A multi-dimensional operation data visualization presentation method based on chart rendering

Abstract

This invention relates to the field of data processing technology, specifically to a method for visualizing multi-dimensional operational data based on chart rendering. The method includes acquiring multi-dimensional operational monitoring data, generating feature tensors, and constructing three-dimensional statistical charts; calculating the fluctuation feature scalars of adjacent windows, calling a spatial connectivity detection algorithm to mark abnormal chart regions, extracting bounding box parameters to calculate observation positions, and driving smooth alignment of the camera viewpoint; collecting video memory and frame rate decay slopes as hardware load parameters, and constructing a visual importance field by combining spatial distance and feature decay rate; calculating the aggregation step size and local fusion weights based on hardware load and the importance field, merging and drawing reduced-order primitives, and intercepting the rendering of other primitives. This invention achieves automatic viewpoint focusing on abnormal regions, adaptive reduction and fusion of non-core primitives based on hardware load, and controls rendering overhead while preserving macroscopic topology, achieving a dynamic balance between chart presentation and hardware load.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method for visualizing and presenting multi-dimensional operational data based on chart rendering. Background Technology

[0002] With the deepening of enterprise digital transformation and the popularization of cloud computing architecture, various business systems, middleware services, and infrastructure foundations generate massive amounts of multi-dimensional operational monitoring data every day. In order to intuitively display the operating status and topology of cross-regional data centers, cluster nodes, and rack arrays, the industry usually uses three-dimensional statistical charts (such as three-dimensional bar chart arrays, scatter plots, voxel heatmaps, etc.) for global visualization monitoring at the front end. However, with the exponential expansion of the scale of connected assets and the continuous increase in data sampling frequency, the number of data primitives that need to be concurrently instantiated in the front-end three-dimensional scene often reaches hundreds of thousands or even millions.

[0003] Most existing 3D chart rendering mechanisms adopt an indiscriminate global full-scale rendering strategy or rely solely on simple static hierarchical detail control. In actual business monitoring scenarios, when local nodes experience data mutations or business alarms, traditional monitoring dashboards cannot intelligently identify anomalies and proactively guide the observation perspective, greatly reducing the efficiency of maintenance personnel in troubleshooting problems. At the same time, full-scale rendering of massive primitives will continuously and indiscriminately consume system computing power, not only occupying a large amount of ineffective resources, but also lacking a flexible scheduling mechanism when facing high-frequency data refreshes. Meanwhile, when massive, multi-source, heterogeneous industrial and operational monitoring data are presented concurrently in three-dimensional space, traditional chart rendering mechanisms lack intelligent guidance capabilities that are linked to the real-time fluctuation of data, as well as adaptive rendering filtering strategies based on the bottlenecks of underlying hardware resources. This makes it easy for the system to cause GPU memory overdraft and severe frame drops and stuttering when encountering high-frequency data surges or local node anomalies. Furthermore, it makes it difficult for users to quickly locate and track core business risks in complex global massive charts. To address this, a multi-dimensional operational data visualization method based on chart rendering is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for visualizing multi-dimensional operational data based on chart rendering, in order to solve the problems raised in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for visualizing multi-dimensional operational data based on chart rendering, including: Acquire multidimensional operational monitoring data, interpolate and align it with a preset rendering cycle, and stitch together spatial coordinates to generate a multidimensional feature tensor, and construct a three-dimensional statistical chart containing data primitives; Calculate the fluctuation feature scalar of the multidimensional feature tensor in adjacent windows, call the spatial connectivity detection algorithm to mark data primitives with adjacent spatial coordinates and fluctuation feature scalars that deviate from the preset benchmark as abnormal chart regions; construct the bounding box of the abnormal chart region and extract the center coordinates and scale, calculate the target observation position in combination with the current field of view, drive the virtual camera viewpoint to smoothly transition to the target observation position through the interpolation algorithm, and align the observation target point with the center coordinates of the bounding box; Real-time acquisition of video memory usage and frame rate decay slope is used as hardware load parameters. The spatial geometric distance from the remaining data primitives of the three-dimensional statistical chart (excluding the abnormal chart area) to the abnormal chart area is calculated, and the decay rate of the fluctuation feature scalar is used to construct a visual importance field. The global aggregation step size is calculated using the hardware load parameters. The visual importance field is then mapped inversely to local fusion weights. Reduced-order primitives are generated using the global aggregation step size and the local fusion weights. Instantiation instructions are called to merge and draw the reduced-order primitives, and the remaining data primitives are intercepted and rendered.

[0006] Preferably, based on the business level, asset logical grouping, and physical region label of the multidimensional operation monitoring data, the corresponding basic grid coordinates are allocated in the virtual three-dimensional space through a three-dimensional grid layout algorithm, and discrete offsets are introduced within each group in combination with the primitive arrangement index to calculate the spatial coordinates of each data primitive in the virtual three-dimensional space. The construction process of the three-dimensional statistical chart is as follows: the acquired multi-dimensional operation monitoring data is interpolated and aligned according to the preset rendering period to generate a discrete timestamp continuous feature sequence; the spatial coordinates and the corresponding feature sequence are matrix-concatenated and mapped to the corresponding vertex cache object to generate the multi-dimensional feature tensor; the multi-dimensional feature tensor is batch-instantiated into the corresponding data primitives to construct the three-dimensional statistical chart; the three-dimensional statistical chart can be any one of a three-dimensional bar chart array, a three-dimensional scatter plot, and a three-dimensional voxel heatmap.

[0007] Preferably, the business layer includes an infrastructure base layer, a middleware service layer, and a front-end business application layer; the asset logical grouping includes a computing resource pool cluster, a distributed storage cluster, and a network load balancing gateway group; the physical region label includes cross-domain data center node identifiers, data center availability zone codes, and rack physical array sequences.

[0008] Preferably, the execution process of the spatial connected component detection algorithm is as follows: the data primitives whose fluctuation feature scalars deviate from the preset benchmark are selected as abnormal seed primitives; starting from the abnormal seed primitives, the algorithm iteratively searches outward in the virtual three-dimensional space, merging the spatially adjacent data primitives whose fluctuation feature scalars also deviate from the preset benchmark into the same candidate connected component; after the iteration terminates, candidate connected components with a total number of primitives less than a preset size are removed, and the remaining candidate connected components are taken as the abnormal graph region.

[0009] Preferably, the calculation process of the fluctuation feature scalar involves extracting the first dynamic feature subsequence of the multidimensional feature tensor in the current time window and the second dynamic feature subsequence in the adjacent historical time window; and calculating the mean vector of the first dynamic feature subsequence and the second dynamic feature subsequence in the time dimension, respectively. Based on the deviations of the first dynamic feature subsequence from the mean vector and the deviations of the second dynamic feature subsequence from the mean vector, the expected value of their outer product is calculated to obtain the covariance matrix; the largest eigenvalue of the covariance matrix is ​​extracted and used as the fluctuation feature scalar.

[0010] Preferably, the process of aligning the observation target point with the center coordinates of the bounding box is as follows: constructing the minimum directed bounding box of the abnormal chart region, extracting its center coordinates as the center coordinates of the bounding box, and using the diagonal length of the minimum directed bounding box as the scale; and calculating the target observation distance by combining the current field of view and the preset anti-clipping coefficient. The target observation position is calculated by translating the bounding box center coordinates along the direction pointing from the bounding box center coordinates to the current virtual camera viewpoint by the target observation distance. A cubic Hermite spline curve with the current viewpoint as the starting point and the target observation position as the ending point, and with both the first and last tangent vectors being zero, is constructed as the translation trajectory. An interpolation step size is generated through a nonlinear easing function to drive the virtual camera to update its position along the translation trajectory. At the same time, a quaternion spherical linear interpolation algorithm is called to smoothly rotate the camera's current pose to the target pose aligned with the bounding box center coordinates according to the interpolation step size.

[0011] Preferably, the construction process of the visual importance field is as follows: The frustum range is extracted based on the current virtual camera's pose; the shortest Euclidean distance from the centroid of the remaining data primitives falling within the frustum to the bounding box of the abnormal chart region is calculated as the spatial geometric distance; the absolute value of the scalar difference in fluctuation characteristics between the remaining data primitives and the data primitives at the boundary of the abnormal chart region is calculated, and the ratio of this absolute value to the spatial geometric distance is taken as the attenuation rate; the distance weight obtained by mapping the spatial geometric distance through the attenuation function is weighted and summed with the feature weight obtained by inverse mapping the normalized attenuation rate to calculate the visual importance value of each of the remaining data primitives; the spatial distribution set of all the visual importance values ​​constitutes the visual importance field.

[0012] Preferably, the process of generating reduced-order primitives through the global aggregation step size and the local fusion weights involves fusing the video memory usage and the frame rate decay slope to calculate and map them into a spatial grid size, which is used as the global aggregation step size; and mapping the visual importance values ​​in the visual importance field inversely to the local fusion weights corresponding to the remaining data primitives. Divide the three-dimensional aggregation grid according to the global aggregation step size, and classify the remaining data primitives falling into the same three-dimensional aggregation grid into a primitive cluster to be fused; use the local fusion weight as the weighting coefficient to perform weighted fusion on the spatial coordinates, geometric scale and color attributes of each of the remaining data primitives in the primitive cluster to be fused, and generate the reduced-order primitive.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention calculates the fluctuation feature scalar of a multidimensional feature tensor in adjacent windows, calls a spatial connectivity detection algorithm to mark abnormal chart regions, and combines bounding box center coordinates with an interpolation algorithm to drive a smooth transition of the virtual camera viewpoint, achieving automatic alignment of the viewpoint to the target observation position. This mechanism enables the front-end monitoring perspective to adaptively migrate according to the fluctuation state of the data itself, changing the conventional interactive method of manually navigating in a global 3D view to find abnormal data nodes, and providing a view focusing function directly driven by underlying data features.

[0014] 2. This invention uses real-time acquisition of video memory usage and frame rate decay slope as hardware load parameters. It constructs a visual importance field by combining the spatial geometric distance from primitives to abnormal regions with the fluctuation decay rate. This field then drives the remaining data primitives within the view frustum to perform spatial centroid calculation and feature mean fusion to generate reduced-order primitives. This mechanism combines the system's real-time hardware operating status with the weight distribution of the visible area, replacing multiple discrete primitives in non-core areas with aggregated entity displays. It provides an execution logic that dynamically adjusts and reduces the number of primitive instances when the system faces high-frequency data access.

[0015] 3. This invention drives the generation of reduced-order entities from adjacent primitives through global aggregation step size and local fusion weights, calls instantiation instructions for merging and rendering, and, in conjunction with rendering interception of other data primitives outside the view frustum, establishes a pipeline flow that allocates rendering resources according to the current field of view and importance. This strategy directly reduces the total number of primitives that the graphics card needs to independently render by intercepting primitives in non-visible areas and reducing and merging secondary primitives within the visible range. At the same time, it maintains the basic topological distribution and macroscopic feature parameters of the original data in the virtual 3D space by using feature mean fusion. Attached Figure Description

[0016] Figure 1 A schematic diagram of a method for visualizing multi-dimensional operational data based on chart rendering, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the scalar calculation process for fluctuation characteristics provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the virtual camera viewpoint smooth alignment process provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some 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.

[0018] Please see Figure 1 This invention provides a method for visualizing multi-dimensional operational data based on chart rendering, the technical solution of which is as follows: A method for visualizing multi-dimensional operational data based on chart rendering, including: Acquire multidimensional operational monitoring data, interpolate and align it with a preset rendering cycle, and stitch together spatial coordinates to generate a multidimensional feature tensor, and construct a three-dimensional statistical chart containing data primitives; Calculate the fluctuation feature scalar of the multidimensional feature tensor in adjacent windows, call the spatial connectivity detection algorithm to mark data primitives with adjacent spatial coordinates and fluctuation feature scalars that deviate from the preset benchmark as abnormal chart regions; construct the bounding box of the abnormal chart region and extract the center coordinates and scale, calculate the target observation position in combination with the current field of view, drive the virtual camera viewpoint to smoothly transition to the target observation position through the interpolation algorithm, and align the observation target point with the center coordinates of the bounding box; Real-time acquisition of video memory usage and frame rate decay slope is used as hardware load parameters. The spatial geometric distance from the remaining data primitives of the three-dimensional statistical chart (excluding the abnormal chart area) to the abnormal chart area is calculated, and the decay rate of the fluctuation feature scalar is used to construct a visual importance field. The global aggregation step size is calculated using the hardware load parameters. The visual importance field is then mapped inversely to local fusion weights. Reduced-order primitives are generated using the global aggregation step size and the local fusion weights. Instantiation instructions are called to merge and draw the reduced-order primitives, and the remaining data primitives are intercepted and rendered.

[0019] Example 1 Acquire multi-dimensional operational monitoring data, interpolate and align it with a preset rendering cycle, and stitch together spatial coordinates to generate a multi-dimensional feature tensor, and construct a three-dimensional statistical chart; Based on the business level, asset logical grouping and physical region label of the multidimensional operation monitoring data, the corresponding basic grid coordinates are allocated in the virtual three-dimensional space through the three-dimensional grid layout algorithm, and discrete offsets are introduced in each group in combination with the primitive arrangement index to calculate the spatial coordinates of each data primitive in the virtual three-dimensional space. The construction process of the three-dimensional statistical chart is as follows: the acquired multi-dimensional operation monitoring data is interpolated and aligned according to the preset rendering period to generate a discrete timestamp continuous feature sequence; the spatial coordinates and the corresponding feature sequence are matrix-concatenated and mapped to the corresponding vertex cache object to generate the multi-dimensional feature tensor; the multi-dimensional feature tensor is batch-instantiated into the corresponding data primitives to construct the three-dimensional statistical chart; the three-dimensional statistical chart can be any one of a three-dimensional bar chart array, a three-dimensional scatter plot, and a three-dimensional voxel heatmap.

[0020] The business layers include an infrastructure base layer, a middleware service layer, and a front-end business application layer; the asset logical groups include a computing resource pool cluster, a distributed storage cluster, and a network load balancing gateway group; the physical region labels include cross-domain data center node identifiers, data center availability zone codes, and rack physical array sequences.

[0021] Specifically, multi-dimensional operation monitoring data is acquired, and the business level, asset logical grouping, and physical region label carried in the multi-dimensional operation monitoring data are extracted; wherein, the business level includes the infrastructure base layer, the middleware service layer, and the front-end business application layer; the asset logical grouping includes the computing resource pool cluster, the distributed storage cluster, and the network load balancing gateway group; the physical region label includes the cross-domain data center node identifier, the data center availability zone code, and the rack physical array sequence.

[0022] The virtual 3D space is established as a 3D coordinate system comprising a horizontal axis, a depth axis, and a height axis. The physical region labels are mapped sequentially according to the hierarchy of cross-domain data center node identifiers, data center availability zone codes, and rack physical array sequences, as the basic coordinates of the horizontal axis. The asset logical groups are mapped sequentially according to the data flow order of network load balancing gateway groups, computing resource pool clusters, and distributed storage clusters, as the basic coordinates of the depth axis. The business layers are mapped sequentially according to the bottom-up architectural order of the infrastructure base layer, middleware service layer, and front-end business application layer, as the basic coordinates of the height axis.

[0023] The total number of multi-dimensional operational monitoring data within each group is counted. The maximum number of columns allowed in the current group is obtained by dividing the total width of the bounding box grid boundary by the physical width of a single primitive. The row offset step is obtained by dividing the primitive arrangement index of each data primitive within the group by the maximum number of columns and rounding down. The column offset step is obtained by taking the primitive arrangement index modulo the maximum number of columns. The row offset step and the column offset step are multiplied by the obtained safety isolation distance between adjacent primitives to obtain the discrete offsets in the horizontal and vertical directions. The discrete offset in the height direction is set to zero to construct a three-dimensional vector of discrete offsets.

[0024] The three-dimensional vector of the allocated basic grid coordinates is added to the three-dimensional vector of the calculated discrete offset to complete the coordinate transformation and output the unique spatial coordinates of each data primitive in the virtual three-dimensional space.

[0025] The construction process of the three-dimensional statistical chart includes a time alignment and spatial instantiation process. This involves reading the physical refresh rate of the current display by calling the system environment interface of the display device, calculating the reciprocal of the physical refresh rate to generate the preset rendering cycle, and using a linear interpolation algorithm to interpolate and align the data states of the original irregular time sampling points in the multi-dimensional operation monitoring data according to the preset rendering cycle with a fixed step size in the time dimension, thereby generating a feature sequence with discrete timestamps.

[0026] A vertex data structure containing spatial location attributes, time series attributes, and business status attributes is constructed in the system main memory. The calculated spatial coordinates are filled into the spatial location attributes. The values ​​of the interpolated and aligned feature sequences are extracted and filled into the time series attributes and business status attributes. The data primitives are then concatenated in the index order to form a feature matrix.

[0027] The process of mapping to the corresponding vertex cache object to generate a multidimensional feature tensor is as follows: the feature matrix is ​​transferred from the system main memory and written to the contiguous memory space of the vertex cache object of the graphics processor through the memory copy instruction of the graphics rendering programming interface. All the graph data in the current scene are integrated in the video memory to generate a three-dimensional array structure. Its first dimension is the total number of graph primitives, the second dimension is the timestamp length, and the third dimension is the feature dimension of a single graph primitive. This three-dimensional array structure is the multidimensional feature tensor.

[0028] The hardware instantiation drawing instruction in the graphics rendering pipeline is invoked in a single drawing call command, passing in the total number of instances parameter in the multidimensional feature tensor. Based on the spatial coordinates within the multidimensional feature tensor, the basic geometric mesh is batch cloned in the virtual three-dimensional space to generate corresponding three-dimensional data primitives and construct a three-dimensional statistical chart. According to the chart type switching instruction input by the user interface, the shape of the basic geometric mesh is controlled so that the generated three-dimensional statistical chart is presented as any one of a three-dimensional bar chart array, a three-dimensional scatter plot, and a three-dimensional voxel heat map.

[0029] By accurately mapping physical regions, asset logic, and business levels to a three-dimensional coordinate system and introducing a discrete offset algorithm, the spatial overlap and visual occlusion of massive primitives are effectively eliminated, achieving intuitive and three-dimensional cross-level fault location. At the same time, by using hardware refresh cycles for temporal interpolation alignment and directly mapping the spliced ​​multi-dimensional feature tensors to vertex cache objects to perform hardware batch instantiation rendering, not only is smooth continuity and real-time rendering ensured during the evolution of massive dynamic monitoring data, but the hierarchical division rules are also highly consistent with the real operation and maintenance scenarios of modern cloud-native and cross-domain multi-data center systems.

[0030] Calculate the fluctuation feature scalar of the multidimensional feature tensor in adjacent windows, and call the spatial connected component detection algorithm to mark the data primitives with adjacent coordinates and fluctuation feature scalars that deviate from the baseline as abnormal graph regions; The execution process of the spatial connected component detection algorithm is as follows: the data primitives whose fluctuation feature scalars deviate from the preset benchmark are selected as abnormal seed primitives; starting from the abnormal seed primitives, the algorithm iteratively searches outward in the virtual three-dimensional space, merging the spatially adjacent data primitives whose fluctuation feature scalars also deviate from the preset benchmark into the same candidate connected component; after the iteration terminates, candidate connected components with a total number of primitives less than the preset size are removed, and the remaining candidate connected components are taken as the abnormal graph region.

[0031] The calculation process of the fluctuation feature scalar is to extract the first dynamic feature subsequence of the multidimensional feature tensor in the current time window and the second dynamic feature subsequence in the adjacent historical time window; and to calculate the mean vector of the first dynamic feature subsequence and the second dynamic feature subsequence in the time dimension respectively. Based on the deviations of the first dynamic feature subsequence from the mean vector and the deviations of the second dynamic feature subsequence from the mean vector, the expected value of their outer product is calculated to obtain the covariance matrix; the largest eigenvalue of the covariance matrix is ​​extracted and used as the fluctuation feature scalar. Specifically, the continuous timestamp sequence of the multidimensional feature tensor is obtained, and the continuous timestamp sequence is divided into the current time window and the adjacent historical time window of equal length according to a fixed time step span; according to the time index of the current time window, the first dynamic feature subsequence of the multidimensional feature tensor in the current time window is extracted; according to the time index of the adjacent historical time window, the second dynamic feature subsequence of the multidimensional feature tensor in the adjacent historical time window is extracted.

[0032] See Figure 2 The mean vectors of the first dynamic feature subsequence and the second dynamic feature subsequence in the time dimension are calculated by summing each component in the time dimension and dividing by the sequence length of the first dynamic feature subsequence.

[0033] Subtracting the corresponding mean vector from each feature vector in the first dynamic feature subsequence yields a first deviation sequence; subtracting the corresponding mean vector from each feature vector in the second dynamic feature subsequence yields a second deviation sequence; let the first deviation sequence be... The second deviation sequence is The first deviation sequence and the transpose of the second deviation sequence are multiplied by a matrix to calculate their outer product. The outer product result is then divided by the sequence length to obtain the expected value, and the covariance matrix is ​​calculated using the following formula: in, Let the covariance matrix be represented. Indicates the length of the sequence. Indicates that the first deviation sequence is in the 1st... A column vector of timestamps, Indicates the second deviation sequence in the 1st... A row vector of timestamps.

[0034] The covariance matrix is ​​decomposed into eigenvalues ​​using the Jacobi iteration method. The covariance matrix is ​​transformed into a diagonal matrix through orthogonal transformation. The elements on the diagonal of the diagonal matrix are extracted as the eigenvalue set. The eigenvalue with the largest value in the eigenvalue set is selected as the fluctuation feature scalar of the multidimensional feature tensor in the adjacent window.

[0035] The fluctuation feature scalar of all data primitives in the current virtual 3D space is collected through traversal. The mathematical mean and standard deviation of the global fluctuation feature scalar are calculated. The result of adding three times the standard deviation to the mathematical mean is used as a preset benchmark. Data primitives whose fluctuation feature scalar exceeds the preset benchmark are selected as abnormal seed primitives.

[0036] A mesh adjacency matrix is ​​established for a virtual 3D space. Spatial adjacent data primitives of any data primitive are defined as the twenty-six neighboring data primitives that share vertices, mesh edges, or mesh faces in the virtual 3D space. Starting from the abnormal seed primitive, a first-in-first-out queue is established, and a breadth-first search algorithm is called to iteratively search outward in the virtual 3D space, sequentially detecting the twenty-six neighboring data primitives. Spatial adjacent data primitives whose fluctuation feature scalar also exceeds the preset benchmark are pushed into the first-in-first-out queue and merged into the same candidate connected component. At the same time, the merged data primitives are marked as visited.

[0037] The outward iterative search is performed repeatedly until the first-in-first-out queue is empty and all adjacent and qualified data primitives have been traversed. The iteration terminates, and the construction of the candidate connected component is completed. The total number of data primitives instantiated in the current virtual 3D space is counted, and the result is multiplied by the total number of data primitives according to the scale benchmark ratio. The calculated value is rounded down to obtain the preset scale. The scale benchmark ratio is the ratio between the mathematical mean of the number of primitives contained in the abnormal connected component in each confirmed abnormal event recorded in the historical preset time period and the mathematical mean of the total number of data primitives instantiated in the virtual 3D space at the corresponding time.

[0038] Traverse all the constructed candidate connected regions, count the total number of graph elements contained in each candidate connected region, compare the total number of graph elements with the preset size, remove candidate connected regions whose total number of graph elements is less than the preset size, and output the remaining candidate connected regions as the abnormal graph areas.

[0039] By calculating the maximum eigenvalue of the covariance matrix of the feature sequences of adjacent time windows as the fluctuation feature scalar, this invention realizes the principal component extraction of multidimensional data fluctuations, which can effectively capture the joint mutation signal of multidimensional features. At the same time, combined with the three-dimensional spatial connected domain detection algorithm, the spatial clustering features and scale thresholds of abnormal primitives are used for screening, which logically eliminates the visual noise caused by isolated sporadic spikes. Thus, an anomaly filtering mechanism based on spatiotemporal coupling is constructed, which can stably locate and present the real business fluctuation area with cluster characteristics in a complex multidimensional chart rendering environment.

[0040] See Figure 3Extract the center coordinates and scale of the bounding box of the abnormal chart region, calculate the target observation position in combination with the current field of view, drive the virtual camera viewpoint to smoothly transition to the observation position through an interpolation algorithm, and align the observation target point with the center coordinates; The process of aligning the observation target point with the center coordinates of the bounding box is as follows: construct the minimum directed bounding box of the abnormal chart region, extract its center coordinates as the center coordinates of the bounding box, and use the diagonal length of the minimum directed bounding box as the scale; combine the current field of view angle with the preset anti-clipping coefficient to calculate the target observation distance; The target observation position is calculated by translating the bounding box center coordinates along the direction pointing from the bounding box center coordinates to the current virtual camera viewpoint by the target observation distance. A cubic Hermite spline curve with the current viewpoint as the starting point and the target observation position as the ending point, and with both the first and last tangent vectors being zero, is constructed as the translation trajectory. An interpolation step size is generated through a nonlinear easing function to drive the virtual camera to update its position along the translation trajectory. At the same time, a quaternion spherical linear interpolation algorithm is called to smoothly rotate the camera's current pose to the target pose aligned with the bounding box center coordinates according to the interpolation step size.

[0041] Specifically, the vertex spatial coordinate set of all data primitives within the abnormal chart area is extracted, the three-dimensional covariance matrix of the vertex spatial coordinate set is calculated, and the Jacobi iteration method is used to obtain the eigenvalues ​​and corresponding eigenvectors of the three-dimensional covariance matrix. The three mutually orthogonal eigenvectors are used as the three principal axes of the local coordinate system. All vertices in the vertex spatial coordinate set are projected onto the three principal axes, and the maximum and minimum projection values ​​are obtained in each principal axis direction. Six boundary planes are constructed based on the maximum and minimum projection values ​​in the three principal axes, and the intersection of the six boundary planes generates the minimum directed bounding box of the abnormal chart area.

[0042] Calculate the mean of the maximum and minimum projection values ​​along the three principal axes, and combine the three mean values ​​into three-dimensional coordinates as the center coordinates; calculate the difference between the maximum and minimum projection values ​​along the three principal axes, and take the square root of the sum of the squares of the three differences to obtain the diagonal length of the minimum directed bounding box, and use the diagonal length as the scale parameter.

[0043] The near clipping plane distance of the current virtual camera is read, divided by the scale parameter, and a fixed safety constant is added to the quotient to obtain the anti-clipping coefficient. The fixed safety constant is a preset empirical value, and its value range can be, for example, between 0.05 and 0.5, preferably 0.1 in this embodiment. The current field of view of the virtual camera in the vertical direction is read, half of the scale parameter is used as the dividend, and the tangent of half of the current field of view is used as the divisor for division. The result is multiplied by the anti-clipping coefficient to calculate the target observation distance. The calculation formula is as follows: in, Indicates the target observation distance, This represents the scale parameter. Indicates the current field of view angle. This represents the anti-cutting coefficient.

[0044] Obtain the three-dimensional vector pointing from the current virtual camera viewpoint to the center coordinates, normalize the three-dimensional vector to obtain the unit direction vector, multiply the unit direction vector by the target observation distance to obtain the reverse translation vector, subtract the reverse translation vector from the center coordinates, and calculate the target observation position.

[0045] Taking the current virtual camera viewpoint as the starting point and the target observation position as the ending point, and setting both the starting point tangent vector and the ending point tangent vector as three-dimensional zero vectors, a cubic Hermitian spline curve is constructed as the translation trajectory, the mathematical expression of which is: in, Normalized time parameter The spatial coordinates below, Indicates the starting point, Indicates the endpoint; The accumulated elapsed time from the start of interpolation is collected by the system clock. This elapsed time is divided by a fixed total animation duration to obtain a linear time variable. A smooth step function is then used as a nonlinear easing function to map the linear time variable to generate the interpolation step size. The mapping formula is as follows: in, This indicates the interpolation step size. The linear time variable is represented by the interpolation step size. The normalized time parameter is replaced by substituting the interpolation step size into the mathematical expression of the cubic Hermite spline curve. The spatial coordinates corresponding to the current frame are calculated, and the virtual camera is driven to update its position along the translation trajectory.

[0046] Calculate the target orientation vector pointing from the virtual camera position to the center coordinates in the current frame. Calculate the target rotation matrix based on the target orientation vector and the global up vector, and convert the target rotation matrix into a target quaternion. Extract the camera's current pose quaternion, and call the quaternion spherical linear interpolation algorithm to calculate the intermediate pose quaternion according to the interpolation step size. The interpolation formula is: in, This represents the intermediate attitude quaternion. This represents the current attitude quaternion. Represents the target quaternion, This indicates the interpolation step size. The intermediate pose quaternion represents the half-angle between the current pose quaternion and the target quaternion; the intermediate pose quaternion is assigned to the virtual camera so that its current pose is smoothly rotated to the target pose aligned with the center coordinates.

[0047] By extracting the diagonal of the bounding box as a scale parameter and dynamically calculating the target observation distance in combination with the current field of view and anti-cropping coefficient, this invention can adaptively include abnormal chart regions of various shapes completely and in the optimal proportion within the visible range of the virtual camera, effectively avoiding the problems of image cropping or target overflow caused by a fixed observation distance. At the same time, by using cubic Hermite spline curves, nonlinear easing functions, and quaternion spherical linear interpolation algorithms to achieve synchronous dynamic interpolation of camera displacement and attitude, the continuity of the viewpoint switching process is guaranteed at the mathematical model level.

[0048] Real-time acquisition of video memory usage and frame rate decay slope is used as hardware load parameters. The spatial geometric distance from the remaining data primitives of the three-dimensional statistical chart (excluding the abnormal chart area) to the abnormal chart area is calculated, and the decay rate of the fluctuation feature scalar is used to construct a visual importance field. The construction process of the visual importance field is as follows: The frustum range is extracted based on the current virtual camera's pose; the shortest Euclidean distance from the centroid of the remaining data primitives falling within the frustum to the bounding box of the anomalous chart region is calculated as the spatial geometric distance; the absolute value of the scalar difference in fluctuation characteristics between the remaining data primitives and the data primitives at the boundary of the anomalous chart region is calculated, and the ratio of this absolute value to the spatial geometric distance is taken as the attenuation rate; the distance weights obtained by mapping the spatial geometric distance through the attenuation function are weighted and summed with the feature weights obtained by inverse mapping the normalized attenuation rate to calculate the visual importance value of each of the remaining data primitives; the spatial distribution set of all the visual importance values ​​constitutes the visual importance field.

[0049] Specifically, the perspective projection matrix and view matrix of the current virtual camera are extracted, and the two are multiplied to extract the equations of the six clipping planes of the view frustum; all data primitives in the 3D statistical chart that are not marked as abnormal chart areas are traversed, and the spatial directed distance from the center coordinates of the bounding sphere of each data primitive to the six clipping planes is calculated; if the spatial directed distance is greater than the negative radius of the bounding sphere, the corresponding data primitive is determined to be located within the view frustum, and it is filtered as the remaining data primitives within the view frustum.

[0050] Extract the three-dimensional centroid coordinates of the remaining data primitives, and simultaneously extract the minimum and maximum boundary values ​​of the bounding box of the abnormal chart region along the three orthogonal coordinate axes; restrict the components of the three-dimensional centroid coordinates on the three orthogonal coordinate axes to the corresponding minimum and maximum boundary values ​​respectively. If the component is less than the minimum boundary value, the minimum boundary value is taken; if it is greater than the maximum boundary value, the maximum boundary value is taken, thus obtaining the coordinates of the nearest point on the surface of the bounding box; calculate the straight line length between the three-dimensional centroid coordinates and the coordinates of the nearest point as the spatial geometric distance.

[0051] Traverse the outermost geometric contour of the abnormal chart region and extract all data primitives at the contour edge to form a boundary primitive set; calculate the distance from the centroid of the remaining data primitives to the centroid of each primitive in the boundary primitive set, and select the primitive with the smallest distance value as the target boundary primitive; read the fluctuation characteristic scalar of the remaining data primitives and the fluctuation characteristic scalar of the target boundary primitive, calculate the absolute value of the difference between the two, and perform a division operation with the absolute value as the dividend and the spatial geometric distance as the divisor to obtain the attenuation rate.

[0052] Read the maximum visible depth value of the current rendering pipeline and take its reciprocal as the scene attenuation constant; using the natural constant as the base, take the negative of the product of the scene attenuation constant and the spatial geometric distance as the exponent for power operation, and use it as the attenuation function to map the spatial geometric distance to calculate the distance weight, the calculation formula of which is: in, This represents the distance weight. Represents the natural constant. This represents the scene attenuation constant. This represents the spatial geometric distance.

[0053] Collect the attenuation rates of all remaining data primitives within the view frustum, and filter out the global maximum attenuation rate and the global minimum attenuation rate; subtract the global minimum attenuation rate from the current attenuation rate of the remaining data primitives to obtain a first difference, and subtract the global minimum attenuation rate from the global maximum attenuation rate to obtain a second difference; divide the first difference by the second difference to obtain the normalized attenuation rate; subtract the normalized attenuation rate from the number one to achieve inverse mapping, and calculate the feature weights.

[0054] Extract the percentage of video memory usage acquired in real time and use it as the spatial distance tendency coefficient; subtract the spatial distance tendency coefficient from the number 1 to obtain the feature fluctuation tendency coefficient; multiply the distance weight by the spatial distance tendency coefficient, multiply the feature weight by the feature fluctuation tendency coefficient, add the products of the two and perform a weighted sum to calculate the visual importance value of each remaining data primitive. The calculation formula is as follows: in, This represents the visual importance value. This represents the spatial distance tendency coefficient. This represents the distance weight. This indicates the current attenuation rate. This represents the global minimum attenuation rate. The global maximum attenuation rate is represented; the visual importance value corresponding to all the remaining data primitives within the view frustum is matrix-bound and stored with its three-dimensional spatial coordinates, and the spatial distribution set of all the visual importance values ​​constitutes the visual importance field; the motivation for introducing the video memory usage percentage as the spatial distance tendency coefficient is that when the underlying hardware load surges, increasing the weight of the spatial distance tendency coefficient can make the visual importance assessment of non-core area primitives more strongly controlled by the exponential attenuation of spatial geometric distance.

[0055] By combining the frustum culling algorithm with the calculation of the shortest Euclidean distance, this invention can filter and quantify the spatial positional relationship between peripheral primitives and abnormal chart regions within the frustum. Utilizing the exponential decay function and normalized inverse mapping mechanism, spatial geometric distance and feature fluctuation decay rate are transformed into a unified weight index. This index is then weighted and summed in conjunction with system memory usage parameters to construct a visual importance field. This field objectively reflects the spatial proximity and feature correlation of each data primitive relative to the abnormal core, providing a definite mathematical basis for subsequent adaptive fusion and reduced-order rendering of primitives.

[0056] The global aggregation step size is calculated using the hardware load parameters. The visual importance field is then mapped inversely to local fusion weights. Reduced-order primitives are generated using the global aggregation step size and the local fusion weights. Instantiation instructions are called to merge and draw the reduced-order primitives. The remaining data primitives are intercepted and rendered.

[0057] The process of generating reduced-order primitives by the global aggregation step size and the local fusion weights involves fusing the video memory usage and the frame rate decay slope to calculate and map them into a spatial grid size, which is used as the global aggregation step size; and mapping the visual importance values ​​in the visual importance field inversely to the local fusion weights corresponding to the other data primitives. The three-dimensional aggregation grid is divided according to the global aggregation step size. The remaining data primitives falling into the same three-dimensional aggregation grid are classified into a primitive cluster to be fused. The spatial coordinates, geometric scale and color attributes of each remaining data primitive in the primitive cluster to be fused are weighted and fused using the local fusion weight as the weighting coefficient to generate the reduced-order primitive. Specifically, all visual importance values ​​in the visual importance field are traversed, and the global maximum importance value and global minimum importance value are extracted. The visual importance values ​​corresponding to the remaining data primitives are subtracted from the global maximum importance value. The difference obtained from the subtraction is used as the dividend, and the difference between the global maximum importance value and the global minimum importance value is used as the divisor for division. This process completes the inversion and normalization mapping, and the local fusion weights corresponding to the remaining data primitives are calculated. The mapping formula is as follows: in, Indicates the local fusion weights, This represents the visual importance value of the remaining data primitives. This represents the global maximum importance value. This represents the global minimum importance value.

[0058] The system calls the underlying performance monitoring interface to read the number of bytes of video memory currently occupied by the rendering process in real time and divide it by the total number of bytes of physical video memory to obtain the video memory utilization rate. Simultaneously, it records the rendering time of multiple consecutive frames and calculates the rate of change in the number of frames rendered per second as the frame rate decay slope. It extracts the basic grid side length allocated during the initialization of the 3D statistical chart as the baseline size, multiplies the video memory utilization rate and the frame rate decay slope by their respective fixed sensitivity coefficients, sums them, and adds 1 to construct a dynamic penalty multiplier. It multiplies the baseline size by the dynamic penalty multiplier, calculates and maps it to a spatial grid size as the global aggregation step size. The fixed sensitivity coefficients are obtained by running rendering stress tests with different data scales, recording the peak video memory utilization and peak frame rate decay slope when the system experiences frame drops or stuttering. It calculates the ratio of 1 to the peak video memory utilization and the ratio of 1 to the peak frame rate decay slope, and uses these two ratios directly as the fixed sensitivity coefficients for the corresponding video memory utilization and frame rate decay slope. Extract the three-dimensional spatial coordinates of each of the remaining data primitives, divide the three orthogonal axis components of the three-dimensional spatial coordinates by the global aggregation step size, and round down the division results to generate three integers as the grid index of the three-dimensional aggregation grid. Divide the three-dimensional aggregation grid according to the global aggregation step size, and group the remaining data primitives that have generated the same grid index, i.e., fall into the same three-dimensional aggregation grid, into a primitive cluster to be merged. Within each constructed 3D aggregated mesh, using the local fusion weight as a weighting coefficient, the spatial coordinates, geometric scale, and color attributes of each of the remaining data primitives within the primitive cluster to be fused are weighted and fused: the local fusion weights of all primitives within the primitive cluster to be fused are summed to obtain the total weight within the group; the local fusion weights of each primitive within the group are divided by the total weight within the group to obtain the relative weight within the group; the spatial coordinates of each primitive within the group are multiplied by the corresponding relative weight within the group and then vector-summed, and the centroid coordinates of the current primitive cluster to be fused are calculated using this weighted summation. The calculation formula is as follows: in, Indicates the centroid coordinates, This indicates the total number of graphic elements within the current group. Indicates the number within the group Spatial coordinates of each primitive Indicates the number within the group The local fusion weights of each primitive are calculated using the same weighting logic. The geometric scale parameter and color attribute parameter of each primitive within the primitive cluster to be fused are multiplied by their corresponding relative weights within the group and summed. This weighted average is then used to calculate the mean geometric scale and mean color attribute of the current primitive cluster to be fused. The calculated centroid coordinates, mean geometric scale, and mean color attribute are then matrix-concatenated and recombined to construct new primitive vertex data. An instantiation rendering command is then issued to generate the reduced-order primitives. In the rendering pipeline of the current rendering frame, the hardware instantiation command of the underlying graphics application programming interface is used to merge and render the reduced-order primitives, directly replacing all independent data primitives within the primitive cluster to be fused corresponding to the original 3D aggregated mesh for display.

[0059] Extract the view projection matrix under the current virtual camera pose, extract the row vectors of the view projection matrix, and parse the plane equations of the six clipping planes of the view frustum by performing addition, subtraction and combination operations on the row vectors; for all other data primitives in the current 3D space, extract the coordinates of the center of the bounding sphere and the radius of the bounding sphere of each primitive.

[0060] For each clipping plane, extract the unit normal vector and constant offset value of the clipping plane; multiply the three spatial components of the bounding sphere center coordinates by the corresponding spatial components of the unit normal vector and sum them up, then add the constant offset value to the summation result to calculate the directed distance from the bounding sphere center to the current clipping plane; sequentially judge the six calculated directed distances, and if any one of the directed distances is a negative value less than the radius of the bounding sphere, then determine that the corresponding data primitive is located outside the view frustum.

[0061] For other data primitives determined to be located outside the view frustum, their visibility status is marked as false. When constructing the drawing instruction queue of the current rendering frame, the memory address of primitive with a visibility status of false is refused to be read, so as to cut off its data transmission channel to the graphics processor's video memory. This completes the interception and rendering of other data primitives outside the view frustum, avoiding redundant graphics computation overhead.

[0062] For all the generated reduced-order primitives and data primitives within the abnormal chart region, the vertex coordinates and topological indexes of the standard basic geometric mesh corresponding to the 3D statistical chart are extracted and passed into the vertex cache object of the graphics processor as shared geometric basic data. Simultaneously, the centroid of the spatial coordinates of the reduced-order primitives is extracted and used as a translation parameter in the fourth-order identity matrix to construct an instance transformation matrix. The geometric scale mean and color attribute mean of the reduced-order primitives are extracted and converted into vertex color vectors and scaling parameters through a pre-established color mapping lookup table. The construction and conversion process of the color mapping lookup table is as follows: During the initialization phase, the standard color space is linearly interpolated with equal steps within a preset numerical range to generate a one-dimensional numerical matrix as the color lookup table, where each index address of the one-dimensional numerical matrix corresponds to storing a four-dimensional vertex color RGBA vector. During conversion, the calculated color attribute mean is used as the index address to perform reference addressing or bilinear interpolation sampling on the one-dimensional numerical matrix, and the corresponding RGBA vector output is directly read as the vertex color vector. The calculated geometric scale mean is used as the scaling parameter of the reduced-order primitive. The constructed instance transformation matrix, vertex color vector, and scaling parameters are concatenated in memory according to the primitive sequence, packaged to generate an instance attribute array, and the instance attribute array is dynamically pushed into the instance cache object of the graphics processor. The vertex cache object and the instance cache object are bound in the graphics pipeline at the same time. The hardware instantiation instruction of the underlying graphics application interface is called, and the batch merge drawing task is issued through a single drawing call. According to the parameters in the instance attribute array, all degraded primitives and abnormal area primitives are batch rendered on the screen to complete the visualization of multi-dimensional operational data.

[0063] This invention extracts the fluctuation characteristics of multidimensional operational data and combines them with spatial connectivity analysis to automatically locate abnormal chart regions. It then uses interpolation calculations to smooth and drive the virtual camera to align with the observation target, thereby providing accurate visual guidance for anomalies in complex chart environments. Simultaneously, by combining real-time acquired hardware load parameters such as memory usage and frame rate decay slope, a visual importance field based on spatial geometric distance and feature decay rate is constructed. The invention uses dynamically calculated step size and weights to cluster and reduce the order of secondary data primitives within the view frustum. Combined with primitive interception and instantiation merging rendering techniques outside the view frustum, this invention objectively controls the computational overhead of the rendering pipeline while ensuring the prominent presentation of core abnormal data, achieving a dynamic balance between the presentation of massive charts and the system hardware load.

[0064] Example 2 This embodiment applies a multi-dimensional operational data visualization method based on chart rendering to the full-link multi-dimensional operational data visualization process of a large e-commerce platform during a major promotion.

[0065] Acquire multi-dimensional operational monitoring data (including server CPU load, real-time network bandwidth, and concurrent transactions per second) from various data center nodes of the e-commerce platform. Map physical location labels such as "Data Center X, Data Center Y" to the horizontal axis of space, map logical asset groups such as "Network Load Balancing Gateway, Application Server Cluster, Distributed Database Cluster" to the depth axis of space, and map business layers such as "Infrastructure Base, Middleware, Front-end Business Applications" to the height axis of space. Combine the element arrangement index and discrete offset within each group to instantiate and generate a massive array of three-dimensional bar charts representing the real-time health status of global assets in a virtual three-dimensional space.

[0066] The time feature sequence is continuously extracted according to the preset rendering cycle, and the maximum eigenvalue of the covariance matrix of the feature sequences of adjacent windows is calculated as the fluctuation feature scalar. When the traffic surge of the promotion arrives, it is detected that the fluctuation feature scalar of multiple data primitives in the "Y Data Center - Distributed Database Cluster" area (such as a sharp increase in transaction read and write latency) exceeds the preset benchmark. Using these excess primitives as anomaly seeds, the 26-neighborhood breadth-first search algorithm is called to merge spatially adjacent database nodes with the same high latency into the same candidate connected domain. After removing isolated noise points, they are marked as "Y Data Center Database Congestion" abnormal chart areas. The rendering engine extracts the 3D covariance matrix of all primitive vertices within the abnormal chart area, constructs the minimum directed bounding box, and extracts its center coordinates and diagonal scale parameters. Combining the anti-clipping coefficient and the current vertical field of view of the virtual camera, the optimal observation distance is calculated, and the target observation position is obtained by translating along the direction from the center coordinates to the current virtual camera viewpoint. A cubic Hermit spline curve with zero tangent vectors at both ends is generated as the translation trajectory, and combined with the nonlinear easing function and quaternion spherical linear interpolation algorithm, the virtual camera viewpoint on the monitoring screen is smoothly and naturally zoomed in and rotated from the global macroscopic perspective, accurately focusing the observation target point on the center of the "Y data center database congestion" area. After locking the abnormal region from the camera's perspective, the current view projection matrix is ​​extracted and the six clipping planes of the view frustum are resolved. Data primitives whose bounding sphere is outside the view frustum (such as the X data center primitive completely behind the field of view) are marked as invisible and their rendering pipeline data transmission is completely blocked. For the remaining data primitives visible within the view frustum (such as the load balancing gateway node of the Y server room in a normal state), the shortest Euclidean distance from its centroid to the database abnormal bounding box and the scalar decay rate of the fluctuation feature are calculated. Distance weights and feature weights are generated through the exponential decay function and normalized inverse mapping, and the two are weighted to calculate the visual importance value of each node, thus constructing a spatially distributed visual importance field. Due to the surge in pressure on the underlying hardware caused by high-frequency concurrent data, the system detected a real-time GPU memory utilization rate as high as 85% and a significant rate drop in frame rate. The hardware load parameters were incorporated into the dynamic penalty multiplier calculation logic, significantly increasing the global aggregation step size of the 3D spatial partitioning. Using this increased aggregation step size, voxel clustering was performed on surrounding normal application nodes within the view frustum that were far from the abnormal core and had low visual importance. Within each group, using local fusion weights as weighting coefficients, the coordinates and features (such as average CPU utilization) of hundreds or thousands of tiny primitives were weighted, summed, and averaged to generate several reduced-order primitives representing the overall state of the local mesh. The system extracts detailed database node primitives within the abnormal chart area, as well as reduced-order primitive entities generated by merging surrounding normal nodes. It extracts their respective spatial coordinate centroids and feature mean values ​​to construct instance transformation matrices and color attributes. These data are then concatenated in memory and dynamically pushed into the instance cache object of the graphics processor. Hardware instantiation instructions are then called to perform a single merge drawing. This approach, under extremely demanding hardware load conditions, not only highlights the core database fault nodes that urgently need to be investigated for maintenance personnel, but also preserves the macroscopic business context topology of the entire chain with minimal rendering overhead.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for visualizing multi-dimensional operational data based on chart rendering, characterized in that, include: Acquire multidimensional operational monitoring data, interpolate and align it with a preset rendering cycle, and stitch together spatial coordinates to generate a multidimensional feature tensor, and construct a three-dimensional statistical chart containing data primitives; Calculate the fluctuation feature scalar of the multidimensional feature tensor in adjacent windows, call the spatial connectivity detection algorithm to mark data primitives with adjacent spatial coordinates and fluctuation feature scalars that deviate from the preset benchmark as abnormal chart regions; construct the bounding box of the abnormal chart region and extract the center coordinates and scale, calculate the target observation position in combination with the current field of view, drive the virtual camera viewpoint to smoothly transition to the target observation position through the interpolation algorithm, and align the observation target point with the center coordinates of the bounding box; Real-time acquisition of video memory usage and frame rate decay slope is used as hardware load parameters. The spatial geometric distance from the remaining data primitives of the three-dimensional statistical chart (excluding the abnormal chart area) to the abnormal chart area is calculated, and the decay rate of the fluctuation feature scalar is used to construct a visual importance field. The global aggregation step size is calculated using the hardware load parameters. The visual importance field is then mapped inversely to local fusion weights. Reduced-order primitives are generated using the global aggregation step size and the local fusion weights. Instantiation instructions are called to merge and draw the reduced-order primitives, and the remaining data primitives are intercepted and rendered.

2. The method for visualizing multi-dimensional operational data based on chart rendering according to claim 1, characterized in that, Based on the business level, asset logical grouping and physical region label of the multidimensional operation monitoring data, the corresponding basic grid coordinates are allocated in the virtual three-dimensional space through the three-dimensional grid layout algorithm, and discrete offsets are introduced in each group in combination with the primitive arrangement index to calculate the spatial coordinates of each data primitive in the virtual three-dimensional space. The construction process of the three-dimensional statistical chart is as follows: the acquired multi-dimensional operation monitoring data is interpolated and aligned according to the preset rendering period to generate a discrete timestamp continuous feature sequence; the spatial coordinates and the corresponding feature sequence are matrix-concatenated and mapped to the corresponding vertex cache object to generate the multi-dimensional feature tensor; the multi-dimensional feature tensor is batch-instantiated into the corresponding data primitives to construct the three-dimensional statistical chart; the three-dimensional statistical chart can be any one of a three-dimensional bar chart array, a three-dimensional scatter plot, and a three-dimensional voxel heatmap.

3. The method for visualizing multi-dimensional operational data based on chart rendering according to claim 2, characterized in that, The business layers include an infrastructure base layer, a middleware service layer, and a front-end business application layer; the asset logical groups include a computing resource pool cluster, a distributed storage cluster, and a network load balancing gateway group; the physical region labels include cross-domain data center node identifiers, data center availability zone codes, and rack physical array sequences.

4. The method for visualizing multi-dimensional operational data based on chart rendering according to claim 1, characterized in that, The execution process of the spatial connected component detection algorithm is to filter the data primitives whose fluctuation feature scalars deviate from the preset benchmark as abnormal seed primitives; Starting from the abnormal seed primitive, an iterative search is performed outward in the virtual three-dimensional space to merge the spatially adjacent data primitives whose fluctuation feature scalars also deviate from the preset benchmark into the same candidate connected domain. After the iteration terminates, candidate connected regions with a total number of primitives less than a preset size are removed, and the remaining candidate connected regions are designated as the abnormal graph regions.

5. The method for visualizing multi-dimensional operational data based on chart rendering according to claim 4, characterized in that, The calculation process of the fluctuation feature scalar is to extract the first dynamic feature subsequence of the multidimensional feature tensor in the current time window and the second dynamic feature subsequence in the adjacent historical time window; and to calculate the mean vector of the first dynamic feature subsequence and the second dynamic feature subsequence in the time dimension respectively. Based on the deviations of the first dynamic feature subsequence from the mean vector and the deviations of the second dynamic feature subsequence from the mean vector, the expected value of their outer product is calculated to obtain the covariance matrix; the largest eigenvalue of the covariance matrix is ​​extracted as the fluctuation feature scalar.

6. The method for visualizing multi-dimensional operational data based on chart rendering according to claim 1, characterized in that, The process of aligning the observation target point with the center coordinates of the bounding box is as follows: construct the minimum directed bounding box of the abnormal chart region, extract its center coordinates as the center coordinates of the bounding box, and use the diagonal length of the minimum directed bounding box as the scale; combine the current field of view and the anti-clipping coefficient to calculate the target observation distance; The target observation position is calculated by translating the bounding box center coordinates along the direction pointing from the bounding box center coordinates to the current virtual camera viewpoint by the target observation distance. A cubic Hermite spline curve with the current viewpoint as the starting point and the target observation position as the ending point, and with both the first and last tangent vectors being zero, is constructed as the translation trajectory. An interpolation step size is generated through a nonlinear easing function to drive the virtual camera to update its position along the translation trajectory. At the same time, a quaternion spherical linear interpolation algorithm is called to smoothly rotate the camera's current pose to the target pose aligned with the bounding box center coordinates according to the interpolation step size.

7. The method for visualizing multi-dimensional operational data based on chart rendering according to claim 1, characterized in that, The construction process of the visual importance field is as follows: The frustum range is extracted based on the current virtual camera's pose; the shortest Euclidean distance from the centroid of the remaining data primitives falling within the frustum to the bounding box of the anomalous chart region is calculated as the spatial geometric distance; the absolute value of the scalar difference in fluctuation characteristics between the remaining data primitives and the data primitives at the boundary of the anomalous chart region is calculated, and the ratio of this absolute value to the spatial geometric distance is taken as the attenuation rate; the distance weights obtained by mapping the spatial geometric distance through the attenuation function are weighted and summed with the feature weights obtained by inverse mapping the normalized attenuation rate to calculate the visual importance value of each of the remaining data primitives; the spatial distribution set of all the visual importance values ​​constitutes the visual importance field.

8. The method for visualizing multi-dimensional operational data based on chart rendering according to claim 1, characterized in that, The process of generating reduced-order primitives by the global aggregation step size and the local fusion weights involves fusing the video memory usage and the frame rate decay slope to calculate and map them into a spatial grid size, which is used as the global aggregation step size; and mapping the visual importance values ​​in the visual importance field inversely to the local fusion weights corresponding to the other data primitives. Divide the three-dimensional aggregation grid according to the global aggregation step size, and classify the remaining data primitives falling into the same three-dimensional aggregation grid into a primitive cluster to be fused; use the local fusion weight as the weighting coefficient to perform weighted fusion on the spatial coordinates, geometric scale and color attributes of each of the remaining data primitives in the primitive cluster to be fused, and generate the reduced-order primitive.

Images