A Dynamic Data Visualization Construction Method Based on Large Language Models
By independently setting up an anomaly protection state model and constraining incremental control commands, the problem of excessive aggregation of the anomaly protection region in the visualization adjustment of the large language model is solved, thereby improving the stability and efficiency of the data visualization process.
Patent Information
- Application Number
- CN202611619202.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to prevent excessive aggregation of abnormal protection areas while maintaining local dynamic updates and overall stability of the visualization state during the dynamic visualization of multi-sampling rate operating data of industrial equipment, especially when executing incremental visualization adjustment commands generated by large language models.
By independently setting up an anomaly protection state model, conflicts between incremental control commands generated by the large language model and the anomaly protection state are detected and constrained. The display granularity is adjusted to meet the maximum allowable aggregation granularity, and the granularity configuration of the ordinary display area is independently updated in the visualization state model.
This effectively avoids excessive aggregation of abnormal protection areas, reduces data processing and rendering load, maintains the continuity and stability of the visualization state, and reduces the possibility of unexpected impacts.
Smart Images

Figure CN123115495A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of data processing and computer visualization technology, and in particular to a dynamic construction method for data visualization based on a large language model. Background Technology
[0002] With the development of Industrial Internet of Things (IIoT), intelligent sensing, and online equipment monitoring technologies, industrial equipment generates multi-source operational data such as temperature, pressure, vibration, and current during operation, and different types of operational data typically have different sampling frequencies. Existing data visualization systems usually filter, aggregate, downsample, and render the operational data based on preset chart rules or user operations to create visualizations of equipment operating trends, abnormal states, and historical changes. In recent years, large language models have been applied to data visualization scenarios. By parsing user-input natural language commands, corresponding visualization adjustment commands are generated, thereby enabling dynamic adjustments to the visualization status, such as data objects, display time range, display granularity, and display layout.
[0003] However, in the dynamic visualization of multi-sampling rate operating data of industrial equipment, when the incremental control instructions generated by the large language model involve adjusting the display granularity, existing technologies usually uniformly adjust the data within the current display range according to the new display granularity. This can easily lead to the anomaly protection area and the ordinary data area adopting the same downsampling or aggregation strategy, resulting in transient abnormal data in the anomaly protection area being covered by lower-granularity data aggregation. At the same time, if all data is displayed with high granularity to preserve anomaly details, it will increase the data processing and visualization rendering load. Therefore, there is a technical problem of how to implement independent granularity constraints on the anomaly protection area when executing the incremental visualization adjustment instructions generated by the large language model, and achieve local dynamic reconstruction without affecting other visualization states. Summary of the Invention
[0004] To overcome the above shortcomings, this invention provides a dynamic data visualization construction method based on a large language model. It aims to improve the problem of avoiding excessive aggregation of abnormal protection areas and achieving local dynamic updates of the visualization status when executing incremental visualization adjustment instructions generated by the large language model during the dynamic visualization of multi-sampling rate operation data of industrial equipment.
[0005] This invention provides the following technical solution: a method for dynamically constructing data visualization based on a large language model, comprising: S1. Collect multi-source operating data of the target device and synchronize the time. Based on the synchronized multi-sampling rate operating data sequence, determine the abnormal protection area and establish an abnormal protection status including the abnormal protection area and the maximum allowable aggregation particle size. S2. Establish a visual state model based on the multi-sampling rate running data sequence, and set the visual state model independently from the anomaly protection state; S3. Obtain the natural language visualization adjustment instructions input by the user, parse the natural language visualization adjustment instructions through the large language model, and generate incremental control instructions relative to the visualization state model. S4. Match the incremental control command with the abnormal protection state. When the incremental control command conflicts with the abnormal protection state, constrain the incremental control command by projection, retain the control parameters that do not cause conflict, and adjust the conflicting control parameters to an executable incremental control command that satisfies the maximum allowable aggregation granularity. S5. Update the visualization state model based on the executable incremental control instruction, and perform regional granular configuration on the multi-sampling rate running data sequence according to the updated visualization state model and the anomaly protection state, so that the normal display area is configured according to the visualization state model and the anomaly protection area is configured according to the anomaly protection state. S6. Based on the regional granularity configuration result, perform local data reconstruction and visualization rendering on the data region where the display state changes to obtain dynamic visualization results, and update the visualization state model based on subsequent natural language visualization adjustment instructions.
[0006] The present invention has the following beneficial effects: 1. In this invention, by setting the abnormal protection state independently outside the visualization state model, and performing conflict detection and constraint projection on the abnormal protection state when executing the incremental control instructions generated by the large language model, the display granularity of the abnormal protection area is constrained by the maximum allowable aggregation granularity, while the ordinary display area can still be adjusted according to the visualization state model. This avoids excessive aggregation of the running data in the abnormal protection area and reduces the data processing and rendering load generated by using high-precision display for all running data.
[0007] 2. In this invention, by enabling the large language model to generate incremental control instructions relative to the current visualization state model, and only updating the state parameters involved in the incremental control instructions, the visualization state not involved in the incremental control instructions remains unchanged, thereby reducing unnecessary changes in the visualization state during continuous natural language interaction and maintaining the continuity of the display state during dynamic visualization.
[0008] 3. In this invention, by maintaining the anomaly protection state and the visualization state model separately, the anomaly protection rules and the visualization display adjustment are made independent of each other, and the anomaly protection state is kept effective during the visualization state update process. This reduces the possibility that the visualization adjustment will have an unexpected impact on the anomaly data protection rules and improves the stability of the dynamic visualization control process. Attached Figure Description
[0009] Figure 1 This is a flowchart illustrating a dynamic data visualization construction method based on a large language model proposed in this invention. Figure 2 This is a schematic diagram of the data acquisition and anomaly protection status establishment process of a data visualization dynamic construction method based on a large language model proposed in this invention; Figure 3 This is a schematic diagram of the process for establishing a visualization state model and generating incremental control instructions in a data visualization dynamic construction method based on a large language model proposed in this invention. Figure 4 This is a schematic diagram of the conflict detection and constraint projection process of a dynamic data visualization construction method based on a large language model proposed in this invention. Figure 5 This diagram illustrates the state update, regional granular configuration, and local reconstruction rendering process of a data visualization dynamic construction method based on a large language model proposed in this invention. Detailed Implementation
[0010] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0011] This invention provides a method for dynamically constructing data visualization based on a large language model, such as... Figures 1-5 As shown, it includes the following steps: S1. Collect multi-source operating data of the target device and synchronize the time. Based on the synchronized multi-sampling rate operating data sequence, determine the abnormal protection area and establish an abnormal protection status including the abnormal protection area and the maximum allowable aggregation particle size.
[0012] Furthermore, in S1, multi-source operational data of the target device is collected and time-synchronized, including: Collect operational data from different data sources in the target device, and obtain the sampling timestamp and sampling frequency corresponding to each operational data. A unified time base is determined based on the sampling timestamps of each operational data, and each operational data is mapped to the unified time base. The mapped running data is time-aligned according to the sampling frequency of each running data to obtain a multi-sampling rate running data sequence corresponding to a unified time base.
[0013] Furthermore, in S1, an anomaly protection region is determined based on the synchronized multi-sampling rate running data sequence, and an anomaly protection state is established, including the anomaly protection region and the maximum allowable aggregation granularity, including: The synchronized multi-sampling rate running data sequence is divided according to a preset time window, and the rate of change, fluctuation degree and mutation degree corresponding to each time window are extracted respectively. The abnormal time windows are determined based on the rate of change, degree of fluctuation, and degree of abrupt change corresponding to each time window, and adjacent abnormal time windows are merged to obtain the abnormal protection area. The maximum allowable aggregation granularity corresponding to the anomaly protection area is determined based on the duration of the anomaly characteristics within the anomaly protection area and the sampling frequency of the corresponding operational data. An abnormal protection status is established based on the abnormal protection area and the maximum allowable particle size.
[0014] Specifically, in this embodiment, the target device is an industrial pump. The target device is equipped with a temperature sensor, a pressure sensor, a vibration sensor, and a current sensor to collect temperature data, pressure data, vibration data, and current data during the operation of the target device, respectively. The sampling frequency of the temperature data is 1 Hz, the sampling frequency of the pressure data is 10 Hz, the sampling frequency of the current data is 100 Hz, and the sampling frequency of the vibration data is 1000 Hz.
[0015] Collect operational data from different data sources in the target device, and obtain the sampling timestamp and sampling frequency corresponding to each data source. For the i-th data source, the operational data is represented as: ; in, This represents the sampling timestamp of the k-th running data in the i-th data source. This represents the running data value of the k-th running data in the i-th data source. This represents the number of data items processed by the i-th data source. This represents the sampling frequency of the i-th data source.
[0016] Synchronization markers are sent to each data source according to a preset synchronization period. Each data source records the corresponding timestamp upon receiving the synchronization marker. The time offset of each data source relative to the unified time base is determined based on the timestamps of the same synchronization marker in each data source. The time offset is expressed as: ; in, This represents the timestamp corresponding to the m-th synchronization marker of the i-th data source. This represents the timestamp corresponding to the m-th synchronization marker under a unified time base, where M represents the number of synchronization markers. In this embodiment, M=3.
[0017] The sampling timestamps of the running data are corrected according to the time offsets corresponding to each data source to obtain the synchronized sampling timestamps: ; in, This indicates that the k-th running data in the i-th data source corresponds to the sampling timestamp of the unified time base. The running data of each data source is sorted according to the synchronized sampling timestamps, and the sampling frequency corresponding to each data source is retained to obtain the synchronized multi-sampling rate running data sequence.
[0018] The synchronized multi-sampling rate running data sequence is divided according to a preset time window. In this embodiment, the time window is set to 1 second, and an overlap time of 0.5 seconds is set between adjacent time windows. For the i-th data source in the w-th time window, the rate of change characteristic is calculated: ; in, This represents the initial running data value of the i-th data source within the w-th time window. This represents the end-of-run data value of the i-th data source within the w-th time window. This represents the length of the w-th time window. The rate of change feature is used to characterize the speed at which the running data changes within the time window.
[0019] Calculate the volatility characteristics: ; in, This represents the number of data items processed by the i-th data source within the w-th time window. This represents the mean of the running data from the i-th data source within the w-th time window. The volatility feature is used to characterize the dispersion of the running data within the time window.
[0020] Calculate the mutation degree characteristics: ; in, and This represents the running data values at adjacent sampling times. The mutation degree feature is used to characterize the degree of local mutation in the running data within the time window.
[0021] The rate of change feature, the fluctuation feature, and the abrupt change feature are each normalized to ensure that their values range from 0 to 1, thus obtaining the normalized rate of change feature, normalized fluctuation feature, and normalized abrupt change feature. The anomaly level of the w-th time window is determined based on the normalized features corresponding to each data source. ; in, , and These represent the maximum rate of change, maximum volatility, and maximum abrupt change of each data source after normalization within the w-th time window, respectively. 0.3, 0.30, and 0.40 represent the weights corresponding to the rate of change, volatility, and abrupt change, respectively. In this embodiment, the anomaly threshold is set to 0.6. When When the w-th time window is identified as an abnormal time window, for example, when the normalized maximum rate of change corresponding to the w-th time window is 0.8, the normalized maximum fluctuation is 0.6, and the normalized maximum abrupt change is 0.9: ; The anomaly level is greater than 0.6, therefore the w-th time window is determined as the abnormal time window. The anomaly level threshold is used to determine the abnormal time window based on the anomaly level of the target device's operating data. In other embodiments, the anomaly level threshold can be adjusted based on the distribution of the target device's normal operating data.
[0022] Adjacent time windows are merged to form anomaly protection zones. When the time interval between adjacent anomaly time windows does not exceed 0.5 seconds, they are merged into the same anomaly protection zone. The duration of the anomaly characteristic corresponding to each anomaly protection zone is then determined. The duration of the abnormal feature The time difference between the termination time and the start time of the anomaly protection zone is determined, and the highest sampling frequency in the operational data associated with the anomaly protection zone is identified. Based on the duration of the aforementioned abnormal characteristics. and the highest sampling frequency Determine the maximum allowable particle size corresponding to the abnormal protection region: ; in, This indicates the maximum allowed aggregation timescale for the anomaly protection region. This represents the upper limit of the aggregation timescale determined based on the duration of the abnormal features. This represents the upper limit of the aggregation time scale determined based on the highest sampling frequency of the associated running data. This indicates the maximum number of sampling points allowed in a single aggregation. This represents the original sampling time interval corresponding to the highest sampling frequency. In this embodiment, When the duration of the abnormal characteristics in the protected area is 0.12 s, and the sampling frequency of the vibration data associated with the protected area is 1000 Hz: ; That is, the maximum allowable polymerization particle size corresponding to the abnormal protection area is 20ms.
[0023] An anomaly protection status is established based on the anomaly protection area and the maximum allowable aggregation granularity. A correspondence is then established between the anomaly protection status and the associated data object corresponding to the anomaly protection area. This allows for constraints on the display granularity of the anomaly protection area when dynamically visualizing the synchronized multi-sampling rate running data sequence.
[0024] S2. Establish a visual state model based on the multi-sampling rate running data sequence, and set the visual state model independently from the anomaly protection state.
[0025] Furthermore, in S2, a visual state model is established based on the multi-sampling-rate running data sequence, including: The data objects to be displayed and their corresponding display time ranges are determined based on the multi-sampling rate running data sequence. Determine the corresponding display granularity and display layout based on the data object and the display time range; Generate a visual state model based on data objects, display time range, display granularity, and display layout; The visualized state model and the anomaly protection state are stored and updated independently, so that changes in the state of the visualized state model do not change the anomaly protection state.
[0026] Specifically, in this embodiment, a visualization state model is established based on the synchronized multi-sampling rate running data sequence obtained in S1. The visualization state model is used to describe the display objects, display time range, data display granularity, display layout, and interaction state of the current multi-sampling rate running data sequence. The anomaly protection state is not included as a component parameter of the visualization state model.
[0027] The data objects to be displayed are determined based on the synchronized multi-sampling-rate running data sequence. Each data source in the synchronized multi-sampling-rate running data sequence is taken as a data object, and the data type, sampling frequency, and time coverage range corresponding to each data object are obtained. In this embodiment, the data objects to be displayed include temperature data, pressure data, vibration data, and current data. The display time range is determined based on the time coverage range of each data object. The latest sampling time of each data object is obtained. and earliest sampling time The display time range is determined according to the common time coverage of the target devices: ; in, Indicates the time range to be displayed. This represents the earliest sampling time of the i-th data object. This represents the latest sampling time of the i-th data object. By determining the common time range covered by all data objects, different data objects can be displayed under the same time coordinate. In this embodiment, the most recent 24 hours are preferentially selected as the display time range; when the common time coverage range is less than 24 hours, the common time coverage range is used as the display time range.
[0028] The data display granularity is determined based on the display time range. Let the time length corresponding to the display time range be... The preset target number of data points to display for a single data object is [number]. The minimum sampling frequency for each data object to be displayed is The time scale corresponding to the data granularity is determined according to the following formula: ; in, This indicates the time scale corresponding to the data granularity. Indicates the duration of the displayed time range. This indicates the number of target display data points corresponding to a single data object. This indicates the lowest sampling frequency in the data object to be displayed. In this embodiment, Set it to 2000.
[0029] The formula is used to ensure that the amount of displayed data is within a preset range while avoiding the time scale corresponding to the data display granularity being smaller than the original sampling time interval corresponding to the lowest sampling frequency. When the display time range is 24 hours and the lowest sampling frequency is 1Hz, we have: ; Therefore, in the initial visualization state, the data display granularity corresponds to a time scale of 43.2 seconds.
[0030] The display layout is determined based on the number and data type of the data objects to be displayed. For continuously running data, corresponding data trend display areas are established for each data object, and arranged in single or multiple columns according to the number of data objects. When the number of data objects to be displayed is no more than four, a single or double column arrangement is used; when the number of data objects to be displayed is greater than four, the data is divided into multiple display areas according to the number of data objects, allowing different data objects to be displayed in independent display areas. A visualization state model is established based on the data objects to be displayed, the display time range, the data display granularity, and the display layout. The visualization state model is represented as follows: ; in, This represents the data object to be displayed. Indicates the time range to be displayed. This indicates the time scale corresponding to the data granularity. Indicates the display layout. Indicates the interactive state.
[0031] When establishing the visual state model, the visual state model and the anomaly protection state obtained from S1 are stored in separate state storage areas. Updates to the visual state model only modify the state parameters in the visual state model, and updates to the anomaly protection state only modify the protection parameters in the anomaly protection state, so that changes in the state of the visual state model do not directly modify the anomaly protection state.
[0032] When the visualization state model is subsequently modified based on the user's input natural language visualization adjustment instructions, only the corresponding state parameters in the visualization state model are updated, and the abnormal protection state is read as an independent constraint state, thereby establishing a visualization state model for subsequent incremental control.
[0033] S3. Obtain the natural language visualization adjustment instructions input by the user, parse the natural language visualization adjustment instructions through the large language model, and generate incremental control instructions relative to the visualization state model.
[0034] Furthermore, in S3, incremental control commands relative to the visualized state model are generated, including: Obtain the user's input of natural language visualization adjustment instructions, and input the natural language visualization adjustment instructions and visualization state model into the large language model; By parsing natural language visualization adjustment instructions through a large language model, the data object to be adjusted, the display time range, the display granularity, or the display layout can be determined. The data objects to be adjusted, the display time range, the display granularity, or the display layout are mapped to the visualization state model to determine the state parameters to be updated in the visualization state model. Generate incremental control commands relative to the visualized state model based on the state parameters to be updated.
[0035] Specifically, in this embodiment, the visualization state model V obtained in S2 acquires the user's input natural language visualization adjustment instructions, and inputs the natural language visualization adjustment instructions and the visualization state model together into the large language model. The visualization state model includes the data objects to be displayed. Display time range Data granularity Display layout and interaction status .
[0036] The large language model employs a pre-trained large language model, performing semantic parsing of the input data during the inference phase without involving model parameter training. The input to the large language model includes natural language visualization adjustment instructions and the current visualization state model. The output includes the data object to be adjusted, the display time range, display granularity, display layout, and the corresponding adjustment value. It also outputs structured state adjustment data containing state parameter identifiers, target parameter values, and adjustment operation types. For example, the current visualization state model is: .in, This represents the data object to be displayed. Indicates the current time range being displayed. This indicates the granularity of the current data display. Indicates the current display layout. Indicates the current interaction state.
[0037] When a user inputs a natural language visualization adjustment command, such as requesting increased pressure data display and magnification of abnormal areas, the large language model performs semantic parsing on the command, identifying the data object adjustment requirements, display area adjustment requirements, and display granularity adjustment requirements, and mapping them to the state parameters in the visualization state model. Based on the state parameter identifiers in the structured state adjustment data, the target parameter value and adjustment operation type are mapped to the corresponding state parameters in the visualization state model to determine the data object, display time range, display granularity, or display layout to be updated. For state parameters for which the user has not specified an adjustment request, no corresponding adjustment parameters are generated. Based on the state parameter identifiers in the structured state adjustment data, the target parameter value is compared with the current parameter value of the corresponding state parameter in the visualization state model to determine the state parameters to be updated. The state parameters to be updated are represented as follows: ; in, This represents the set of state parameters to be updated. This represents the j-th state parameter. This represents the parameter value of the j-th state parameter in the current visualized state model. This represents the target parameter value of the j-th state parameter after being determined by the natural language visualization adjustment instructions. The formula is used to filter the state parameters in the current visualization state model that actually need to change.
[0038] Incremental control instructions are generated based on the set of state parameters to be updated. Each incremental control instruction includes at least a state parameter identifier, a target parameter value, and the corresponding adjustment operation. The incremental control instruction is represented as follows: ; in, This indicates an incremental control command. This indicates the status parameter to be updated. This represents the target parameter value corresponding to the state parameter to be updated. This indicates the adjustment operation corresponding to the status parameter to be updated.
[0039] For example, in the current visualization state model, the data object is temperature data, the display time range is the last 24 hours, and the data display granularity corresponds to a time scale of 43.2 seconds. After the user inputs a natural language visualization adjustment command to increase pressure data and zoom in on the abnormal area, the large language model outputs the following adjustment parameters: data object adjustment parameter is to increase pressure data, display area adjustment parameter is to add an abnormal protection area, and display granularity adjustment parameter is to improve the display accuracy of the abnormal protection area. After comparing these adjustment parameters with the current visualization state model, only the newly added pressure data, the abnormal protection area, and the display granularity of the abnormal protection area are selected as the state parameters to be updated.
[0040] Based on the state parameters to be updated, an incremental control command relative to the current visualization state model is generated, and the incremental control command is output to the subsequent abnormal protection state matching process, so that the incremental control command can be conflict detected and constraint projected according to the abnormal protection state.
[0041] S4. Match the incremental control command with the abnormal protection state. When the incremental control command conflicts with the abnormal protection state, constrain the incremental control command by projection, retain the control parameters that do not cause conflict, and adjust the conflicting control parameters to the executable incremental control command that meets the maximum allowable aggregation granularity.
[0042] Furthermore, in S4, incremental control commands are matched with abnormal protection states, including: Extract the data object to be adjusted, the display time range, and the display granularity from the incremental control command, and obtain the abnormal protection area, associated data object, and maximum allowable aggregation granularity from the abnormal protection status; Based on the correspondence between the data object to be adjusted and the associated data object, as well as the time range relationship between the display time range and the anomaly protection area, the anomaly impact area corresponding to the incremental control command is determined. When an abnormal influence area exists, the displayed granularity in the incremental control command is compared with the maximum allowable aggregation granularity; When the displayed granularity is greater than the maximum allowable aggregation granularity, it is determined that there is a conflict between the incremental control command and the abnormal protection state; otherwise, it is determined that there is no conflict between the incremental control command and the abnormal protection state.
[0043] Furthermore, in S4, the incremental control instructions are constrained and projected, retaining control parameters that do not cause conflicts, and adjusting conflicting control parameters to meet the executable incremental control instructions that satisfy the maximum allowable aggregation granularity, including: The incremental control command is decomposed according to the status parameters corresponding to the incremental control command to obtain multiple control parameters, and the conflict control parameters related to the abnormal protection area are determined according to the abnormal protection status. Among multiple control parameters, retain the non-conflicting control parameters that do not affect the abnormal protection status, and determine the non-conflicting control parameters as the control parameters that are allowed to be executed; For the conflict control parameters, adjust their corresponding display granularity to a display granularity no greater than the maximum allowed aggregation granularity; The non-conflicting control parameters are recombined with the adjusted conflicting control parameters to generate executable incremental control instructions that satisfy the abnormal protection state.
[0044] Specifically, in this embodiment, based on the incremental control command obtained in S3 and the abnormal protection state obtained in S1, abnormal protection constraint matching and constraint projection are performed on the incremental control command.
[0045] The incremental control command reads the data object to be adjusted, the display time range, and the display granularity. The anomaly protection status reads the anomaly protection area, associated data objects, and the maximum allowable aggregation granularity. The display granularity is represented by an aggregation time scale; the larger the aggregation time scale, the higher the degree of data aggregation. For each data object to be adjusted in the incremental control command, it is matched with the associated data objects in the anomaly protection status. When the data object to be adjusted is the same as the associated data object, the display time range is further compared with the time range of the anomaly protection area. When there is a time overlap, the corresponding anomaly protection area is determined as the anomaly impact area. The anomaly impact area can be represented as: ; in, Represents the set of areas affected by the anomaly. This represents the set of data objects to be adjusted corresponding to the incremental control command. This represents the set of associated data objects corresponding to the r-th anomaly protection region. This indicates the display time range corresponding to the incremental control command. This represents the time range of the r-th anomaly protection zone. The formula is used to determine the anomaly protection zone actually affected by the current incremental control command.
[0046] When the affected area is empty, it is determined that the incremental control instruction does not affect the abnormal protection status, and the control parameters in the incremental control instruction retain their original values and are output as executable incremental control instructions. When the affected area is not empty, the maximum allowable aggregation granularity corresponding to each affected area is obtained. For cases where the same incremental control instruction simultaneously affects multiple abnormal protection areas, the minimum value among the maximum allowable aggregation granularities corresponding to each affected area is determined as the constraint granularity corresponding to the current incremental control instruction. ; in, This indicates the constraint granularity corresponding to the current incremental control command. This represents the maximum allowable aggregation granularity corresponding to the r-th anomaly protection region. The formula is used to ensure that when the same incremental control command is applied to multiple anomaly protection regions, the granularity constraints of each anomaly protection region are satisfied simultaneously.
[0047] The target display granularity and constraint granularity in the incremental control command are compared. If the time scale corresponding to the target display granularity is greater than the constraint granularity, a conflict is determined between the incremental control command and the abnormal protection state; if the time scale corresponding to the target display granularity is not greater than the constraint granularity, no conflict is determined. When a conflict is determined, the incremental control command is decomposed according to its contained state parameters to obtain multiple control parameters. Based on the data object and display time range corresponding to each control parameter, it is determined whether it affects the abnormal protection state. For non-conflicting control parameters that do not affect the abnormal protection state, their target parameter values remain unchanged. For conflicting control parameters that affect the abnormal protection state, the target display granularity corresponding to the conflicting control parameter is compared with the maximum permissible aggregation granularity. If the time scale corresponding to the target display granularity is greater than the maximum permissible aggregation granularity, the target display granularity is adjusted to the maximum permissible aggregation granularity.
[0048] The display granularity after constraint projection is expressed as follows: ; in, This indicates the executable display granularity after constraint projection. This indicates the target display granularity in the incremental control command. This indicates the constraint granularity corresponding to the current incremental control command. The formula is used to limit the time scale corresponding to the target display granularity to the maximum aggregate time scale allowed by the abnormal protection state.
[0049] For example, if the incremental control command requires adjusting the current display granularity to 5 minutes, and the maximum allowable aggregation granularity of the current anomaly protection area is 20 ms, then: ; Therefore, the display granularity corresponding to the abnormal impact area is adjusted to 20 ms, while the normal display area unrelated to the abnormal protection area is still executed according to the 5 min display granularity in the incremental control instruction.
[0050] The retained non-conflicting control parameters are recombine with the conflicting control parameters after constraint projection to generate executable incremental control instructions that satisfy the abnormal protection state. These executable incremental control instructions are then output to S5 to update the visualized state model. In this way, when an incremental control instruction conflicts with the abnormal protection state, only the conflicting control parameters are adjusted, without changing the non-conflicting control parameters.
[0051] S5. Update the visualization state model based on the executable incremental control instructions, and perform regional granular configuration on the multi-sampling rate running data sequence according to the updated visualization state model and the abnormal protection state, so that the ordinary display area is configured according to the visualization state model and the abnormal protection area is configured according to the abnormal protection state.
[0052] Furthermore, in S5, the visualized state model is updated based on executable incremental control instructions, including: Extract the state parameters to be updated and their corresponding values from the executable incremental control instructions, and determine the corresponding state parameters in the visualized state model. Update the corresponding state parameter value to the corresponding value to obtain the updated visualized state parameter; Keep the state parameters in the visual state model that are not involved by the executable incremental control instructions unchanged, and generate an updated visual state model based on the updated visual state parameters.
[0053] Furthermore, in S5, based on the updated visualization state model and anomaly protection state, the multi-sampling rate running data sequences are configured with regional granularity, including: The current display time range is determined based on the updated visualization status model, and the multi-sampling rate running data sequence is divided into a normal display area and an anomaly protection area based on the relationship between the current display time range and the time range of the anomaly protection area. The display granularity corresponding to the normal display area is determined based on the updated visualization state model, and the maximum allowable aggregation granularity corresponding to the abnormal protection area is determined based on the abnormal protection state. The normal display area is configured with granularity according to the display granularity, and the abnormal protection area is configured with granularity according to the display granularity not greater than the maximum allowable aggregation granularity, thus obtaining the regional granularity configuration result.
[0054] Specifically, based on the executable incremental control instructions obtained in S4, the visual state model established in S2 is updated, and the updated visual state model and the anomaly protection state established in S1 are used to determine the regional granularity configuration of the current multi-sampling rate running data sequence. The state parameters to be updated and their corresponding values are read from the executable incremental control instructions, and the state parameters to be updated are matched with the corresponding state parameters in the visual state model according to their names. State parameters include at least one of data objects, display time ranges, display granularity, and display layout.
[0055] The matched state parameters are updated to the parameter values corresponding to the executable incremental control instructions, forming updated visual state parameters. For state parameters not included in the executable incremental control instructions, their original parameter values are kept unchanged, and an updated visual state model is generated based on the updated visual state parameters. The updated visual state model can be represented as: ; in, This represents the updated visual state model. This represents the updated data object. This indicates the updated display time range. This indicates the updated display granularity. This indicates the updated display layout. This indicates the updated interaction state.
[0056] The updated visualization status model is used to determine the multi-sampling rate running data sequence, display time range, display granularity, and display layout that need to be displayed. Abnormal protection states continue to be stored independently in their respective protection state storage areas. Subsequently, based on the updated display time range... The time range is matched with the abnormal protection area in the abnormal protection state. When there is an intersection between the abnormal protection area and the updated display time range, the intersection is determined as the abnormal protection area within the current display time range; the portion of the updated display time range that does not fall into the abnormal protection area is determined as the normal display area.
[0057] For multiple anomaly protection regions, the corresponding time interval for each anomaly protection region within the updated display time range is determined, and the non-overlapping time intervals between anomaly protection regions are designated as the corresponding normal display regions. This yields the normal display regions and anomaly protection regions within the current display time range. The display granularity corresponding to the normal display regions is then determined based on the updated visualization state model. For the normal display area, with As the corresponding data aggregation time scale, the granularity of the multi-sampling rate running data sequence is configured. For each anomaly protection region, the maximum allowable aggregation granularity corresponding to the anomaly protection state is obtained. Where r represents the number of the anomaly protection region. The maximum allowable aggregation granularity is used to limit the maximum aggregation time scale allowed for the anomaly protection region, and the aggregation time scale corresponding to the actual display granularity of the anomaly protection region is not greater than the maximum allowable aggregation granularity. The actual display granularity of the anomaly protection region is determined based on the updated display granularity and the maximum allowable aggregation granularity: ; in, This represents the actual display granularity corresponding to the r-th anomaly protection region. This indicates the aggregation timescale corresponding to the display granularity determined by the updated visualization state model. This represents the maximum allowable aggregation granularity corresponding to the r-th anomaly protection region. The formula is used to ensure that the aggregation timescale corresponding to the actual display granularity of the anomaly protection region does not exceed the maximum allowable aggregation granularity, and to keep the display granularity unchanged when the aggregation timescale corresponding to the updated display granularity is not greater than the maximum allowable aggregation granularity.
[0058] For example, if the updated visualization state model determines the display granularity to be 5 minutes, and the maximum allowable aggregation granularity corresponding to a certain anomaly protection area is 20 ms, then: ; Therefore, the abnormal protection area uses an actual display granularity of 20 ms, while the normal display area still uses a display granularity of 5 min.
[0059] For both the normal display area and the anomaly protection area, the granularity of the multi-sampling rate running data sequence is configured according to the corresponding display granularity. For the normal display area, continuous data is aggregated over time according to the corresponding display granularity; for the anomaly protection area, running data is aggregated over time within the corresponding maximum allowable aggregation granularity, and the multi-sampling rate running data sequence corresponding to the anomaly protection area is retained.
[0060] The regional granularity configuration results are generated based on the display granularity corresponding to each normal display area and each abnormal protection area. The regional granularity configuration results include at least the area location, associated data object and corresponding display granularity, and establish a correspondence with the updated visualization state model so that S6 can perform local data reconstruction and visualization rendering on the data areas where the display state has changed based on the regional granularity configuration results.
[0061] S6. Based on the regional granularity configuration results, perform local data reconstruction and visualization rendering on the data areas where the display status changes to obtain dynamic visualization results, and update the visualization status model based on subsequent natural language visualization adjustment instructions.
[0062] Furthermore, in S6, local data reconstruction and visualization rendering are performed on the data areas where the display state changes, resulting in dynamic visualization results. The visualization state model is then updated based on subsequent natural language visualization adjustment instructions, including: Determine the data areas where the display status has changed based on the visualization status model before and after the update; Based on the regional granularity configuration results, extract the multi-sampling rate running data sequence corresponding to the data region where the display status changes, and perform corresponding granularity processing on the multi-sampling rate running data sequence to obtain local visualization data; Based on the updated visualization state model, the local visualization data is reconstructed to generate corresponding local visualization components, while keeping the visualization components corresponding to data areas whose display state has not changed unchanged. By combining and rendering local visualization components with visualization components corresponding to data areas whose display states have not changed, a dynamic visualization result is obtained. After obtaining subsequent natural language visualization adjustment instructions, the natural language visualization adjustment instructions are executed based on the visualization state model corresponding to the dynamic visualization results, and the visualization state model is updated.
[0063] Specifically, in this embodiment, based on the regional granularity configuration results obtained in S5, the data region corresponding to the updated visualization state model is locally reconstructed and visualized, and the visualization state model is further updated using subsequent natural language visualization adjustment instructions.
[0064] The visualization state model before and after the S5 update is compared to determine the changed state parameters. Based on the data objects, display time ranges, and display layouts corresponding to these changed state parameters, the data regions where the display state has changed are identified. These data regions include those corresponding to newly added data objects, those where the display time range has changed, and those where the display granularity has changed. When only the display layout changes, the corresponding data region is not identified as a region where the display state has changed; instead, the display position and size of the corresponding visualization components are adjusted according to the updated visualization state model. For the r-th data region where the display state has changed, the data object, data time range, and display granularity corresponding to the r-th data region are obtained from the sub-regional granularity configuration results. The corresponding running data is then extracted from the multi-sampling rate running data sequence based on the data object and data time range.
[0065] The extracted runtime data is processed according to the granularity configuration results for different regions. The aggregation timescale corresponding to the display granularity is divided into multiple continuous time aggregation intervals. The runtime data within the q-th time aggregation interval is aggregated to obtain locally visualized data. For continuous runtime data, this embodiment uses the average value of the time interval as the aggregation result, which is expressed as follows: ; in, This represents the local visualization data corresponding to the q-th time aggregation interval within the r-th data region where the display status changes. This represents the number of running data within the q-th time aggregation interval. This represents the k-th running data point within the q-th time aggregation interval. The formula is used to convert multiple running data points within the same time aggregation interval into corresponding visual data according to the configured data granularity.
[0066] For anomaly protection areas, the time scale of the time aggregation interval does not exceed the maximum allowable aggregation granularity determined by S1, in order to maintain the data granularity constraints corresponding to the anomaly protection state. For normal display areas, the time scale of the time aggregation interval is set according to the display granularity determined by the updated visualization state model. Based on the data objects, display time range, display granularity, and display layout in the updated visualization state model, the local visualization data is mapped to the corresponding display coordinates, generating local visualization components corresponding to the data areas where the display state has changed. For newly added data objects, new local visualization components are generated; for data areas where the display granularity has changed, the corresponding local visualization components are regenerated according to the updated display granularity; for data areas where only the display time range has changed, only the data parts that have been added or moved out of the current display time range are updated.
[0067] For data areas whose display state has not changed, their corresponding visualization components and display states remain unchanged. The updated local visualization components are combined with the visualization components corresponding to the data areas whose display state has not changed according to the updated visualization state model, and visualization rendering is performed to obtain dynamic visualization results. After the dynamic visualization results are generated, the user's input natural language visualization adjustment instructions are obtained. The natural language visualization adjustment instructions and the current visualization state model are input into the large language model for parsing to generate new incremental control instructions relative to the current visualization state model. The new incremental control instructions are processed according to the anomaly protection state matching and constraint projection process in S4, and the current visualization state model is updated according to the processing results, so that subsequent dynamic visualization construction continues to be based on the updated visualization state model.
[0068] Therefore, whenever the visualization state changes, only the local visualization data of the corresponding data area is reorganized and local visualization rendering is performed, while the data areas whose display state has not changed retain their original visualization state.
[0069] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for dynamically constructing data visualization based on a large language model, characterized in that, include: S1. Collect multi-source operating data of the target device and synchronize the time. Based on the synchronized multi-sampling rate operating data sequence, determine the abnormal protection area and establish an abnormal protection status including the abnormal protection area and the maximum allowable aggregation particle size. S2. Establish a visual state model based on the multi-sampling rate running data sequence, and set the visual state model independently from the anomaly protection state; S3. Obtain the natural language visualization adjustment instructions input by the user, parse the natural language visualization adjustment instructions through the large language model, and generate incremental control instructions relative to the visualization state model. S4. Match the incremental control command with the abnormal protection state. When the incremental control command conflicts with the abnormal protection state, constrain the incremental control command by projection, retain the control parameters that do not cause conflict, and adjust the conflicting control parameters to an executable incremental control command that satisfies the maximum allowable aggregation granularity. S5. Update the visualization state model based on the executable incremental control instruction, and perform regional granular configuration on the multi-sampling rate running data sequence according to the updated visualization state model and the anomaly protection state, so that the normal display area is configured according to the visualization state model and the anomaly protection area is configured according to the anomaly protection state. S6. Based on the regional granularity configuration result, perform local data reconstruction and visualization rendering on the data region where the display state changes to obtain dynamic visualization results, and update the visualization state model based on subsequent natural language visualization adjustment instructions.
2. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S1, multi-source operational data of the target device is collected and time-synchronized, including: Collect operational data from different data sources in the target device, and obtain the sampling timestamp and sampling frequency corresponding to each operational data. A unified time base is determined based on the sampling timestamps of each of the aforementioned operational data, and each of the aforementioned operational data is mapped to the unified time base; The mapped running data is time-aligned according to the sampling frequency of each running data to obtain a multi-sampling rate running data sequence corresponding to the unified time base.
3. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In S1, an anomaly protection region is determined based on the synchronized multi-sampling rate running data sequence, and an anomaly protection state is established including the anomaly protection region and the maximum allowable aggregation granularity, including: The synchronized multi-sampling rate running data sequence is divided according to a preset time window, and the rate of change, fluctuation degree and mutation degree corresponding to each time window are extracted respectively. The abnormal time window is determined based on the rate of change, fluctuation degree and abrupt change degree corresponding to each time window, and adjacent abnormal time windows are merged to obtain the abnormal protection area. The maximum allowable aggregation granularity corresponding to the anomaly protection area is determined based on the duration of the anomaly characteristics within the anomaly protection area and the sampling frequency of the corresponding operating data. The abnormal protection state is established based on the abnormal protection area and the maximum allowable aggregation particle size.
4. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S2, a visual state model is established based on the multi-sampling-rate running data sequence, including: The data objects to be displayed and their corresponding display time ranges are determined based on the multi-sampling rate running data sequence. The corresponding display granularity and display layout are determined based on the data object and the display time range; The visualization state model is generated based on the data object, the display time range, the display granularity, and the display layout. The visualization state model and the anomaly protection state are stored and updated independently, so that changes in the state of the visualization state model do not change the anomaly protection state.
5. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S3, incremental control instructions are generated relative to the visualized state model, including: Obtain the user's input of natural language visualization adjustment instructions, and input the natural language visualization adjustment instructions and the visualization state model into the large language model; The natural language visualization adjustment instructions are analyzed using the large language model to determine the data object to be adjusted, the display time range, the display granularity, or the display layout. The data object to be adjusted, the display time range, the display granularity, or the display layout is mapped to the visualization state model to determine the state parameters to be updated in the visualization state model. Incremental control instructions relative to the visualized state model are generated based on the state parameters to be updated.
6. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S4, matching the incremental control command with the abnormal protection state includes: Extract the data object to be adjusted, the display time range, and the display granularity from the incremental control command, and obtain the abnormal protection area, associated data object, and maximum allowable aggregation granularity from the abnormal protection status; Based on the correspondence between the data object to be adjusted and the associated data object, and the time range relationship between the display time range and the anomaly protection area, the anomaly impact area corresponding to the incremental control command is determined; When the abnormal influence area exists, the display granularity in the incremental control instruction is compared with the maximum allowable aggregation granularity; When the display granularity is greater than the maximum allowable aggregation granularity, it is determined that there is a conflict between the incremental control command and the abnormal protection state; otherwise, it is determined that there is no conflict between the incremental control command and the abnormal protection state.
7. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S4, the incremental control command is constrained and projected, retaining control parameters that do not conflict, and adjusting conflicting control parameters to an executable incremental control command that satisfies the maximum allowable aggregation granularity, including: The incremental control instruction is decomposed according to the state parameters corresponding to the incremental control instruction to obtain multiple control parameters, and conflict control parameters related to the abnormal protection area are determined according to the abnormal protection state. Among the multiple control parameters, retain the non-conflicting control parameters that do not affect the abnormal protection state, and determine the non-conflicting control parameters as the control parameters that are allowed to be executed; For the conflict control parameters, the corresponding display granularity is adjusted to a display granularity that is no greater than the maximum allowable aggregation granularity; The non-conflict control parameters are recombined with the adjusted conflict control parameters to generate an executable incremental control instruction that satisfies the abnormal protection state.
8. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S5, updating the visualized state model based on the executable incremental control instruction includes: Extract the state parameters to be updated and their corresponding values from the executable incremental control instructions, and determine the corresponding state parameters of the state parameters in the visualized state model; The parameter value of the corresponding state parameter is updated to the corresponding parameter value to obtain the updated visualized state parameter; Keep the state parameters in the visualization state model that are not involved by the executable incremental control instruction unchanged, and generate an updated visualization state model based on the updated visualization state parameters.
9. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S5, based on the updated visualization state model and the anomaly protection state, the multi-sampling rate running data sequence is configured with regional granularity, including: The current display time range is determined based on the updated visualization state model, and the multi-sampling rate running data sequence is divided into a normal display area and an anomaly protection area based on the relationship between the current display time range and the time range of the anomaly protection area. The display granularity corresponding to the normal display area is determined based on the updated visualization state model, and the maximum allowable aggregation granularity corresponding to the abnormal protection area is determined based on the abnormal protection state. The normal display area is configured with granularity according to the display granularity, and the abnormal protection area is configured with granularity according to a display granularity no greater than the maximum allowable aggregation granularity, to obtain the regional granularity configuration result.
10. The method for dynamically constructing data visualization based on a large language model according to claim 1, characterized in that, In step S6, local data reconstruction and visualization rendering are performed on the data area where the display state changes to obtain a dynamic visualization result. The visualization state model is then updated based on subsequent natural language visualization adjustment instructions, including: Determine the data areas where the display status has changed based on the visualization status model before and after the update; According to the regional granularity configuration result, extract the multi-sampling rate running data sequence corresponding to the data region where the display state changes, and perform corresponding granularity processing on the multi-sampling rate running data sequence to obtain local visualization data; Based on the updated visualization state model, the local visualization data is reconstructed to generate corresponding local visualization components, while keeping the visualization components corresponding to data areas whose display state has not changed unchanged. The local visualization component is combined with the visualization component corresponding to the data area whose display state has not changed and then rendered to obtain a dynamic visualization result; After obtaining subsequent natural language visualization adjustment instructions, the natural language visualization adjustment instructions are executed based on the visualization state model corresponding to the dynamic visualization results, and the visualization state model is updated.
