.

[0080] S108: If Alert loss =1, or Alert gpu =1. The multi-scenario visualization engine layer calls the Datart warning rendering sub-interface to generate corresponding visual warning labels. After the warning labels are generated, they are synchronized to the Web visualization portal of the user interaction layer through the Vue.js component communication interface to ensure that the data dashboard displays warning information in real time.

[0081] The dynamic adaptation layer includes a hardware architecture adaptation unit, a scheduling system adaptation unit, a business scenario adaptation unit, and an adaptation rule update unit.

[0082] The hardware architecture adaptation unit automatically adjusts the data collection granularity to the second level based on the resource characteristics of CPU / GPU / AI chips, and supplements the corresponding hardware-specific display dimensions; the scheduling system adaptation unit has a built-in scheduling system feature library containing Slurm / K8s / YARN, and automatically matches the job status definition and resource indicator caliber in the feature library when accessing a new scheduling system, without the need for manual code modification; the adaptation rule update unit supports receiving user-defined adaptation rules through the API interface, and after updating, it is synchronized to the hardware architecture adaptation unit and the scheduling system adaptation unit in real time. The adaptation rule update unit also supports users to import new adaptation rules through Excel, and automatically verifies the legality of the rules after import. After the verification is passed, the rule synchronization is completed within 10 seconds without restarting the system;

[0083] The business scenario adaptation unit dynamically selects core display indicators based on the differences between scientific research computing and commercial computing power leasing scenarios. The specific logical steps are as follows:

[0084] S201: The business scenario adaptation unit obtains the current scenario type S through two methods: proactive configuration recognition and behavioral feature-assisted judgment. Proactive configuration recognition reads the scenario types preset by the intelligent computing platform administrator in the "scenario customization interface". When using behavioral feature-assisted judgment, if the scenario type is not proactively configured, the judgment is supplemented by analyzing the tenant's operational behavior characteristics over the past 30 days. If the tenant's operation includes "job script upload frequency ≥ 5 times / day" and "no billing query operation", it is judged as S=1, which is scientific research computing. If the tenant's operation includes "billing query frequency ≥ 3 times / day" and "core time recharge record ≥ 1", it is judged as S=2, which is commercial computing power leasing. After the judgment is completed, the current scenario type S is output as the basis for subsequent indicator screening.

[0085] S202: Based on the scenario type S determined in step S1, retrieve the corresponding scenario's indicator weight configuration W from the "Scene Feature Library". S,i If S = 1, call the weight set {W} 1,1 W 1,2 W 1,3 W 1,4 W 1,5 If S = 2, call the weight set {W}. 2,1 W 2,2 W 2,3 W 2,4 W 2,5 Meanwhile, tenants can customize and adjust weights through the user interaction layer. Custom weights must meet the condition that "the sum of all indicator weights is ≤ 5". After adjustment, the weight set W for the current scenario will be automatically updated. S,i ;

[0086] S203: Use the "weighted threshold screening method" to select the core display indicators under scenario S: I s,core ={I i |W S,i ≥T W ,I i ∈I}, where I represents the set of all indices traversed, and the weights W under scenario S are... S,i Greater than or equal to the filtering threshold T W Indicators, incorporated into the core display indicator set I S,core ;

[0087] S204: For the selected core indicators I S,core Sort by weight in descending order to determine display priority. The formula used is: Rank(I i =DescendingOrder(W S,i |I i ∈I S,core ), according to its weight W in scene S S,i Sort from largest to smallest;

[0088] S205: The business scenario adaptation unit will integrate "core indicator set I" S,core The "Display priority sorting results" option is pushed to the multi-scenario visualization engine layer through the Datart template configuration interface to achieve control over the visibility of indicators, adjustment of layout priority, and real-time synchronization updates.

[0089] S206: If the intelligent computing platform supports tenants switching business scenarios during use, the following steps must be performed:

[0090] S2061: Triggering Scene Switching and Resetting Weights: The tenant clicks the "Scene Switching" button in the user interaction layer and selects the target scene S. new The system automatically resets the indicator weight set to The tenant will be prompted that "The scene switch will update the visualization dashboard metrics. Do you confirm?". After confirmation, proceed to the next step.

[0091] S2062: Execute historical data adaptation and display: Call the historical data interface of the cross-platform data interaction layer to obtain the tenant's full historical indicator data for the past 7 days, and display it according to the new scenario S. new core indicator set Filter the corresponding indicator data from the historical data, redraw the visualization charts, and refresh the visualization dashboard after completing the historical data adaptation to ensure that the core indicators in the new scenario are displayed in a consistent manner with the historical data.

[0092] The business scenario adaptation unit is also equipped with a scenario dynamic switching submodule and a historical data adaptation submodule. The scenario dynamic switching submodule allows tenants to select the target scenario for scientific research computing or commercial computing power leasing in real time through the "Scenario Switching" function button in the user interaction layer. When switching, a secondary verification of the scenario type is automatically triggered. By comparing the matching degree between the tenant's current operation behavior and the characteristics of the target scenario, if the matching degree is ≥80%, the switch is completed directly. If the matching degree is <80%, a scenario switching risk warning will pop up. After the tenant confirms, the indicator weight set of the current scenario is reset.

[0093] After a scenario switch, the historical data adaptation submodule obtains the tenant's full historical metrics data for the past 7 days through the historical data interface of the cross-platform data interaction layer, and then adapts it to the core metric set of the target scenario. According to the formula Filter the historical data required for the target scenario, and then redraw the historical trend charts using the Datart data re-rendering interface to ensure that the historical and real-time data of core indicators are displayed consistently after the scenario switch. To adapt to historical data, D total,i This is the complete historical data;

[0094] The user interaction layer includes cross-platform operation jump units, scenario-customized interfaces, and multi-terminal synchronization units.

[0095] The cross-platform operation jump unit enables one-click jump from the Datart dashboard to the Slurm job management interface and the K8s container console via the URL mapping interface; the scenario-based customization interface supports dragging and dropping visualization components, adjusting data display dimensions, and saving customized templates to the cloud. The scenario-based customization interface has a built-in template reuse submodule, which allows users to export customized templates as JSON format files. Other users can import the files and reuse the templates quickly after permission verification.

[0096] The multi-terminal synchronization unit is linked through user accounts to achieve real-time synchronization of customized templates and operation records between different terminals;

[0097] The deployment and operation layer uses Docker containerization to encapsulate the core modules of the system, and is configured with container orchestration unit, dynamic scaling unit, cross-platform monitoring unit and one-click deployment unit;

[0098] The container orchestration unit, based on Kubernetes, enables cross-environment deployment across physical machines, virtual machines, public clouds, and private clouds. The dynamic scaling unit automatically adjusts the number of container instances for the visualization service based on concurrent access volume. The cross-platform monitoring unit collects container CPU / memory usage and interface response time operation and maintenance data, pushes it to the multi-scenario visualization engine layer through the data interface, and generates system operation and maintenance visualization charts. The cross-platform monitoring unit is configured with an operation and maintenance alarm sub-module, which automatically triggers multi-channel alarms when container CPU usage exceeds 80% or interface response time exceeds 500ms. The alarm content includes the abnormal module, the current value of the abnormal indicator, and the suggested handling solution.

[0099] The one-click deployment unit provides automated deployment scripts that include environment checks, dependency installation, and configuration initialization. This embodiment achieves full visualization coverage of diverse scenarios such as AI training and commercial computing power leasing by building a multi-scenario visualization template library and supporting dynamic adaptation to business scenarios and custom component development. With the help of multi-protocol adaptation, heterogeneous hardware identification, and containerized deployment technologies, it breaks the binding restrictions of x86 architecture, Linux system, and Slurm scheduling system, and supports flexible adaptation and deployment across hardware, system, and cloud environments. At the same time, it provides scenario-based customized interfaces, template reuse, and dedicated component development functions, effectively meeting the personalized monitoring needs of different intelligent computing services and comprehensively solving the technical pain points of existing systems.

[0100] In this embodiment, the cross-platform data interaction layer serves as the system's data input hub. Through a protocol adaptation unit, it supports dynamic switching between multiple protocols such as SSH, RESTful API, gRPC, and JDBC, adapting to the communication specifications of different intelligent computing scheduling systems such as Slurm, K8s, and YARN. The multi-source data access unit, leveraging the hardware architecture identification subunit, automatically distinguishes between x86 / ARM heterogeneous hardware architectures and Linux / Windows Server operating systems, and specifically matches communication protocols to complete data acquisition. The data format conversion unit normalizes heterogeneous data (JSON / XML / CSV) into a preset intelligent computing data model format, eliminating data format differences. The data caching and synchronization unit uses a Redis distributed cache to store 72 hours of collected data when the intelligent computing node is offline, automatically retransmitting it after network recovery to ensure the integrity and continuity of cross-platform data. Finally, this layer transmits standardized, complete intelligent computing data to the dynamic adaptation layer through a standardized data interface.

[0101] After receiving standardized data from the cross-platform data interaction layer, the dynamic adaptation layer achieves precise data adaptation through various units. The hardware architecture adaptation unit automatically adjusts the data collection granularity and supplements exclusive display dimensions based on the resource characteristics of CPU / GPU / AI chips. The scheduling system adaptation unit relies on the built-in Slurm / K8s / YARN scheduling system feature library to automatically match job status definitions and resource indicator standards without manual code modification. The business scenario adaptation unit determines the current business scenario through proactive configuration recognition and behavioral feature-assisted judgment, and then dynamically filters and sorts core display indicators using the "weight threshold filtering method," while also supporting dynamic scenario switching and historical data adaptation. The adaptation rule update unit allows users to import custom rules via API or Excel, completes rule synchronization and application to each adaptation unit within 10 seconds, and pushes the adapted structured data to the multi-scenario visualization engine layer through the Datart platform data access interface.

[0102] The multi-scenario visualization engine layer is based on the open-source Datart data dashboard platform. It first calls the scenario-based template library and combines it with adapted data for visualization rendering: For AI training job monitoring scenarios, the Datart time-series data rendering interface is used to draw real-time curves of indicators such as loss function values ​​and GPU memory usage. When indicators exceed limits or are abnormal, a visual warning is triggered and a pop-up notification is displayed. The component development unit supports users in developing custom components such as GPU computing power heatmaps and AI job iteration progress curves based on WebGL / Canvas technology. These components are accessed through the Datart plugin interface to extend visualization capabilities. Finally, the rendered visualization chart data is transmitted to the user through a web front-end rendering interface. The user interaction layer presents a multi-scenario visual dashboard to users through a responsive web portal: the cross-platform operation jump unit uses a URL mapping interface to enable one-click jump from the Datart dashboard to the Slurm job management interface and the K8s container console, connecting the operation links of various systems; the scenario-based customized interface allows users to drag and drop components and adjust data dimensions, and customized templates can be saved to the cloud and exported as JSON files through the template reuse submodule for other users to reuse; the multi-terminal synchronization unit achieves real-time synchronization of customized templates and operation records on PC / tablet / mobile terminals through user account association, ensuring a consistent interactive experience across multiple terminals. At the same time, user operation commands are transmitted to the deployment and maintenance layer through the operation and maintenance control interface;

[0103] The container orchestration unit in the deployment and operation layer is based on Kubernetes to achieve cross-environment deployment on physical machines / virtual machines / public clouds / private clouds; the dynamic scaling unit automatically adjusts the number of container instances according to concurrent access volume to cope with peak access; the cross-platform monitoring unit collects operation and maintenance data such as container CPU / memory usage and interface response time, pushes them to the multi-scenario visualization engine layer to generate operation and maintenance charts, and triggers multi-channel alarms through the operation and maintenance alarm submodule when the indicators exceed the limits; the one-click deployment unit provides automated scripts including environment checks, dependency installation, and configuration initialization, which greatly shortens the cross-platform deployment and adaptation time.

[0104] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-dimensional data visualization management system based on an intelligent computing platform, characterized in that, It includes a cross-platform data interaction layer, a multi-scenario visualization engine layer, a dynamic adaptation layer, a user interaction layer, and a deployment and maintenance layer. The cross-platform data interaction layer and the dynamic adaptation layer are connected through a standardized data interface, the multi-scenario visualization engine layer and the user interaction layer are connected through a Web front-end rendering interface, the user interaction layer and the deployment and maintenance layer are connected through an operation and maintenance control interface, and the dynamic adaptation layer and the multi-scenario visualization engine layer are connected through a Datart platform data access interface. The cross-platform data interaction layer includes a protocol adaptation unit, a multi-source data access unit, a data format conversion unit, and a data cache synchronization unit. The multi-scenario visualization engine layer integrates the open-source Datart data dashboard platform, constructs a scenario-based template library including AI training job monitoring, heterogeneous resource scheduling, multi-tenant resource metering, and hybrid cloud intelligent computing overview, and configures a custom visualization component development unit. The dynamic adaptation layer includes a hardware architecture adaptation unit, a scheduling system adaptation unit, a business scenario adaptation unit, and an adaptation rule update unit. The user interaction layer includes a cross-platform operation jump unit, a scenario-based customized interface, and a multi-terminal synchronization unit. The deployment and operation layer uses Docker containerization to encapsulate the core modules of the system, and is configured with a container orchestration unit, a dynamic scaling unit, a cross-platform monitoring unit, and a one-click deployment unit.

2. The multi-dimensional data visualization management system based on an intelligent computing platform according to claim 1, characterized in that, The protocol adaptation unit supports dynamic switching between multiple protocols such as SSH, RESTfulAPI, gRPC, and JDBC, and adapts to the communication specifications of different intelligent computing scheduling systems such as Slurm, K8s, and YARN. The multi-source data access unit can simultaneously connect to heterogeneous hardware architectures of x86 / ARM, multiple operating systems such as Linux / Windows Server, and intelligent computing nodes deployed in a hybrid public / private cloud environment. The multi-source data access unit is configured with a hardware architecture identification subunit, which automatically matches the corresponding communication protocol and data acquisition method by reading the CPU architecture identifier of the intelligent computing node. x86 nodes use the SSH protocol for data acquisition, and ARM nodes use the gRPC protocol for data acquisition. The data format conversion unit, based on a preset intelligent computing data model, normalizes heterogeneous data such as JSON / XML / CSV into a standardized format; the data cache synchronization unit uses Redis distributed caching to save the collected data within 72 hours when the intelligent computing node is offline, and automatically retransmits it to the dynamic adaptation layer after the network is restored, ensuring cross-platform data integrity.

3. The multi-dimensional data visualization management system based on an intelligent computing platform according to claim 1, characterized in that, The AI ​​training job monitoring template has a built-in anomaly detection submodule. By comparing the real-time loss function value of the AI ​​job with the historical average, a red warning sign is triggered when the deviation exceeds 20%, and an anomaly cause analysis pop-up window is displayed in the user interaction layer. The component development unit supports users in developing dedicated components for GPU force heatmaps and AI job iteration progress curves based on WebGL / Canvas technology. After the component development is completed, it can be connected to the scenario-based template library through the Datart plugin interface. The multi-scenario visualization engine layer can use the Datart time-series data rendering interface to draw real-time curves of AI job loss function values ​​and GPU memory usage indicators. When the indicators exceed preset thresholds, a visual warning sign is automatically triggered. The specific operation logic steps are as follows: S101: The multi-scenario visualization engine layer receives AI job time-series data pushed by the dynamic adaptation layer through the Datart standardized data access interface. The data must conform to the preset intelligent computing data model specifications. The multi-scenario visualization engine layer performs legality verification on the received data and removes invalid data that is abnormal or missing. S102: Write the validated data into the time-series database and build a three-level index structure to improve query efficiency. The first-level index is JobID + GPUID, used to distinguish data from different GPU devices in different jobs. The second-level index is t. collect Sort by collection time in ascending order, supports fast query within time range, and has a three-level index for metric types to distinguish between "loss function" and "GPU memory" metrics for easy classification and querying. After storage is completed, a data storage log is generated to record the data write time, data volume, and index building status for subsequent data traceability and troubleshooting. S103: The multi-scenario visualization engine layer calls the Datart time-series data rendering interface to initialize curve drawing parameters according to the AI ​​job type; S104: The specific steps for filtering the data to be drawn are as follows: Based on the sliding time window algorithm, the following steps are taken: S1041: Determine window boundaries: Assume the sliding window duration is T = 300 seconds, and the current system timestamp is t. current Then the window start timestamp t start =t current -T×1000, window end timestamp t end =t current ; S1042: Data Query: Filter by "JobID + GPUID" using the time-series database index, querying time ranges within [t]. start ,t end All time-series data within ]; S1043: Data Sorting: Sort the query results by t collect Sort in ascending order to ensure the temporal continuity of curve plotting; S105: Based on the relative threshold and basic parameters in the received data, the "relative threshold" is converted into a directly comparable "absolute threshold" using a mathematical formula. The specific calculation is as follows: (1) Based on the historical mean of the loss function Compared with the relative warning threshold Thr loss,rel According to the formula Calculate the absolute threshold; (2) Based on the total GPU memory M gpu,total Compared with the relative warning threshold Thr gpu,rel According to the formula Calculate the absolute threshold; After the calculation is completed, the absolute threshold is bound to the corresponding job and GPU device and stored for subsequent real-time metric comparison. S106: The multi-scene visualization engine layer passes the time-series data filtered in S104 to the Datart rendering center, and dynamically draws curves according to the following logic: S1061: Coordinate Mapping: Mapping the timestamp t collect Convert L to x-axis coordinates current M gpu,current Mapped to the left and right y-axis coordinates respectively; S1062: Vertex generation: Connect the data points in time sequence to generate curve vertices, and use the Bézier curve smoothing algorithm to optimize the curve transition effect and avoid the stuttering of the broken line. S1063: Real-time update: Repeat the S104-S106 process every 1 second to refresh the sliding window data and update the curve to ensure the real-time display of the curve and the data collection. S107: Compare the latest collected time-series data within the current window with the absolute threshold calculated in S105 to determine whether to trigger an alert: If L is satisfied... current >Thr loss,abs or L current If the value is less than 0, then the loss function alert status is set to Alert. loss =1, trigger an alert; otherwise, an alert will be issued. loss =0, no warning; If M is satisfied gpu,current >Thr gpu,abs The GPU memory warning status will then be displayed as an Alert. gpu =1 triggers an alert; otherwise, an alert is issued. gpu =0, no warning; At the same time, calculate the GPU memory utilization. S108: If Alert loss =1, or Alert gpu =1. The multi-scenario visualization engine layer calls the Datart warning rendering sub-interface to generate corresponding visual warning labels. After the warning labels are generated, they are synchronized to the Web visualization portal of the user interaction layer through the Vue.js component communication interface to ensure that the data dashboard displays warning information in real time.

4. The multi-dimensional data visualization management system based on an intelligent computing platform according to claim 1, characterized in that, The hardware architecture adaptation unit automatically adjusts the data collection granularity to the second level based on the resource characteristics of CPU / GPU / AI chips and supplements the corresponding hardware-specific display dimensions. The scheduling system adaptation unit has a built-in scheduling system feature library containing Slurm / K8s / YARN. When a new scheduling system is connected, it automatically matches the job status definition and resource indicator caliber in the feature library without requiring manual code modification. The adaptation rule update unit supports receiving user-defined adaptation rules through the API interface. After updating, it is synchronized to the hardware architecture adaptation unit and the scheduling system adaptation unit in real time. The adaptation rule update unit also supports users importing new adaptation rules through Excel. After import, the rule validity is automatically verified, and the rule synchronization is completed within 10 seconds after the verification is passed, without the need to restart the system.

5. A multi-dimensional data visualization management system based on an intelligent computing platform according to claim 1, characterized in that, The container orchestration unit is based on Kubernetes to achieve cross-environment deployment of physical machines / virtual machines / public clouds / private clouds; the dynamic scaling unit automatically adjusts the number of container instances for the visualization service according to the concurrent access volume; the cross-platform monitoring unit collects container CPU / memory usage and interface response time operation and maintenance data, pushes it to the multi-scenario visualization engine layer through the data interface, and generates system operation and maintenance visualization charts. The cross-platform monitoring unit is configured with an operation and maintenance alarm sub-module. When the container CPU usage exceeds 80% and the interface response time exceeds 500ms, it automatically triggers multi-channel alarms. The alarm content includes the abnormal module, the current value of the abnormal indicator, and the suggested handling solution. The one-click deployment unit provides automated deployment scripts that include environment checks, dependency installation, and configuration initialization.

6. The multi-dimensional data visualization management system based on an intelligent computing platform according to claim 1, characterized in that, The cross-platform operation jump unit enables one-click jump from the Datart dashboard to the Slurm job management interface and the K8s container console via a URL mapping interface; the scenario-based customization interface supports dragging and dropping visualization components, adjusting data display dimensions, and saving customized templates to the cloud. The scenario-based customization interface has a built-in template reuse submodule, which allows users to export customized templates as JSON format files. Other users can import the files and reuse the templates quickly after permission verification. The multi-terminal synchronization unit is linked through user accounts to achieve real-time synchronization of customized templates and operation records between different terminals.

7. A multi-dimensional data visualization management system based on an intelligent computing platform according to claim 1, characterized in that, The business scenario adaptation unit dynamically selects core display indicators based on the differences between scientific research computing and commercial computing power leasing scenarios. The specific logical steps are as follows: S201: The business scenario adaptation unit obtains the current scenario type S through two methods: proactive configuration recognition and behavioral feature-assisted judgment. Proactive configuration recognition reads the scenario types preset by the intelligent computing platform administrator in the "scenario customization interface". When using behavioral feature-assisted judgment, if the scenario type is not proactively configured, the judgment is supplemented by analyzing the tenant's operational behavior characteristics over the past 30 days. If the tenant's operation includes "job script upload frequency ≥ 5 times / day" and "no billing query operation", it is judged as S=1, which is scientific research computing. If the tenant's operation includes "billing query frequency ≥ 3 times / day" and "core time recharge record ≥ 1", it is judged as S=2, which is commercial computing power leasing. After the judgment is completed, the current scenario type S is output as the basis for subsequent indicator screening. S202: Based on the scenario type S determined in step S1, retrieve the corresponding scenario's indicator weight configuration W from the "Scene Feature Library". S,i If S = 1, call the weight set {W} 1,1 W 1,2 W 1,3 W 1,4 W 1,5 If S = 2, call the weight set {W}. 2,1 W 2,2 W 2,3 W 2,4 W 2,5 Meanwhile, tenants can customize and adjust weights through the user interaction layer. Custom weights must meet the condition that "the sum of all indicator weights is ≤ 5". After adjustment, the weight set W for the current scenario will be automatically updated. S,i ; S203: Use the "weighted threshold screening method" to select the core display indicators under scenario S: I S,core ={I i |W S,i ≥T W ,I i ∈I}, where I represents the set of all indices traversed, and the weights W under scenario S are... S,i Greater than or equal to the filtering threshold T W Indicators, incorporated into the core display indicator set I S,core ; S204: For the selected core indicators I S,core Sort by weight in descending order to determine display priority. The formula used is: Rank(I i =DescendingOrder(W S,i |I i ∈I S,core ), according to its weight W in scene S S,i Sort from largest to smallest; S205: The business scenario adaptation unit will integrate "core indicator set I" S,core The "Display priority sorting results" option is pushed to the multi-scenario visualization engine layer through the Datart template configuration interface to achieve control over the visibility of indicators, adjustment of layout priority, and real-time synchronization updates. S206: If the intelligent computing platform supports tenants switching business scenarios during use, the following steps must be performed: S2061: Triggering Scene Switching and Resetting Weights: The tenant clicks the "Scene Switching" button in the user interaction layer and selects the target scene S. new The system automatically resets the indicator weight set to The tenant will be prompted that "The scene switch will update the visualization dashboard metrics. Do you confirm?". After confirmation, proceed to the next step. S2062: Execute historical data adaptation and display: Call the historical data interface of the cross-platform data interaction layer to obtain the tenant's full historical indicator data for the past 7 days, and display it according to the new scenario S. new core indicator set Filter the corresponding indicator data from the historical data, redraw the visualization charts, and refresh the visualization dashboard after completing the historical data adaptation to ensure that the core indicators in the new scenario are displayed in a consistent manner with the historical data.

8. A multi-dimensional data visualization management system based on an intelligent computing platform according to claim 7, characterized in that, The business scenario adaptation unit is also equipped with a scenario dynamic switching submodule and a historical data adaptation submodule. The scenario dynamic switching submodule allows tenants to select the target scenario for scientific research computing or commercial computing power leasing in real time through the "Scenario Switching" function button in the user interaction layer. During the switch, a secondary verification of the scenario type is automatically triggered. By comparing the matching degree between the tenant's current operation behavior and the characteristics of the target scenario, if the matching degree is ≥80%, the switch is completed directly. If the matching degree is <80%, a scenario switching risk warning pops up. After the tenant confirms, the indicator weight set of the current scenario is reset. After a scenario switch, the historical data adaptation submodule obtains the tenant's full set of historical metrics data for the past 7 days through the historical data interface of the cross-platform data interaction layer, and then adapts it to the core metric set of the target scenario. According to the formula Filter the historical data required for the target scenario, and then redraw the historical trend charts using the Datart data re-rendering interface to ensure that the historical and real-time data of core indicators are displayed consistently after the scenario switch. To adapt to historical data, D total,i This is the complete historical data.

Images